Lecture Notes in Macroeconomics

Christian Groth

September 27, 2015

ii

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Contents

Preface xvii

I THE FIELD AND BASIC CATEGORIES 1

1 Introduction 3
1.1 Macroeconomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 The …eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 The di¤erent “runs” . . . . . . . . . . . . . . . . . . . . . 5
1.2 Components of macroeconomic models . . . . . . . . . . . . . . . 8
1.2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 The time dimension of input and output . . . . . . . . . . 11
1.3 Macroeconomic models and national income accounting . . . . . . 13
1.4 Some terminological points . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Brief history of macroeconomics . . . . . . . . . . . . . . . . . . . 15
1.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Review of technology and …rms 17

2.1 The production technology . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 A neoclassical production function . . . . . . . . . . . . . 18
2.1.2 Returns to scale . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.3 Properties of the production function under CRS . . . . . 27
2.2 Technological change . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 The concepts of a representative …rm and an aggregate production
function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 The neoclassical competitive one-sector setup . . . . . . . . . . . 38
2.4.1 Pro…t maximization . . . . . . . . . . . . . . . . . . . . . . 38
2.4.2 Clearing in factor markets . . . . . . . . . . . . . . . . . . 42
2.5 More complex model structures* . . . . . . . . . . . . . . . . . . . 47
2.5.1 Convex capital installation costs . . . . . . . . . . . . . . . 47
2.5.2 Long-run vs. short-run production functions . . . . . . . . 48

iii
iv CONTENTS

2.5.3 A simple portrayal of price-making …rms . . . . . . . . . . 50

2.5.4 The …nancing of …rms’operations . . . . . . . . . . . . . . 53
2.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

II LOOKING AT THE LONG RUN 63

3 The basic OLG model: Diamond 65
3.1 Motives for saving . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2 The model framework . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 The saving by the young . . . . . . . . . . . . . . . . . . . . . . . 70
3.4 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.5 The dynamic path of the economy . . . . . . . . . . . . . . . . . . 83
3.5.1 Technically feasible paths . . . . . . . . . . . . . . . . . . 84
3.5.2 A temporary equilibrium . . . . . . . . . . . . . . . . . . . 84
3.5.3 An equilibrium path . . . . . . . . . . . . . . . . . . . . . 87
3.6 The golden rule and dynamic ine¢ ciency . . . . . . . . . . . . . . 101
3.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.8 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

4 A growing economy 123

4.1 Harrod-neutrality and Kaldor’s stylized facts . . . . . . . . . . . . 124
4.2 The Diamond OLG model with Harrod-neutral technological progress134
4.3 The golden rule under Harrod-neutral technological progress . . . 139
4.4 The functional distribution of income . . . . . . . . . . . . . . . . 142
4.5 The CES production function* . . . . . . . . . . . . . . . . . . . . 146
4.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.7 Literature notes and discussion . . . . . . . . . . . . . . . . . . . 152
4.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

5 Applying and extending the Diamond model 163

5.1 Pension schemes and aggregate saving . . . . . . . . . . . . . . . 163
5.2 Endogenous labor supply . . . . . . . . . . . . . . . . . . . . . . . 175
5.2.1 The intensive margin: A simple one-period model . . . . . 175
5.2.2 Endogenous labor supply in an extended Diamond model . 182
5.3 Early retirement with transfer income . . . . . . . . . . . . . . . . 187

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS v

5.4 Intertemporal substitution of labor supply . . . . . . . . . . . . . 196

5.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.6 Appendix: The extended Slutsky equation . . . . . . . . . . . . . 200

6 Long-run aspects of …scal policy and public debt 203

6.1 An overview of government spending and …nancing issues . . . . . 204
6.2 The government budget . . . . . . . . . . . . . . . . . . . . . . . 205
6.3 Government solvency and …scal sustainability . . . . . . . . . . . 210
6.3.1 The critical role of the growth-corrected interest factor . . 210
6.3.2 Sustainable …scal policy . . . . . . . . . . . . . . . . . . . 213
6.4 Debt arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
6.4.1 The required primary budget surplus . . . . . . . . . . . . 215
6.4.2 Case study: The Stability and Growth Pact of the EMU . 223
6.5 Solvency, the NPG condition, and the intertemporal government
budget constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.5.1 When is the NPG condition necessary for solvency? . . . . 230
6.5.2 Equivalence of NPG and GIBC . . . . . . . . . . . . . . . 233
6.6 A proper accounting of public investment* . . . . . . . . . . . . . 237
6.7 Ricardian equivalence? . . . . . . . . . . . . . . . . . . . . . . . . 240
6.7.1 Two di¤ering views . . . . . . . . . . . . . . . . . . . . . . 241
6.7.2 A small open OLG economy with a temporary budget de…cit242
6.8 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.9 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254

7 Bequests and the modi…ed golden rule 257

7.1 Bequests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
7.2 Barro’s dynasty model . . . . . . . . . . . . . . . . . . . . . . . . 258
7.2.1 A forward-looking altruistic parent . . . . . . . . . . . . . 259
7.2.2 Case 1: the bequest motive operative (bt+1 > 0 optimal) . . 262
7.2.3 Case 2: the bequest motive not operative (bt+1 = 0 optimal) 268
7.2.4 Necessary and su¢ cient conditions for the bequest motive
to be operative . . . . . . . . . . . . . . . . . . . . . . . . 269
7.3 Bequests and Ricardian equivalence . . . . . . . . . . . . . . . . . 272
7.4 The modi…ed golden rule when there is technological progress* . . 278
7.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
7.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

vi CONTENTS

8 Optimal capital accumulation 291

8.1 Command optimum . . . . . . . . . . . . . . . . . . . . . . . . . . 291
8.1.1 A social planner . . . . . . . . . . . . . . . . . . . . . . . . 292
8.1.2 The modi…ed golden rule of the command optimum . . . . 301
8.1.3 The turnpike property . . . . . . . . . . . . . . . . . . . . 302
8.2 Optimal control theory and the social planner’s problem* . . . . . 304
8.2.1 Decomposing the social planner’s problem . . . . . . . . . 304
8.2.2 Applying the Maximum Principle . . . . . . . . . . . . . . 308
8.3 The overtaking and catching-up criteria* . . . . . . . . . . . . . . 319
8.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 323
8.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
8.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

9 The intertemporal consumption-saving problem in discrete and

continuous time 343
9.1 Market conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
9.2 Maximizing discounted utility in discrete time . . . . . . . . . . . 346
9.3 Transition to continuous time analysis . . . . . . . . . . . . . . . 357
9.4 Maximizing discounted utility in continuous time . . . . . . . . . 363
9.4.1 The saving problem in continuous time . . . . . . . . . . . 363
9.4.2 Solving the saving problem . . . . . . . . . . . . . . . . . . 366
9.4.3 The Keynes-Ramsey rule . . . . . . . . . . . . . . . . . . . 371
9.4.4 Mangasarian’s su¢ cient conditions . . . . . . . . . . . . . 374
9.5 The consumption function . . . . . . . . . . . . . . . . . . . . . . 374
9.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 379
9.7 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
9.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
9.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388

10 The basic representative agent model: Ramsey 391

10.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
10.2 The agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
10.2.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . 393
10.2.2 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
10.3 General equilibrium and dynamics . . . . . . . . . . . . . . . . . . 399
10.4 Comparative analysis . . . . . . . . . . . . . . . . . . . . . . . . . 411
10.4.1 The role of key parameters . . . . . . . . . . . . . . . . . . 411
10.4.2 Solow’s growth model as a special case . . . . . . . . . . . 413
10.5 A social planner’s problem . . . . . . . . . . . . . . . . . . . . . . 414
10.5.1 The equivalence theorem . . . . . . . . . . . . . . . . . . . 415

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS vii

10.5.2 Ramsey’s original zero discount rate and the overtaking

criterion* . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
10.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 421
10.7 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
10.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
10.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430

11 Applications of the Ramsey model 431

11.1 Market economy with a public sector . . . . . . . . . . . . . . . . 431
11.1.1 Public consumption …nanced by lump-sum taxes . . . . . . 432
11.1.2 Income taxation . . . . . . . . . . . . . . . . . . . . . . . . 437
11.1.3 E¤ects of shifts in the capital income tax rate . . . . . . . 439
11.1.4 Ricardian equivalence . . . . . . . . . . . . . . . . . . . . . 445
11.2 Learning by investing and investment-enhancing policy . . . . . . 448
11.2.1 The common framework . . . . . . . . . . . . . . . . . . . 448
11.2.2 The arrow case: < 1 . . . . . . . . . . . . . . . . . . . . 452
11.2.3 Romer’s limiting case: = 1; n = 0 . . . . . . . . . . . . . 457
11.3 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 465
11.4 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
11.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
11.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470

12 Overlapping generations in continuous time 471

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
12.2 The model of perpetual youth . . . . . . . . . . . . . . . . . . . . 473
12.2.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . 473
12.2.2 Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . 482
12.2.3 The representative …rm . . . . . . . . . . . . . . . . . . . . 484
12.2.4 General equilibrium (closed economy) . . . . . . . . . . . . 485
12.3 Adding retirement . . . . . . . . . . . . . . . . . . . . . . . . . . 494
12.4 The rate of return in the long run . . . . . . . . . . . . . . . . . . 500
12.5 National wealth and foreign debt . . . . . . . . . . . . . . . . . . 504
12.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 513
12.7 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
12.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
12.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524

13 General equilibrium analysis of public and foreign debt 525

13.1 Reconsidering the issue of Ricardian equivalence . . . . . . . . . . 525
13.2 Dynamic general equilibrium e¤ects of lasting budget de…cits . . . 532
13.3 Public and foreign debt: a small open economy . . . . . . . . . . 545

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

viii CONTENTS

13.4 Government debt when taxes are distortionary* . . . . . . . . . . 559

13.5 Public debt policy . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
13.6 Credibility problems due to time inconsistency . . . . . . . . . . . 567
13.7 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
13.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 568
13.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570

III MODELING FIXED CAPITAL INVESTMENT 571

14 Fixed capital investment and Tobin’s q 573
14.1 Convex capital installation costs . . . . . . . . . . . . . . . . . . . 574
14.1.1 The decision problem of the …rm . . . . . . . . . . . . . . 577
14.1.2 The implied investment function . . . . . . . . . . . . . . . 582
14.1.3 A not implausible special case . . . . . . . . . . . . . . . . 583
14.2 Marginal q and average q . . . . . . . . . . . . . . . . . . . . . . . 585
14.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
14.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 594
14.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
14.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
14.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 606

15 Further applications of adjustment cost theory 607

15.1 Oil price shock in a small oil importing economy . . . . . . . . . . 607
15.1.1 Three inputs: capital, labor, and raw material . . . . . . . 608
15.1.2 General equilibrium and dynamics . . . . . . . . . . . . . . 613
15.1.3 National income accounting for an open economy with cap-
ital installation costs . . . . . . . . . . . . . . . . . . . . . 617
15.1.4 Household behavior and …nancial wealth . . . . . . . . . . 619
15.1.5 General aspects of modeling a small open economy . . . . 624
15.2 Housing and the macroeconomy . . . . . . . . . . . . . . . . . . . 625
15.2.1 The housing service market and the house market . . . . . 626
15.2.2 Construction activity . . . . . . . . . . . . . . . . . . . . . 629
15.2.3 Equilibrium dynamics . . . . . . . . . . . . . . . . . . . . 632
15.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 635
15.3 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 636
15.4 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637
15.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 640

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS ix

IV MODELING MONEY 641

16 Money in macroeconomics 643
16.1 What is money? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
16.1.1 The concept of money . . . . . . . . . . . . . . . . . . . . 643
16.1.2 Historical remarks . . . . . . . . . . . . . . . . . . . . . . 645
16.1.3 The functions of money . . . . . . . . . . . . . . . . . . . 646
16.2 The money supply . . . . . . . . . . . . . . . . . . . . . . . . . . 647
16.2.1 Di¤erent measures of the money supply . . . . . . . . . . . 647
16.2.2 The money multiplier . . . . . . . . . . . . . . . . . . . . . 648
16.3 Money demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
16.4 What is then the “money market”? . . . . . . . . . . . . . . . . . 652
16.5 Key questions in monetary theory and policy . . . . . . . . . . . . 656
16.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 656
16.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 657

17 In‡ation and capital accumulation: The Sidrauski model 659

17.1 The agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660
17.2 Equilibrium and evolution over time . . . . . . . . . . . . . . . . . 666
17.3 Theoretical implications . . . . . . . . . . . . . . . . . . . . . . . 670
17.3.1 Money neutrality and superneutrality . . . . . . . . . . . . 670
17.3.2 Milton Friedman’s zero interest rate rule . . . . . . . . . . 673
17.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 675
17.4 Are in‡ation and de‡ation bubbles possible? . . . . . . . . . . . . 677
17.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
17.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
17.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685

18 Wider perspectives on monetary economies 687

18.1 Money growth and in‡ation in the long run . . . . . . . . . . . . 687
18.2 Are neutrality and superneutrality of money theoretically robust
properties? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
18.3 In‡ationary public …nance . . . . . . . . . . . . . . . . . . . . . . 694
18.3.1 The seigniorage La¤er curve . . . . . . . . . . . . . . . . . 695
18.3.2 Hyperin‡ation . . . . . . . . . . . . . . . . . . . . . . . . . 698
18.4 Bridging the gap between the short and the long run . . . . . . . 704
18.4.1 The monetary transmission mechanism in the short and the
long run . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
18.4.2 In‡ation - social costs and bene…ts . . . . . . . . . . . . . 707
18.5 Theory of “the level of interest rates” . . . . . . . . . . . . . . . . 711
18.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

x CONTENTS

18.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 720

18.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 720

V LOOKING AT THE SHORT RUN 721

19 The theory of e¤ective demand 723
19.1 Stylized facts about the short run . . . . . . . . . . . . . . . . . . 724
19.2 A simple short-run model . . . . . . . . . . . . . . . . . . . . . . 725
19.2.1 Elements of the model . . . . . . . . . . . . . . . . . . . . 726
19.2.2 The case of fully ‡exible W and P . . . . . . . . . . . . . 728
19.2.3 The case of W and P …xed in the short run . . . . . . . . 731
19.2.4 Short-run adjustment dynamics . . . . . . . . . . . . . . . 743
19.3 Price setting and menu costs . . . . . . . . . . . . . . . . . . . . . 746
19.3.1 Imperfect competition with price setters . . . . . . . . . . 747
19.3.2 Price adjustment costs . . . . . . . . . . . . . . . . . . . . 751
19.3.3 Menu costs in action . . . . . . . . . . . . . . . . . . . . . 756
19.4 Abundant capacity . . . . . . . . . . . . . . . . . . . . . . . . . . 760
19.4.1 Putty-clay technology . . . . . . . . . . . . . . . . . . . . 760
19.4.2 Capacity utilization and monopolistic-competitive equilib-
rium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 762
19.4.3 Aggregation over di¤erent regimes . . . . . . . . . . . . . . 766
19.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 769
19.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 771
19.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773
19.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773

20 General equilibrium under monopolistic competition 775

20.1 The emergence of new-Keynesian economics . . . . . . . . . . . . 775
20.2 The Blanchard-Kiyotaki model of monopolistic competition . . . . 777
20.2.1 Overview of agents’decision problems . . . . . . . . . . . . 778
20.2.2 The resulting behavior . . . . . . . . . . . . . . . . . . . . 780
20.3 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . 792
20.3.1 The case with ‡exible wages and prices . . . . . . . . . . . 792
20.3.2 The case with sticky wages and prices . . . . . . . . . . . . 797
20.4 Spillover complementarity and multiple equilibria . . . . . . . . . 804
20.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 806
20.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 807
20.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 807
20.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 807

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS xi

21 The IS-LM model 809

21.1 The building blocks . . . . . . . . . . . . . . . . . . . . . . . . . . 810
21.2 Keynesian equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 815
21.3 Alternative monetary policy regimes . . . . . . . . . . . . . . . . 817
21.3.1 Money stock rule . . . . . . . . . . . . . . . . . . . . . . . 818
21.3.2 Fixed interest rate rule . . . . . . . . . . . . . . . . . . . . 824
21.3.3 Counter-cyclical interest rate rule . . . . . . . . . . . . . . 832
21.3.4 Further aspects . . . . . . . . . . . . . . . . . . . . . . . . 833
21.4 Some robustness checks . . . . . . . . . . . . . . . . . . . . . . . . 835
21.4.1 Presence of an interest rate spread (banks’lending rate =
i + > i): . . . . . . . . . . . . . . . . . . . . . . . . . . . 835
21.4.2 What if households are in…nitely-lived? . . . . . . . . . . . 835
21.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 837
21.6 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 838
21.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839
21.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839

22 IS-LM dynamics with forward-looking expectations 841

22.1 A dynamic IS-LM model . . . . . . . . . . . . . . . . . . . . . . . 842
22.2 Monetary policy regimes . . . . . . . . . . . . . . . . . . . . . . . 846
22.2.1 Policy regime m: Money stock rule . . . . . . . . . . . . . 847
22.2.2 Policy regime i: Fixed short-term interest rate . . . . . . . 860
22.2.3 Policy regime i0 : A counter-cyclical interest rate rule . . . 865
22.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865
22.4 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866
22.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 867

23 The open economy and di¤erent exchange rate regimes 873

23.1 The Mundell-Fleming model . . . . . . . . . . . . . . . . . . . . . 876
23.1.1 The basic elements . . . . . . . . . . . . . . . . . . . . . . 876
23.1.2 Fixed exchange rate . . . . . . . . . . . . . . . . . . . . . 878
23.1.3 Floating exchange rate . . . . . . . . . . . . . . . . . . . . 880
23.1.4 Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . 882
23.2 Dynamics under a …xed exchange rate . . . . . . . . . . . . . . . 883
23.3 Dynamics under a ‡oating exchange rate: overshooting . . . . . . 885
23.3.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . 886
23.3.2 Unanticipated rise in the real money supply . . . . . . . . 890
23.3.3 Anticipated rise in the money supply . . . . . . . . . . . . 895
23.3.4 Monetary policy tightening . . . . . . . . . . . . . . . . . . 898
23.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 902
23.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

xii CONTENTS

23.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904

23.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 905

24 A closer look at the labor market 907

24.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 907
24.2 Themes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 907
24.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908

VI UNCERTAINTY AND EXPECTATIONS 909

25 Uncertainty, rational expectations, and staggered wage setting 911
25.1 Simple expectation formation hypotheses . . . . . . . . . . . . . . 912
25.2 The rational expectations hypothesis . . . . . . . . . . . . . . . . 913
25.2.1 Two model classes . . . . . . . . . . . . . . . . . . . . . . 914
25.2.2 Rational expectations . . . . . . . . . . . . . . . . . . . . . 916
25.2.3 Solving a simple RE model . . . . . . . . . . . . . . . . . . 918
25.3 Wage setting in advance . . . . . . . . . . . . . . . . . . . . . . . 922
25.4 A benchmark model with synchronous wage setting . . . . . . . . 923
25.4.1 Wage setting one period in advance . . . . . . . . . . . . . 924
25.4.2 Solving the benchmark model . . . . . . . . . . . . . . . . 927
25.5 Asynchronous wage setting for several periods: Fischer’s approach 928
25.5.1 The original Fischer model . . . . . . . . . . . . . . . . . . 929
25.5.2 A modi…ed Fischer model . . . . . . . . . . . . . . . . . . 933
25.6 Asynchronous wage setting with constant wage level for several
periods: Taylor’s model . . . . . . . . . . . . . . . . . . . . . . . . 935
25.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 942
25.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944
25.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 951

26 Forward-looking rational expectations 953

26.1 Expectational di¤erence equations . . . . . . . . . . . . . . . . . . 953
26.2 Solutions when jaj < 1 . . . . . . . . . . . . . . . . . . . . . . . . 956
26.2.1 Repeated forward substitution . . . . . . . . . . . . . . . . 956
26.2.2 The fundamental solution . . . . . . . . . . . . . . . . . . 957
26.2.3 Bubble solutions . . . . . . . . . . . . . . . . . . . . . . . 960
26.2.4 When rational bubbles in asset prices can or can not be
ruled out . . . . . . . . . . . . . . . . . . . . . . . . . . . . 967
26.2.5 Time-dependent coe¢ cients . . . . . . . . . . . . . . . . . 972
26.2.6 Three classes of bubble processes . . . . . . . . . . . . . . 973
26.3 Solutions when jaj > 1 . . . . . . . . . . . . . . . . . . . . . . . . 975

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS xiii

26.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 976

26.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 977
26.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 978
26.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 982

27 Applications to New Classical and Keynesian models 987

27.1 New Classical Macroeconomics . . . . . . . . . . . . . . . . . . . . 988
27.1.1 The New Classical school . . . . . . . . . . . . . . . . . . . 988
27.1.2 An NCM model . . . . . . . . . . . . . . . . . . . . . . . . 988
27.1.3 Weak and strong policy ine¤ectiveness . . . . . . . . . . . 998
27.1.4 Alternative speci…cations and extensions . . . . . . . . . . 1004
27.1.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 1009
27.2 The Lucas critique of econometric policy evaluation . . . . . . . . 1010
27.3 Paper money and hyperin‡ation . . . . . . . . . . . . . . . . . . . 1013
27.4 Announcement e¤ects . . . . . . . . . . . . . . . . . . . . . . . . . 1016
27.5 Is increased wage ‡exibility stabilizing? . . . . . . . . . . . . . . . 1019
27.5.1 Dynamic AD-AS model with uncertainty and nominal rigidi-
ties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1020
27.5.2 Dynamic responses . . . . . . . . . . . . . . . . . . . . . . 1023
27.6 Depression economics . . . . . . . . . . . . . . . . . . . . . . . . . 1025
27.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025
27.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1026

28 Can rational bubbles be ruled out in general equilibrium? 1029

28.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1029
28.2 Finite number of agents . . . . . . . . . . . . . . . . . . . . . . . 1031
28.3 In…nite number of agents: OLG models . . . . . . . . . . . . . . . 1032
28.3.1 No bubbles . . . . . . . . . . . . . . . . . . . . . . . . . . 1032
28.3.2 The Diamond-Tirole model . . . . . . . . . . . . . . . . . 1034
28.3.3 Stochastic bubbles . . . . . . . . . . . . . . . . . . . . . . 1040
28.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 1041
28.5 Notes on the literature . . . . . . . . . . . . . . . . . . . . . . . . 1041
28.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1041
28.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1044

VII FITTING THE PARTS TOGETHER: THE MEDIUM

RUN 1045
29 Business cycle ‡uctuations 1047
29.1 Some business cycle facts . . . . . . . . . . . . . . . . . . . . . . . 1047

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

xiv CONTENTS

29.2 Key terms from the business cycle vocabulary . . . . . . . . . . . 1049

29.3 A quick glance at the Great Recession and its aftermath . . . . . 1050
29.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
29.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
29.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053

30 The real business cycle theory 1055

30.1 A simple RBC model . . . . . . . . . . . . . . . . . . . . . . . . . 1056
30.2 A deterministic steady state . . . . . . . . . . . . . . . . . . . . . 1065
30.3 On the approximate solution and numerical simulation . . . . . . 1067
30.3.1 Log-linearization . . . . . . . . . . . . . . . . . . . . . . . 1067
30.3.2 Numerical simulation . . . . . . . . . . . . . . . . . . . . . 1070
30.4 The two basic propagation mechanisms . . . . . . . . . . . . . . . 1072
30.5 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073
30.6 Technological change as a random walk with drift . . . . . . . . . 1075
30.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 1076
30.8 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077
30.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077

31 Keynesian perspectives on business cycles 1079

31.1 A short period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1080
31.2 The dynamic links . . . . . . . . . . . . . . . . . . . . . . . . . . 1082
31.2.1 Changes in expectations . . . . . . . . . . . . . . . . . . . 1083
31.2.2 Phillips curve/wage curve . . . . . . . . . . . . . . . . . . 1083
31.2.3 Other dynamic links . . . . . . . . . . . . . . . . . . . . . 1085
31.2.4 Aren’t desirable adjustments automatic and fast? . . . . . 1086
31.3 Vicious and virtuous circles . . . . . . . . . . . . . . . . . . . . . 1088
31.4 Precautionary saving . . . . . . . . . . . . . . . . . . . . . . . . . 1091
31.4.1 Consumption/saving under di¤erent forms of uncertainty . 1092
31.4.2 Precautionary saving in a macroeconomic perspective . . . 1098
31.5 Literature notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1098
31.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1099
31.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1099

32 The New Keynesian workhorse model 1101

32.1 Main text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1101
32.2 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1101

33 Credit and business cycles 1103

34 Issues in monetary and …scal policy 1105

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

CONTENTS xv

35 Outlook 1107

VIII SUPPLEMENTS 1109

36 Hints and solutions for selected exercise problems 1111

37 Math tools 1113

38 Equation systems and causal analysis 1115

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

xvi CONTENTS

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 1

Introduction

The art of successful theorizing is to make the inevitable simplifying

assumptions in such a way that the …nal results are not very sensitive.
Robert M. Solow (1956, p. 65)

1.1 Macroeconomics
1.1.1 The …eld
Economics is the social science that studies the production and distribution of
goods and services in society. Then, what de…nes the branch of economics named
macroeconomics? There are two de…ning characteristics. First, macroeconomics
is the systematic study of the economic interactions in society as a whole. This
could also be said of microeconomic general equilibrium theory, however. The
second de…ning characteristic of macroeconomics is that it aims at understanding
the empirical regularities in the behavior of aggregate economic variables such
as aggregate production, investment, unemployment, the general price level for
goods and services, the in‡ation rate, the level of interest rates, the level of real
wages, the foreign exchange rate, productivity growth etc. Thus macroeconomics
focuses on the major lines of the economics of a society.
The aspiration of macroeconomics is three-fold:
1. to explain the levels of the aggregate variables as well as their movement
over time in the short run and the long run;
2. to make well-founded forecasts possible;
3. to provide foundations for rational economic policy applicable to macroeco-
nomic problems, be they short-run distress in the form of economic recession
or problems of a more long-term, structural character.

3
4 CHAPTER 1. INTRODUCTION

We use economic models to make our complex economic environment accessi-

ble for theoretical analysis. What is an economic model? It is a way of organizing
one’s thoughts about the economic functioning of a society. A more speci…c an-
swer is to de…ne an economic model as a conceptual structure based on a set of
mathematically formulated assumptions which have an economic interpretation
and from which empirically testable predictions can be derived. In particular,
a macroeconomic model is an economic model concerned with macroeconomic
phenomena, i.e., the short-run ‡uctuations of aggregate variables as well as their
long-run trend.
Any economic analysis is based upon a conceptual framework. Formulating
this framework as a precisely stated economic model helps to break down the issue
into assumptions about the concerns and constraints of households and …rms and
the character of the market environment within which these agents interact. The
advantage of this approach is that it makes rigorous reasoning possible, lays bare
where the underlying disagreements behind di¤erent interpretations of economic
phenomena are, and makes sensitivity analysis of the conclusions amenable. By
being explicit about agents’concerns, the technological constraints, and the social
structures (market forms, social conventions, and legal institutions) conditioning
their interactions, this approach allows analysis of policy interventions, including
the use of well-established tools of welfare economics. Moreover, mathematical
modeling is a simple necessity to keep track of the many mutual dependencies
and to provide a consistency check of the many accounting relationships involved.
And mathematical modeling opens up for use of powerful mathematical theorems
from the mathematical toolbox. Without these math tools it would in many cases
be impossible to reach any conclusion whatsoever.
Undergraduate students of economics are often perplexed or even frustrated by
macroeconomics being so preoccupied with composite theoretical models. Why
not study the issues each at a time? The reason is that the issues, say housing
prices and changes in unemployment, are not separate, but parts of a complex
system of mutually dependent variables. This also suggests that macroeconomics
must take advantage of theoretical and empirical knowledge from other branches
of economics, including microeconomics, industrial organization, game theory,
political economy, behavioral economics, and even sociology and psychology.
At the same time models necessarily give a simpli…ed picture of the economic
reality. Ignoring secondary aspects and details is indispensable to be able to
focus on the essential features of a given problem. In particular macroeconomics
deliberately simpli…es the description of the individual actors so as to make the
analysis of the interaction between di¤erent types of actors manageable.
The assessment of and choice between competing simplifying frameworks
should be based on how well they perform in relation to the three-fold aim of

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.1. Macroeconomics 5

macroeconomics listed above, given the problem at hand. A necessary condition

for good performance is the empirical tenability of the model’s predictions. A
guiding principle in the development of useful models therefore lies in confronta-
tion of the predictions as well as the crucial assumptions with data. This can be
based on a variety of methods ranging from sophisticated econometric techniques
to qualitative case studies.
Three constituents make up an economic theory: 1) the union of connected
and non-contradictory economic models, 2) the theorems derived from these, and
3) the conceptual system de…ning the correspondence between the variables of
the models and the social reality to which they are to be applied. Being about
the interaction of human beings in societies, the subject matter of economic the-
ory is extremely complex and at the same time history dependent. The overall
political, social, and economic institutions (“rules of the game”in a broad sense)
evolve. These circumstances explain why economic theory is far from the natural
sciences with respect to precision and undisputable empirical foundation. Espe-
cially in macroeconomics, to avoid confusion one should be aware of the existence
of di¤ering conceptions and in several matters con‡icting theoretical schools.

1.1.2 The di¤erent “runs”

This textbook is about the macroeconomics of the industrialized market economies
of today. We study basic concepts, models, and analytical methods of rele-
vance for understanding macroeconomic processes where sometimes centripetal
and sometimes centrifugal forces are dominating. A simplifying device is the
distinction between “short-run”, “medium-run”, and “long-run” analysis. The
…rst concentrates on the behavior of the macroeconomic variables within a time
horizon of a few years, whereas “long-run” analysis deals with a considerably
longer time horizon indeed, long enough for changes in the capital stock, pop-
ulation, and technology to have a dominating in‡uence on changes in the level of
production. The “medium run”is then something in between.
To be more speci…c, long-run macromodels study the evolution of an econ-
omy’s productive capacity over time. Typically a time span of at least 15 years
is considered. The analytical framework is by and large supply-dominated. That
is, variations in the employment rate for labor and capital due to demand ‡uctu-
ations are abstracted away. This can to a …rst approximation be justi…ed by the
fact that these variations, at least in advanced economies, tend to remain within
a fairly narrow band. Therefore, under “normal” circumstances the economic
outcome after, say, a 30 years’ interval re‡ects primarily the change in supply
side factors such as the labor force, the capital stock, and the technology. The
‡uctuations in demand and monetary factors tend to be of limited quantitative

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6 CHAPTER 1. INTRODUCTION

importance within such a time horizon.

By contrast, when we speak of short-run macromodels, we think of models
concentrating on mechanisms that determine how fully an economy uses its pro-
ductive capacity at a given point in time. The focus is on the level of output and
employment within a time horizon less than, say, four years. These models are
typically demand-dominated. In this time perspective the demand side, mone-
tary factors, and price rigidities matter signi…cantly. Shifts in aggregate demand
(induced by, e.g., changes in …scal or monetary policy, exports, interest rates,
the general state of con…dence, etc.) tend to be accommodated by changes in
the produced quantities rather than in the prices of manufactured goods and ser-
vices. By contrast, variations in the supply of production factors and technology
are diminutive and of limited importance within this time span. With Keynes’
words the aim of short-run analysis is to explain “what determines the actual
employment of the available resources”(Keynes 1936, p. 4).
The short and the long run make up the traditional subdivision of macro-
economics. It is convenient and fruitful, however, to include also a medium run,
referring to a time interval of, say, four-to-…fteen years.1 We shall call models
attempting to bridge the gap between the short and the long run medium-run
macromodels. These models deal with the regularities exhibited by sequences of
short periods. However, in contrast to long-run models which focus on the trend
of the economy, medium-run models attempt to understand the pattern charac-
terizing the ‡uctuations around the trend. In this context, variations at both
the demand and supply side are important. Indeed, at the centre of attention
is the dynamic interaction between demand and supply factors, the correction
of expectations, and the time-consuming adjustment of wages and prices. Such
models are also sometimes called business cycle models.
Returning to the “long run”, what does it embrace in this book? Well, since
the surge of “new growth theory”or “endogenous growth theory”in the late 1980s
and early 1990s, growth theory has developed into a specialized discipline study-
ing the factors and mechanisms that determine the evolution of technology and
productivity (Paul Romer 1987, 1990; Phillipe Aghion and Peter Howitt, 1992).
An attempt to give a systematic account of this expanding line of work within
macroeconomics would take us too far. When we refer to “long-run macromod-
els”, we just think of macromodels with a time horizon long enough such that
changes in the capital stock, population, and technology matter. Apart from a
taste of “new growth theory” in Chapter 11, we leave the sources of changes in
technology out of consideration, which is tantamount to regarding these changes

1
These number-of-years …gures are only a rough indication. The di¤erent “runs”are relative
concepts and their appropriateness depends on the speci…c problem and circumstances at hand.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.1. Macroeconomics 7

as exogenous.2

Figure 1.1: Quarterly Industrial Production Index in six major countries (Q1-1958 to
Q2-2013; index Q1-1961=100). Source: OECD Industry and Service Statistics. Note:
Industrial production includes manufacturing, mining and quarrying, electricity, gas,
and water, and construction.

In addition to the time scale dimension, the national-international dimension

is important for macroeconomics. Most industrialized economies participate in
international trade of goods and …nancial assets. This results in considerable
mutual dependency and co-movement of these economies. Downturns as well as
upturns occur at about the same time, as indicated by Fig. 1.1. In particular the
economic recessions triggered by the oil price shocks in 1973 and 1980 and by the
disruption of credit markets in the outbreak 2007 of the Great Financial Crisis
are visible across the countries, as also shown by the evolution of GDP, cf. Fig.
1.2. Many of the models and mechanisms treated in this text will therefore be
considered not only in a closed economy setup, but also from the point of view
of open economies.

2
References to textbooks on economic growth are given in Literature notes at the end of this
chapter.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

8 CHAPTER 1. INTRODUCTION

Denmark
Eurozone
100 United States

90

80

70

1996 1998 2000 2002 2004 2006 2008 2010 2012

Figure 1.2: Indexed real GDP for Denmark, Eurozone and US, 1995-2012 (2007=100).
Source: EcoWin and Statistics Denmark.

1.2 Components of macroeconomic models

1.2.1 Basics
(Incomplete)

Basic categories

Agents: We use simple descriptions of the economic agents: A household is

an abstract entity making consumption, saving and labor supply decisions.
A …rm is an abstract entity making decisions about production and sales.
The administrative sta¤ and sales personnel are treated along with the
production workers as an undi¤erentiated labor input.

Technological constraints.

Goods, labor, and assets markets.

The institutions and social norms regulating the economic interactions (for-
mal and informal “rules of the game”).

Types of variables
Endogenous vs. exogenous variables.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.2. Components of macroeconomic models 9

Stocks vs. ‡ows.

State variables vs. control variables (decision variables). Closely related to
this distinction is that between a predetermined variable and a jump variable. The
former is a variable whose value is determined historically at any point in time.
For example, the stock (quantity) of water in a bathtub at time t is historically
determined as the accumulated quantity of water stemming from the previous
in‡ow and out‡ow. But if yt is a variable which is not tied down by its own past
but, on the contrary, can immediately adjust if new conditions or new information
emerge, then yt is a non-predetermined variable, also called a jump variable. A
decision about how much to consume and how much to save or dissave in
a given month is an example of a jump variable. Returning to our bath tub
example: in the moment we pull out the waste plug, the out‡ow of water per
time unit will jump from zero to a positive value it is a jump variable.

Types of basic model relations

Although model relations can take di¤erent forms, in macroeconomics they
often have the form of equations. A taxonomy for macroeconomic model relations
is the following:

1. Technology equations describe relations between inputs and output (pro-

duction functions and similar).
P
2. Preference equations express preferences, e.g. U = Tt=0 (1+
u(ct )
)t
; > 0; u0 >
0; u00 < 0:

3. Budget constraints, whether in the form of an equation or an inequality.

4. Institutional equations refer to relationships required by law (e.g., how the

tax levied depends on income) and similar.

5. Behavioral equations describe the behavioral response to the determinants

of behavior. This includes an agent’s optimizing behavior written as a func-
tion of its determinants. A consumption function is an example. Whether
…rst-order conditions in optimization problems should be considered behav-
ioral equations or just separate …rst-order conditions is a matter of taste.

6. Identity equations are true by de…nition of the variables involved. National

income accounting equations are an example.

7. Equilibrium equations de…ne the condition for equilibrium (“state of rest”)

of some kind, for instance equality of Walrasian demand and Walrasian
supply. No-arbitrage conditions for the asset markets also belong under the
heading equilibrium condition.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10 CHAPTER 1. INTRODUCTION

8. Initial conditions are equations …xing the initial values of the state variables
in a dynamic model

Types of analysis
Statics vs. dynamics. Comparative dynamics vs. study of dynamic e¤ects of
a parameter shift in historical time.
Macroeconomics studies processes in real time. The emphasis is on dynamic
models, that is, models that establishes a link from the state of the economic
system to the subsequent state. A dynamic model thus allows a derivation of
the evolution over time of the endogenous variables. A static model is a model
where time does not enter or where all variables refer to the same point in time.
Occasionally we consider static models, or more precisely quasi-static models. The
modi…er “quasi-”is meant to indicate that although the model is a framework for
analysis of only a single period, the model considers some variables as inherited
from the past and some variables that involve expectations about the future.
What we call temporary equilibrium models are of this type. Their role is to serve
as a prelude to a more elaborate dynamic model dealing with the same elements.
Dynamic analysis aims at establishing dynamic properties of an economic
system: is the system stable or unstable, is it asymptotically stable, if so, is it
globally or only locally asymptotically stable, is it oscillatory? If the system is
asymptotically stable, how fast is the adjustment?
Partial equilibrium vs. general equilibrium:
We say that a given single market is in partial equilibrium at a given point in
time if for arbitrarily given prices and quantities in the other markets, the agents’
chosen actions in this market are mutually compatible. In contrast the concept of
general equilibrium take the mutual dependencies between markets into account.
We say that a given economy is in general equilibrium at a given point in time if
in all markets the actions chosen by all the agents are mutually compatible.
An analyst trying to clarify a partial equilibrium problem is doing partial
equilibrium analysis. Thus partial equilibrium analysis does not take into account
the feedbacks from these actions to the rest of the economy and the feedbacks
from these feedbacks and so on. In contrast, an analyst trying to clarify a
general equilibrium problem is doing general equilibrium analysis. This requires
considering the mutual dependencies in the system of markets as a whole.
Sometimes even the analysis of the constrained maximization problem of a
single decision maker is called partial equilibrium analysis. Consider for instance
the consumption-saving decision of a household. Then the analytical derivation
of the saving function of the household is by some authors included under the
heading partial equilibrium analysis, which may seem natural since the real wage
and real interest rate appearing as arguments in the derived saving function are

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.2. Components of macroeconomic models 11

arbitrary. Indeed, what the actual saving of the young will be in the end, depends
on the real wage and real interest rate formed in the general equilibrium.
In this book we call the analysis of a single decision maker’s problem partial
analysis, not partial equilibrium analysis. The motivation for this is that trans-
parency is improved if one preserves the notion of equilibrium for a state of a
market or a state of a system of markets .

1.2.2 The time dimension of input and output

In macroeconomic theory the production of a …rm, a sector, or the economy as a
whole is often represented by a two-inputs-one-output production function,

Y = F (K; L); (1.1)

where Y is output (value added in real terms), K is capital input, and L is

labor input (K 0; L 0). The idea is that for several issues it is useful to
think of output as a homogeneous good which is produced by two inputs, one of
which is capital, by which we mean a producible durable means of production, the
other being labor, usually considered a non-producible human input. Of course,
thinking of these variables as representing one-dimensional entities is a drastic
abstraction, but may nevertheless be worthwhile in a …rst approach.
Simple as it looks, an equation like (1.1) is not always interpreted in the
right way. A key issue here is: how are the variables entering (1.1) denominated,
that is, in what units are the variables measured? It is most satisfactory, both
from a theoretical and empirical point of view, to think of both outputs and
inputs as ‡ows: quantities per unit of time. This is generally recognized as far
as Y is concerned. Unfortunately, it is less recognized concerning K and L; a
circumstance which is probably related to a tradition in macroeconomic notation,
as we will now explain.
Let the time unit be one year. Then the K appearing in the production
function should be seen as the number of machine hours per year. Similarly, L
should be seen as the number of labor hours per year. Unless otherwise speci…ed,
it should be understood that the rate of utilization of the production factors is
constant over time; for convenience, one can then normalize the rate of utilization
of each factor to equal one. Thus, with one year as our time unit, we imagine
that “normally” a machine is in operation in h hours during a year. Then, we
de…ne one machine-year as the service of a machine in operation h hours a year.
If K machines are in operation and on average deliver one machine year per year,
then the total capital input is K machine-years per year:

K (machine-yrs/yr) = K (machines) 1 ((machine-yrs/yr)/machine), (1.2)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

12 CHAPTER 1. INTRODUCTION

where the denomination of the variables is indicated in brackets. Similarly, if

the stock of laborers is L men and on average they deliver one man-year (say h
hours) per year, then the total labor input is L man-years per year:

L(man-yrs/yr) = L(men) 1((man-yrs/yr)/man). (1.3)

One of the reasons that confusion of stocks and ‡ows may arise is the tradition
in macroeconomics to use the same symbol, K; for the capital input (the number
of machine hours per year), in (1.1) as for the capital stock in an accumulation
equation like
Kt+1 = Kt + It Kt : (1.4)
Here the interpretation of Kt is as a capital stock (number of machines) at the
beginning of period t; It is gross investment, and is the rate of physical capital
depreciation due to wear and tear (0 1): In (1.4) there is no role for the
rate of utilization of the capital stock, which is, however, of key importance in
(1.1). Similarly, there is a tradition in macroeconomics to denote the number of
heads in the labor force by L and write, for example, Lt = L0 (1 + n)t ; where n
is a constant growth rate of the labor force. Here the interpretation of Lt is as a
stock (number of persons). There is no role for the average rate of utilization in
actual employment of this stock over the year.
This text will not attempt a break with this tradition of using the same symbol
for two in principle di¤erent variables. But we insist on interpretations such that
the notation is consistent. This requires normalization of the utilization rates for
capital and labor in the production function to equal one, as indicated in (1.2)
and (1.3) above. We are then allowed to use the same symbol for a stock and the
corresponding ‡ow because the values of the two variables will coincide.
An illustration of the importance of being aware of the distinction between
stock and ‡ows appears when we consider the following measure of per capita
income in a given year:
GDP GDP #hours of work #employed workers #workers
= ;
N #hours of work #employed workers #workers N
(1.5)
where N; #workers, and #employed workers indicate, say, the average size of the
population, the workforce (including the unemployed), and the employed work-
force, respectively, during the year. That is, aggregate per capita income equals
average labor productivity times average labor intensity times the crude employ-
ment rate times the workforce participation rate.3 An increase from one year to
3
By the crude employment rate is meant the number of employed individuals, without
weighting by the number of hours they work per week, divided by the total number of individuals
in the labor force.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.3. Macroeconomic models and national income accounting 13

the next in the ratio on the left-hand side of the equation re‡ects the net e¤ect
of changes in the four ratios on the right-hand side. Similarly, a fall in per capita
income (a ratio between a ‡ow and a stock) need not re‡ect a fall in productiv-
ity (GDP=#hours of work, a ratio of two ‡ows), but may re‡ect, say, a fall in
the number of hours per member of the workforce (#hours of work/#workers)
due to a rise in unemployment (fall in #employed workers/workers) or an ageing
population (fall in #workers/N ).
A second conceptual issue concerning the production function in (1.1) re-
lates to the question: what about land and other natural resources? As farming
requires land and factories and o¢ ce buildings require building sites, a third
argument, a natural resource input, should in principle appear in (1.1). In theo-
retical macroeconomics for industrialized economies this third factor is often left
out because it does not vary much as an input to production and tends to be of
secondary importance in value terms.
A third conceptual issue concerning the production function in (1.1) relates to
the question: what about intermediate goods? By intermediate goods we mean
non-durable means of production like raw materials and energy. Certainly, raw
materials and energy are generally necessary inputs at the micro level. Then
it seems strange to regard output as produced by only capital and labor. The
point is that in macroeconomics we often abstract from the engineering input-
output relations, involving intermediate goods. We imagine that at a lower stage
of production, raw materials and energy are continuously produced by capital
and labor, but are then immediately used up at a higher stage of production,
again using capital and labor. The value of these materials are not part of value
added in the sector or in the economy as a whole. Since value added is what
macroeconomics usually focuses at and what the Y in (1.1) represents, materials
therefore are often not explicit in the model.
On the other hand, if of interest for the problems studied, the analysis should,
of course, take into account that at the aggregate level in real world situations,
there will generally be a minor di¤erence between produced and used-up raw
materials which then constitute net investment in inventories of materials.
To further clarify this point as well as more general aspects of how macro-
economic models are related to national income and product accounts, the next
section gives a review of national income accounting.

1.3 Macroeconomic models and national income

accounting
Stylized national income and product accounts

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14 CHAPTER 1. INTRODUCTION

(very incomplete)
We give here a stylized picture of national income and product accounts with
emphasis on the conceptual structure. The basic point to be aware of is that
national income accounting looks at output from three sides:

the production side (value added),

the use side,

the income side.

These three “sides”refer to di¤erent approaches to the practical measurement

of production and income: the “output approach”, the “expenditure approach”,
and the “income approach”.
Consider a closed economy with three production sectors. Sector 1 produces
raw materials (or energy) in the amount Q1 per time unit, Sector 2 produces
durable capital goods in the amount Q2 per time unit, and the third sector pro-
duces consumption goods in the amount Q3 per time unit. It is common to distin-
guish between three basic production factors available ex ante a given production
process. These are land (or, more generally, non-producible natural resources),
labor, and capital (producible durable means of production). In practice also raw
materials are a necessary production input. Traditionally, this input has been
regarded as itself produced at an early stage within the production process and
then used up during the remainder of the production process. In formal dynamic
analysis, however, both capital and raw materials are considered produced prior
to the production process in which the latter are used up. This is why we include
raw materials as a fourth production factor in the production functions of the
three sectors.
....

1.4 Some terminological points

On the vocabulary used in this book:
(Incomplete)
Economic terms
Physical capital refers to stocks of reproducible durable means of production
such as machines and structures. Reproducible non-durable means of production
include raw materials and energy and are sometimes called intermediate goods.
Non-reproducible means of production, such as land and other natural resources,
are in this book not included under the heading “capital”but just called natural
resources.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

1.5. Brief history of macroeconomics 15

We follow the convention in macroeconomics and, unless otherwise speci…ed,

use “capital”for physical capital, that is, a production factor. In other branches
of economics and in everyday language “capital”may mean the funds (sometimes
called “…nancial capital”) that …nance purchases of physical capital.
By a household’s wealth (sometimes denoted net wealth), W; we mean the
value of the total stock of resources possessed by the household at a given point in
time. This wealth generally has two main components, the human wealth, which
is the present value of the stream of future labor income, and the non-human
wealth. The latter is the sum of the value of the household’s physical assets (also
called real assets) and its net …nancial assets. Typically, housing wealth is the
dominating component in households’physical assets. By net …nancial assets is
meant the di¤erence between the value of …nancial assets and the value of …nancial
liabilities. Financial assets include cash as well as paper claims that entitles the
owner to future transfers from the issuer of the claim, perhaps conditional on
certain events. Bonds and shares are examples. And a …nancial liability of a
household (or other type of agent) is an obligation to transfer resources to others
in the future. A mortgage loan is an example.
In spite of this distinction between what is called physical assets and what is
called …nancial assets, often in macroeconomics (and in this book unless other-
wise indicated) the household’s “…nancial wealth” is used synonymous with its
non-human wealth, that is, including purely physical assets like land, house, car,
machines, and other equipment. Somewhat at odds with this convention macro-
economics (including this book) generally uses “investment”as synonymous with
“physical capital investment”, that is, procurement of new machines and plants
by …rms and new houses or apartments by households. Then, when having pur-
chases of …nancial assets in mind, macroeconomists talk of …nancial investment.
...
Saving (‡ow) vs. savings (stock).
...

1.5 Brief history of macroeconomics

Text not yet available.
—
Akerlof and Shiller (2009)
Gali (2008)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16 CHAPTER 1. INTRODUCTION

1.6 Literature notes

....
The modern theory of economic growth (“new growth theory”, “endogenous
growth theory”) is extensively covered in dedicated textbooks like Aghion and
Howitt (1998), Jones (2002), Barro and Sala-i Martin (2004), Acemoglu (2009),
and Aghion and Howitt (2009). A good introduction to analytical development
economics is Basu (1997).
Snowdon and Vane (1997), Blanchard (2000), and Woodford (2000) present
useful overviews of the history of macroeconomics. For surveys on recent devel-
opments on the research agenda within theory as well as practical policy analysis,
see Mankiw (2006), Blanchard (2008), and Woodford (2009). Somewhat di¤erent
perspectives, from opposite poles, are o¤ered by Chari et al. (2009) and Colander
et al. (2008).
To be incorporated in the preface:
Two textbooks that have been a great inspiration for the one in your hands
are Blanchard and Fischer, Lectures in Macroeconomics, 1989, and Malinvaud,
Macroeconomic Theory, vol. A and B, 1998, both of which dig deeper into a
lot of the stu¤. Compared with Blanchard and Fischer the present book on the
one hand of course includes some more recent contributions to macroeconomics,
while on ther hand it is more elementary. It is intended to be accessible for third-
year undergraduates with a good background in calculus and …rst-year graduate
students. Compared with Malinvaud the emphasis in this book is more on formu-
lating complete dynamic models and analyze their applications and implications.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 2

Review of technology and …rms

The aim of this chapter is threefold. First, we shall introduce this book’s vocabu-
lary concerning …rms’technology and technological change. Second, we shall re-
fresh our memory of key notions from microeconomics relating to …rms’behavior
and factor market equilibrium under simplifying assumptions, including perfect
competition. Finally, to prepare for the many cases where perfect competition
and other simplifying assumptions are not good approximations to reality, we
give an introduction to …rms’behavior under more realistic conditions including
monopolistic competition.
The vocabulary pertaining to other aspects of the economy, for instance house-
holds’preferences and behavior, is better dealt with in close connection with the
speci…c models to be discussed in the subsequent chapters. Regarding the dis-
tinction between discrete and continuous time analysis, most of the de…nitions
contained in this chapter are applicable to both.

2.1 The production technology

Consider a two-input-one-output production function given by

Y = F (K; L); (2.1)

where Y is output (value added) per time unit, K is capital input per time unit,
and L is labor input per time unit (K 0; L 0). We may think of (2.1)
as describing the output of a …rm, a sector, or the economy as a whole. It is
in any case a very simpli…ed description, ignoring the heterogeneity of output,
capital, and labor. Yet, for many macroeconomic questions it may be a useful
…rst approach.
Note that in (2.1) not only Y but also K and L represent ‡ows, that is,
quantities per unit of time. If the time unit is one year, we think of K as

17
18 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

measured in machine hours per year. Similarly, we think of L as measured in

labor hours per year. Unless otherwise speci…ed, it is understood that the rate of
utilization of the production factors is constant over time and normalized to one
for each production factor. As explained in Chapter 1, we can then use the same
symbol, K; for the ‡ow of capital services as for the stock of capital. Similarly
with L:

2.1.1 A neoclassical production function

By de…nition, Y; K and L are non-negative. It is generally understood that a
production function, Y = F (K; L); is continuous and that F (0; 0) = 0 (no input,
no output). Sometimes, when a production function is speci…ed by a certain for-
mula, that formula may not be de…ned for K = 0 or L = 0 or both. In such a case
we adopt the convention that the domain of the function is understood extended
to include such boundary points whenever it is possible to assign function values
to them such that continuity is maintained. For instance the function F (K; L)
= L + KL=(K + L); where > 0 and > 0; is not de…ned at (K; L) = (0; 0):
But by assigning the function value 0 to the point (0; 0); we maintain both con-
tinuity and the “no input, no output”property, cf. Exercise 2.4.
We call the production function neoclassical if for all (K; L); with K > 0 and
L > 0; the following additional conditions are satis…ed:

(a) F (K; L) has continuous …rst- and second-order partial derivatives satisfying:

FK > 0; FL > 0; (2.2)

FKK < 0; FLL < 0: (2.3)

(b) F (K; L) is strictly quasiconcave (i.e., the level curves, also called isoquants,
are strictly convex to the origin).

In words: (a) says that a neoclassical production function has continuous

substitution possibilities between K and L and the marginal productivities are
positive, but diminishing in own factor. Thus, for a given number of machines,
adding one more unit of labor, adds to output, but less so, the higher is already
the labor input. And (b) says that every isoquant, F (K; L) = Y ; has a strictly
convex form qualitatively similar to that shown in Fig. 2.1.1 When we speak
of for example FL as the marginal productivity of labor, it is because the “pure”
1
For any …xed Y 0; the associated isoquant is the level set f(K; L) 2 R+ j F (K; L) = Y :
A refresher on mathematical terms such as level set, boundary point, convex function, etc. is
contained in Math Tools.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 19

partial derivative, @Y =@L = FL ; has the denomination of a productivity (out-

put units/yr)/(man-yrs/yr). It is quite common, however, to refer to FL as the
marginal product of labor. Then a unit marginal increase in the labor input is
understood: Y (@Y =@L) L = @Y =@L when L = 1: Similarly, FK can
be interpreted as the marginal productivity of capital or as the marginal prod-
uct of capital. In the latter case it is understood that K = 1; so that Y
(@Y =@K) K = @Y =@K:
The de…nition of a neoclassical production function can be extended to the
case of n inputs. Let the input quantities be X1 ; X2 ; : : : ; Xn and consider a
production function Y = F (X1 ; X2 ; : : : ; Xn ): Then F is called neoclassical if all
the marginal productivities are positive, but diminishing in own factor, and F is
strictly quasiconcave (i.e., the upper contour sets are strictly convex, cf. Appendix
A). An example where n = 3 is Y = F (K; L; J); where J is land, an important
production factor in an agricultural economy.
Returning to the two-factor case, since F (K; L) presumably depends on the
level of technical knowledge and this level depends on time, t; we might want to
replace (2.1) by
Yt = F (Kt ; Lt ; t); (2.4)

where the third argument indicates that the production function may shift over
time, due to changes in technology. We then say that F is a neoclassical produc-
tion function if for all t in a certain time interval it satis…es the conditions (a)
and (b) w.r.t its …rst two arguments. Technological progress can then be said to
occur when, for Kt and Lt held constant, output increases with t:
For convenience, to begin with we skip the explicit reference to time and level
of technology.

The marginal rate of substitution Given a neoclassical production function

F; we consider the isoquant de…ned by F (K; L) = Y ; where Y is a positive con-
stant. The marginal rate of substitution, M RSKL , of K for L at the point (K; L)
is de…ned as the absolute slope of the isoquant (K; L) 2 R2++ F (K; L) = Y at
that point, cf. Fig. 2.1. For some reason (unknown to this author) the tradition
in macroeconomics is to write Y = F (K; L) and in spite of ordering the argu-
ments of F this way, nonetheless have K on the vertical and L on the horizontal
axis when considering an isoquant. At this point we follow the tradition.
The equation F (K; L) = Y de…nes K as an implicit function K = '(L) of L:
By implicit di¤erentiation we get FK (K; L)dK=dL +FL (K; L) = 0; from which
follows
dK FL (K; L)
M RSKL = '0 (L) = > 0: (2.5)
dL jY =Y FK (K; L)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

20 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

So M RSKL equals the ratio of the marginal productivities of labor and capital,
respectively.2 The economic interpretation of M RSKL is that it indicates (ap-
proximately) the amount of K that can be saved by applying an extra unit of
labor.
Since F is neoclassical, by de…nition F is strictly quasi-concave and so the
marginal rate of substitution is diminishing as substitution proceeds, i.e., as the
labor input is further increased along a given isoquant. Notice that this feature
characterizes the marginal rate of substitution for any neoclassical production
function, whatever the returns to scale (see below).

Figure 2.1: M RSKL as the absolute slope of the isoquant representing F (K; L) = Y .

When we want to draw attention to the dependency of the marginal rate

of substitution on the factor combination considered, we write M RSKL (K; L):
Sometimes in the literature, the marginal rate of substitution between two pro-
duction factors, K and L; is called the technical rate of substitution (or the
technical rate of transformation) in order to distinguish from a consumer’s mar-
ginal rate of substitution between two consumption goods.
As is well-known from microeconomics, a …rm that minimizes production costs
for a given output level and given factor prices, will choose a factor combination
such that M RSKL equals the ratio of the factor prices. If F (K; L) is homogeneous
of degree q, then the marginal rate of substitution depends only on the factor
proportion and is thus the same at any point on the ray K = (K=L)L: In this
case the expansion path is a straight line.
2
The subscript Y = Y in (2.5) signi…es that “we are moving along a given isoquant F (K; L)
= Y ”, i.e., we are considering the relation between K and L under the restriction F (K; L) = Y :
Expressions like FL (K; L) or F2 (K; L) mean the partial derivative of F w.r.t. the second
argument, evaluated at the point (K; L):

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 21

The Inada conditions A continuously di¤erentiable production function is

said to satisfy the Inada conditions 3 if

lim FK (K; L) = 1; lim FK (K; L) = 0; (2.6)

K!0 K!1
lim FL (K; L) = 1; lim FL (K; L) = 0: (2.7)
L!0 L!1

In this case, the marginal productivity of either production factor has no upper
bound when the input of the factor becomes in…nitely small. And the marginal
productivity is gradually vanishing when the input of the factor increases without
bound. Actually, (2.6) and (2.7) express four conditions, which it is preferable to
consider separately and label one by one. In (2.6) we have two Inada conditions
for M P K (the marginal productivity of capital), the …rst being a lower, the
second an upper Inada condition for M P K. And in (2.7) we have two Inada
conditions for M P L (the marginal productivity of labor), the …rst being a lower,
the second an upper Inada condition for M P L. In the literature, when a sentence
like “the Inada conditions are assumed”appears, it is sometimes not made clear
which, and how many, of the four are meant. Unless it is evident from the context,
it is better to be explicit about what is meant.
The de…nition of a neoclassical production function we have given is quite
common in macroeconomic journal articles and convenient because of its ‡exibil-
ity. Yet there are textbooks that de…ne a neoclassical production function more
narrowly by including the Inada conditions as a requirement for calling the pro-
duction function neoclassical. In contrast, in this book, when in a given context
we need one or another Inada condition, we state it explicitly as an additional
assumption.

2.1.2 Returns to scale

If all the inputs are multiplied by some factor, is output then multiplied by the
same factor? There may be di¤erent answers to this question, depending on
circumstances. We consider a production function F (K; L) where K > 0 and
L > 0: Then F is said to have constant returns to scale (CRS for short) if it is
homogeneous of degree one, i.e., if for all (K; L) 2 R2++ and all > 0;

F ( K; L) = F (K; L):

As all inputs are scaled up or down by some factor, output is scaled up or down
by the same factor.4 The assumption of CRS is often defended by the replication
3
After the Japanese economist Ken-Ichi Inada, 1925-2002.
4
In their de…nition of a neoclassical production function some textbooks add constant re-
turns to scale as a requirement besides (a) and (b) above. This book follows the alternative

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

22 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

argument saying that “by doubling all inputs we are always able to double the
output since we are essentially just replicating a viable production activity”.
Before discussing this argument, lets us de…ne the two alternative “pure”cases.
The production function F (K; L) is said to have increasing returns to scale
(IRS for short) if, for all (K; L) 2 R2++ and all > 1,

F ( K; L) > F (K; L):

That is, IRS is present if, when increasing the scale of operations by scaling up
every input by some factor > 1, output is scaled up by more than this factor. One
argument for the plausibility of this is the presence of equipment indivisibilities
leading to high unit costs at low output levels. Another argument is that gains
by specialization and division of labor, synergy e¤ects, etc. may be present, at
least up to a certain level of production. The IRS assumption is also called the
economies of scale assumption.
Another possibility is decreasing returns to scale (DRS). This is said to occur
when for all (K; L) 2 R2++ and all > 1;

F ( K; L) < F (K; L):

That is, DRS is present if, when all inputs are scaled up by some factor, output
is scaled up by less than this factor. This assumption is also called the disec-
onomies of scale assumption. The underlying hypothesis may be that control and
coordination problems con…ne the expansion of size. Or, considering the “repli-
cation argument” below, DRS may simply re‡ect that behind the scene there
is an additional production factor, for example land or a irreplaceable quality
of management, which is tacitly held …xed, when the factors of production are
varied.
EXAMPLE 1 The production function

Y = AK L ; A > 0; 0 < < 1; 0 < < 1; (2.8)

where A; ; and are given parameters, is called a Cobb-Douglas production

function. The parameter A depends on the choice of measurement units; for a
given such choice it re‡ects e¢ ciency, also called the “total factor productivity”.
Exercise 2.2 asks the reader to verify that (2.8) satis…es (a) and (b) above and
is therefore a neoclassical production function. The function is homogeneous of
degree + . If + = 1; there are CRS. If + < 1; there are DRS, and if
terminology where, if in a given context an assumption of constant returns to scale is needed,
this is stated as an additional assumption and we talk about a CRS-neoclassical production
function.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 23

+ > 1; there are IRS. Note that and must be less than 1 in order not to
violate the diminishing marginal productivity condition.
EXAMPLE 2 The production function
1
Y =A K + (1 )L ; (2.9)

where A; ; and are parameters satisfying A > 0, 0 < < 1; and < 1; 6= 0;
is called a CES production function (CES for Constant Elasticity of Substitution).
For a given choice of measurement units, the parameter A re‡ects e¢ ciency (or
“total factor productivity”) and is thus called the e¢ ciency parameter. The
parameters and are called the distribution parameter and the substitution
parameter, respectively. The latter name comes from the property that the higher
is ; the more sensitive is the cost-minimizing capital-labor ratio to a rise in
the relative factor price. Equation (2.9) gives the CES function for the case of
constant returns to scale; the cases of increasing or decreasing returns to scale
are presented in Chapter 4.5. A limiting case of the CES function (2.9) gives the
Cobb-Douglas function with CRS. Indeed, for …xed K and L;
1
lim A K + (1 )L = AK L1 :
!0

This and other properties of the CES function are shown in Chapter 4.5. The
CES function has been used intensively in empirical studies.
EXAMPLE 3 The production function

Y = min(AK; BL); A > 0; B > 0; (2.10)

where A and B are given parameters, is called a Leontief production function 5

(or a …xed-coe¢ cients production function; A and B are called the technical coef-
…cients. The function is not neoclassical, since the conditions (a) and (b) are not
satis…ed. Indeed, with this production function the production factors are not
substitutable at all. This case is also known as the case of perfect complementarity
between the production factors. The interpretation is that already installed pro-
duction equipment requires a …xed number of workers to operate it. The inverse
of the parameters A and B indicate the required capital input per unit of output
and the required labor input per unit of output, respectively. Extended to many
inputs, this type of production function is often used in multi-sector input-output
models (also called Leontief models). In aggregate analysis neoclassical produc-
tion functions, allowing substitution between capital and labor, are more popular
5
After the Russian-American economist and Nobel laureate Wassily Leontief (1906-99) who
used a generalized version of this type of production function in what is known as input-output
analysis.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

24 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

than Leontief functions. But sometimes the latter are preferred, in particular in
short-run analysis with focus on the use of already installed equipment where the
substitution possibilities tend to be limited.6 As (2.10) reads, the function has
CRS. A generalized form of the Leontief function is Y = min(AK ; BL ); where
> 0. When < 1; there are DRS, and when > 1; there are IRS.

The replication argument The assumption of CRS is widely used in macro-

economics. The model builder may appeal to the replication argument. This is
the argument saying that by doubling all the inputs, we should always be able
to double the output, since we are just “replicating”what we are already doing.
Suppose we want to double the production of cars. We may then build another
factory identical to the one we already have, man it with identical workers and
deploy the same material inputs. Then it is reasonable to assume output is dou-
bled.
In this context it is important that the CRS assumption is about technology in
the sense of functions linking outputs to inputs. Limits to the availability of input
resources is an entirely di¤erent matter. The fact that for example managerial
talent may be in limited supply does not preclude the thought experiment that
if a …rm could double all its inputs, including the number of talented managers,
then the output level could also be doubled.
The replication argument presupposes, …rst, that all the relevant inputs are
explicit as arguments in the production function; second, that these are changed
equiproportionately. This, however, exhibits the weakness of the replication argu-
ment as a defence for assuming CRS of our present production function, F: One
could easily make the case that besides capital and labor, also land is a necessary
input and should appear as a separate argument.7 If an industrial …rm decides
to duplicate what it has been doing, it needs a piece of land to build another
plant like the …rst. Then, on the basis of the replication argument, we should in
fact expect DRS w.r.t. capital and labor alone. In manufacturing and services,
empirically, this and other possible sources for departure from CRS w.r.t. capital
and labor may be minor and so many macroeconomists feel comfortable enough
with assuming CRS w.r.t. K and L alone, at least as a …rst approximation.
This approximation is, however, less applicable to poor countries, where natural
resources may be a quantitatively important production factor.
There is a further problem with the replication argument. By de…nition, CRS
is present if and only if, by changing all the inputs equiproportionately by any
positive factor (not necessarily an integer), the …rm is able to get output changed
6
Cf. Section 2.5.2.
7
Recall from Chapter 1 that we think of “capital”as producible means of production, whereas
“land” refers to non-producible natural resources, including for instance building sites.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 25

by the same factor. Hence, the replication argument requires that indivisibilities
are negligible, which is certainly not always the case. In fact, the replication
argument is more an argument against DRS than for CRS in particular. The
argument does not rule out IRS due to synergy e¤ects as scale is increased.
Sometimes the replication line of reasoning is given a more subtle form. This
builds on a useful local measure of returns to scale, named the elasticity of scale.

The elasticity of scale*8 To allow for indivisibilities and mixed cases (for
example IRS at low levels of production and CRS or DRS at higher levels), we
need a local measure of returns to scale. One de…nes the elasticity of scale,
(K; L); of F at the point (K; L); where F (K; L) > 0; as

dF ( K; L) F ( K; L)=F (K; L)
(K; L) = ; evaluated at = 1:
F (K; L) d =
(2.11)
So the elasticity of scale at a point (K; L) indicates the (approximate) percentage
increase in output when both inputs are increased by 1 percent. We say that
8
< > 1; then there are locally IRS,
if (K; L) = 1; then there are locally CRS, (2.12)
:
< 1; then there are locally DRS.

The production function may have the same elasticity of scale everywhere. This
is the case if and only if the production function is homogeneous of some degree
h > 0. In that case (K; L) = h for all (K; L) for which F (K; L) > 0; and h
indicates the global elasticity of scale. The Cobb-Douglas function, cf. Example
1, is homogeneous of degree + and has thereby global elasticity of scale equal
to + :
Note that the elasticity of scale at a point (K; L) will always equal the sum
of the partial output elasticities at that point:

FK (K; L)K FL (K; L)L

(K; L) = + : (2.13)
F (K; L) F (K; L)

This follows from the de…nition in (2.11) by taking into account that

dF ( K; L)
= FK ( K; L)K + FL ( K; L)L
d
= FK (K; L)K + FL (K; L)L; when evaluated at = 1:
8
A section headline marked by * indicates that in a …rst reading the section can be skipped
- or at least just skimmed through.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

26 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

Fig. 2.2 illustrates a popular case from introductory economics, an average

cost curve which from the perspective of the individual …rm is U-shaped: at low
levels of output there are falling average costs (thus IRS), at higher levels rising
average costs (thus DRS).9 Given the input prices wK and wL and a speci…ed
output level F (K; L) = Y ; we know that the cost-minimizing factor combination
(K; L) is such that FL (K; L)=FK (K; L) = wL =wK : It is shown in Appendix A
that the elasticity of scale at (K; L) will satisfy:

LAC(Y )
(K; L) = ; (2.14)
LM C(Y )

where LAC(Y ) is average costs (the minimum unit cost associated with producing
Y ) and LM C(Y ) is marginal costs at the output level Y . The L in LAC and
LM C stands for “long-run”, indicating that both capital and labor are considered
variable production factors within the period considered. At the optimal plant
size, Y ; there is equality between LAC and LM C, implying a unit elasticity
of scale. That is, locally we have CRS. That the long-run average costs are
here portrayed as rising for Y > Y ; is not essential for the argument but may
re‡ect either that coordination di¢ culties are inevitable or that some additional
production factor, say the building site of the plant, is tacitly held …xed.

Figure 2.2: Locally CRS at optimal plant size.

Anyway, on this basis Robert Solow (1956) came up with a more subtle repli-
cation argument for CRS at the aggregate level. Even though technologies may
di¤er across plants, the surviving plants in a competitive market will have the
same average costs at the optimal plant size. In the medium and long run, changes
in aggregate output will take place primarily by entry and exit of optimal-size
9
By a “…rm” is generally meant the company as a whole. A company may have several
“manufacturing plants” placed at di¤erent locations.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 27

plants. Then, with a large number of relatively small plants, each producing at
approximately constant unit costs for small output variations, we can without
substantial error assume constant returns to scale at the aggregate level. So the
argument goes. Notice, however, that even in this form the replication argument
is not entirely convincing since the question of indivisibility remains. The opti-
mal, i.e., cost-minimizing, plant size may be large relative to the market and
is in fact so in many industries. Besides, in this case also the perfect competition
premise breaks down.

2.1.3 Properties of the production function under CRS

The empirical evidence concerning returns to scale is mixed (see the literature
notes at the end of the chapter). Notwithstanding the theoretical and empirical
ambiguities, the assumption of CRS w.r.t. capital and labor has a prominent
role in macroeconomics. In many contexts it is regarded as an acceptable ap-
proximation and a convenient simple background for studying the question at
hand.
Expedient inferences of the CRS assumption include:

(i) marginal costs are constant and equal to average costs (so the right-hand
side of (2.14) equals unity);

(ii) if production factors are paid according to their marginal productivities,

factor payments exactly exhaust total output so that pure pro…ts are neither
positive nor negative (so the right-hand side of (2.13) equals unity);

(iii) a production function known to exhibit CRS and satisfy property (a) from
the de…nition of a neoclassical production function above, will automatically
satisfy also property (b) and consequently be neoclassical;

(iv) a neoclassical two-factor production function with CRS has always FKL > 0;
i.e., it exhibits “direct complementarity”between K and L;

(v) a two-factor production function that has CRS and is twice continuously
di¤erentiable with positive marginal productivity of each factor everywhere
in such a way that all isoquants are strictly convex to the origin, must
have diminishing marginal productivities everywhere and thereby be neo-
classical.10

A principal implication of the CRS assumption is that it allows a reduction

of dimensionality. Considering a neoclassical production function, Y = F (K; L)
10
Proof of claim (iii) is in Appendix A and proofs of claim (iv) and (v) are in Appendix B.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

28 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

with L > 0; we can under CRS write F (K; L) = LF (K=L; 1) Lf (k); where
k K=L is called the capital-labor ratio (sometimes the capital intensity) and
f (k) is the production function in intensive form (sometimes named the per capita
production function). Thus output per unit of labor depends only on the capital
intensity:
Y
y = f (k):
L
When the original production function F is neoclassical, under CRS the expres-
sion for the marginal productivity of capital simpli…es:

@Y @ [Lf (k)] @k
FK (K; L) = = = Lf 0 (k) = f 0 (k): (2.15)
@K @K @K
And the marginal productivity of labor can be written

@Y @ [Lf (k)] @k
FL (K; L) = = = f (k) + Lf 0 (k)
@L @L @L
= f (k) + Lf 0 (k)K( L 2 ) = f (k) f 0 (k)k: (2.16)

A neoclassical CRS production function in intensive form always has a positive

…rst derivative and a negative second derivative, i.e., f 0 > 0 and f 00 < 0: The
property f 0 > 0 follows from (2.15) and (2.2). And the property f 00 < 0 follows
from (2.3) combined with

@f 0 (k) @k 1
FKK (K; L) = = f 00 (k) = f 00 (k) :
@K @K L
For a neoclassical production function with CRS, we also have

f (k) f 0 (k)k > 0 for all k > 0; (2.17)

in view of f (0) 0 and f 00 < 0: Moreover,

lim [f (k) f 0 (k)k] = f (0): (2.18)

k!0

Indeed, from the mean value theorem 11 we know there exists a number a 2 (0; 1)
such that for any k > 0 we have f (k) f (0) = f 0 (ak)k: From this follows f (k)
f 0 (ak)k = f (0) < f (k) f 0 (k)k; since f 0 (ak) > f 0 (k) by f 00 < 0. In view of
f (0) 0; this establishes (2.17): And from f (k) > f (k) f 0 (k)k > f (0) and
continuity of f follows (2.18).
11
This theorem says that if f is continuous in [ ; ] and di¤erentiable in ( ; ); then there
exists at least one point in ( ; ) such that f 0 ( ) = (f ( ) f ( ))=( ):

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.1. The production technology 29

Under CRS the Inada conditions for M P K can be written

lim f 0 (k) = 1; lim f 0 (k) = 0: (2.19)
k!0 k!1

In this case standard parlance is just to say that “f satis…es the Inada conditions”.
An input which must be positive for positive output to arise is called an
essential input; an input which is not essential is called an inessential input. The
second part of (2.19), representing the upper Inada condition for M P K under
CRS, has the implication that labor is an essential input; but capital need not
be, as the production function f (k) = a + bk=(1 + k); a > 0; b > 0; illustrates.
Similarly, under CRS the upper Inada condition for M P L implies that capital
is an essential input. These claims are proved in Appendix C. Combining these
results, when both the upper Inada conditions hold and CRS obtain, then both
capital and labor are essential inputs.12
Fig. 2.3 is drawn to provide an intuitive understanding of a neoclassical
CRS production function and at the same time illustrate that the lower Inada
conditions are more questionable than the upper Inada conditions. The left panel
of Fig. 2.3 shows output per unit of labor for a CRS neoclassical production
function satisfying the Inada conditions for M P K. The f (k) in the diagram
could for instance represent the Cobb-Douglas function in Example 1 with =
1 ; i.e., f (k) = Ak : The right panel of Fig. 2.3 shows a non-neoclassical
case where only two alternative Leontief techniques are available, technique 1: y
= min(A1 k; B1 ); and technique 2: y = min(A2 k; B2 ): In the exposed case it is
assumed that B2 > B1 and A2 < A1 (if A2 A1 at the same time as B2 > B1 ;
technique 1 would not be e¢ cient, because the same output could be obtained
with less input of at least one of the factors by shifting to technique 2). If the
available K and L are such that k K=L < B1 =A1 or k > B2 =A2 , some of either
L or K; respectively, is idle. If, however, the available K and L are such that
B1 =A1 < k < B2 =A2 ; it is e¢ cient to combine the two techniques and use the
fraction of K and L in technique 1 and the remainder in technique 2, where
= (B2 =A2 k)=(B2 =A2 B1 =A1 ): In this way we get the “labor productivity
curve” OPQR (the envelope of the two techniques) in Fig. 2.3. Note that for
k ! 0; M P K stays equal to A1 < 1; whereas for all k > B2 =A2 ; M P K = 0:
A similar feature remains true, when we consider many, say n; alternative
e¢ cient Leontief techniques available. Assuming these techniques cover a con-
siderable range w.r.t. the B=A ratios, we get a labor productivity curve looking
more like that of a neoclassical CRS production function. On the one hand, this
gives some intuition of what lies behind the assumption of a neoclassical CRS
production function. On the other hand, it remains true that for all k > Bn =An ;
12
Given a Cobb-Douglas production function, both production factors are essential whether
we have DRS, CRS, or IRS.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

30 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

Figure 2.3: Two labor productivity curves based on CRS technologies. Left: neoclas-
sical technology with Inada conditions for MPK satis…ed; the graphical representation
of MPK and MPL at k = k0 as f 0 (k0 ) and f (k0 ) f 0 (k0 )k0 are indicated. Right: the
line segment PQ makes up an e¢ cient combination of two e¢ cient Leontief techniques.

M P K = 0;13 whereas for k ! 0; M P K stays equal to A1 < 1; thus questioning

the lower Inada condition.
The implausibility of the lower Inada conditions is also underlined if we look
at their implication in combination with the more reasonable upper Inada condi-
tions. Indeed, the four Inada conditions taken together imply, under CRS, that
output has no upper bound when either input goes towards in…nity for …xed
amount of the other input (see Appendix C).

2.2 Technological change

When considering the movement over time of the economy, we shall often take
into account the existence of technological change. When technological change
occurs, the production function becomes time-dependent. Over time the produc-
tion factors tend to become more productive: more output for given inputs. To
put it di¤erently: the isoquants move inward. When this is the case, we say that
the technological change displays technological progress.

Concepts of neutral technological change

A …rst step in taking technological change into account is to replace (2.1) by
(2.4). Empirical studies often specialize (2.4) by assuming that technological
change take a form known as factor-augmenting technological change:

Yt = F (At Kt ; Bt Lt ); (2.20)
13
Here we assume the techniques are numbered according to ranking with respect to the size
of B:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.2. Technological change 31

where F is a (time-independent) neoclassical production function, Yt ; Kt ; and

Lt are output, capital, and labor input, respectively, at time t; while At and
Bt are time-dependent “e¢ ciencies” of capital and labor, respectively, re‡ecting
technological change.
In macroeconomics an even more speci…c form is often assumed, namely the
form of Harrod-neutral technological change.14 This amounts to assuming that At
in (2.20) is a constant (which we can then normalize to one). So only Bt ; which
is then conveniently denoted Tt ; is changing over time, and we have

Yt = F (Kt ; Tt Lt ): (2.21)

The e¢ ciency of labor, Tt ; is then said to indicate the technology level. Although
one can imagine natural disasters implying a fall in Tt ; generally Tt tends to rise
over time and then we say that (2.21) represents Harrod-neutral technological
progress. An alternative name often used for this is labor-augmenting technolog-
ical progress. The names “factor-augmenting”and, as here, “labor-augmenting”
have become standard and we shall use them when convenient, although they
may easily be misunderstood. To say that a change in Tt is labor-augmenting
might be understood as meaning that more labor is required to reach a given
output level for given capital. In fact, the opposite is the case, namely that Tt
has risen so that less labor input is required. The idea is that the technological
change a¤ects the output level as if the labor input had been increased exactly
by the factor by which T was increased, and nothing else had happened. (We
might be tempted to say that (2.21) re‡ects “labor saving”technological change.
But also this can be misunderstood. Indeed, keeping L unchanged in response to
a rise in T implies that the same output level requires less capital and thus the
technological change is “capital saving”.)
If the function F in (2.21) is homogeneous of degree one (so that the technol-
ogy exhibits CRS w.r.t. capital and labor), we may write

Yt Kt
y~t = F( ; 1) = F (k~t ; 1) f (k~t ); f 0 > 0; f 00 < 0:
Tt Lt Tt Lt

where k~t Kt =(Tt Lt ) kt =Tt (habitually called the “e¤ective” capital intensity
or, if there is no risk of confusion, just the capital intensity). In rough accordance
with a general trend in aggregate productivity data for industrialized countries
we often assume that T grows at a constant rate, g; so that in discrete time Tt
= T0 (1 + g)t and in continuous time Tt = T0 egt ; where g > 0: The popularity
in macroeconomics of the hypothesis of labor-augmenting technological progress
derives from its consistency with Kaldor’s “stylized facts”, cf. Chapter 4.
14
After the English economist Roy F. Harrod, 1900-1978.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

32 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

There exists two alternative concepts of neutral technological progress. Hicks-

neutral technological progress is said to occur if technological development is such
that the production function can be written in the form
Yt = Tt F (Kt ; Lt ); (2.22)
where, again, F is a (time-independent) neoclassical production function, while
Tt is the growing technology level.15 The assumption of Hicks-neutrality has been
used more in microeconomics and partial equilibrium analysis than in macroeco-
nomics. If F has CRS, we can write (2.22) as Yt = F (Tt Kt ; Tt Lt ): Comparing
with (2.20), we see that in this case Hicks-neutrality is equivalent to At = Bt in
(2.20), whereby technological change is said to be equally factor-augmenting.
Finally, in a symmetric analogy with (2.21), what is known as capital-augmenting
technological progress is present when
Yt = F (Tt Kt ; Lt ): (2.23)
Here technological change acts as if the capital input were augmented. For some
reason this form is sometimes called Solow-neutral technological progress.16 This
association of (2.23) to Solow’s name is misleading, however. In his famous growth
model,17 Solow assumed Harrod-neutral technological progress. And in another
famous contribution, Solow generalized the concept of Harod-neutrality to the
case of embodied technological change and capital of di¤erent vintages, see below.
It is easily shown (Exercise 2.5) that the Cobb-Douglas production function
(2.8) (with time-independent output elasticities w.r.t. K and L) satis…es all three
neutrality criteria at the same time, if it satis…es one of them (which it does if
technological change does not a¤ect and ). It can also be shown that within
the class of neoclassical CRS production functions the Cobb-Douglas function is
the only one with this property (see Exercise 4.??).
Note that the neutrality concepts do not say anything about the source of
technological progress, only about the quantitative form in which it materializes.
For instance, the occurrence of Harrod-neutrality should not be interpreted as
indicating that the technological change emanates speci…cally from the labor
input in some sense. Harrod-neutrality only means that technological innovations
predominantly are such that not only do labor and capital in combination become
more productive, but this happens to manifest itself in the form (2.21), that is,
as if an improvement in the quality of the labor input had occurred. (Even when
improvement in the quality of the labor input is on the agenda, the result may be
a reorganization of the production process ending up in a higher Bt along with,
or instead of, a higher At in the expression (2.20).)
15
After the English economist and Nobel Prize laureate John R. Hicks, 1904-1989.
16
After the American economist and Nobel Prize laureate Robert Solow (1924-).
17
Solow (1956).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.2. Technological change 33

Rival versus nonrival goods

When a production function (or more generally a production possibility set) is
speci…ed, a given level of technical knowledge is presumed. As this level changes
over time, the production function changes. In (2.4) this dependency on the level
of knowledge was represented indirectly by the time dependency of the production
function. Sometimes it is useful to let the knowledge dependency be explicit by
perceiving knowledge as an additional production factor and write, for instance,

Yt = F (Xt ; Tt ); (2.24)

where Tt is now an index of the amount of knowledge, while Xt is a vector

of ordinary inputs like raw materials, machines, labor etc. In this context the
distinction between rival and nonrival inputs or more generally the distinction
between rival and nonrival goods is important. A good is rival if its character is
such that one agent’s use of it inhibits other agents’use of it at the same time.
A pencil is thus rival. Many production inputs like raw materials, machines,
labor etc. have this property. They are elements of the vector Xt : By contrast,
however, technical knowledge is a nonrival good. An arbitrary number of factories
can simultaneously use the same piece of technical knowledge in the sense of a list
of instructions about how di¤erent inputs can be combined to produce a certain
output. An engineering principle or a farmaceutical formula are examples. (Note
that the distinction rival-nonrival is di¤erent from the distinction excludable-
nonexcludable. A good is excludable if other agents, …rms or households, can be
excluded from using it. Other …rms can thus be excluded from commercial use of
a certain piece of technical knowledge if it is patented. The existence of a patent
concerns the legal status of a piece of knowledge and does not interfere with its
economic character as a nonrival input.).
What the replication argument really says is that by, conceptually, doubling
all the rival inputs, we should always be able to double the output, since we
just “replicate” what we are already doing. This is then an argument for (at
least) CRS w.r.t. the elements of Xt in (2.24). The point is that because of its
nonrivalry, we do not need to increase the stock of knowledge. Now let us imagine
that the stock of knowledge is doubled at the same time as the rival inputs are
doubled. Then more than a doubling of output should occur. In this sense we
may speak of IRS w.r.t. the rival inputs and T taken together.

The perpetual inventory method

Before proceeding, a brief remark about how the capital stock Kt can be measured
While data on gross investment, It ; is typically available in o¢ cial national income
and product accounts, data on Kt usually is not. It has been up to researchers

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

34 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

and research institutions to make their own time-series for capital. One approach
to the measurement of Kt is the perpetual inventory method which builds upon
the accounting relationship
Kt = It 1 + (1 )Kt 1 : (2.25)
Assuming a constant capital depreciation rate ; backward substitution gives
X
N
Kt = It 1 + (1 ) [It 2 + (1 )Kt 2 ] = . . . = (1 )i 1 It i + (1 )T Kt N:
i=1
(2.26)
Based on a long time series for I and an estimate of ; one can insert these
observed values in the formula and calculate Kt , starting from a rough conjec-
ture about the initial value Kt N : The result will not be very sensitive to this
conjecture since for large N the last term in (2.26) becomes very small.

Embodied vs. disembodied technological progress*

An additional taxonomy of technological change is the following. We say that
technological change is embodied, if taking advantage of new technical knowledge
requires construction of new investment goods. The new technology is incorpo-
rated in the design of newly produced equipment, but this equipment will not
participate in subsequent technological progress. An example: only the most
recent vintage of a computer series incorporates the most recent advance in in-
formation technology. Then investment goods produced later (investment goods
of a later “vintage”) have higher productivity than investment goods produced
earlier at the same resource cost. Thus investment becomes an important driving
force in productivity increases.
We may formalize embodied technological progress by writing capital accu-
mulation in the following way:
Kt+1 Kt = Qt It Kt ; (2.27)
where It is gross investment in period t, i.e., It = Yt Ct ; and Qt measures the
“quality” (productivity) of newly produced investment goods. The rising level
of technology implies rising Q so that a given level of investment gives rise to
a greater and greater addition to the capital stock, K; measured in e¢ ciency
units. In aggregate models C and I are produced with the same technology, the
aggregate production function. From this together with (2.27) follows that Q
capital goods can be produced at the same minimum cost as one consumption
good. Hence, the equilibrium price, p; of capital goods in terms of the consump-
tion good must equal the inverse of Q; i.e., p = 1=Q: The output-capital ratio in
value terms is Y =(pK) = QY =K:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.3. The concepts of a representative …rm and an aggregate production function35

Note that even if technological change does not directly appear in the produc-
tion function, that is, even if for instance (2.21) is replaced by Yt = F (Kt ; Lt );
the economy may experience a rising standard of living when Q is growing over
time.
In contrast, disembodied technological change occurs when new technical and
organizational knowledge increases the combined productivity of the production
factors independently of when they were constructed or educated. If the Kt
appearing in (2.21), (2.22), and (2.23) above refers to the total, historically ac-
cumulated capital stock as calculated by (2.26), then the evolution of T in these
expressions can be seen as representing disembodied technological change. All
vintages of the capital equipment bene…t from a rise in the technology level Tt :
No new investment is needed to bene…t.
Based on data for the U.S. 1950-1990, and taking quality improvements into
account, Greenwood et al. (1997) estimate that embodied technological progress
explains about 60% of the growth in output per man hour. So, empirically,
embodied technological progress seems to play a dominant role. As this tends not
to be fully incorporated in national income accounting at …xed prices, there is
a need to adjust the investment levels in (2.26) to better take estimated quality
improvements into account. Otherwise the resulting K will not indicate the
capital stock measured in e¢ ciency units.
For most issues dealt with in this book the distinction between embodied and
disembodied technological progress is not very important. Hence, unless explicitly
speci…ed otherwise, technological change is understood to be disembodied.

2.3 The concepts of a representative …rm and

an aggregate production function
Many macroeconomic models make use of the simplifying notion of a represen-
tative …rm. By this is meant a …ctional …rm whose production “represents”
aggregate production (value added) in a sector or in society as a whole.
Suppose there are n …rms in the sector considered or in society as a whole.
Let F i be the production function for …rm i so that Yi = F i (Ki ; Li ); where Yi ,
Ki ; and Li are output, capital input, and labor input, respectively, i = 1; 2; : : : ; n.
Further, let Y = ni=1 Yi , K = ni=1 Ki ; and L = ni=1 Li . Ignoring technological
change, suppose the aggregate variables are related through some function, F ;
such that we can write
Y = F (K; L);
and such that the choices of a single …rm facing this production function coincide
with the aggregate outcomes, ni=1 Yi , ni=1 Ki ; and ni=1 Li ; in the original econ-

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

36 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

omy. Then F (K; L) is called the aggregate production function or the production
function of the representative …rm. It is as if aggregate production is the result
of the behavior of such a single …rm.
A simple example where the aggregate production function is well-de…ned is
the following. Suppose that all …rms have the same production function so that
Yi = F (Ki ; Li ); i = 1; 2; : : : ; n: If in addition F has CRS, we have

Yi = F (Ki ; Li ) = Li F (ki ; 1) Li f (ki );

where ki Ki =Li : Hence, facing given factor prices, cost-minimizing …rms will
choose
P the same
P capital intensity ki = k for all i: From Ki = kLi then follows
i Ki = k i Li so that k = K=L: Thence,
X X X
Y Yi = Li f (ki ) = f (k) Li = f (k)L = F (k; 1)L = F (K; L):

In this (trivial) case the aggregate production function is well-de…ned and turns
out to be exactly the same as the identical CRS production functions of the
individual …rms. Moreover, given CRS and ki = k for all i; we have @Yi =@Ki
= f 0 (ki ) = f 0 (k) = FK (K; L) for all i: So each …rm’s marginal productivity of
capital is the same as the marginal productivity of capital on the basis of the
aggregate production function.
Allowing for the existence of di¤erent production functions at …rm level, we
may de…ne the aggregate production function as

F (K; L) = max F 1 (K1 ; L1 ) + + F n (Kn ; Ln )

(K1 ;L1 ;:::;Kn ;Ln ) 0
X X
s.t. Ki K; Li L:
i i

Here it is no longer generally true that @Yi =@Ki (= FKi (Ki ; Li ) = @Y =@K (=
FK (K; L):
A next step is to allow also for the existence of di¤erent output goods, dif-
ferent capital goods, and di¤erent types of labor. This makes the issue even
more intricate, of course. Yet, if …rms are price taking pro…t maximizers and
face nonincreasing returns to scale, we at least know from microeconomics that
the aggregate outcome is as if, for given prices, the …rms jointly maximize aggre-
gate pro…t on the basis of their combined production technology. The problem
is, however, that the conditions needed for this to imply existence of an aggre-
gate production function which is well-behaved (in the sense of inheriting simple
qualitative properties from its constituent parts) are restrictive.
Nevertheless macroeconomics often treats aggregate output as a single homo-
geneous good and capital and labor as being two single and homogeneous inputs.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.3. The concepts of a representative …rm and an aggregate production function37

There was in the 1960s a heated debate about the problems involved in this,
with particular emphasis on the aggregation of di¤erent kinds of equipment into
one variable, the capital stock “K”. The debate is known as the “Cambridge
controversy”because the dispute was between a group of economists from Cam-
bridge University, UK, and a group from Massachusetts Institute of Technology
(MIT), which is located in Cambridge, USA. The former group questioned the
theoretical robustness of several of the neoclassical tenets, including the propo-
sition that a higher aggregate capital intensity is induced by a lower rate of
interest. Starting at the disaggregate level, an association of this sort is not a
logical necessity because, with di¤erent production functions across the indus-
tries, the relative prices of produced inputs tend to change, when the interest
rate changes. While acknowledging the possibility of “paradoxical”relationships,
the MIT group maintained that in a macroeconomic context they are likely to
cause devastating problems only under exceptional circumstances. In the end this
is a matter of empirical assessment.18
To avoid complexity and because, for many important issues in macroeco-
nomics, there is today no well-tried alternative, this book is about models that
use aggregate constructs like “Y ”, “K”, and “L” as simplifying devices, assum-
ing they are, for a broad class of cases, acceptable in a …rst approximation. Of
course there are cases where some disaggregation is pertinent. When for example
the role of imperfect competition is in focus, we shall be ready to (modestly)
disaggregate the production side of the economy into several product lines, each
producing its own di¤erentiated product (cf. Section 2.5.3).
Like the representative …rm, the representative household and the aggregate
consumption function are simplifying notions that should be applied only when
they do not get in the way of the issue to be studied. The role of budget con-
straints may make it even more di¢ cult to aggregate over households than over
…rms. Yet, if (and that is a big if) all households have the same constant propen-
sity to consume out of income or wealth, aggregation is straightforward and the
representative household is a meaningful simplifying concept. On the other hand,
if we aim at understanding, say, the interaction between lending and borrowing
households, perhaps via …nancial intermediaries, the representative household is
not a useful starting point. Similarly, if the theme is con‡icts of interests between
…rm owners and employees, the existence of di¤erent types of households should
be taken into account. Or if we want to assess the welfare costs of business cycle
‡uctuations, we have to take heterogeneity into account in view of the fact that
exposure to unemployment risk tends to be very unevenly distributed.

18
In his review of the Cambridge controversy Mas-Colell (1989) concluded that: “What the
‘paradoxical’comparative statics [of disaggregate capital theory] has taught us is simply that
modelling the world as having a single capital good is not a priori justi…ed. So be it.”

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

38 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

2.4 The neoclassical competitive one-sector setup

Many long-run macromodels, including those in the …rst chapters to follow, share
the same abstract setup regarding the …rms and the market environment in which
they are placed. We give an account here which will serve as a reference point
for these later chapters.
The setup is characterized by the following simpli…cations:

(a) There is only one produced good, an all-purpose good that can be used for
consumption as well as investment. Physical capital is just the accumulated
amount of what is left of the produced good after consumption. Models
using this simpli…cation are called one-sector models. One may think of
“corn”, a good that can be used for consumption as well as investment in
the form of seed to yield corn next period.

(b) A representative …rm maximizes pro…t subject to a neoclassical production

function under non-increasing returns to scale.

(c) Capital goods become productive immediately upon purchase or renting (so
installation costs and similar features are ignored).

(d) In all markets perfect competition rules and so the economic actors are price
takers, perceiving no constraint on how much they can sell or buy at the
going market price. It is understood that market prices are ‡exible and
adjust quickly to levels required for market clearing.

(e) Factor supplies are inelastic.

(f) There is no uncertainty. When a choice of action is made, the consequences

are known.

We call such a setup the neoclassical competitive one-sector setup. In many

respects it is an abstraction. Nevertheless, the outcome under these conditions is
of theoretical interest. Think of Galilei’s discovery that a falling body falls with
a uniform acceleration as long as it is falling through a perfect vacuum.

2.4.1 Pro…t maximization

We consider a single period. Let the representative …rm have the neoclassical
production function
Y = F (K; L); (2.28)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.4. The neoclassical competitive one-sector setup 39

where technological change is ignored. Although in this book often CRS will be
assumed, we may throw the CRS outcome in relief by starting with a broader
view.
From microeconomics we know that equilibrium with perfect competition is
compatible with producers operating under the condition of locally nonincreasing
returns to scale (cf. Fig. 2.2). In standard macroeconomics it is common to
accept a lower level of generality and simply assume that F is a concave function.
This allows us to carry out the analysis as if there were non-increasing returns
to scale everywhere (see Appendix D).19
Since F is neoclassical, we have FKK < 0 and FLL < 0 everywhere. To
guarantee concavity it is then necessary and su¢ cient to add the assumption
that
D FKK (K; L)FLL (K; L) FKL (K; L)2 0; (2.29)
holds for all (K; L): This is a simple application of a general theorem on concave
functions (see Math Tools).
We consider both K and L as variable production factors. Let the factor
prices be denoted wK and wL ; respectively. For the time being we assume the
…rm rents the machines it uses; then the price, wK ; of capital services is called
the rental price or the rental rate. As numeraire (unit of account) we apply the
output good. So all prices are measured in terms of the output good which itself
has the price 1. Then pro…t, de…ned as revenue minus costs, is

= F (K; L) wK K wL L: (2.30)

We assume both production inputs are variable inputs. Taking the factor prices
as given from the factor markets, the …rm’s problem is to choose (K; L); where
K 0 and L 0, so as to maximize . An interior solution will satisfy the
…rst-order conditions
@
= FK (K; L) wK = 0 or FK (K; L) = wK ; (2.31)
@K
@
= FL (K; L) wL = 0 or FL (K; L) = wL : (2.32)
@L
Since F is concave, so is the pro…t function. The …rst-order conditions are then
su¢ cient for (K; L) to be a solution.
It is now convenient to proceed by considering the two cases, DRS and CRS,
separately.
19
By de…nition, concavity means that by applying a weighted average of two factor combina-
tions, (K1 ; L1 ) and (K2 ; L2 ); the obtained output is at least as large as the weighted average
of the original outputs, Y1 and Y2 . So, if 0 < < 1 and (K; L) = (K1 ; L1 ) +(1 )(K2 ; L2 ),
then F (K; L) F (K1 ; L1 ) +(1 )F (K2 ; L2 ).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

40 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

The DRS case

Suppose the production function satis…es (2.29) with strict inequality everywhere,
i.e.,
D > 0:
In combination with the neoclassical property of diminishing marginal productiv-
ities, this implies that F is strictly concave which in turn implies DRS everywhere.
The factor demands will now be unique. Indeed, the equations (2.31) and (2.32)
de…ne the factor demands K d and Ld (“d” for demand) as implicit functions of
the factor prices:

K d = K(wK ; wL ); Ld = L(wK ; wL ):

An easy way to …nd the partial derivatives of these functions is to …rst take the
di¤erential20 of both sides of (2.31) and (2.32), respectively:

FKK dK d + FKL dLd = dwK ;

FLK dK d + FLL dLd = dwL :

Then we interpret these conditions as a system of two linear equations with two
unknowns, the variables dK d and dLd : The determinant of the coe¢ cient matrix
equals D in (2.29) and is in this case positive everywhere. Using Cramer’s rule
(see Math Tools), we …nd

FLL dwK FKL dwL

dK d = ;
D
FKK dwL FLK dwK
dLd = ;
D
so that

@K d FLL @K d FKL
= < 0; = < 0 if FKL > 0; (2.33)
@wK D @wL D
@Ld FKL @Ld FKK
= < 0 if FKL > 0; = < 0; (2.34)
@wK D @wL D
20
The di¤ erential of a di¤erentiable function is a convenient tool for deriving results like
(2.33) and (2.34). For a function of one variable, y = f (x); the di¤erential is denoted dy (or df )
and is de…ned as f 0 (x)dx; where dx is some arbitrary real number (interpreted as the change in
x): For a di¤erentiable function of two variables, z = g(x; y) ; the di¤ erential of the function is
denoted dz (or dg) and is de…ned as dz = gx (x; y)dx +gy (x; y)dy; where dx and dy are arbitrary
real numbers.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.4. The neoclassical competitive one-sector setup 41

in view of FLK = FKL :21

In contrast to the cases of CRS and IRS, here we cannot be sure that direct
complementarity (FKL > 0) holds everywhere. In any event, the rule for both
factors is that when a factor price increases, the demand for the factor in question
decreases and under direct complementarity also the demand for the other factor
will decrease. Although there is a substitution e¤ect towards higher demand for
the factor whose price has not been increased, this is more than o¤set by the
negative output e¤ect, which is due to the higher marginal costs. This is an
implication of perfect competition. In a di¤erent market structure output may
be determined from the demand side (think of a Keynesian short-run model) and
then only the substitution e¤ect will be operative. An increase in one factor price
will then increase the demand for the other factor.

The CRS case

Under CRS, D in (2.29) takes the value

D=0

everywhere, as shown in Appendix B. Then the factor prices no longer determine

the factor demands uniquely. But the relative factor demand, k d K d =Ld ; is
determined uniquely by the relative factor price, wL =wK : Indeed, by (2.31) and
(2.32),
FL (K; L) f (k) f 0 (k)k wL
M RS = = 0
mrs(k) = ; (2.35)
FK (K; L) f (k) wK
where the second equality comes from (2.15) and (2.16). By straightforward
calculation,
f (k)f 00 (k) kf 00 (k)=f 0 (k)
mrs0 (k) = = > 0;
f 0 (k)2 (k)
where (k) kf 0 (k)=f (k) is the elasticity of f w.r.t. k and the numerator is the
elasticity of f 0 w.r.t. k: For instance, in the Cobb-Douglas case f (k) = Ak ; we
get mrs0 (k) = (1 )= : Given wL =wK , the last equation in (2.35) gives k d as
an implicit function k d = k(wL =wK ); where k 0 (wL =wK ) = 1=mrs0 (k) > 0: The
solution is illustrated in Fig. 2.4. Under CRS (indeed, for any homogeneous
neoclassical production function) the desired capital-labor ratio is an increasing
function of the inverse factor price ratio and independent of the output level.
21
Applying the full content of the implicit function theorem (see Math tools), one could
directly have written down the results (2.33) and (2.34) and would not need the procedure
outlined here, based on di¤erentials. On the other hand the present procedure is probably
more intuitive and easier to remember.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

42 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

Figure 2.4: Constancy of MRS along rays when the production function is homogeneous
of degree h (the cost-minimizing capital intensity is the same at all output levels).

To determine K d and Ld separately we need to know the level of output. And

here we run into the general problem of indeterminacy under perfect competition
combined with CRS. Saying that the output level is so as to maximize pro…t is
pointless. Well, if at the going factor prices attainable pro…t is negative, exit
from the market is pro…t maximizing (or rather loss minimizing), which amounts
to K d = Ld = 0. But if the pro…t is positive, there will be no upper bound to the
factor demands. Owing to CRS, doubling the factor inputs will double the pro…ts
of a price taking …rm. An equilibrium with positive production is only possible if
pro…t is zero. And then the …rm is indi¤erent w.r.t. the level of output. Solving
the indeterminacy problem requires a look at the factor markets.

2.4.2 Clearing in factor markets

Considering a closed economy, we denote the available supplies of physical capital
and labor K s and Ls ; respectively, and assume these supplies are inelastic. W.r.t.
capital this is a “natural” assumption since in a closed economy in the short
term the available amount of capital will be predetermined, that is, historically
determined by the accumulated previous investment in the economy. W.r.t. labor
supply it is just a simplifying assumption introduced because the question about
possible responses of labor supply to changes in factor prices is a secondary issue
in the present context.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.4. The neoclassical competitive one-sector setup 43

The factor markets clear when

K d = K s; (2.36)
Ld = Ls : (2.37)
Achieving this equilibrium (state of “rest”) requires that the factor prices adjust
to their equilibrium levels, which are
wK = FK (K s ; Ls ); (2.38)
wL = FL (K s ; Ls ); (2.39)
by (2.31) and (2.32). This says that in equilibrium the real factor prices are
determined by the marginal productivities of the respective factors at full utiliza-
tion of the given supplies. This holds under DRS as well as CRS. So, under
non-increasing returns to scale there is, at the macroeconomic level, a unique
equilibrium (wK ; wL ; K d ; Ld ) given by the above four equilibrium conditions for
the factor markets.22 It is an equilibrium in the sense that no agent has an
incentive to “deviate”.
As to comparative statics, since FKK < 0, a larger capital supply implies a
lower wK ; and since FLL < 0, a larger labor supply implies a lower wL .
The intuitive mechanism behind the attainment of equilibrium is that if, for
example, for a short moment wK < FK (K s ; Ls ); then K d > K s and so competi-
tion between the …rms will generate an upward pressure on wK until equality is
obtained. And if for a short moment wK > FK (K s ; Ls ); then K d < K s and so
competition between the suppliers of capital will generate a downward pressure
on wK until equality is obtained.
Looking more carefully at the matter, however, we see that this intuitive
reasoning …ts at most the DRS case. In the CRS case we have FK (K s ; Ls ) = f (k s );
where k s K s =Ls : Here we can only argue that for instance wK < FK (K s ; Ls )
implies k d > k s : And even if this leads to upward pressure on wK until k d = k s
is achieved, and even if both factor prices have obtained their equilibrium levels
given by (2.38) and (2.39), there is nothing to induce the representative …rm (or
the many …rms in the actual economy taken together) to choose the “right”input
levels so as to satisfy the clearing conditions (2.36) and (2.37). In this way the
indeterminacy under CRS pops up again, this time as a problem endangering
stability of the equilibrium.

Stability not guaranteed*

To substantiate the point that the indeterminacy under CRS may endanger sta-
bility of competitive equilibrium, let us consider a Walrasian tâtonnement ad-
22
At the microeconomic level, under CRS, industry structure remains indeterminate in that
…rms are indi¤erent as to their size.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

44 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

justment process.23 We imagine that our period is sub-divided into many short
time intervals (t; t + t): In the initial short time interval the factor markets
may not be in equilibrium. It is assumed that no capital or labor is hired out
of equilibrium. To allow an analysis in continuous time, we let t ! 0: A dot
over a variable denotes the time derivative, i.e., x(t)
_ = dx(t)=dt. The adjustment
process assumed is the following:
K_ d (t) = 1 FK (K d (t); Ld (t)) wK (t) ; 1 > 0;
L_ d (t) = 2
d d
FL (K (t); L (t)) wL (t) ; 2 > 0;
d s
w_ K (t) = K (t) K ;
w_ L (t) = Ld (t) Ls ;
where the initial values, K d (0); Ld (0); wK (0); and wL (0); are given. The parame-
ters 1 and 2 are constant adjustment speeds. The corresponding adjustment
speeds for the factor prices are set equal to one by choice of measurement units of
the inputs. Of course, the four endogenous variables should be constrained to be
nonnegative, but that is not important for the discussion here. The system has
a unique stationary state: K d (t) = K s ; Ld (t) = Ls ; wK (t) = KK (K s ; Ls ); wL (t)
= KL (K s ; Ls ):
A widespread belief, even in otherwise well-informed circles, seems to be that
with such adjustment dynamics, the stationary state is at least locally asymptot-
ically stable. By this is meant that there exists a (possibly only small) neigh-
borhood, N , of the stationary state with the property that if the initial state,
(K d (0); Ld (0); wK (0); wL (0)); belongs to N ; then the solution (K d (t); Ld (t);
wK (t); wL (t)) converges to the stationary state for t ! 1?
Unfortunately, however, this stability property is not guaranteed. To bear
1 1
this out, it is enough to present a counterexample. Let F (K; L) = K 2 L 2 ; 1
= 2 = K s = Ls = 1; and suppose K d (0) = Ld (0) > 0 and wK (0) = wL (0) > 0:
All this symmetry implies that K d (t) = Ld (t) = x(t) > 0 and wK (t) = wL (t)
= w(t) for all t 0: So FK (K d (t); Ld (t)) = 0:5x(t) 0:5 x(t)0:5 = 0:5; and similarly
FL (K d (t); Ld (t)) = 0:5 for all t 0: Now the system is equivalent to the two-
dimensional system,
x(t)
_ = 0:5 w(t); (2.40)
w(t)
_ = x(t) 1: (2.41)
Using the theory of coupled linear di¤erential equations, the solution is24
x(t) = 1 + (x(0) 1) cos t (w(0) 0:5) sin t; (2.42)
w(t) = 0:5 + (w(0) 0:5) cos t + (x(0) 1) sin t: (2.43)
23
Tâtonnement is a French word meaning “groping”.
24
For details, see hints in Exercise 2.6.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.4. The neoclassical competitive one-sector setup 45

The solution exhibits undamped oscillations and never settles down at the sta-
tionary state, (1; 0:5); if not being there from the beginning. In fact, the solution
curves in the (x; w) plane will be circles around
p the stationary state. This is
so whatever the size of the initial distance, (x(0) 1)2 + (w(0) 0:5)2 ; to the
stationary point.
The economic mechanism is as follows. Suppose for instance that x(0) < 1
and w(0) < 0:5: Then to begin with there is excess supply and so w will be falling
while, with w below marginal products, x will be increasing. When x reaches its
potential equilibrium value, 1, w is at its trough and so induces further increases
in the factor demands, thus bringing about a phase where x > 1: This excess
demand causes w to begin an upturn. When w reaches its potential equilibrium
value, 0.5, however, excess demand, x 1; is at its peak and this induces further
increases in factor prices, w: This brings about a phase where w > 0:5 so that
factor prices exceed marginal products, which leads to declining factor demands.
But as x comes back to its potential equilibrium value, w is at its peak and drives
x further down. Thus excess supply arises which in turn triggers a downturn of w:
This continues in never ending oscillations where the overreaction of one variable
carries the seed to an overreaction of the other variable soon after and so on.
This possible outcome underlines that the theoretical existence of equilibrium
is one thing and stability of the equilibrium is another. In particular under CRS,
where demand functions for inputs are absent, the issue of stability can be more
intricate than one might at …rst glance think.

The link between capital costs and the interest rate*

Returning to the description of equilibrium, we shall comment on the relationship
between the factor price wK and the more everyday concept of an interest rate.
The factor price wK is the cost per unit of capital service. It has di¤erent names
in the literature such as the rental price, the rental rate, the unit capital cost, or
the user cost. It is related to the interest and depreciation costs that the owner of
the capital good in question defrays. In the simple neoclassical setup considered
here, it does not matter whether the …rm rents the capital it uses or owns it;
in the latter case, wK ; is the imputed capital cost, i.e., the forgone interest plus
depreciation.
As to depreciation it is common in simple macroeconomics to apply the ap-
proximation that, due to wear and tear, a constant fraction (where 0 1)
of a given capital stock evaporates per period. If for instance the period length
is one year and = 0:1; this means that a given machine in the next year has
only the fraction 0.9 of its productive capacity in the current year. Otherwise the
productive characteristics of a capital good are assumed to be the same whatever
its time of birth. Sometimes is referred to as the rate of physical capital depre-

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

46 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

ciation or the deterioration rate. When changes in relative prices can occur, this
must be distinguished from the economic depreciation of capital which refers to
the loss in economic value of a machine after one year.
Let pt 1 be the price of a certain type of machine bought at the end of period
t 1: Let prices be expressed in the same numeraire as that in which the interest
rate, r; is measured. And let pt be the price of the same type of machine one
period later. Then the economic depreciation in period t is

pt 1 (1 )pt = pt (pt pt 1 ):

The economic depreciation thus equals the value of the physical wear and tear
minus the capital gain (positive or negative) on the machine.
By holding the machine the owner faces an opportunity cost, namely the
forgone interest on the value pt 1 placed in the machine during period t: If rt is
the interest rate on a loan from the end of period t 1 to the end of period t; this
interest cost is rt pt 1 : The bene…t of holding the (new) machine is that it can be
rented out to the representative …rm and provide the return wKt at the end of
the period. Since there is no uncertainty, in equilibrium we must then have wKt
= rt pt 1 + pt (pt pt 1 ); or

wKt p t + pt pt 1
= rt : (2.44)
pt 1

This is a no-arbitrage condition saying that the rate of return on holding the
machine equals the rate of return obtainable in the loan market (no pro…table
arbitrage opportunities are available).25
In the simple setup considered so far, the capital good and the produced good
are physically identical and thus have the same price. As the produced good
is our numeraire, we have pt 1 = pt = 1: This has two implications. First, the
interest rate, rt ; is a real interest rate so that 1 + rt measures the rate at which
future units of output can be traded for current units of output. Second, (2.44)
simpli…es to
wKt = rt :
Combining this with equation (2.38), we see that in the simple neoclassical setup
the equilibrium real interest rate is determined as

rt = FK (Kts ; Lst ) ; (2.45)

25
In continuous time analysis the rental rate, the interest rate, and the price of the machine
are considered as di¤erentiable functions of time, wK (t); r(t); and p(t), respectively. In analogy
with (2.44) we then get wK (t) = (r(t) + )p(t) p(t);
_ where p(t)
_ denotes the time derivative of
the price p(t).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.5. More complex model structures* 47

where KtS and Lst are predetermined. Under CRS this takes the form rt = f 0 (kts )

; where kts Kts =Lst :

We have assumed that the …rms rent capital goods from their owners, presum-
ably the households. But as long as there is no uncertainty, no capital adjustment
costs, and no taxation, it will have no consequences for the results if instead we
assume that the …rms own the physical capital they use and …nance capital invest-
ment by issuing bonds or shares. Then such bonds and shares would constitute
…nancial assets, owned by the households and o¤ering a rate of return rt as given
by (2.45).

2.5 More complex model structures*

The neoclassical setup described above may be useful as a …rst way of organizing
one’s thoughts about the production side of the economy. To come closer to
a model of how modern economies function, however, many modi…cations and
extensions are needed.

2.5.1 Convex capital installation costs

In the real world the capital goods used by a production …rm are usually owned
by the …rm itself rather than rented for single periods on rental markets. This is
because inside the speci…c plant in which these capital goods are an integrated
part, they are generally worth much more than outside. So in practice …rms ac-
quire and install …xed capital equipment with a view on maximizing discounted
expected pro…ts in the future. The cost associated with this …xed capital in-
vestment not only includes the purchase price of new equipment, but also the
installation costs (the costs of setting up the new …xed equipment in the …rm and
the associated costs of reorganizing work processes).
Assuming the installation costs are strictly convex in the level of investment,
the …rm has to solve an intertemporal optimization problem. Forward-looking
expectations thus become important and this has implications for how equilib-
rium in the output market is established and how the equilibrium interest rate is
determined. Indeed, in the simple neoclassical setup above, the interest rate equi-
librates the market for capital services. The value of the interest rate is simply
tied down by the equilibrium condition (2.39) in this market and what happens
in the output market is a trivial consequence of this. But with convex capital
installation costs the …rm’s capital stock is given in the short run and the interest
rate(s) become(s) determined elsewhere in the model, as we shall see in chapters
14 and 15.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

48 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

2.5.2 Long-run vs. short-run production functions

In the discussion of production functions up to now we have been silent about the
distinction between “ex ante”and “ex post”substitutability between capital and
labor. By ex ante is meant “when plant and machinery are to be decided upon”
and by ex post is meant “after the equipment is designed and constructed”. In the
standard neoclassical competitive setup like in (2.35) there is a presumption that
also after the construction and installation of the equipment in the …rm, the ratio
of the factor inputs can be fully adjusted to a change in the relative factor price.
In practice, however, when some machinery has been constructed and installed,
its functioning will often require a more or less …xed number of machine operators.
What can be varied is just the degree of utilization of the machinery. That is,
after construction and installation of the machinery, the choice opportunities are
no longer described by the neoclassical production function but by a Leontief
production function,

Y = min(AuK; BL); A > 0; B > 0; (2.46)

where K is the size of the installed machinery (a …xed factor in the short run)
measured in e¢ ciency units, u is its utilization rate (0 u 1); and A and B
are given technical coe¢ cients measuring e¢ ciency (cf. Section 2.1.2).
So in the short run the choice variables are u and L: In fact, essentially only
u is a choice variable since e¢ cient production trivially requires L = AuK=B:
Under “full capacity utilization” we have u = 1 (each machine is used 24 hours
per day seven days per week). “Capacity” is given as AK per week. Producing
e¢ ciently at capacity requires L = AK=B and the marginal product by increasing
labor input is here nil. But if demand, Y d ; is less than capacity, satisfying this
demand e¢ ciently requires L = Y d =B and u = BL=(AK) < 1: As long as u < 1;
the marginal productivity of labor is a constant, B:
The various e¢ cient input proportions that are possible ex ante may be ap-
proximately described by a neoclassical CRS production function. Let this func-
tion on intensive form be denoted y = f (k): When investment is decided upon
and undertaken, there is thus a choice between alternative e¢ cient pairs of the
technical coe¢ cients A and B in (2.46). These pairs satisfy

f (k) = Ak = B: (2.47)

So, for an increasing sequence of k’s, k1 ; k2 ;. . . , ki ;. . . , the corresponding pairs are

(Ai ; Bi ) = (f (ki )=ki ; f (ki )); i = 1; 2;. . . .26 We say that ex ante, depending on the
relative factor prices as they are “now”and are expected to evolve in the future,
26
The points P and Q in the right-hand panel of Fig. 2.3 can be interpreted as constructed
this way from the neoclassical production function in the left-hand panel of the …gure.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.5. More complex model structures* 49

a suitable technique, (Ai ; Bi ); is chosen from an opportunity set described by the

given neoclassical production function. But ex post, i.e., when the equipment
corresponding to this technique is installed, the production opportunities are
described by a Leontief production function with (A; B) = (Ai ; Bi ):
In the picturesque language of Phelps (1963), technology is in this case putty-
clay. Ex ante the technology involves capital which is “putty” in the sense of
being in a malleable state which can be transformed into a range of various
machinery requiring capital-labor ratios of di¤erent magnitude. But once the
machinery is constructed, it enters a “hardened”state and becomes ”clay”. Then
factor substitution is no longer possible; the capital-labor ratio at full capacity
utilization is …xed at the level k = Bi =Ai ; as in (2.46). Following the terminology
of Johansen (1972), we say that a putty-clay technology involves a “long-run
production function”which is neoclassical and a “short-run production function”
which is Leontief.

Table 1. Technologies classi…ed according to

factor substitutability ex ante and ex post.
Ex post substitution
Ex ante substitution possible impossible
possible putty-putty putty-clay
impossible clay-clay

In contrast, the standard neoclassical setup assumes the same range of sub-
stitutability between capital and labor ex ante and ex post. Then the technology
is called putty-putty. This term may also be used if ex post there is at least some
substitutability although less than ex ante. At the opposite pole of putty-putty
we may consider a technology which is clay-clay. Here neither ex ante nor ex post
is factor substitution possible. Table 1 gives an overview of the alternative cases.
The putty-clay case is generally considered the realistic case. As time pro-
ceeds, technological progress occurs. To take this into account, we may replace
(2.47) and (2.46) by f (kt ; t) = At kt = Bt and Yt = min(At ut Kt ; Bt Lt ); respec-
tively. If a new pair of Leontief coe¢ cients, (At2 ; Bt2 ); e¢ ciency-dominates its
predecessor (by satisfying At2 At1 and Bt2 Bt1 with at least one strict equal-
ity), it may pay the …rm to invest in the new technology at the same time as
some old machinery is scrapped. Real wages tend to rise along with technolog-
ical progress and the scrapping occurs because the revenue from using the old
machinery in production no longer covers the associated labor costs.
The clay property ex-post of many technologies is important for short-run
analysis. It implies that there may be non-decreasing marginal productivity of

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

50 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

labor up to a certain point. It also implies that in its investment decision the
…rm will have to take expected future technologies and future factor prices into
account. For many issues in long-run analysis the clay property ex-post may be
less important, since over time adjustment takes place through new investment.

2.5.3 A simple portrayal of price-making …rms

Another modi…cation which is important in short- and medium-run analysis,
relates to the assumed market forms. Perfect competition is not a good approx-
imation to market conditions in manufacturing and service industries. To bring
perfect competition in the output market in perspective, we give here a brief re-
view of …rms’behavior under a form of monopolistic competition that is applied
in many short-run models.
Suppose there is a large number of di¤erentiated goods, i = 1; 2; : : : ; n; each
produced by a separate …rm. In the short run n is given. Each …rm has monopoly
on its own good (supported, say, by a trade mark, patent protection, or simply
secrecy regarding the production recipe). The goods are imperfect substitutes to
each other and so indirect competition prevails. Each …rm is small in relation to
the “sum”of competing …rms and perceives that these other …rms do not respond
to its actions.
In the given period let …rm i face a given downward-sloping demand curve for
its product,
"
Pi Y
Yi D(Pi ); " > 1: (2.48)
P n
Here Yi is the produced quantity and the expression on the right-hand side of the
inequality is the demand as a function of the price Pi chosen by the …rm.27 The
“general price level”P (a kind of average across the di¤erent goods, cf. Chapter
22) and the “general demand level”, given by the index Y , matter for the position
of the demand curve in the (Yi ; Pi ) plan, cf. Fig. 2.5. The price elasticity
of demand, "; is assumed constant and higher than one (otherwise there is no
solution to the monopolist’s decision problem). Variables that the monopolist
perceives as exogenous are implicit in the demand function symbol D: We imagine
prices are expressed in terms of money (so they are “nominal” prices, hence
denoted by capital letters whereas we generally use small letters for “real”prices).
For simplicity, factor markets are still assumed competitive. Given the nomi-
nal factor prices, WK and WL ; …rm i wants to maximize its pro…t

i = Pi Y i WK Ki WL Li ;
27
We ignore production for inventory holding.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.5. More complex model structures* 51

subject to (2.48) and the neoclassical production function Yi = F (Ki ; Li ): For the
purpose of simple comparison with the case of perfect competition as described
in Section 2.4, we return to the case where both labor and capital are variable
inputs in the short run.28 It is no serious restriction on the problem to assume
the monopolist will want to produce the amount demanded so that Yi = D(Pi ):
It is convenient to solve the problem in two steps.

Figure 2.5: Determination of the monopolist price and output.

Step 1. Imagine the monopolist has already chosen the output level Yi : Then
the problem is to minimize cost:
min WK Ki + WL Li s.t. F (Ki ; Li ) = Yi :
Ki ;Li

An interior solution (Ki ; Li ) will satisfy the …rst-order conditions

FK (Ki ; Li ) = WK ; FL (Ki ; Li ) = WL ; (2.49)
where is the Lagrange multiplier. Since F is neoclassical and thereby strictly
quasiconcave, the …rst-order conditions are not only necessary but also su¢ cient
for (Ki ; Li ) to be a solution, and (Ki ; Li ) will be unique so that we can write
these conditional factor demands as functions, Kid = K(WK ; WL ; Yi ) and Ldi =
L(WK ; WL ; Yi ): This gives rise to the cost function C(Yi ) = WK K(WK ; WL ; Yi )
+WL L(WK ; WL ; Yi ):
Step 2. Solve
max (Yi ) = R(Yi ) C(Yi ) = P(Yi )Yi C(Yi ):
Yi

28
Generally, the technology would di¤er across the di¤erent product lines and F should thus
be replaced by F i , but for notational convenience we ignore this.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

52 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

We have here introduced “total revenue” R(Yi ) = P(Yi )Yi , where P(Yi ) is the
inverse demand function de…ned by P(Yi ) D 1 (Yi ) = [Yi =(Y =n)] 1=" P from
(2.48). The …rst-order condition is

R0 (Yi ) = P(Yi ) + P 0 (Yi )Yi = C 0 (Yi ); (2.50)

where the left-hand side is marginal revenue and the right-hand side is marginal
cost.
A su¢ cient second-order condition is that 00 (Yi ) = R00 (Yi ) C 00 (Yi ) < 0; i.e.,
the marginal revenue curve crosses the marginal cost curve from above. In the
present case this is surely satis…ed if we assume C 00 (Yi ) 0; which also ensures
existence and uniqueness of a solution to (2.50). Substituting this solution, which
we denote Yis ; cf. Fig. 2.5, into the conditional factor demand functions from
Step 1, we …nd the factor demands, Kid and Ldi : Owing to the downward-sloping
demand curves the factor demands are unique whether the technology exhibits
DRS, CRS, or IRS. Thus, contrary to the perfect competition case, neither CRS
nor IRS pose particular problems.
From the de…nition R(Yi ) = P (Yi )Yi follows

Yi 0 1 " 1
R0 (Yi ) = Pi 1 + P (Yi ) = Pi 1 = Pi :
Pi " "

So the pricing rule is Pi = (1 + )C 0 (Yi ); where Yi is the pro…t maximizing output

level and "=(" 1) 1 > 0 is the mark-up on marginal cost. An analytical
very convenient feature is that the markup is thus a constant.
In parallel with (2.31) and (2.32) the solution to …rm i’s decision problem is
characterized by the marginal revenue productivity conditions

R0 (Yis )FK (Kid ; Ldi ) = WK ; (2.51)

R0 (Yis )FL (Kid ; Ldi ) = WL ; (2.52)

where Yis = F (Kid ; Ldi ): These conditions follow from (2.49), since the Lagrange
multiplier equals marginal cost (see Appendix A), which equals marginal revenue.
That is, at pro…t maximum the marginal revenue products of capital and labor,
respectively, equal the corresponding factor prices. Since Pi > R0 (Yis ); the factor
prices are below the value of the marginal productivities. This re‡ects the market
power of the …rms.
In macro models a lot of symmetry is often assumed. If there is complete
symmetry across product lines and if factor markets clear as in (2.36) and (2.37)
with inelastic factor supplies, K s and Ls ; then Kid = K s =n and Ldi = Ls =n:
Furthermore, all …rms will choose the same price so that Pi = P; i = 1; 2; : : : ; n:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.5. More complex model structures* 53

Then the given factor supplies, together with (2.51) and (2.52), determine the
equilibrium real factor prices:

WK 1 K s Ls
wK = FK ( ; );
P 1+ n n
WL 1 K s Ls
wL = FL ( ; );
P 1+ n n

where we have used that R0 (Yis ) = P=(1+ ) under these circumstances. As under
perfect competition, the real factor prices are proportional to the corresponding
marginal productivities, although with a factor of proportionality less than one,
namely equal to the inverse of the markup. This observation is sometimes used
as a defence for applying the simpler perfect-competition framework for studying
certain long-run aspects of the economy. For these aspects, the size of the pro-
portionality factor may be immaterial, at least as long as it is relatively constant
over time. Indeed, the constant markups open up for a simple transformation of
many of the perfect competition results to monopolistic competition results by
inserting the markup factor 1 + the relevant places in the formulas.
If in the short term only labor is a variable production factor, then (2.51)
need not hold. As claimed by Keynesian and New Keynesian thinking, also the
prices chosen by the …rms may be more or less …xed in the short run because
the …rms face price adjustment costs (“menu costs”) and are reluctant to change
prices too often, at least vis-a-vis changes in demand. Then in the short run only
the produced quantity will adjust to changes in demand. As long as the output
level is within the range where marginal cost is below the price, such adjustments
are still bene…cial to the …rm. As a result, even (2.52) may at most hold “on
average” over the business cycle. These matters are dealt with in Part V of this
book.
In practice, market power and other market imperfections also play a role in
the factor markets, implying that further complicating elements enter the pic-
ture. One of the tasks of theoretical and empirical macroeconomics is to clarify
the aggregate implications of market imperfections and sort out which market
imperfections are quantitatively important in di¤erent contexts.

2.5.4 The …nancing of …rms’operations

We have so far talked about aspects related to production and pricing. What
about the …nancing of a …rm’s operations? To acquire not only its …xed capital
(structures and machines) but also its raw material and other intermediate inputs,
a …rm needs funds (there are expenses before the proceeds from sale arrive). These
funds ultimately come from the accumulated saving of households. In long-run

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

54 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

macromodels to be considered in the next chapters, uncertainty as well as non-

neutrality of corporate taxation are ignored; in that context the capital structure
(the debt-equity ratio) of …rms is indeterminate and irrelevant for production
outcomes.29 In those chapters we shall therefore concentrate on the latter. Later
chapters, dealing with short- and medium-run issues, touch upon cases where
capital structure and bankruptcy risk matter and …nancial intermediaries enter
the scene.

2.6 Literature notes

As to the question of the empirical validity of the constant returns to scale as-
sumption, Malinvaud (1998) o¤ers an account of the econometric di¢ culties as-
sociated with estimating production functions. Studies by Basu (1996) and Basu
and Fernald (1997) suggest returns to scale are about constant or decreasing.
Studies by Hall (1990), Caballero and Lyons (1992), Harris and Lau (1992),
Antweiler and Tre- er (2002), and Harrison (2003) suggest there are quantita-
tively signi…cant increasing returns, either internal or external. On this back-
ground it is not surprising that the case of IRS (at least at industry level), to-
gether with market forms di¤erent from perfect competition, has in recent years
received more attention in macroeconomics and in the theory of economic growth.
Macroeconomists’use of the value-laden term “technological progress”in con-
nection with technological change may seem suspect. But the term should be
interpreted as merely a label for certain types of shifts of isoquants in an abstract
universe. At a more concrete and disaggregate level analysts of course make use
of more re…ned notions about technological change, recognizing not only bene…ts
of new technologies, but for instance also the risks, including risk of fundamental
mistakes (think of the introduction and later abandonment of asbestos in the
construction industry). For history of technology see, e.g., Ruttan (2001) and
Smil (2003).
When referring to a Cobb-Douglas (or CES) production function some au-
thors implicitly assume that the partial output elasticities w.r.t. inputs time-
independent and thereby independent of technological change. For the case where
the inputs in question are renewable and nonrenewable natural resources, Growiec
and Schumacher (2008) study cases of time-dependency of the partial output elas-
ticities.
When technical change is not “neutral”in one of the senses described, it may
be systematically “biased” in alternative “directions”. The reader is referred to
the specialized literature on economic growth, cf. literature notes to Chapter 1.
29
In chapter 14 we return to this irrelevance proposition, called the Modigliani-Miller theorem.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.6. Literature notes 55

Embodied technological progress, sometimes called investment-speci…c tech-

nological progress, is explored in, for instance, Solow (1960), Greenwood et al.
(1997), and Groth and Wendner (2014).
Time series for di¤erent countries’ aggregate and to some extent sectorial
capital stocks are available from Penn World Table, ..., EU KLEMS, ...., and the
AMECO database, .
The concept of Gorman preferences and conditions ensuring that a represen-
tative household is admitted are surveyed in Acemoglu (2009). Another source,
also concerning the conditions for the representative …rm to be a meaningful no-
tion, is Mas-Colell et al. (1995). For general discussions of the limitations of
representative agent approaches, see Kirman (1992) and Gallegati and Kirman
(1999). Reviews of the “Cambridge Controversy” are contained in Mas-Colell
(1989) and Felipe and Fisher (2003). The last-mentioned authors …nd the condi-
tions required for the well-behavedness of these constructs so stringent that it is
di¢ cult to believe that actual economies are in any sense close to satisfy them.
For less distrustful views and constructive approaches to the issues, see for in-
stance Johansen (1972), Malinvaud (1998), Jorgenson et al. (2005), and Jones
(2005).
Scarf (1960) provided a series of examples of lack of dynamic stability of an
equilibrium price vector in an exchange economy. Mas-Colell et al. (1995) survey
the later theoretical development in this …eld.
The counterexample to guaranteed stability of the neoclassical factor market
equilibrium presented towards the end of Section 2.4 is taken from Bliss (1975),
where further perspectives are discussed. It may be argued that this kind of
stability questions should be studied on the basis of adjustment processes of a
less mechanical nature than a Walrasian tâtonnement process. The view would be
that trade out of equilibrium should be incorporated in the analysis and agents’
behavior out of equilibrium should be founded on some kind of optimization
or “satis…cing”, incorporating adjustment costs and imperfect information. The
…eld is complicated and the theory not settled. Yet it seems fair to say that the
studies of adjustment processes out of equilibrium indicate that the equilibrating
force of Adam Smith’s invisible hand is not without its limits. See Fisher (1983),
Osborne and Rubinstein (1990), and Negishi (2008) for reviews and elaborate
discussion of these issues.
We introduced the assumption that physical capital depreciation can be de-
scribed as geometric (in continuous time exponential) evaporation of the capital
stock. This formula is popular in macroeconomics, more so because of its simplic-
ity than its realism. An introduction to more general approaches to depreciation
is contained in, e.g., Nickell (1978).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

56 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

2.7 Appendix
A. Strict quasiconcavity
Consider a function f : A ! R, where A is a convex set, A Rn .30 Given a
real number a; if f (x) = a, the upper contour set is de…ned as fx 2 Aj f (x) ag
(the set of input bundles that can produce at least the amount a of output). The
function f (x) is called quasiconcave if its upper contour sets, for any constant
a; are convex sets. If all these sets are strictly convex, f (x) is called strictly
quasiconcave.

Average and marginal costs To show that (2.14) holds with n production
inputs, n = 1; 2;. . . , we derive the cost function of a …rm with a neoclassical
production function, Y = F (X1 ; X2 ; : : : ; Xn ): Given a vector of strictly positive
input prices w = (w1 ; : : : ; wn ) >> 0; the …rm faces the problem of …nding a cost-
minimizing way to produce a given positive output level Y within the range of
F: The problem is
Xn
min wi Xi s.t. F (X1 ; : : : ; Xn ) = Y and Xi 0; i = 1; 2; : : : ; n:
i=1

An interior solution, X = (X1 ; : : : ; Xn ); to this problem satis…es the …rst-order

conditions Fi0 (X ) = wi , where is the Lagrange multiplier, i = 1; : : : ; n:31 Since
F is neoclassical and thereby strictly quasiconcave in the interior of Rn+ , the …rst-
order conditions are not only necessary but also su¢ cient for the vector X to be
a solution, and X will be unique32 so that we can write it as a function, P X (Y ) =
(X1 (Y ); : : : ; Xn (Y )). This gives rise to the cost function C(Y ) = ni=1 wi Xi (Y ).
So average cost is C(Y )=Y : We …nd marginal cost to be
X n X n
0 0
C (Y ) = wi Xi (Y ) = Fi0 (X )Xi 0 (Y ) = ;
i=1 i=1

where the third equality comes from the …rst-order conditions, and the last equal-
ity is due to the constraint
Pn F0 (X (Y )) = Y ; which, by taking the total derivative
0
on both sides, gives i=1 Fi (X )Xi (Y ) = 1: Consequently, the ratio of average
to marginal costs is
Pn Pn 0
C(Y )=Y i=1 wi Xi (Y ) i=1 Fi (X )Xi (Y )
= = ;
C 0 (Y ) Y F (X )
30
Recall that a set S is said to be convex if x; y 2 S and 2 [0; 1] implies x + (1 )y 2 S:
31
Since in this section we use a bit of vector notation, we exceptionally mark …rst-order partial
derivatives by a prime in order to clearly distinguish from the elements of a vector (so we write
Fi0 instead of our usual Fi ):
32
See Sydsaeter et al. (2008), pp. 74, 75, and 125.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.7. Appendix 57

which in analogy with (2.13) is the elasticity of scale at the point X : This proves
(2.14).

Su¢ cient conditions for strict quasiconcavity The claim (iii) in Section
2.1.3 was that a continuously di¤erentiable two-factor production function F (K; L)
with CRS, satisfying FK > 0; FL > 0; and FKK < 0; FLL < 0; will automatically
also be strictly quasi-concave in the interior of R2 and thus neoclassical.
To prove this, consider a function of two variables, z = f (x, y); that is twice
continuously di¤erentiable with f1 @z=@x > 0 and f2 @z=@y > 0; everywhere.
Then the equation f (x, y) = a; where a is a constant, de…nes an isoquant,
y = g(x); with slope g 0 (x) = f1 (x; y)=f2 (x; y): Substitute g(x) for y in this
equation and take the derivative w.r.t. x. By straightforward calculation we …nd

f12 f22 2f1 f2 f21 + f22 f11

g 00 (x) = (2.53)
f23

If the numerator is negative, then g 00 (x) > 0; that is, the isoquant is strictly
convex to the origin. And if this holds for all (x, y), then f is strictly quasi-
concave in the interior of R2 . A su¢ cient condition for a negative numerator is
that f11 < 0, f22 < 0 and f21 0. All these conditions, including the last three
are satis…ed by the given function F: Indeed, FK ; FL , FKK ; and FLL have the
required signs. And when F has CRS, F is homogeneous of degree 1 and thereby
FKL > 0; see Appendix B. Hereby claim (iii) in Section 2.1.3 is proved.

B. Homogeneous production functions

The claim (iv) in Section 2.1.3 was that a two-factor production function with
CRS, satisfying FK > 0; FL > 0; and FKK < 0; FLL < 0; has always FKL > 0;
i.e., there is direct complementarity between K and L: This assertion is implied
by the following observations on homogeneous functions.
Let Y = F (K, L) be a twice continuously di¤erentiable production function
with FK > 0 and FL > 0 everywhere. Assume F is homogeneous of degree h > 0;
that is, for all possible (K; L) and all > 0; F ( K; L) = h F (K; L). According
to Euler’s theorem (see Math Tools) we then have:
CLAIM 1 For all (K, L); where K > 0 and L > 0;

KFK (K; L) + LFL (K; L) = hF (K; L): (2.54)

Euler’s theorem also implies the inverse:

CLAIM 2 If (2.54) is satis…ed for all (K, L), where K > 0 and L > 0; then
F (K; L) is homogeneous of degree h:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

58 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

Partial di¤erentiation w.r.t. K and L; respectively, gives, after ordering,

KFKK + LFLK = (h 1)FK (2.55)

KFKL + LFLL = (h 1)FL : (2.56)

In (2.55) we can substitute FLK = FKL (by Young’s theorem). In view of Claim
2 this shows:
CLAIM 3 The marginal products, FK and FL , considered as functions of K and
L, are homogeneous of degree h 1.
We see also that when h 1 and K and L are positive; then

FKK < 0 implies FKL > 0; (2.57)

FLL < 0 implies FKL > 0: (2.58)

For h = 1 this establishes the direct complementarity result, (iv) in Section 2.1.3,
to be proved. A by-product of the derivation is that also when a neoclassical
production function is homogeneous of degree h > 1 (which implies IRS), does
direct complementarity between K and L hold.
Remark. The terminology around complementarity and substitutability may eas-
ily lead to confusion. In spite of K and L exhibiting direct complementarity when
FKL > 0; K and L are still substitutes in the sense that cost minimization for a
given output level implies that a rise in the price of one factor results in higher
demand for the other factor.
The claim (v) in Section 2.1.3 was the following. Suppose we face a CRS
production function, Y = F (K; L); that has positive marginal products, FK and
FL ; everywhere and isoquants, K = g(L); satisfying the condition g 00 (L) > 0
everywhere (i.e., F is strictly quasi-concave). Then the partial second derivatives
must satisfy the neoclassical conditions:

FKK < 0; FLL < 0: (2.59)

The proof is as follows. The …rst inequality in (2.59) follows from (2.53) combined
with (2.55). Indeed, for h = 1; (2.55) and (2.56) imply FKK = FLK L=K
= FKL L=K and FKL = FLL L=K, i.e., FKK = FLL (L=K)2 (or, in the notation
of Appendix A, f22 = f11 (x=y)2 ), which combined with (2.53) gives the conclusion
FKK < 0, when g 00 > 0. The second inequality in (2.59) can be veri…ed in a similar
way.
Note also that for h = 1 the equations (2.55) and (2.56) entail

KFKK = LFLK and KFKL = LFLL ; (2.60)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.7. Appendix 59

respectively. By dividing the left- and right-hand sides of the …rst of these equa-
tions with those of the second we conclude that FKK FLL = FKL 2
in the CRS case.
We see also from (2.60) that, under CRS, the implications in (2.57) and (2.58)
can be turned round.
Finally, we asserted in § 2.1.1 that when the neoclassical production function
Y = F (K, L) is homogeneous of degree h, then the marginal rate of substitution
between the production factors depends only on the factor proportion k K=L:
Indeed,

FL (K; L) Lh 1 FL (k; 1) FL (k; 1)

M RSKL (K; L) = = h 1 = mrs(k); (2.61)
FK (K; L) L FK (k; 1) FK (k; 1)

where k K=L: The result (2.61) follows even if we only assume F (K; L) is
homothetic. When F (K; L) is homothetic, by de…nition we can write F (K, L)
'(G(K; L)), where G is homogeneous of degree 1 and ' is an increasing function.
In view of this, we get

'0 GL (K; L) GL (k; 1)

M RSKL (K; L) = 0
= ;
' GK (K; L) GK (k; 1)

where the last equality is implied by Claim 3 for h = 1:

C. The Inada conditions combined with CRS

We consider a neoclassical production function, Y = F (K; L); exhibiting CRS.
De…ning k K=L; we can then write Y = LF (k; 1) Lf (k); where f (0)
0; f 0 > 0; and f 00 < 0:

Essential inputs In Section 2.1.2 we claimed that the upper Inada condition
for M P L together with CRS implies that without capital there will be no output:

F (0; L) = 0 for any L > 0:

In other words: in this case capital is an essential input. To prove this claim, let
K > 0 be …xed and let L ! 1: Then k ! 0; implying, by (2.16) and (2.18),
that FL (K; L) = f (k) f 0 (k)k ! f (0): But from the upper Inada condition for
M P L we also have that L ! 1 implies FL (K; L) ! 0: It follows that

the upper Inada condition for M P L implies f (0) = 0: (2.62)

Since under CRS, for any L > 0; F (0; L) = LF (0; 1) Lf (0); we have hereby
shown our claim.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

60 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

Similarly, we can show that the upper Inada condition for M P K together
with CRS implies that labor is an essential input. Consider the output-capital
ratio x Y =K: When F has CRS, we get x = F (1; `) g(`); where ` L=K;
g 0 > 0; and g 00 < 0: Thus, by symmetry with the previous argument, we …nd that
under CRS, the upper Inada condition for M P K implies g(0) = 0: Since under
CRS F (K; 0) = KF (1; 0) Kg(0); we conclude that the upper Inada condition
for M P K together with CRS implies

F (K; 0) = 0 for any K > 0;

that is, without labor, no output.

Su¢ cient conditions for output going to in…nity when either input goes
to in…nity Here our …rst claim is that when F exhibits CRS and satis…es the
upper Inada condition for M P L and the lower Inada condition for M P K, then

lim F (K; L) = 1 for any K > 0:

L!1

To prove this, note that Y can be written Y = Kf (k)=k; since K=k = L: Here,

lim f (k) = f (0) = 0;

k!0

by continuity and (2.62), presupposing the upper Inada condition for M P L.

Thus, for any given K > 0;

f (k) f (k) f (0)

lim F (K; L) = K lim = K lim = K lim f 0 (k) = 1;
L!1 L!1 k k!0 k k!0

by the lower Inada condition for M P K. This veri…es the claim.

Our second claim is symmetric with this and says: when F exhibits CRS and
satis…es the upper Inada condition for M P K and the lower Inada condition for
M P L, then
lim F (K; L) = 1 for any L > 0:
K!1

The proof is analogue. So, in combination, the four Inada conditions imply, under
CRS, that output has no upper bound when either input goes to in…nity.

D. Concave neoclassical production functions

Two claims made in Section 2.4 are proved here.
CLAIM 1 When a neoclassical production function F (K; L) is concave, it has
non-increasing returns to scale everywhere.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

2.8. Exercises 61

Proof. We consider a concave neoclassical production function, F . Let x =

(x1 ; x2 ) = (K; L): Then we can write F (K;
PL) as F (x): By concavity, for all pairs
0 2 0 2 0 0
x ; x 2 R+ ; we have F (x ) F (x) i=1 Fi (x)(xi xi ): In particular, for
0 0
x = (0; 0); since F (x ) = F (0; 0) = 0; we have

X
2
F (x) Fi0 (x)xi : (2.63)
i=1

Suppose x 2R2++ . Then F (x) > 0 in view of F being neoclassical so that FK > 0
and FL > 0: From (2.63) we now …nd the elasticity of scale to be

X
2
Fi0 (x)xi =F (x) 1: (2.64)
i=1

In view of (2.13) and (2.12), this implies non-increasing returns to scale every-
where.
CLAIM 2 When a neoclassical production function F (K; L) is strictly concave,
it has decreasing returns to scale everywhere.
Proof. The argument is analogue to that above, but in view of strict concavity
the inequalities in (2.63) and (2.64) become strict. This implies that F has DRS
everywhere.

2.8 Exercises
2.1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

62 CHAPTER 2. REVIEW OF TECHNOLOGY AND FIRMS

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 Part II

LOOKING AT THE LONG RUN

63
Chapter 3

The basic OLG model: Diamond

There exists two main analytical frameworks for analyzing the basic intertemporal
choice, consumption versus saving, and the dynamic long-run implications of
this choice: overlapping generations models and representative agent models. In
the …rst class of models the focus is on (a) the interaction between di¤erent
generations alive at the same time, and (b) the never-ending entrance of new
generations. In the second class of models the household sector is modelled as
consisting of a …nite number of in…nitely-lived agents. One interpretation is that
these agents are dynasties where parents take the utility of their descendants fully
into account by leaving bequests. This approach, which is also called the Ramsey
approach (after the British mathematician and economist Frank Ramsey, 1903-
1930), will be described in Chapter 8 (discrete time) and Chapter 10 (continuous
time).
In the present chapter we introduce the overlapping generations approach
which has shown its usefulness for analysis of questions associated with public
debt problems, taxation of capital income, …nancing of social security (pensions),
design of educational systems, non-neutrality of money, and the possibility of
speculative bubbles. Our focus will be on the overlapping generations model
called Diamond’s OLG model1 after the American economist and Nobel Prize
laureate Peter A. Diamond (1940-).
Among the strengths of the model are:
The life-cycle aspect of human behavior is taken into account. Although
the economy is in…nitely-lived, the individual agents have …nite time hori-
zons. During lifetime one’s educational level, working capacity, income, and
needs change and this is re‡ected in the individual labor supply and saving
behavior. The aggregate implications of the life-cycle behavior of coexisting
individual agents at di¤erent stages in their life is at the centre of attention.
1
Diamond (1965).

65
66 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

The model takes elementary forms of heterogeneity in the population into

account there are “old”and there are “young”, there are currently-alive
people and there are as yet unborn whose preferences are not re‡ected
in current market transactions. Questions relating to the distribution of
income and wealth across generations can be studied. For example, how
does the investment in capital and environmental protection by current
generations a¤ect the conditions for succeeding generations?

3.1 Motives for saving

Before going into the speci…cs of Diamond’s model, let us brie‡y consider what
may motivate people to save:

(a) The consumption-smoothing motive for saving. Individuals go through a life

cycle where individual income typically has a hump-shaped time pattern; by
saving and dissaving the individual attempts to obtain the desired smooth-
ing of consumption across lifetime. This is the essence of the life-cycle
saving hypothesis put forward by Nobel laureate Franco Modigliani (1918-
2003) and associates in the 1950s. This hypothesis states that consumers
plan their saving and dissaving in accordance with anticipated variations
in income and needs over lifetime. Because needs vary less over lifetime
than income, the time pro…le of saving tends to be hump-shaped with some
dissaving early in life (while studying etc.), positive saving during the years
of peak earnings and then dissaving after retirement.

(b) The precautionary motive for saving. Income as well as needs may vary
due to conditions of uncertainty: sudden unemployment, illness, or other
kinds of bad luck. By saving, the individual can obtain a bu¤er against
such unwelcome events.

Horioka and Watanabe (1997) …nd that empirically, the saving motives (a)
and (b) are of dominant importance (Japanese data). Yet other motives include:

(c) Saving enables the purchase of durable consumption goods and owner-occupied
housing as well as repayment of debt.

(d) Saving may be motivated by the desire to leave bequests to heirs.

(e) Saving may simply be motivated by the fact that …nancial wealth may lead
to social prestige and economic or political power.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.2. The model framework 67

Diamond’s OLG model aims at simplicity and concentrates on motive (a).

Only one aspect of motive (a) is in fact considered, namely the saving for re-
tirement. People live for two periods only, as “young”, working full-time, and as
“old”, having retired and living by their savings. The Diamond model abstracts
from a possible bequest motive.
Now to the details.

3.2 The model framework

The ‡ow of time is divided into successive periods of equal length, taken as the
time unit. Given the two-period lifetime of (adult) individuals, the period length
is understood to be around, say, 30 years. The main assumptions are:

1. The number of young people in period t, denoted Lt ; changes over time

according to Lt = L0 (1 + n)t ; t = 0; 1; 2; :::, where n is a constant, n > 1:
Indivisibility is ignored and so Lt is just considered a positive real number.

2. Only the young work. Each young supplies one unit of labor inelastically.
The division of available time between work and leisure is thereby considered
as exogenous.

3. Output is homogeneous and can be used for consumption as well as invest-

ment in physical capital. Physical capital is the only non-human asset in
the economy; it is owned by the old and rented out to the …rms. Output is
the numeraire (unit of account) used in trading. Money (means of payment)
is ignored.2

4. The economy is closed (no foreign trade).

5. Firms’technology has constant returns to scale.

6. In each period three markets are open, a market for output, a market for
labor services, and a market for capital services. Perfect competition rules
in all markets. Uncertainty is absent; when a decision is made, its conse-
quences are known.

7. Agents have perfect foresight.

Assumption 7 entails the following. First, the agents are assumed to have
“rational expectations”or, with a better name, “model-consistent expectations”.
2
As to the disregard of money we may imagine that agents have safe electronic accounts in
a …ctional central bank allowing costless transfers between accounts.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

68 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

This means that forecasts made by the agents coincide with the forecasts that
can be calculated on the basis of the model. Second, as there are no stochastic
elements in the model (no uncertainty), the forecasts are point estimates rather
than probabilistic forecasts. Thereby the model-consistent expectations take the
extreme form of perfect foresight: the agents agree in their expectations about
the future evolution of the economy and these expectations are point estimates
that coincide with the subsequent actual evolution of the economy.

Figure 3.1: The two-period model’s time structure.

Of course, this is an unrealistic assumption. The model makes this assumption

in order to simplify in a …rst approach. The results that emerge will be the
outcome of economic mechanisms in isolation from expectational errors. In this
sense the model constitutes a “pure”case (benchmark case).
The time structure of the model is illustrated in Fig. 3.1. In every period
two generations are alive and interact with each other as indicated by the arrows.
The young supply labor to the …rms, earn a labor income part of which they
consume and part of which they save for retirement. The young thereby o¤set
the dissaving by the old and possibly bring about positive net investment in the
economy. At the end of the period the savings by the young is converted into
direct ownership of new capital goods which constitute the non-consumed part
of aggregate output plus capital goods left over from the previous period. In the
next period the now old owners of the capital goods rent them out to the …rms.
We may imagine that the …rms are owned by the old, but this ownership is not
visible in the equilibrium allocation because pure pro…ts will be nil due to the
combination of perfect competition and constant returns to scale.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.2. The model framework 69

Let the output good be the numeraire and let r^t denote the rental rate for
capital in period t; that is, r^t is the real price a …rm has to pay at the end of
period t for the right to use one unit of someone else’s physical capital through
period t. So the owner of Kt units of physical capital receives a

r^t Kt Kt
real (net) rate of return on capital = = r^t ; (3.1)
Kt

where is the rate of physical capital depreciation which is assumed constant,

0 1.
Suppose there is also a market for loans, the “credit market”. Assume you
have lent out one unit of output from the end of period t 1 to the end of period
t. If the real interest rate in the loan market is rt ; then, at the end of period t you
should get back 1 + rt units of output. In the absence of uncertainty, equilibrium
requires that capital and loans give the same rate of return,

r^t = rt : (3.2)

This no-arbitrage condition indicates how the rental rate for capital and the more
everyday concept, the interest rate, would be related in an equilibrium where
both the market for capital services and a credit market were active. We shall
see, however, that in this model no credit market will be active in an equilibrium.
Nevertheless we will follow the tradition and call the right-hand side of (3.2) the
interest rate.
Table 3.1 provides an overview of the notation. As to our timing convention,
notice that any stock variable dated t indicates the amount held at the beginning
of period t: That is, the capital stock accumulated by the end of period t 1
and available for production in period t is denoted Kt : We therefore write Kt
= (1 )Kt 1 + It 1 and Yt = F (Kt ; Lt ); where F is an aggregate production
function. In this context it is useful to think of “period t”as running from date t
to date t + 1: So period t is the time interval [t; t + 1) on a continuous time axis.
Still, all decisions are made at discrete points in time t = 0; 1; 2; ::: (“dates”). We
imagine that receipts for work and lending as well as payment for the consumption
in period t occur at the end of the period. These timing conventions are common
in discrete-time growth and business cycle theory;3 they are convenient because
they make switching between discrete and continuous time analysis fairly easy.

3
In contrast, in accounting and …nance literature, typically Kt would denote the end-of-
period-t stock that begins to yield its services next period.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

70 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Table 3.1. List of main variable symbols

Symbol Meaning
Lt the number of young people in period t
n generation growth rate
Kt aggregate capital available in period t
c1t consumption as young in period t
c2t consumption as old in period t
wt real wage in period t
rt real interest rate (from end of per. t 1 to end of per. t)
rate of time preference (impatience)
elasticity of marginal utility
st saving of each young in period t
Yt aggregate output in period t
Ct = c1t Lt + c2t Lt 1 aggregate consumption in period t
St = Yt Ct aggregate gross saving in period t
2 [0; 1] capital depreciation rate
Kt+1 Kt = It Kt aggregate net investment in period t

3.3 The saving by the young

Suppose the preferences of the young can be represented by the lifetime utility
function speci…ed in (3.3). Given wt and rt+1 ; the decision problem of the young
in period t then is:

max U (c1t ; c2t+1 ) = u(c1t ) + (1 + ) 1 u(c2t+1 ) s.t. (3.3)

c1t ;c2t+1

c1t + st = wt (wt > 0); (3.4)

c2t+1 = (1 + rt+1 )st (rt+1 > 1); (3.5)
c1t 0; c2t+1 0: (3.6)

The interpretation of the variables is given in Table 3.1 above. We may think
of the “young” as a household consisting of one adult and 1 + n children whose
consumption is included in c1t : Note that “utility” appears at two levels. There
is a lifetime utility function, U; and a period utility function, u:4 The latter is
assumed to be the same in both periods of life (this has no e¤ects on the qualita-
tive results and simpli…es the exposition). The period utility function is assumed
continuous and twice continuously di¤erentiable with u0 > 0 and u00 < 0 (positive,
but diminishing marginal utility of consumption). Many popular speci…cations
4
Other names for these two functions are the intertemporal utility function and the subutility
function, respectively.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.3. The saving by the young 71

of u, e.g., u(c) = ln c; have the property that limc!0 u(c) = 1; then we de…ne
u(0) = 1:
The parameter is called the rate of time preference. It acts as a utility
discount rate, whereas (1 + ) 1 is a utility discount factor. Thus indicates the
degree of impatience w.r.t. the “arrival” of utility. By de…nition, > 1; but
> 0 is often assumed. When preferences can be represented in this additive way,
they are called time-separable. In principle, as seen from period t the interest rate
appearing in (3.5) should be interpreted as an expected real interest rate. But
as long as we assume perfect foresight, there is no need to distinguish between
actual and expected magnitudes.

Box 3.1. Discount rates and discount factors

By a discount rate is meant an interest rate applied in the construction of a dis-

count factor. A discount factor is a factor by which future bene…ts or costs, mea-
sured in some unit of account, are converted into present equivalents. The higher
the discount rate the lower the discount factor.
One should bear in mind that a discount rate depends on what is to be dis-
counted. In (3.3) the unit of account is “utility” and acts as a utility discount rate.
In (3.7) the unit of account is the consumption good and rt+1 acts as a consump-
tion discount rate. If people also work as old, the right-hand side of (3.7) would
read wt + (1 + rt+1 ) 1 wt+1 and thus rt+1 would act as an earnings discount rate.
This will be the same as the consumption discount rate if we think of real income
measured in consumption units. But if we think of nominal income, that is, income
measured in monetary units, there would be a nominal earnings discount rate,
namely the nominal interest rate, which in an economy with in‡ation will exceed
the consumption discount rate. Unfortunately, confusion of di¤erent discount rates
is not rare.

In (3.5) the interest rate rt+1 acts as a (net) rate of return on saving.5 An
interest rate may also be seen as a discount rate relating to consumption over time.
Indeed, by isolating st in (3.5) and substituting into (3.4), we may consolidate
5
While st in (3.4) appears as a ‡ow (non-consumed income), in (3.5) st appears as a stock
(the accumulated …nancial wealth at the end of period t). This notation is legitimate because
the magnitude of the two is the same when the time unit is the same as the period length.
In real life the gross payo¤ of individual saving may sometimes be nil (if invested in a project
that completely failed). Unless otherwise indicated, it is in this book understood that an interest
rate is a number exceeding 1 as indicated in (3.5). Thereby the discount factor 1=(1 + rt+1 )
is well-de…ned. In general equilibrium, the condition 1 + rt+1 > 0 is always met in the present
model.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

72 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

the two period budget constraints of the individual into one budget constraint,
1
c1t + c2t+1 = wt : (3.7)
1 + rt+1
In this intertemporal budget constraint the interest rate appears as the discount
rate entering the discount factor converting future amounts of consumption into
present equivalents, cf. Box 3.1.

Solving the saving problem

To avoid the possibility of corner solutions, we impose the No Fast Assumption

lim u0 (c) = 1: (A1)

c!0

In view of the sizeable period length in the model, this is de…nitely plausible.
Inserting the two budget constraints into the objective function in (3.3), we get
U (c1t ; c2t+1 ) = u(wt st ) +(1+ ) 1 u((1+rt+1 )st ) U~t (st ); a function of only one
decision variable, st : According to the non-negativity constraint on consumption
in both periods, (3.6), st must satisfy 0 st wt . Maximizing w.r.t. st gives
the …rst-order condition

dU~t
= u0 (wt st ) + (1 + ) 1 u0 ((1 + rt+1 )st )(1 + rt+1 ) = 0: (FOC)
dst

The second derivative of U~t is

d2 U~t
= u00 (wt st ) + (1 + ) 1 u00 ((1 + rt+1 )st )(1 + rt+1 )2 < 0: (SOC)
ds2t

Hence there can at most be one st satisfying (FOC). Moreover, for a positive
wage income there always exists such an st : Indeed:
LEMMA 1 Let wt > 0 and suppose the No Fast Assumption (A1) applies. Then
the saving problem of the young has a unique solution st = s(wt ; rt+1 ). The
solution is interior, i.e., 0 < st < wt ; and st satis…es (FOC).
Proof. Assume (A1). For any s 2 (0; wt ); dU~t (s)=ds > 1: Now consider the
endpoints s = 0 and s = wt : By (FOC) and (A1),

dU~t
lim = u0 (wt ) + (1 + ) 1 (1 + rt+1 ) lim u0 ((1 + rt+1 )s) = 1;
s!0 ds s!0

dU~t
lim = lim u0 (wt s) + (1 + ) 1 (1 + rt+1 )u0 ((1 + rt+1 )wt ) = 1:
s!w ds s!wt

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.3. The saving by the young 73

By continuity of U~t , it follows that there exists an st 2 (0; wt ) such that at s = st ;

dU~t =ds = 0; This is an application of the intermediate value theorem. It follows
that (FOC) holds for this st : By (SOC); st is unique and can therefore be written
as an implicit function, s(wt ; rt+1 ); of the exogenous variables in the problem, wt
and rt+1 .
Inserting the solution for st into the two period budget constraints, (3.4) and
(3.5), immediately gives the optimal consumption levels, c1t and c2t+1 :
The simple optimization method we have used here is called the substitution
method: by substitution of the constraints into the objective function an uncon-
strained maximization problem is obtained.6

The consumption Euler equation

The …rst-order condition (FOC) can conveniently be written

u0 (c1t ) = (1 + ) 1 u0 (c2t+1 )(1 + rt+1 ): (3.8)

This is known as an Euler equation, after the Swiss mathematician L. Euler (1707-
1783) who was the …rst to study dynamic optimization problems. In the present
context the condition is called a consumption Euler equation.
Intuitively, in an optimal plan the marginal utility cost of saving must equal
the marginal utility bene…t obtained by saving. The marginal utility cost of
saving is the opportunity cost (in terms of current utility) of saving one more
unit of account in the current period (approximately). This one unit of account
is transferred to the next period with interest so as to result in 1 + rt+1 units of
account in that period. An optimal plan requires that the utility cost equals the
utility bene…t of having rt+1 more units of account in the next period. And this
utility bene…t is the discounted value of the extra utility that can be obtained
next period through the increase in consumption by rt+1 units.
It may seem odd to attempt an intuitive interpretation this way, that is, in
terms of “utility units”. The utility concept is just a convenient mathematical de-
vice used to represent the assumed preferences. Our interpretation is only meant
as an as-if interpretation: as if utility were something concrete. An interpretation
in terms of concrete measurable quantities goes like this. We rewrite (3.8) as

u0 (c1t )
= 1 + rt+1 : (3.9)
(1 + ) 1 u0 (c2t+1 )

The left-hand side measures the marginal rate of substitution, MRS, of consump-
tion as old for consumption as young, evaluated at the point (c1 ; c2 ): MRS is
6
Alternatively, one could use the Lagrange method.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

74 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

de…ned as the increase in period-t + 1 consumption needed to compensate for a

one-unit marginal decrease in period-t consumption. That is,
dc2t+1 u0 (c1t )
M RSc2 c1 = jU =U = ; (3.10)
dc1t (1 + ) 1 u0 (c2t+1 )
where we have used implicit di¤erentiation in U (c1t ; c2t+1 ) = U : The right-hand
side of (3.9) indicates the marginal rate of transformation, MRT, which is the
rate at which saving allows an agent to shift consumption from period t to period
t + 1 via the market. In an optimal plan MRS must equal MRT.
Even though interpretations in terms of “MRS equal to MRT”are more sat-
isfactory, we will often use “as if”interpretations like the one before. They are a
convenient short-hand for the more elaborate interpretation.
The Euler equation (3.8) implies that
Q rt+1 causes u0 (c1t ) R u0 (c2t+1 ); i.e., c1t Q c2t+1 ;
respectively, in the optimal plan (because u00 < 0): That is, absent uncertainty
the optimal plan entails either increasing, constant or decreasing consumption
over time according to whether the rate of time preference is below, equal to, or
above the market interest rate, respectively. For example, when < rt+1; the
plan is to start with relatively low consumption in order to take advantage of the
relatively high rate of return on saving.
Note that there are in…nitely many pairs (c1t ; c2t+1 ) satisfying the Euler equa-
tion (3.8). Only when requiring the two period budget constraints, (3.4) and
(3.5), satis…ed, do we get the unique solution st and thereby the unique solution
for c1t and c2t+1 .

Properties of the saving function

The …rst-order condition (FOC), where the two budget constraints are inserted,
determines the saving as an implicit function of the market prices faced by the
young decision maker, i.e., st = s(wt ; rt+1 ):
The partial derivatives of this function can be found by applying the implicit
function theorem on (FOC). A practical procedure is the following. We …rst write
dU~t =dst as a function, f; of the variables involved, st ; wt ; and rt+1 ; i.e.,
dU~t
= u0 (wt st ) + (1 + ) 1 u0 ((1 + rt+1 )st )(1 + rt+1 ) f (st ; wt ; rt+1 ):
dst
By (FOC), f (st ; wt ; rt+1 ) = 0 and so the implicit function theorem (see Math
tools) implies
@st @f =@wt @st @f =@rt+1
= and = ;
@wt D @rt+1 D

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.3. The saving by the young 75

where D @f =@st d2 U~t =ds2t < 0 by (SOC). We …nd

@f
= u00 (c1t ) > 0;
@wt
@f
= (1 + ) 1 [u0 (c2t+1 ) + u00 (c2t+1 )st (1 + rt+1 )] :
@rt+1

Consequently, the partial derivatives of the saving function st = s(wt ; rt+1 ) are

@st u00 (c1t )

sw = > 0 (but < 1); (3.11)
@wt D
@st (1 + ) 1 [u0 (c2t+1 ) + u00 (c2t+1 )c2t+1 ]
sr = ; (3.12)
@rt+1 D

where in the last expression we have used (3.5).7

We see that 0 < sw < 1; which implies that 0 < @c1t =@wt < 1 and 0 < @c2t =@wt
< 1 + rt+1 : The positive sign of these two derivatives indicate that consumption
in each of the periods is a normal good (which certainly is plausible since we are
talking about the total consumption by the individual in each period).8 The sign
of sr is seen to be ambiguous. This ambiguity re‡ects that the Slutsky substi-
tution and income e¤ects on consumption as young of a rise in the interest rate
are of opposite signs. To understand this, it is useful to keep the intertempo-
ral budget constraint, (3.7), in mind. The substitution e¤ect on c1t is negative
because the higher interest rate makes future consumption cheaper in terms of
current consumption. And the income e¤ect on c1t is positive because with a
higher interest rate, a given budget can buy more consumption in both periods,
cf. (3.7). Generally there would be a third Slutsky e¤ect, a wealth e¤ect of a
rise in the interest rate. But such an e¤ect is ruled out in this model. This is
because there is no labor income in the second period of life. Indeed, as indicated
7
A perhaps more straightforward procedure, not requiring full memory of the exact content
of the implicit function theorem, is based on “implicit di¤erentiation”. First, keeping rt+1 …xed,
one calculates the total derivative w.r.t. wt on both sides of (FOC). Next, keeping wt …xed,
one calculates the total derivative w.r.t. rt+1 on both sides of (FOC).
Yet another possible procedure is based on “total di¤erentiation” in terms of di¤ erentials.
Taking the di¤erential w.r.t. st ; wt ; and rt+1 on both sides of (FOC) gives u00 (c1t )(dwt dst )+
+(1+ ) 1 fu00 (c2t+1 ) [(1 + rt+1 )dst + st drt+1 ] (1 + rt+1 ) + u0 (c2t+1 )drt+1 g = 0: By rearranging
we …nd the ratios dst =dwt and dst =drt+1 , which will indicate the value of the partial derivatives
(3.11) and (3.12).
8
Recall, a consumption good is called normal for given consumer preferences if the demand
for it is an increasing function of the consumer’s wealth. Since in this model the consumer is
born without any …nancial wealth, the consumer’s wealth at the end of period t is simply the
present value of labor earnings through life, which here, evaluated at the beginning of period t;
is wt =(1 + rt ) as there is no labor income in the second period of life, cf. (3.7).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

76 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

by (3.4), the human wealth of a member of generation t; evaluated at the end of

period t; is simply wt , which is independent of rt+1 :
Rewriting (3.12) gives
(1 + ) 1 u0 (c2t+1 )[ (c2t+1 ) 1]
sr = T 0 for (c2t+1 ) S 1; (3.13)
D
respectively, where D < 0; and where (c2t+1 ) is the absolute elasticity of marginal
utility of consumption in the second period, that is,
c2t+1 00 u0 (c2t+1 )=u0 (c2t+1 )
(c2t+1 ) u (c2t+1 ) > 0;
u0 (c2t+1 ) c2t+1 =c2t+1
where the approximation is valid for a “small” increase, c2t+1 ; in c2t+1 : The
inequalities in (3.13) show that when the absolute elasticity of marginal utility is
below one, then the substitution e¤ect on consumption as young of an increase in
the interest rate dominates the income e¤ect and saving increases. The opposite
is true if the elasticity of marginal utility is above one.
The reason that (c2t+1 ) has this role is that (c2t+1 ) re‡ects how sensitive
marginal utility of c2t+1 is to a rise in c2t+1 . To see the intuition, consider the
case where consumption as young, and thus saving, happens to be una¤ected by
an increase in the interest rate. Even in this case, consumption as old, c2t+1 ; is
automatically increased (in view of the higher income as old through the higher
rate of return on the unchanged saving); and the marginal utility of c2t+1 is thus
decreased in response to a higher interest rate. The point is that this outcome can
only be optimal if the elasticity of marginal utility of c2t+1 is of “medium” size.
A very high absolute elasticity of marginal utility of c2t+1 would result in a sharp
decline in marginal utility so sharp that not much would be lost by dampening
the automatic rise in c2t+1 and instead increase c1t ; thus reducing saving. On the
other hand, a very low elasticity of marginal utility of c2t+1 would result in only a
small decline in marginal utility so small that it is bene…cial to take advantage
of the higher rate of return and save more, thus accepting a …rst-period utility
loss brought about by a lower c1t :
We see from (3.12) that an absolute elasticity of marginal utility equal to
exactly one is the case leading to the interest rate being neutral vis-a-vis the
saving of the young. What is the intuition behind this? Neutrality vis-a-vis
the saving of the young of a rise in the interest rate requires that c1t remains
unchanged since c1t = wt st . In turn this requires that the marginal utility,
u0 (c2t+1 ); on the right-hand side of (3.8) falls by the same percentage as 1 + rt+1
rises. At the same time the budget (3.5) as old tells us that c2t+1 has to rise by
the same percentage as 1 + rt+1 if st remains unchanged. Altogether we thus need
that u0 (c2t+1 ) falls by the same percentage as c2t+1 rises. But this requires that
the absolute elasticity of u0 (c2t+1 ) w.r.t. c2t+1 is exactly one.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.3. The saving by the young 77

The elasticity of marginal utility, also called the marginal utility ‡exibility,
will generally depend on the level of consumption, as implicit in the notation
(c2t+1 ): There exists a popular special case, however, where the elasticity of
marginal utility is constant.
EXAMPLE 1 The CRRA utility function. If we impose the requirement that
u(c) should have an absolute elasticity of marginal utility of consumption equal
to a constant > 0; then one can show (see Appendix A) that the utility function
must be of the CRRA form:
c1 1
; when 6= 1;
u(c) = 1 ; (3.14)
ln c; when = 1:

It may seem odd that in the upper case we subtract the constant 1=(1 )
1
from c =(1 ): But adding or subtracting a constant from a utility function
does not a¤ect the marginal rate of substitution and consequently not behavior.
Notwithstanding that we could do without this constant, its presence in (3.14)
has two advantages. One is that in contrast to c1 =(1 ); the expression
(c1 1)=(1 ) can be interpreted as valid even for = 1; namely as identical
to ln c: This is because (c1 1)=(1 ) ! ln c for ! 1 (by L’Hôpital’s
rule for “0/0”). Another advantage is that the kinship between the di¤erent
members, indexed by ; of the CRRA family becomes more transparent. Indeed,
by de…ning u(c) as in (3.14), all graphs of u(c) will go through the same point as
the log function, namely (1; 0); cf. Fig. 3.2.
The higher is ; the more “curvature”does the corresponding curve in Fig. 3.2
have. In turn, more “curvature”re‡ects a higher incentive to smooth consumption
across time. The reason is that a large curvature means that the marginal utility
will drop sharply if consumption rises and will increase sharply if consumption
falls. Consequently, not much utility is lost by lowering consumption when it
is relatively high but there is a lot of utility to be gained by raising it when it
is relatively low. So the curvature indicates the degree of aversion towards
variation in consumption. Or we may say that indicates the strength of the
preference for consumption smoothing.9
Suppose the period utility is of CRRA form as given in (3.14). (FOC) then
yields an explicit solution for the saving of the young:
1
st = 1 wt : (3.15)
1 + (1 + )( 1+r
1+
t+1
)
9
The name CRRA is a shorthand for Constant Relative Risk Aversion and comes from the
theory of behavior under uncertainty. Also in that theory does the CRRA function constitute an
important benchmark case. And is in that context called the degree of relative risk aversion.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

78 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

u(c)

θ = 0
θ = 0.5
θ = 1
θ = 2
θ = 5

0 c
1

−1

Figure 3.2: The CRRA family of utility functions.

We see that the signs of @st =@wt and @st =@rt+1 shown in (3.11) and (3.13), re-
spectively, are con…rmed. Moreover, the saving of the young is in this special
case proportional to income with a factor of proportionality that depends on the
interest rate (as long as 6= 1). But in the general case the saving-income ratio
depends also on the income level.
A major part of the attempts at empirically estimating suggests that > 1:
Based on U.S. data, Hall (1988) provides estimates above 5; while Attanasio and
Weber (1993) suggest 1:25 3:33: For Japanese data Okubo (2011) suggests
2:5 5:0: As these studies relate to much shorter time intervals than the
implicit time horizon of about 2 30 years in the Diamond model, we should be
cautious. But if the estimates were valid also to that model, we should expect
the income e¤ect on current consumption of an increase in the interest rate to
dominate the substitution e¤ect, thus implying sr < 0 as long as there is no

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.3. The saving by the young 79

wealth e¤ect of a rise in the interest rate.

When the elasticity of marginal utility of consumption is a constant, ; its
inverse, 1= ; equals the elasticity of intertemporal substitution in consumption.
This concept refers to the willingness to substitute consumption over time when
the interest rate changes. Under certain conditions the elasticity of intertemporal
substitution re‡ects the elasticity of the ratio c2t+1 =c1t w.r.t. 1 + rt+1 when we
move along a given indi¤erence curve. The next subsection, which can be omitted
in a …rst reading, goes more into detail with the concept.

Digression: The elasticity of intertemporal substitution*

Consider a two-period consumption problem like the one above. Fig. 3.3 depicts
a particular indi¤erence curve, u(c1 ) + (1 + ) 1 u(c2 ) = U . At a given point,
(c1 ; c2 ); on the curve, the marginal rate of substitution of period-2 consumption
for period-1 consumption, M RS, is given by

dc2
M RS = j ;
dc1 U =U

that is, M RS at the point (c1 ; c2 ) is the absolute value of the slope of the tangent
to the indi¤erence curve at that point.10 Under the “normal” assumption of
“strictly convex preferences” (as for instance in the Diamond model), M RS is
rising along the curve when c1 decreases (and thereby c2 increases). Conversely,
we can let M RS be the independent variable and consider the corresponding
point on the indi¤erence curve, and thereby the ratio c2 =c1 , as a function of
M RS. If we raise M RS along the indi¤erence curve, the corresponding value of
the ratio c2 =c1 will also rise.
The elasticity of intertemporal substitution in consumption at a given point is
de…ned as the elasticity of the ratio c2 =c1 w.r.t. the marginal rate of substitution
of c2 for c1 ; when we move along the indi¤erence curve through the point (c1 ; c2 ).
Letting the elasticity w.r.t. x of a di¤erentiable function f (x) be denoted E`x f (x);
the elasticity of intertemporal substitution in consumption can be written
(c2 =c1 )
c2 M RS d (c2 =c1 ) c2 =c1
E`M RS = j ;
c1 c2 =c1 dM RS U =U M RS
M RS

where the approximation is valid for a “small”increase, M RS; in M RS:

A more concrete understanding is obtained when we take into account that
in the consumer’s optimal plan, M RS equals the ratio of the discounted prices
10
When the meaning is clear from the context, to save notation we just write M RS instead
of the more precise M RSc2 c1 :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

80 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Figure 3.3: Substitution of period 2-consumption for period 1-consumption as M RS

increases to M RS 0 .

of good 1 and good 2, that is, the ratio 1=(1=(1 + r)) given in (3.7). Indeed, from
(3.10) and (3.9), omitting the time indices, we have
dc2 u0 (c1 )
M RS = jU =U = =1+r R: (3.16)
dc1 (1 + ) 1 u0 (c2 )
Letting (c1 ; c2 ) denote the elasticity of intertemporal substitution, evaluated at
the point (c1 ; c2 ); we then have
(c2 =c1 )
R d (c2 =c1 ) c2 =c1
(c1 ; c2 ) = jU =U R
: (3.17)
c2 =c1 dR R

Consequently, the elasticity of intertemporal substitution can here be interpreted

as the approximate percentage increase in the consumption ratio, c2 =c1 , triggered
by a one percentage increase in the inverse price ratio, holding the utility level
unchanged.11
Given u(c); we let (c) be the absolute elasticity of marginal utility of con-
sumption, i.e., (c) cu00 (c)=u0 (c): As shown in Appendix B, we then …nd the
elasticity of intertemporal substitution to be
c2 + Rc1
(c1 ; c2 ) = : (3.18)
c2 (c1 ) + Rc1 (c2 )
11
This characterization is equivalent to saying that the elasticity of substitution between two
consumption goods indicates the approximate percentage decrease in the ratio of the chosen
quantities of the goods (when moving along a given indi¤erence curve) induced by a one-
percentage increase in the corresponding price ratio.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.4. Production 81

We see that if u(c) belongs to the CRRA class and thereby (c1 ) = (c2 ) = ;
then (c1 ; c2 ) = 1= : In this case (as well as whenever c1 = c2 ) the elasticity of
marginal utility and the elasticity of intertemporal substitution are simply the
inverse of each other.

3.4 Production
Output is homogeneous and can be used for consumption as well as investment
in physical capital. The capital stock is thereby just accumulated non-consumed
output. We may imagine a “corn economy” where output is corn, part of which
is eaten (‡our) while the remainder is accumulated as capital (seed corn).
The speci…cation of technology and production conditions follows the sim-
ple competitive one-sector setup discussed in Chapter 2. Although the Diamond
model is a long-run model, we shall in this chapter for simplicity ignore techno-
logical change.

The representative …rm

There is a representative …rm with a neoclassical production function and con-
stant returns to scale (CRS). Omitting the time argument t when not needed for
clarity, we have

Y = F (K; L) = LF (k; 1) Lf (k); f 0 > 0; f 00 < 0; (3.19)

where Y is output (GNP) per period, K is capital input, L is labor input, and
k K=L is the capital-labor ratio. The derived function, f; is the production
function in intensive form. Capital installation and other adjustment costs are
ignored. Hence pro…t is F (K; L) r^K wL. The …rm maximizes under
perfect competition. This gives, …rst, @ =@K = FK (K; L) r^ = 0; that is,

@ [Lf (k)]
FK (K; L) = = f 0 (k) = r^: (3.20)
@K
Second, @ =@L = FL (K; L) w = 0; that is,

@ [Lf (k)]
FL (K; L) = = f (k) kf 0 (k) = w: (3.21)
@L
The interpretation is that the …rm will in every period use capital up to the point
where the marginal productivity of capital equals the rental rate given from the
market. Similarly, the …rm will employ labor up to the point where the marginal
productivity of labor equals the wage rate given from the market.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

82 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

In view of f 00 < 0; a k > 0 satisfying (3.20) is unique. Let us call it the

desired capital-labor ratio. Owing to CRS, however, at this stage the separate
factor inputs, K and L; are indeterminate; only their ratio, k; is determinate.12
We will now see how the equilibrium conditions for the factor markets select the
factor prices and the level of factor inputs consistent with equilibrium.

Factor prices in equilibrium

Let the aggregate demand for capital services and labor services be denoted K d
and Ld ; respectively: Clearing in factor markets in period t implies

Kt d = Kt ; (3.22)
Lt d = Lt = L0 (1 + n)t ; (3.23)

where Kt is the aggregate supply of capital services and Lt the aggregate supply of
labor services. As was called attention to in Chapter 1, unless otherwise speci…ed
it is understood that the rate of utilization of each production factor is constant
over time and normalized to one. So the quantity Kt will at one and the same time
measure both the capital input, a ‡ow, and the available capital stock. Similarly,
the quantity Lt will at one and the same time measure both the labor input, a
‡ow, and the size of the labor force as a stock (= the number of young people).
The aggregate input demands, K d and Ld , are linked through the desired
capital-labor ratio, k d : In equilibrium we have Ktd =Ldt = kt d = Kt =Lt kt , by
(3.22) and (3.23). The k in (3.20) and (3.21) can thereby be identi…ed with the
ratio of the stock supplies, kt Kt =Lt > 0; which is a predetermined variable.
Interpreted this way, (3.20) and (3.21) determine the equilibrium factor prices r^t
and wt in each period. In view of the no-arbitrage condition (3.2), the real interest
rate satis…es rt = r^t , where is the capital depreciation rate, 0 1; and
so in equilibrium we end up with

rt = f 0 (kt ) r(kt ) (r0 (kt ) = f 00 (kt ) < 0); (3.24)

0
wt = f (kt ) kt f (kt ) w(kt ) (w0 (kt ) = kt f 00 (kt ) > 0); (3.25)

where causality is from the right to the left in the two equations. In line with
our general perception of perfect competition, cf. Section 2.4 of Chapter 2, it is
understood that the factor prices, r^t and wt ; adjust quickly to the market-clearing
levels.
12
It might seem that k is overdetermined because we have two equations, (3.20) and (3.21),
but only one unknown. This reminds us that for arbitrary factor prices, r^ and w; there will not
exist a k satisfying both (3.20) and (3.21). But in equilibrium the factor prices faced by the
…rm are not arbitrary. They are equilibrium prices, i.e., they are adjusted so that (3.20) and
(3.21) become consistent.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 83

Technical Remark. In these formulas it is understood that L > 0; but we may

allow K = 0; i.e., k = 0: In case f 0 (0) is not immediately well-de…ned, we interpret
f 0 (0) as limk!0+ f 0 (k) if this limit exists. If it does not, it must be because
we are in a situation where limk!0+ f 0 (k) = 1; since f 00 (k) < 0 (an example
is the Cobb-Douglas function, f (k) = Ak , 0 < < 1; where limk!0+ f 0 (k)
= limk!0+ A k 1 = +1): In this situation we simply include +1 in the range
of r(k) and de…ne r(0) 0 limk!0+ (f 0 (k) )k = 0; where the last equality
comes from the general property of a neoclassical CRS production function that
limk!0+ kf 0 (k) = 0; cf. (2.18) of Chapter 2. Letting r(0) 0 = 0 also …ts well
with intuition since, when k = 0, nobody receives capital income anyway. Note
that since 2 [0; 1] ; r(k) > 1 for all k 0: What about w(0)? We interpret
w(0) as limk!0 w(k): From (2.18) of Chapter 2 we have that limk!0+ w(k) = f (0)
F (0; 1) 0: If capital is essential, F (0; 1) = 0: Otherwise, F (0; 1) > 0: Finally,
since w0 > 0; we have, for k > 0; w(k) > 0 as also noted in Chapter 2:

To …x ideas we have assumed that households (here the old) own the physical
capital and rent it out to the …rms. In view of perfect competition and constant
returns to scale, pure pro…t is nil in equilibrium. As long as the model ignores
uncertainty and capital installation costs, the results will be una¤ected if instead
we let the …rms themselves own the physical capital and …nance capital investment
by issuing bonds and shares. These bonds and shares would then be accumulated
by the households and constitute their …nancial wealth instead of the capital
goods themselves. The equilibrium rate of return, rt , would be the same.

3.5 The dynamic path of the economy

As in other …elds of economics, it is important to distinguish between the set of

technically feasible allocations and an allocation brought about, within this set,
by a speci…c economic institution (the rules of the game). The economic institu-
tion assumed by the Diamond model is the private-ownership perfect-competition
market institution.
We shall in the next subsections introduce three di¤erent concepts concerning
allocations over time in this economy. The three concepts are: technically feasible
paths, temporary equilibrium, and equilibrium path. These concepts are mutually
related in the sense that there is a whole set of technically feasible paths, within
which there may exist a unique equilibrium path, which in turn is a sequence of
states that have certain properties, including the temporary equilibrium property.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

84 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

3.5.1 Technically feasible paths

When we speak of technically feasible paths, the focus is merely upon what is
feasible from the point of view of the given technology as such and available initial
resources. That is, we disregard the agents’preferences, their choices given the
constraints, their interactions in markets, the market forces etc.
The technology is represented by (3.19) and there are two exogenous resources,
the labor force, Lt = L0 (1 + n)t ; and the initial capital stock, K0 : From na-
tional income accounting aggregate consumption can be written Ct Yt St =
F (Kt ; Lt ) St , where St denotes aggregate gross saving, and where we have
inserted (3.19). In a closed economy aggregate gross saving equals (ex post)
aggregate gross investment, Kt+1 Kt + Kt : So

Ct = F (Kt ; Lt ) (Kt+1 Kt + Kt ): (3.26)

Let ct denote aggregate consumption per unit of labor in period t; i.e.,

Ct c1t Lt + c2t Lt 1 c2t

ct = = c1t + :
Lt Lt 1+n

Combining this with (3.26) and using the de…nitions of k and f (k); we obtain the
dynamic resource constraint of the economy:
c2t
c1t + = f (kt ) + (1 )kt (1 + n)kt+1 : (3.27)
1+n

DEFINITION 1 Let k0 0 be the historically given initial ratio of available

capital and labor. The path f(kt ; c1t ; c2t )g1t=0 is called technically feasible if it has
k0 = k0 and for all t = 0; 1; 2;. . . , (3.27) has kt 0; c1t 0; and c2t 0.
The next subsections consider how, for given household preferences, the private-
ownership market institution with pro…t-maximizing …rms under perfect competi-
tion generates a selection within the set of technically feasible paths. A member
of this selection (which may but need not have just one member) is called an
equilibrium path. It constitutes a sequence of states with certain properties, one
of which is the temporary equilibrium property.

3.5.2 A temporary equilibrium

Standing in a given period, it is natural to think of next period’s interest rate as
an expected interest rate that provisionally can deviate from the ex post realized
e
one. We let rt+1 denote the expected real interest rate of period t + 1 as seen
from period t:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 85

Essentially, by a temporary equilibrium in period t is meant a state where for

e
a given rt+1 , all markets clear in the period. There are three markets, namely
two factor markets and a market for produced goods. We have already described
the two factor markets. In the market for produced goods the representative …rm
supplies the amount Yts = F (Ktd ; Ldt ) in period t: The demand side in this market
has two components, consumption, Ct , and gross investment, It : Equilibrium in
the goods market requires that demand equals supply, i.e.,

Ct + It = c1t Lt + c2t Lt 1 + It = Yts = F (Ktd ; Ldt ); (3.28)

where consumption by the young and old, c1t and c2t , respectively, were deter-
mined in Section 3.
By de…nition, aggregate gross investment equals aggregate net investment,
N
It ; plus capital depreciation, i.e.,

It = ItN + Kt N
I1t N
+ I2t + Kt N
S1t N
+ S2t + Kt = st Lt + ( Kt ) + Kt : (3.29)

The …rst equality follows from the de…nition of net investment and the assump-
tion that capital depreciation equals Kt : Next comes an identity re‡ecting that
aggregate net investment is the sum of net investment by the young and net in-
vestment by the old. In turn, saving in this model is directly an act of acquiring
N N
capital goods. So the net investment by the young, I1t ; and the old, I2t ; are
N N
identical to their net saving, S1t and S2t ; respectively. As we have shown, the
net saving by the young in the model equals st Lt : And the net saving by the
old is negative and equals Kt : Indeed, because they have no bequest motive,
the old consume all they have and leave nothing as bequests. Hence, the young
in any period enter the period with no non-human wealth. Consequently, any
non-human wealth existing at the beginning of a period must belong to the old
in that period and be the result of their saving as young in the previous period.
As Kt constitutes the aggregate non-human wealth in our closed economy at the
beginning of period t; we therefore have

st 1 Lt 1 = Kt : (3.30)

Recalling that the net saving of any group is by de…nition the same as the increase
in its non-human wealth, the net saving of the old in period t is Kt : Aggregate
net saving in the economy is thus st Lt + ( Kt ); and (3.29).is thereby explained.
DEFINITION 2 For a given period t with capital stock Kt 0 and labor supply
e
Lt > 0; let the expected real interest rate be given as rt+1 > 1: With kt Kt =Lt ,
a temporary equilibrium in period t is a state (kt ; c1t ; c2t ; wt ; rt ) of the economy
such that (3.22), (3.23), (3.28), and (3.29) hold (i.e., all markets clear) for c1t
e
= wt st and c2t = (kt + rt kt )(1 + n); where st = s(wt ; rt+1 ); as de…ned in

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

86 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Lemma 1, while wt = w(kt ) > 0 and rt = r(kt ); as de…ned in (3.25) and (3.24),
respectively.
The reason for the requirement wt > 0 in the de…nition is that if wt = 0;
people would have nothing to live on as young and nothing to save from for
retirement. The system would not be economically viable in this case. With
regard to the equation for c2t in the de…nition, note that (3.30) gives st 1 =
Kt =Lt 1 = (Kt =Lt )(Lt =Lt 1 ) = kt (1 + n); which is the wealth of each old at
the beginning of period t. Substituting into c2t = (1 + rt )st 1 , we get c2t =
(1 + rt )kt (1 + n); which can also be written c2t = (kt + rt kt )(1 + n): This last way
of writing c2t has the advantage of being applicable even if kt = 0; cf. Technical
Remark in Section 3.4. The remaining conditions for a temporary equilibrium
are self-explanatory.
PROPOSITION 1 Suppose the No Fast Assumption (A1) applies. Consider a
e
given period t with a given kt 0: Then for any rt+1 > 1;
(i) if kt > 0, there exists a temporary equilibrium, (kt ; c1t ; c2t ; wt ; rt ); and c1t and
c2t are positive;
(ii) if kt = 0, a temporary equilibrium exists if and only if capital is not essential;
in that case, wt = w(kt ) = w(0) = f (0) > 0 and c1t and st are positive (while
c2t = 0);
(iii) whenever a temporary equilibrium exists, it is unique.
Proof. We begin with (iii). That there is at most one temporary equilibrium is
immediately obvious since wt and rt are functions of the given kt : wt = w(kt )
e
and rt = r(kt ): And given wt , rt ; and rt+1 ; c1t and c2t are uniquely determined.
(i) Let kt > 0. Then, by (3.25), w(kt ) > 0: We claim that the state (kt ; c1t ; c2t ; wt ; rt );
e
with wt = w(kt ); rt = r(kt ); c1t = w(kt ) s(w(kt ); rt+1 ); and c2t = (1+r(kt ))kt (1+
n); is a temporary equilibrium. Indeed, Section 3.4 showed that the factor prices
wt = w(kt ) and rt = r(kt ) are consistent with clearing in the factor markets in
period t. Given that these markets clear (by price adjustment), it follows by Wal-
ras’law (see Appendix C) that also the third market, the goods market, clears
in period t. So all criteria in De…nition 2 are satis…ed. That c1t > 0 follows from
w(kt ) > 0 and the No Fast Assumption (A1), in view of Lemma 1. That c2t > 0
follows from c2t = (1 + r(kt ))kt (1 + n) when kt > 0; since r(kt ) > 1 always:
(ii) Let kt = 0. Suppose f (0) > 0: Then, by Technical Remark in Section 3.4,
e
wt = w(0) = f (0) > 0 and c1t = wt s(wt ; rt+1 ) is well-de…ned, positive, and less
e
than wt ; in view of Lemma 1; so st = s(wt ; rt+1 ) > 0. The old in period 0 will
starve since c2t = (0 + 0)(1 + n); in view of r(0) 0 = 0; cf. Technical Remark in
Section 3.4. Even though this is a bad situation for the old, it is consistent with
the criteria in De…nition 2. On the other hand, if f (0) = 0; we get wt = f (0) = 0;
which violates one of the criteria in De…nition 2.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 87

Point (ii) of the proposition says that a temporary equilibrium may exist even
in a period where k = 0: The old in this period will starve and not survive. But if
capital is not essential, the young get positive labor income out of which they will
save a part for their old age and be able to maintain life also next period which
will be endowed with positive capital. Then, by our assumptions the economy is
viable forever.13
Generally, the term “equilibrium”is used to denote a state of “rest”, possibly
only “temporary rest”. The temporary equilibrium in the present model is an
example of a state of “temporary rest” in the following sense: (a) the agents
optimize, given their expectations and the constraints they face; and (b) the
aggregate demands and supplies in the given period are mutually consistent,
i.e., markets clear. The quali…cation “temporary” is motivated by two features.
First, in the next period circumstances may be di¤erent, among other things as a
consequence of the currently chosen actions. Second, the given expectations may
turn out wrong.

3.5.3 An equilibrium path

The concept of an equilibrium path, also called an intertemporal equilibrium,
requires more conditions satis…ed. The concept refers to a sequence of temporary
equilibria such that expectations of the agents are ful…lled in every period:
DEFINITION 3 An equilibrium path is a technically feasible path f(kt ; c1t ; c2t )g1 t=0
such that for t = 0; 1; 2;. . . , the state (kt ; c1t ; c2t ; wt ; rt ) is a temporary equilibrium
e
with rt+1 = r (kt+1 ).
To characterize such a path, we forward (3.30) one period and rearrange so
as to get
Kt+1 = st Lt : (3.31)
Since Kt+1 kt+1 Lt+1 = kt+1 Lt (1 + n); this can be written

s (w (kt ) ; r (kt+1 ))
kt+1 = ; (3.32)
1+n
e e
using that st = s(wt ; rt+1 ); wt = w(kt ), and rt+1 = rt+1 = r (kt+1 ) in a sequence of
temporary equilibria with ful…lled expectations. Equation (3.32) is a …rst-order
di¤erence equation, known as the fundamental di¤erence equation or the law of
motion of the Diamond model.
PROPOSITION 2 Suppose the No Fast Assumption (A1) applies. Then,
13
For simplicity, the model ignores that in practice a certain minimum per capita consumption
level (the subsistence minimum) is needed for viability.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

88 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

(i) for any k0 > 0 there exists at least one equilibrium path;
(ii) if k0 = 0; an equilibrium path exists if and only if f (0) > 0 (i.e., capital not
essential);
(iii) in any case, an equilibrium path has a positive real wage in all periods and
positive capital in all periods except possibly the …rst;
(iv) an equilibrium path satis…es the …rst-order di¤erence equation (3.32).
Proof. (i) and (ii): see Appendix D. (iii) For a given t; let kt 0: Then,
since an equilibrium path is a sequence of temporary equilibria, we have wt =
e e
w(kt ) > 0 and st = s(w (kt ) ; rt+1 ), where rt+1 = r (kt+1 ) : Hence, by Lemma 1,
e
s(w (kt ) ; rt+1 ) > 0; which implies kt+1 > 0; in view of (3.32). This shows that
only for t = 0 is kt = 0 possible along an equilibrium path. (iv) This was shown
in the text above.
The formal proofs of point (i) and (ii) of the proposition are placed in appendix
because they are quite technical. But the graphs in the ensuing …gures 3.4-3.7
provide an intuitive veri…cation. The “only if” part of point (ii) re‡ects the not
very surprising fact that if capital were an essential production factor, no capital
“now”would imply no income “now”, hence no saving and investment and thus no
capital in the next period and so on. On the other hand, the “if”part of point (ii)
says that when capital is not essential, an equilibrium path can set o¤ even from
an initial period with no capital. Then point (iii) adds that an equilibrium path
will have positive capital in all subsequent periods. Finally, as to point (iv), note
that the fundamental di¤erence equation, (3.32), rests on equation (3.31). Recall
from the previous subsection that the economic logic behind this key equation
is that since capital is the only non-human asset in the economy and the young
are born without any inheritance, the aggregate capital stock at the beginning of
period t + 1 must be owned by the old generation in that period. It must thereby
equal the aggregate saving these people had in the previous period where they
were young.

The transition diagram

To be able to further characterize equilibrium paths, we construct a transition
diagram in the (kt ; kt+1 ) plane. The transition curve is de…ned as the set of points
(kt ; kt+1 ) satisfying (3.32). Its form and position depends on the households’
preferences and the …rms’technology. Fig. 3.4 shows one possible, but far from
necessary con…guration of this curve. A complicating circumstance is that the
equation (3.32) has kt+1 on both sides. Sometimes we are able to solve the
equation explicitly for kt+1 as a function of kt ; but sometimes we can do so only
implicitly. What is even worse is that there are cases where kt+1 is not unique
for a given kt : We will proceed step by step.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 89

First, what can we say about the slope of the transition curve? In general a
point on the transition curve has the property that at least in a small neighbor-
hood of this point the equation (3.32) will de…ne kt+1 as an implicit function of
kt .14 Taking the total derivative w.r.t. kt on both sides of (3.32), we get
dkt+1 1 dkt+1
= sw w0 (kt ) + sr r0 (kt+1 ) : (3.33)
dkt 1+n dkt
By ordering, the slope of the transition curve within this small neighborhood can
be written
dkt+1 sw (w (kt ) ; r (kt+1 )) w0 (kt )
= ; (3.34)
dkt 1 + n sr (w (kt ) ; r (kt+1 )) r0 (kt+1 )
when sr (w(kt ); r(kt+1 ))r0 (kt+1 ) 6= 1+n: Since sw > 0 and w0 (kt ) = kt f 00 (kt ) > 0;
the numerator in (3.34) is always positive and we have
dkt+1 1+n
? 0 for sr (w(kt ); r(kt+1 )) ? 0 ;
dkt r (kt+1 )
respectively (recall that r0 (kt+1 ) = f 00 (kt+1 ) < 0):

Figure 3.4: Transition curve and the resulting dynamics in the log-utility Cobb-Douglas
case.

It follows that the transition curve is universally upward-sloping if and only if

sr (w(kt ); r(kt+1 )) > (1 + n)=r0 (kt+1 ) everywhere along the transition curve. The
14
An exception occurs if the denominator in (3.34) below vanishes.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

90 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

intuition behind this becomes visible by rewriting (3.34) in terms of small changes
in kt and kt+1 : Since kt+1 = kt dkt+1 =dkt for kt “small”, (3.34) implies

[1 + n sr ( ) r0 (kt+1 )] kt+1 sw ( ) w0 (kt ) kt : (*)

Let kt > 0: This rise in kt will always raise wage income and, via the resulting
rise in st ; raise kt+1 ; everything else equal. Everything else is not equal, however,
since a rise in kt+1 implies a fall in the rate of interest. There are four cases to
consider:
Case 1: sr ( ) = 0: Then there is no feedback e¤ect from the fall in the rate of
interest. So the tendency to a rise in kt+1 is neither o¤set nor forti…ed.
Case 2: sr ( ) > 0: Then the tendency to a rise in kt+1 will be partly o¤set
through the dampening e¤ect on saving resulting from the fall in the interest
rate. This negative feedback can not fully o¤set the tendency to a rise in kt+1 .
The reason is that the negative feedback on the saving of the young will only
be there if the interest rate falls in the …rst place. We cannot in a period have
both a fall in the interest rate triggering lower saving and a rise in the interest
rate (via a lower kt+1 ) because of the lower saving. So a su¢ cient condition for
a universally upward-sloping transition curve is that the saving of the young is a
non-decreasing function of the interest rate.
Case 3: (1 + n)=r0 (kt+1 ) < sr ( ) < 0: Then the tendency to a rise in kt+1 will
be forti…ed through the stimulating e¤ect on saving resulting from the fall in the
interest rate.
Case 4: sr ( ) < (1 + n)=r0 (kt+1 ) < 0: Then the expression in brackets on
the left-hand side of (*) is negative and requires therefore that kt+1 < 0 in
order to comply with the positive right-hand side. This is a situation of multiple
temporary equilibria, a situation where self-ful…lling expectations operate. We
shall explore this case in the next sub-section.
Another feature of the transition curve is the following:
LEMMA 2 (the transition curve is nowhere ‡at) For all kt > 0; dkt+1 =dkt 6= 0:
Proof. Since sw > 0 and w0 (kt ) > 0 always, the numerator in (3.34) is always
positive.
The implication is that no part of the transition curve can be horizontal.15
When the transition curve crosses the 45 degree line for some kt > 0, as in
the example in Fig. 3.4, we have a steady state at this kt : Formally:
DEFINITION 4 An equilibrium path f(kt ; c1t ; c2t )g1
t=0 is in a steady state with
capital-labor ratio k > 0 if the fundamental di¤erence equation, (3.32), is satis-
…ed with kt as well as kt+1 replaced by k .
15
This would not necessarily hold if the utility function were not time-separable.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 91

This exempli…es the notion of a steady state as a stationary point in a dy-

namic process. Some economists use the term “dynamic equilibrium” instead of
“steady state”. As in this book the term “equilibrium”refers to situations where
the constraints and decided actions of the market participants are mutually com-
patible, an economy can be in “equilibrium” without being in a steady state. A
steady state is seen as a special sequence of temporary equilibria with ful…lled
expectations, namely one with the property that the dynamic variable, here k,
entering the fundamental di¤erence equation does not change over time.
EXAMPLE 2 (the log utility Cobb-Douglas case) Let u(c) = ln c and Y =
AK L1 ; where A > 0 and 0 < < 1: Since u(c) = ln c is the case = 1
in Example 1, by (3.15) we have sr = 0: Indeed, with logarithmic utility the
substitution and income e¤ects on st o¤set each other; and, as discussed above,
in the Diamond model there can be no wealth e¤ect of a rise in rt+1 . Further,
the equation (3.32) reduces to a transition function,
(1 )Akt
kt+1 = : (3.35)
(1 + n)(2 + )
The associated transition curve is shown in Fig. 3.4 and there is for k0 > 0 both
a unique equilibrium path and a unique steady state with capital-labor ratio
1=(1 )
(1 )A
k = > 0:
(2 + )(1 + n)
At kt = k the slope of the transition curve is necessarily less than one. The
dynamics therefore lead to convergence to the steady state as illustrated in the
…gure.16 In the steady state the interest rate is r = f 0 (k ) = (1 + n)(2 +
)=(1 ) : Note that a higher n results in a lower k ; hence a higher r :
Because the Cobb-Douglas production function implies that capital is essen-
tial, (3.35) implies kt+1 = 0 if kt = 0: The state kt+1 = kt = 0 is thus a stationary
point of the di¤erence equation (3.35) considered in isolation. This state is not,
however, an equilibrium path as de…ned above (not a steady state of an economic
system since there is no production). We may call it a trivial steady state in
contrast to the economically viable steady state kt+1 = kt = k > 0 which is then
called a non-trivial steady state.
Theoretically, there may be more than one (non-trivial) steady state. Non-
existence of a steady state is also possible. But before considering these possibil-
ities, the next subsection (which may be skipped in a …rst reading) addresses an
even more de…ant feature which is that for a given k0 there may exist more than
one equilibrium path.
16
A formal proof can be based on the mean value theorem.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

92 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

The possibility of multiple equilibrium paths*

It turns out that a backward-bending transition curve like that in Fig. 3.5 is
possible within the model. Not only are there two steady states but for kt 2 (k; k)
there are three temporary equilibria with self-ful…lling expectations. That is, for a
given kt in this interval, there are three di¤erent values of kt+1 that are consistent
with self-ful…lling expectations. Exercise 3.3 at the end of the chapter documents
this possibility by way of a numerical example.

Figure 3.5: A backward-bending transition curve leads to multiple temporary equilibria

with self-ful…lling expectations.

The theoretical possibility of multiple equilibria with self-ful…lling expecta-

tions requires that there is at least one interval on the horizontal axis where a
section of the transition curve has negative slope. Let us see if we can get an
intuitive understanding of why in this situation multiple equilibria can arise. Con-
sider the speci…c con…guration in Fig. 3.5 where k 0 ; k 00 ; and k 000 are the possible
values for the capital-labor ratio next period when kt 2 (k; k): In a neighbor-
hood of the point P associated with the intermediate value, k 00 ; the slope of the
transition curve is negative. As we saw above, this requires not only that in this
neighborhood sr (wt ; r(kt+1 )) < 0, but that the stricter condition sr (wt ; r(kt+1 ))
< (1 + n)=f 00 (k 00 ) holds (we take wt as given since kt is given and wt = w(kt )).
That the point P with coordinates (kt ; k 00 ) is on the transition curve indicates that
e
given wt = w(kt ) and an expected interest rate rt+1 = r(k 00 ); the induced saving

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 93

by the young, s(wt ; r(k 00 ); will be such that kt+1 = k 00 ; that is, the expectation
is ful…lled. The fact that also the point (kt ; k 0 ); where k 0 > k 00 , is on transition
curve indicates that also a lower interest rate, r(k 0 ); can be self-ful…lling. By this
is meant that if an interest rate at the level r(k 0 ) is expected, then this expecta-
tion induces more saving by the young, just enough more to make kt+1 = k 0 > k 00 ,
thus con…rming the expectation of the lower interest rate level r(k 0 ): What makes
e
this possible is exactly the negative dependency of st on rt+1 : The fact that also
000 000 00
the point (kt ; k ); where k < k , is on the transition curve can be similarly
interpreted. It is also sr < 0 that makes it possible that less saving by the young
than at P can be induced by an expected higher interest rate, r(k 000 ); than at P.
These ambiguities point to a serious problem with the assumption of perfect
foresight. The model presupposes that all the young agree in their expectations.
Only then will one of the three mentioned temporary equilibria appear. But the
model is silent about how the needed coordination of expectations is brought
about, and if it is, why this coordination ends up in one rather than another of
the three possible equilibria with self-ful…lling expectations. Each single young is
isolated in the market and will not know what the others will expect. The market
mechanism in the model provides no coordination of expectations. As it stands,
the model cannot determine how the economy will evolve in this situation.
This is of course a weakness. Yet the encountered phenomenon itself that
multiple self-ful…lling equilibrium paths are theoretically possible is certainly
of interest and plays an important role in certain business cycle theories of booms
and busts.
For now we plainly want to circumvent non-uniqueness. There are at least
two ways to rule out the possibility of multiple equilibrium paths. One simple
approach is to discard the assumption of perfect foresight. Instead, some kind
of adaptive expectations may be assumed, for example in the form of myopic
foresight, also called static expectations. This means that the expectation formed
by the agents in the current period about the value of a variable next period
is that it will stay the same as in the current period. So here the assumption
e
would be that the young have the expectation rt+1 = rt . Then, given k0 > 0;
a unique sequence of temporary equilibria f(kt ; c1t ; c2t ; wt ; rt )g1
t=0 is generated by
the model. Oscillations in the sense of repetitive movements up and down of kt
are possible. Even chaotic trajectories are possible (see Exercise 3.6).
Outside steady state the agents will experience that their expectations are
systematically wrong. And the assumption of myopic foresight rules out that
learning occurs. This may be too simplistic, although it can be argued that
human beings to a certain extent have a psychological disposition to myopic
foresight.
Another approach to the indeterminacy problem in the Diamond model is

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

94 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

motivated by the presumption that the possibility of multiple equilibria is basi-

cally due to the rough time structure of the model. Each period in the model
corresponds to half of an adult person’s lifetime. Moreover, in the …rst period of
life there is no capital income, in the second there is no labor income. This coarse
notion of time may arti…cially generate a multiplicity of equilibria or, with my-
opic foresight, oscillations. An expanded model where people live many periods
may “smooth”the responses of the system to the events impinging on it. Indeed,
with working life stretching over more than one period, wealth e¤ects of changes
in the interest rate arise, thereby reducing the likelihood of a backward-bending
transition curve.
Anyway, in a …rst approach the analyst may want to stay with a rough time
structure because of its analytical convenience and then make the best of it by
imposing conditions on the utility function, the production function, and/or pa-
rameter values so as to rule out multiple equilibria. Following this approach we
stay with the assumption of perfect foresight, but assume that circumstances are
such that multiple temporary equilibria with self-ful…lling expectations do not
arise.

Conditions for uniqueness of the equilibrium path

Su¢ cient for the equilibrium path to be unique is that preferences and technology
in combination are such that the slope of the transition curve is everywhere
positive. Hence we impose the Positive Slope Assumption that
1+n
sr (w(kt ); r(kt+1 )) > (A2)
f 00 (kt+1 )
everywhere along an equilibrium path. This condition is of course always satis…ed
when sr 0 (re‡ecting an elasticity of marginal utility of consumption not above
one) and can be satis…ed even if sr < 0 (as long as sr is small in absolute value).
Essentially, it is an assumption that the income e¤ect on consumption as young
of a rise in the interest rate does not dominate the substitution e¤ect “too much”.
Unfortunately, a condition like (A2) is not in itself very informative. This is
because it is expressed in terms of an endogenous variable, kt+1 ; for given kt : A
model assumption should preferably be stated in terms of what is given, also called
the “primitives”of the model, that is, the exogenous elements which in this model
comprise the assumed preferences, demography, technology, and the market form.
We can state su¢ cient conditions, however, in terms of the “primitives”, such that
(A2) is ensured. Here we state two such su¢ cient conditions, both involving a
CRRA period utility function with parameter as de…ned in (3.14):

(a) If 0 < 1; then (A2) holds for all kt > 0 along an equilibrium path.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 95

(b) If the production function is of CES-type,17 i.e., f (k) = A( k + 1 )1= ;

A > 0; 0 < < 1; 1 < < 1; then (A2) holds along an equilibrium path
even for > 1; if the elasticity of substitution between capital and labor,
1=(1 ); is not too small, i.e., if
1 1 1=
> 1=
(3.36)
1 1 + (1 + ) (1 + f 0 (k) )(1 )=

1 1
for all k > 0: In turn, su¢ cient for this is that (1 ) >1 :

That (a) is su¢ cient for (A2) is immediately visible in (3.15). The su¢ ciency
of (b) is proved in Appendix D. The elasticity of substitution between capital
and labor is a concept analogue to the elasticity of intertemporal substitution
in consumption. It is a measure of the sensitivity of the chosen k = K=L with
respect to the relative factor price. The next chapter goes more into detail with
the concept and shows, among other things, that the Cobb-Douglas production
function corresponds to = 0: So the Cobb-Douglas production function will
1
satisfy the inequality (1 ) 1>1 (since > 0); hence also the inequality
(3.36).
With these or other su¢ cient conditions in the back of our mind we shall now
proceed imposing the Positive Slope Assumption (A2). To summarize:
PROPOSITION 3 (uniqueness of an equilibrium path) Suppose the No Fast and
Positive Slope assumptions, (A1) and (A2), apply. Then, if k0 > 0, there exists
a unique equilibrium path.
(i) if k0 > 0, there exists a unique equilibrium path;
(ii) if k0 = 0; an equilibrium path exists if and only if f (0) > 0 (i.e., capital not
essential).
When the conditions of Proposition 3 hold, the fundamental di¤erence equa-
tion, (3.32), of the model de…nes kt+1 as an implicit function of kt ;

kt+1 = '(kt );

for all kt > 0; where '(kt ) is called a transition function. The derivative of this
implicit function is given by (3.34) with kt+1 on the right-hand side replaced by
'(kt ); i.e.,
sw (w (kt ) ; r ('(kt ))) w0 (kt )
'0 (kt ) = > 0: (3.37)
1 + n sr (w (kt ) ; r ('(kt ))) r0 ('(kt ))
The positivity for all kt > 0 is due to (A2). Example 2 above leads to a transition
function.
17
CES stands for Constant Elasticity of Substitution. CES production functions are consid-
ered in detail in Chapter 4.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

96 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Having determined the evolution of kt ; we have in fact determined the evolu-

tion of “everything”in the economy: the factor prices w(kt ) and r(kt ); the saving
of the young st = s(w(kt ); r(kt+1 )); and the consumption by both the young and
the old. The mechanism behind the evolution of the economy is the Walrasian (or
Classical) mechanism where prices, here wt and rt ; always adjust so as to generate
market clearing as if there were a Walrasian auctioneer and where expectations
always adjust so as to be model consistent.

Existence and stability of a steady state?

Possibly the equilibrium path converges to a steady state. To address this issue,
we examine the possible con…gurations of the transition curve in more detail. In
addition to being positively sloped everywhere, the transition curve will always,
for kt > 0; be situated strictly below the solid curve, kt+1 = w(kt )=(1 + n), shown
in Fig. 3.6. In turn, the latter curve is always, for kt > 0; strictly below the
stippled curve, kt+1 = f (kt )=(1 + n), in the …gure. To be precise:
LEMMA 3 (ceiling and roof) Suppose the No Fast Assumption (A1) applies.
Along an equilibrium path, whenever kt > 0;
w(kt ) f (kt )
0 < kt+1 < < ; t = 0; 1; . . . .
1+n 1+n
Proof. From (iii) of Proposition 2, an equilibrium path has wt = w(kt ) > 0 and
kt+1 > 0 for t = 0; 1; 2;. . . : Thus,
st wt w(kt ) f (kt ) f 0 (kt )kt f (kt )
0 < kt+1 = < = = < ;
1+n 1+n 1+n 1+n 1+n
where the …rst equality comes from (3.32), the second inequality from Lemma
1 in Section 3.3, and the last inequality from the fact that f 0 (kt )kt > 0 when
kt > 0.
We will call the graph (kt ; w(kt )=(1 + n)) in Fig. 3.6 a ceiling. It acts as a
ceiling on kt+1 simply because the saving of the young cannot exceed the income
of the young, w(kt ): And we will call the graph (kt ; f (kt )=(1 + n)) a roof, because
“everything of interest” occurs below it. The roof can be drawn directly on the
basis of the production function f (kt ):
To characterize the position of the roof relative to the 45 line, we consider
the lower Inada condition, limk!0 f 0 (k) = 1.
LEMMA 4 The roof, R(k) f (k)=(1+n); has positive slope everywhere, crosses
the 45 line for at most one k > 0 and can only do that from above. A necessary
and su¢ cient condition for the roof to be above the 45 line for small k is that
either limk!0 f 0 (k) > 1 + n or f (0) > 0 (capital not essential).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 97

Proof. Since f 0 > 0; the roof has positive slope. Since f 00 < 0; it can only cross
the 45 line once and only from above. If and only if limk!0 f 0 (k) > 1 + n, then
for small kt , the roof is steeper than the 45 line. Obviously, if f (0) > 0; then
close to the origin, the roof will be above the 45 line.

Figure 3.6: A case where both the roof and the ceiling cross the 45 line, but the
transition curve does not (no steady state exists).

LEMMA 5 Given w(k) = f (k) f 0 (k)k for all k 0, where f (k) satis…es
f (0) 0, f 0 > 0; f 00 < 0; the following holds:
(i) limk!1 w(k)=k = 0;
(ii) the ceiling, C(k) w(k)=(1 + n); is positive and has positive slope for all
k > 0; moreover, there exists k > 0 such that C(k) < k for all k > k:
Proof. (i) In view of f (0) 0 combined with f 00 < 0; we have w(k) > 0 for
all k > 0. Hence, limk!1 w(k)=k 0 if this limit exists. Consider an arbitrary
k1 > 0: We have f 0 (k1 ) > 0: For all k > k1 ; it holds that 0 < f 0 (k) < f 0 (k1 ); in
view of f 0 > 0 and f 00 < 0; respectively. Hence, limk!1 f 0 (k) exists and

0 lim f 0 (k) < f 0 (k1 ): (3.38)

k!1

We have
w(k) f (k)
lim = lim lim f 0 (k): (3.39)
k!1 k k!1 k k!1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

98 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

There are two cases to consider. Case 1: f (k) has an upper bound. Then,
limk!1 f (k)=k = 0 so that limk!1 w(k)=k = limk!1 f 0 (k) = 0; by (3.39)
and (3.38), as w(k)=k > 0 for all k > 0. Case 2: limk!1 f (k) = 1: Then,
by L’Hôpital’s rule for “1=1”, limk!1 (f (k)=k) = limk!1 f 0 (k) so that (3.39)
implies limk!1 w(k)=k = 0:
(ii) As n > 1 and w(k) > 0 for all k > 0; C(k) > 0 for all k > 0: From
w0 (k) = kf 00 (k) > 0 follows C 0 (k) = kf 00 (k)=(1 + n) > 0 for all k > 0; that is,
the ceiling has positive slope everywhere. For k > 0; the inequality C(k) < k is
equivalent to w(k)=k < 1 + n. By (i) follows that for all " > 0; there exists k" > 0
such that w(k)=k < " for all k > k" : Now, letting " = 1 + n and k = k" proves
that there exists k > 0 such that w(k)=k < 1 + n for all k > k:
While the roof can be above the 45 line for all kt > 0; the ceiling can not.
Indeed, (ii) of the lemma implies that if for small kt the ceiling is above the 45
line, the ceiling will necessarily cross the 45 line at least once for larger kt :
In view of the ceiling being always an upper bound on kt+1 ; what is the point
of introducing also the roof? The point is that the roof is a more straightforward
construct since it is directly given by the production function and is always strictly
concave. The ceiling is generally a more complex construct. It can have convex
sections and for instance cross the 45 line at more than one point if at all. .
A necessary condition for existence of a (non-trivial) steady state is that the
roof is above the 450 line for small kt : But this is not su¢ cient for also the
transition curve to be above the 450 line for small kt . Fig. 3.6 illustrates this. Here
the transition curve is in fact everywhere below the 450 line. In this case no steady
state exists and the dynamics imply convergence towards the “catastrophic”point
(0; 0): Given the rate of population growth, the saving of the young is not su¢ cient
to avoid famine in the long run. This will for example happen if the technology
implies so low productivity that even if all income of the young were saved, we
would have kt+1 < kt for all kt > 0; cf. Exercise 3.2. The Malthusian mechanism
will be at work and bring down n (outside the model). This exempli…es that even
a trivial steady state (the point (0,0)) may be of interest in so far as it may be
the point the economy is heading to without ever reaching it.
To help existence of a steady state we will impose the condition that either
capital is not essential or preferences and technology …t together in such a way
that the slope of the transition curve is larger than one for small kt . That is, we
assume that either
(i) f (0) > 0 or (A3)
(ii) lim '0 (k) > 1;
k!0

where '0 (k) is implicitly given in (3.37). Whether condition (i) of (A3) holds in
a given situation can be directly checked from the production function. If it does

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.5. The dynamic path of the economy 99

not, we should check condition (ii). But this condition is less amenable because
the transition function ' is not one of the “primitives” of the model. There
exist cases, though, where we can …nd an explicit transition function and try out
whether (ii) holds (like in Example 2 above). But generally we can not. Then we
have to resort to su¢ cient conditions for (ii) of (A3), expressed in terms of the
“primitives”. For example, if the period utility function belongs to the CRRA
class and the production function is Cobb-Douglas at least for small k, then (ii)
of (A3) holds (see Appendix E). Anyway, as (i) and (ii) of (A3) can be interpreted
as re‡ecting two di¤erent kinds of “early steepness” of the transition curve, we
shall call (A3) the Early Steepness Assumption.18
PROPOSITION 4 (existence and stability of a steady state) Assume that the
No Fast Assumption (A1) and the Positive Slope assumption (A2) apply as well
as the Early Steepness Assumption (A3). Then there exists at least one steady
state k > 0 that is locally asymptotically stable. Oscillations do not occur.
Proof. By (A1), Lemma 3 applies. From Proposition 2 we know that if (i) of
(A3) holds, then kt+1 = st =(1 + n) > 0 even for kt = 0: Alternatively, (ii) of (A3)
is enough to ensure that the transition curve lies above the 45 line for small kt :
By Lemma 4 the roof then does the same. According to (ii) of Lemma 5, for
large kt the ceiling is below the 45 line. Being below the ceiling, cf. Lemma
3, the transition curve must therefore cross the 45 line at least once. Let k
denote the smallest kt at which it crosses. Then k > 0 is a steady state with the
property 0 < '0 (k ) < 1: By graphical inspection we see that this steady state
is asymptotically stable. For oscillations to come about there must exist a steady
state, k ; with '0 (k ) < 0; but this is impossible in view of (A2).
From Proposition 4 we conclude that, given k0 ; the assumptions (A1) - (A3)
ensure existence and uniqueness of an equilibrium path; moreover, the equilibrium
path converges towards some steady state. Thus with these assumptions, for any
k0 > 0; sooner or later the system settles down at some steady state k > 0. For
the factor prices we therefore have

rt = f 0 (kt ) ! f 0 (k ) r ; and
0
wt = f (kt ) kt f (kt ) ! f (k ) k f 0 (k ) w ;

for t ! 1: But there may be more than one steady state and therefore only
local stability is guaranteed. This can be shown by examples, where the utility
function, the production function, and parameters are speci…ed in accordance
with the assumptions (A1) - (A3) (see Exercise 3.5 and ...).
18
In (i) of (A3), the “steepness” is rather a “hop” at k = 0 if we imagine k approaching nil
from below.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

100 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Figure 3.7: A case of multiple steady states (and capital being not essential).

Fig. 3.7 illustrates such a case (with f (0) > 0 so that capital is not essential).
Moving West-East in the …gure, the …rst steady state, k1 ; is stable, the second,
k2 ; unstable, and the third, k3 ; stable. In which of the two stable steady states
the economy ends up depends on the initial capital-labor ratio, k0 : The lower
steady state, k1 ; is known as a poverty trap. If 0 < k0 < k2 ; the economy is
caught in the trap and converges to the low steady state. But with high enough
k0 (k0 > k2 ); perhaps obtained by foreign aid, the economy avoids the trap and
converges to the high steady state. Looking back at Fig. 3.6, we can interpret
that …gure’s scenario as exhibiting an inescapable poverty trap.
It turns out that CRRA utility combined with a Cobb-Douglas production
function ensures both that (A1) - (A3) hold and that a unique (non-trivial)
steady state exists. So in this case global asymptotic stability of the steady state
is ensured.19 Example 2 and Fig. 3.4 above display a special case of this, the
case = 1:
This is of course a convenient case for the analyst. A Diamond model sat-
isfying assumptions (A1) - (A3) and featuring a unique steady state is called a
well-behaved Diamond model.
We end this section with the question: Is it possible that aggregate consump-
tion, along an equilibrium path, for some periods exceeds aggregate income? We
19
See last section of Appendix E.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.6. The golden rule and dynamic ine¢ ciency 101

shall see that this is indeed possible in the model if K0 (wealth of the old in the
initial period) is large enough. Indeed, from national accounting we have:

C10 + C20 = F (K0 ; L0 ) I0 > F (K0 ; L0 ) , I0 < 0

, K1 < (1 )K0 , K0 K1 > K0 :

So aggregate consumption in period 0 being greater than aggregate income is

equivalent to a fall in the capital stock from period 0 to period 1 greater than
the capital depreciation in period 0: Consider the log utility Cobb-Douglas case
in Fig. 3.4 and suppose < 1 and Lt = L0 = 1; i.e., n = 0: Then kt = Kt for all
t and by (3.35), Kt+1 = (12+ )A Kt : Thus K1 < (1 )K0 for

1=(1 )
(1 )A
K0 > :
(2 + )(1 )

As initial K is arbitrary, this situation is possible. When it occurs, it re‡ects

that the …nancial wealth of the old is so large that their consumption (recall
they consume all their …nancial wealth as well as the interest on this wealth)
exceeds what is left of current aggregate production after subtracting the amount
consumed by the young. So aggregate gross investment in the economy will be
negative. Of course this is only feasible if capital goods can be “eaten”or at least
be immediately (without further resources) converted into consumption goods.
As it stands, the model has implicitly assumed this to be the case. And this is in
line with the general setup since the output good is homogeneous and can either
be consumed or piled up as capital.
We now turn to e¢ ciency problems.

3.6 The golden rule and dynamic ine¢ ciency

An economy described by the Diamond model has the property that even though
there is perfect competition and no externalities, the outcome brought about
by the market mechanism may not be Pareto optimal.20 Indeed, the economy
may overaccumulate forever and thus su¤er from a distinctive form of production
ine¢ ciency.
A key element in understanding the concept of overaccumulation is the con-
cept of a golden-rule capital-labor ratio. Overaccumulation occurs when aggregate
20
Recall that a Pareto optimal path is a technically feasible path with the property that
no other technically feasible path will make at least one individual better o¤ without making
someone else worse o¤. A technically feasible path which is not Pareto optimal is called Pareto
inferior.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

102 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

saving maintains a capital-labor ratio above the golden-rule value forever. Let us
consider these concepts in detail.
In the present section generally the period length is arbitrary except when
we relate to the Diamond model and the period length therefore is half of adult
lifetime.

The golden-rule capital-labor ratio

The golden rule is a principle that relates to technically feasible paths. The
principle does not depend on the market form.
Consider the economy-wide resource constraint Ct = Yt St = F (Kt ; Lt )
(Kt+1 Kt + Kt ); where we assume that F is neoclassical with CRS. Accordingly,
aggregate consumption per unit of labor can be written
Ct F (Kt ; Lt ) (Kt+1 Kt + Kt )
ct = = f (kt ) + (1 )kt (1 + n)kt+1 ;
Lt Lt
(3.40)
where k is the capital-labor ratio K=L: Note that Ct will generally be greater than
the workers’consumption. One should simply think of Ct as the ‡ow of produced
consumption goods in the economy and ct as this ‡ow divided by aggregate em-
ployment, including the labor that in period t produces investment goods. How
the consumption goods are distributed to di¤erent members of society is not our
concern here.
DEFINITION 5 By the golden-rule capital-labor ratio, kGR ; is meant that value
of the capital-labor ratio k; which results in the highest possible sustainable level
of consumption per unit of labor.
Sustainability requires replicability forever. We therefore consider a steady
state. In a steady state kt+1 = kt = k so that (3.40) simpli…es to
c = f (k) ( + n)k c(k): (3.41)
Maximizing gives the …rst-order condition
c0 (k) = f 0 (k) ( + n) = 0: (3.42)
In view of c00 (k) = f 00 (k) < 0; the condition (3.42) is both necessary and su¢ cient
for an interior maximum. Let us assume that + n > 0 and that f satis…es the
condition
lim f 0 (k) < + n < lim f 0 (k):
k!1 k!0

Then (3.42) has a solution in k; and it is unique because c00 (k) < 0. The solution
is called kGR so that
f 0 (kGR ) = n:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.6. The golden rule and dynamic ine¢ ciency 103

That is:
PROPOSITION 5 (the golden rule) The highest sustainable consumption level
per unit of labor in society is obtained when in steady state the net marginal
productivity of capital equals the growth rate of the economy.

Figure 3.8: A steady state with overaccumulation.

It follows that if a society aims at the highest sustainable level of consumption

and initially has k0 < kGR , it should increase its capital-labor ratio up to the
point where the extra output obtainable by a further small increase is exactly
o¤set by the extra gross investment needed to maintain the capital-labor ratio
at that level. The intuition is visible from (3.41). The golden-rule capital-labor
ratio, kGR ; strikes the right balance in the trade-o¤ between high output per unit
of labor and a not too high investment requirement. Although a steady state
with k > kGR would imply higher output per unit of labor, it would also imply
that a large part of that output is set aside for investment (namely the amount
( + n)k per unit of labor) to counterbalance capital depreciation and growth in
the labor force; without this investment the high capital-labor ratio k would not
be maintained. With k > kGR this feature would dominate the …rst e¤ect so that
consumption per unit of labor ends up low. Fig. 3.8 illustrates.
The name golden rule hints at the golden rule from the Bible: “Do unto others
as you would have them to do unto you.” We imagine that God asks the newly
born generation: “What capital-labor ratio would you prefer to be presented

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

104 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

with, given that you must hand over the same capital-labor ratio to the next
generation?”The appropriate answer is: the golden-rule capital-labor ratio.

The possibility of overaccumulation in a competitive market economy

The equilibrium path in the Diamond model with perfect competition implies an
interest rate r = f 0 (k ) in a steady state: As an implication,

r T n , f 0 (k ) T n , k S kGR ; respectively,

in view of f 00 < 0: Hence, a long-run interest rate below the growth rate of the
economy indicates that k > kGR : This amounts to a Pareto-inferior state of
a¤airs. Indeed, everyone can be made better o¤ if by a coordinated reduction of
saving and investment, k is reduced. A formal demonstration of this is given in
connection with Proposition 6 in the next subsection. Here we give an account
in more intuitive terms.
Consider Fig. 3.8. Let k be gradually reduced to the level kGR by refrain-
ing from investment in period t0 and forward until this level is reached. When
this happens, let k be maintained at the level kGR forever by providing for the
needed investment per young, ( + n)kGR : Then there would be higher aggregate
consumption in period t0 and every future period. Both the immediate reduction
of saving and a resulting lower capital-labor ratio to be maintained contribute to
this result. There is thus scope for both young and old to consume more in every
future period.
In the Diamond model a simple policy implementing such a Pareto improve-
ment in the case where k > kGR (i.e., r < n) is to incur a lump-sum tax on
the young, the revenue of which is immediately transferred lump sum to the old,
hence, fully consumed. Suppose this amounts to a transfer of one good from each
young to the old. Since there are 1 + n young people for each old person, every
old receives in this way 1 + n goods in the same period. Let this transfer be
repeated every future period. By decreasing their saving by one unit, the young
can maintain unchanged consumption in their youth, and when becoming old,
they receive 1 + n goods from the next period’s young and so on. In e¤ect, the
“return” on the tax payment by the young is 1 + n next period. This is more
than the 1 + r that could be obtained via the market through own saving.21
21
In this model with no utility of leisure, a tax on wage income, or a mandatory pay-as-you-go
pension contribution (see Chapter 5) would act like a lump-sum tax on the young.
The described tax-transfers policy will a¤ect the equilibrium interest rate negatively. By
choosing an appropriate size of the tax this policy, combined with competitive markets, will
under certain conditions (see Chapter 5.1) bring the economy to the golden-rule steady state
where overaccumulation has ceased and r = n:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.6. The golden rule and dynamic ine¢ ciency 105

A proof that k > kGR is indeed theoretically possible in the Diamond model
can be based on the log utility-Cobb-Douglas case from Example 2 in Section
3.5.3. As indicated by the formula for r in that example, the outcome r < n,
which is equivalent to k > kGR , can always be obtained by making the parameter
2 (0; 1) in the Cobb-Douglas function small enough. The intuition is that a
small implies a high 1 ; that is, a high wage income wL = (1 )K L L
= (1 )Y ; this leads to high saving by the young, since sw > 0: The result is
a high kt+1 which generates a high real wage also next period and may in this
manner be sustained forever.
An intuitive understanding of the fact that the perfectly competitive market
mechanism may thus lead to overaccumulation, can be based on the following
argument. Assume, …rst, that sr < 0. In this case, if the young in period t
expects the rate of return on their saving to end up small (less than n), the
decided saving will be large in order to provide for consumption after retirement.
But the aggregate result of this behavior is a high kt+1 and therefore a low f 0 (kt+1 ):
In this way the expectation of a low rt+1 is con…rmed by the actual events. The
young persons each do the best they can as atomistic individuals, taking the
market conditions as given. Yet the aggregate outcome is an equilibrium with
overaccumulation, hence a Pareto-inferior outcome.
Looking at the issue more closely, we see that sr < 0 is not crucial for this
outcome. Suppose sr = 0 (the log utility case) and that in the current period,
kt is, for some historical reason, at least temporarily considerably above kGR .
Thus, current wages are high, hence, st is relatively high (there is in this case no
o¤setting e¤ect on st from the relatively low expected rt+1 ): Again, the aggregate
result is a high kt+1 and thus the expectation is con…rmed. Consequently, the
situation in the next period is the same and so on. By continuity, even if sr > 0;
the argument goes through as long as sr is not too large.

Dynamic ine¢ ciency and the double in…nity

Another name for the overaccumulation phenomenon is dynamic ine¢ ciency.
DEFINITION 6 A technically feasible path f(ct ; kt )g1 t=0 with the property that
there does not exist another technically feasible path with higher ct in some
periods without smaller ct in other periods is called dynamically e¢ cient. A
technically feasible path f(ct ; kt )g1
t=0 which is not dynamically e¢ cient is called
dynamically ine¢ cient.
PROPOSITION 6 A technically feasible path f(ct ; kt )g1
t=0 with the property that
for t ! 1; kt ! k > kGR ; is dynamically ine¢ cient.
Proof. Let k > kGR : Then there exists an " > 0 such that k 2 (k 2"; k + 2")

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

106 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

implies f 0 (k) < n since f 00 < 0: By concavity of f;

f (k) f (k ") f 0 (k ")": (3.43)

Consider a technically feasible path f(ct ; kt )g1 t=0 with kt ! k for t ! 1 (the
reference path): Then there exists a t0 such that for t t0 ; kt 2 (k "; k + ");
fn0 (kt ) o < n and f 0 (kt ") < n: Consider an alternative feasible path
1
(^ct ; k^t ) ; where a) for t = t0 consumption is increased relative to the reference
t=0
path such that k^t0 +1 = kt0 "; and b) for all t > t0 ; consumption is such that
k^t+1 = kt ": We now show that after period t0 ; c^t > ct : Indeed, for all t > t0 , by
(3.40),

c^t = f (k^t ) + (1 )k^t (1 + n)k^t+1

= f (kt ") + (1 )(kt ") (1 + n)(kt+1 ")
0
f (kt ) f (kt ")" + (1 )(kt ") (1 + n)(kt+1 ") (by (3.43))
> f (kt ) ( + n)" + (1 )kt (1 + n)kt+1 + ( + n)"
= f (kt ) + (1 )kt (1 + n)kt+1 = ct ;

by (3.40).
Moreover, it can be shown22 that:
PROPOSITION 7 A technically feasible path f(ct ; kt )g1
t=0 such that for t ! 1;
kt ! k kGR ; is dynamically e¢ cient.
Accordingly, a steady state with k < kGR is never dynamically ine¢ cient.
This is because increasing k from this level always has its price in terms of a
decrease in current consumption; and at the same time decreasing k from this
level always has its price in terms of lost future consumption. But a steady state
with k > kGR is always dynamically ine¢ cient. Intuitively, staying forever with
k = k > kGR ; implies that society never enjoys its great capacity for producing
consumption goods.
The fact that k > kGR ; and therefore dynamic ine¢ ciency, cannot be ruled
out might seem to contradict the First Welfare Theorem from the microeconomic
theory of general equilibrium. This is the theorem saying that under certain con-
ditions (essentially that increasing returns to scale are absent are absent, markets
are competitive, no goods are of public good character, and there are no exter-
nalities, then market equilibria are Pareto optimal. In fact, however, the First
Welfare Theorem also presupposes a …nite number of periods or, if the number of
periods is in…nite, then a …nite number of agents. In contrast, in the OLG model
22
See Cass (1972).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.7. Concluding remarks 107

there is a double in…nity: an in…nite number of periods and agents. Hence, the
First Welfare Theorem breaks down. Indeed, the case r < n, i.e., k > kGR ; can
arise under laissez-faire. Then, as we have seen, everyone can be made better
o¤ by a coordinated intervention by some social arrangement (a government for
instance) such that k is reduced:
The essence of the matter is that the double in…nity opens up for technically
feasible reallocations which are de…nitely bene…cial when r < n and which a
central authority can accomplish but the market can not. That nobody need
loose by the described kind of redistribution is due to the double in…nity: the
economy goes on forever and there is no last generation. Nonetheless, some kind
of centralized coordination is required to accomplish a solution.
There is an analogy in “Gamow’s bed problem”: There are an in…nite number
of inns along the road, each with one bed. On a certain rainy night all innkeepers
have committed their beds. A late guest comes to the …rst inn and asks for a
bed. “Sorry, full up!” But the minister of welfare hears about it and suggests
that each incumbent guest move down the road one inn.23
Whether the theoretical possibility of overaccumulation should be a matter of
practical concern is an empirical question about the relative size of rates of return
and economic growth. To answer the question meaningfully, we need an extension
of the criterion for overaccumulation so that the presence of technological progress
and rising per capita consumption in the long run can be taken into account. This
is one of the topics of the next chapter. At any rate, we can already here reveal
that there exists no indication that overaccumulation has ever been an actual
problem in industrialized market economies.
A …nal remark before concluding. Proposition 5 about the golden rule can be
generalized to the case where instead of one there are n di¤erent capital goods in
the economy. Essentially the generalization says that assuming CRS-neoclassical
production functions with n di¤erent capital goods as inputs, one consumption
good, no technological change, and perfectly competitive markets, a steady state
in which per-unit-of labor consumption is maximized has interest rate equal to
the growth rate of the labor force when technological progress is ignored (see,
e.g., Mas-Colell, 1989).

3.7 Concluding remarks

(Un…nished)
In several respects the conclusions we get from OLG models are di¤erent than
those from representative agent models to be studied later. In OLG models the
23
George Gamow (1904-1968) was a Russian physicist. The problem is also known as Hilbert’s
hotel problem, after the German mathematician David Hilbert (1862-1943).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

108 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

aggregate quantities are the outcome of the interplay of …nite-lived agents at

di¤erent stages in their life cycle. The turnover in the population plays a crucial
role. In this way the OLG approach lays bare the possibility of coordination
failure on a grand scale. In contrast, in a representative agent model, aggregate
quantities are just a multiple of the actions of the representative household.
Regarding analytical tractability, the complexity implied by having in every
period two di¤erent coexisting generations is in some respects more than compen-
sated by the fact that the …nite time horizon of the households make the dynamics
of the model one-dimensional: we end up with a …rst-order di¤erence equation
in the capital-labor ratio, kt ; in the economy. In contrast, the dynamics of the
basic representative agent model (Chapter 8 and 10) is two-dimensional (owing
to the assumed in…nite horizon of the households considered as dynasties).
Miscellaneous notes:
OLG gives theoretical insights concerning macroeconomic implications of life
cycle behavior, allows heterogeneity, provides training in seeing the economy as
consisting of a heterogeneous population where the distribution of agent charac-
teristics matters for the aggregate outcome.
Farmer (1993), p. 125, notes that OLG models are di¢ cult to apply and
for this reason much empirical work in applied general equilibrium theory has
regrettably instead taken the representative agent approach.
Outlook: Rational speculative bubbles in general equilibrium, cf. Chapter ?.

3.8 Literature notes

1. The Nobel Laureate Paul A. Samuelson (1915-2009) is one of the pioneers
of OLG models. Building on the French economist and Nobel laureate Maurice
Allais (1911-2010), Samuelson’s famous article, Samuelson (1958), was concerned
with a missing market problem. Imagine a two-period OLG economy where,
as in the Diamond model, only the young have an income (which in turn is,
by Samuelson, assumed exogenous). Contrary to the Diamond model, however,
there is neither capital nor other stores of value. Then, in the laissez-faire market
economy the old have to starve. This is clearly a Pareto-inferior allocation; if
each member of the young generation hands over to the old generation one unit
of account, and this is repeated every period, everyone will be better o¤. Since
for every old there are 1 + n young, the implied rate of return would be n; the
population growth rate. Such transfers do not arise under laissez-faire. A kind
of social contract is required. As Samuelson pointed out, a government could in
period 0 issue paper notes, “money”, and transfer them to the members of the
old generation who would then use them to buy goods from the young. Provided
the young believed the notes to be valuable in the next period, they would accept

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.8. Literature notes 109

them in exchange for some of their goods in order to use them in the next period
for buying from the new young generation and so on. We have here an example of
how a social institution can solve a coordination problem. This gives a ‡avour of
Samuelson’s contribution although in his original article he assumed three periods
of life.
2. Diamond (1965) extended Samuelson’s contribution by adding capital ac-
cumulation. Because of its antecedents Diamonds OLG model is sometimes called
the Samuelson-Diamond model or the Allais-Samuelson-Diamond model. In our
exposition we have drawn upon clari…cations by Galor and Ryder (1989) and
de la Croix and Michel (2002). The last mentioned contribution is an extensive
exploration of discrete-time OLG models and their applications.
3. The life-cycle saving hypothesis was put forward by Franco Modigliani
(1918-2003) and associates in the 1950s. See for example Modigliani and Brum-
berg (1954). Numerous extensions of the framework, relating to the motives (b)
- (e) in the list of Section 3.1, see for instance de la Croix and Michel (2002).
4. A review of the empirics of life-cycle behavior and attempts at re…ning
life-cycle models are given in Browning and Crossley (2001).
5. Regarding the dynamic e¢ ciency issue, both the propositions 6 and 7 were
shown in a stronger form by the American economist David Cass (1937-2008).
Cass established the general necessary and su¢ cient condition for a feasible path
f(ct ; kt )g1
t=0 to be dynamically e¢ cient (Cass 1972). Our propositions 6 and 7 are
more restrictive in that they are limited to paths that converge. Partly intuitive
expositions of the deeper aspects of the theory are given by Shell (1971) and
Burmeister (1980).
6. Diamond has also contributed to other …elds of economics, including search
theory for labor markets. In 2010 Diamond, together with Dale Mortensen and
Christopher Pissarides, was awarded the Nobel price in economics.
From here very incomplete:
The two-period structure of Diamonds OLG model leaves little room for con-
sidering, e.g., education and dissaving in the early years of life. This kind of
issues is taken up in three-period extensions of the Diamond model, see ...
Multiple equilibria, self-ful…lling expectations, optimism and pessimism..
Dynamic ine¢ ciency, see also Burmeister (1980).
Bewley 1977, 1980.
Two-sector OLG: Galor (1992). Galor’s book??
On the golden rule in a general setup, see Mas-Colell (1989).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

110 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

3.9 Appendix
A. On the CRRA utility function
Derivation of the CRRA function Consider a utility function u(c); de…ned
for all c > 0 and satisfying u0 (c) > 0; u00 (c) < 0: Let the absolute value of
the elasticity of marginal utility be denoted (c); that is, (c) cu00 (c)=u0 (c)
> 0: We claim that if (c) is a positive constant, ; then up to a positive linear
transformation u(c) must be of the form

c1
; when 6= 1;
u(c) = 1 (*)
ln c; when = 1;

i.e., of CRRA form.

Proof. Suppose (c) = > 0: Then, u00 (c)=u0 (c) = =c: By integration, ln u0 (c)
= ln c + A; where A is an arbitrary constant. Take the antilogarithm function
on both sides to get u0 (c) = eA e ln c = eA c : By integration we get
1
eA c1 + B; when 6 1;
=
u(c) =
eA ln c + B; when = 1;

where B is an arbitrary constant. This proves the claim. Letting A = B = 0; we

get (*).
When we want to make the kinship between the members of the CRRA family
transparent, we maintain A = 0 and for = 1 also B = 0; whereas for 6= 1 we
set B = 1=(1 ). In this way we achieve that all members of the CRRA family
will be represented by curves going through the same point as the log function,
namely the point (1; 0), cf. Fig. 3.2. And adding or subtracting a constant does
not a¤ect marginal rates of substitution and consequently not behavior.

The domain of the CRRA function We want to extend the domain to

include c = 0: If 1; the CRRA function, whether in the form u(c) = (c1
1)=(1 ) or in the form (*), is de…ned only for c > 0; not for c = 0: This is
because for c ! 0 we get u(c) ! 1: In this case we simply de…ne u(0) = 1:
This will create no problems since the CRRA function anyway has the property
that u0 (c) ! 1; when c ! 0 (whether is larger or smaller than one). The
marginal utility thus becomes very large as c becomes very small, that is, the
No Fast Assumption is satis…ed. This will ensure that the chosen c is strictly
positive whenever there is a positive budget. So throughout this book we de…ne
the domain of the CRRA function to be [0; 1) :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.9. Appendix 111

The range of the CRRA function Considering the CRRA function u(c)
c1 1 (1 ) 1 for c 2 [0; 1) ; we have:

for 0 < < 1; the range of u(c) is (1 ) 1; 1 ;

for = 1; the range of u(c) is [ 1; 1) ;
for > 1; the range of u(c) is [ 1; (1 ) 1 ):

Thus, in the latter case u(c) is bounded from above and so allows asymptotic
“saturation”to occur.

B. Deriving the elasticity of intertemporal substitution in consumption

Referring to Section 3.3, we here show that the de…nition of (c1 ; c2 ) in (3.17)
gives the result (3.18). Let x c2 =c1 and (1 + ) 1 : Then the …rst-order
condition (3.16) and the equation describing the considered indi¤erence curve
constitute a system of two equations

u0 (c1 ) = u0 (xc1 )R;

u(c1 ) + u(xc1 ) = U :

For a …xed utility level U = U these equations de…ne c1 and x as implicit functions
of R; c1 = c(R) and x = x(R). We calculate the total derivative w.r.t. R in both
equations and get, after ordering,

[u00 (c1 ) Ru00 (xc1 )x] c0 (R) Ru00 (xc1 )c1 x0 (R)
= u0 (xc1 ); (3.44)
[u0 (c1 ) + u0 (xc1 )x] c0 (R) = u0 (xc1 )c1 x0 (R): (3.45)

Substituting c0 (R) from (3.45) into (3.44) and ordering now yields

c1 u00 (c1 ) xc1 u00 (xc1 ) R 0

x + R x (R) = x + R:
u0 (c1 ) u0 (xc1 ) x

Since cu00 (c)=u0 (c) (c), this can be written

R 0 x+R
x (R) = :
x x (c1 ) + R (xc1 )

Finally, in view of xc1 = c2 and the de…nition of (c1 ; c2 ), this gives (3.18).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

112 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

C. Walras’law
In the proof of Proposition 1 we referred to Walras’law. Here is how Walras’law
works in each period in a model like this. We consider period t; but for simplicity
we skip the time index t on the variables. There are three markets, a market
for capital services, a market for labor services, and a market for output goods.
Suppose a “Walrasian auctioneer”calls out the price vector (^ r; w; 1); where r^ > 0
and w > 0; and asks all agents, i.e., the young, the old, and the representative
…rm, to declare their supplies and demands.
The supplies of capital and labor are by assumption inelastic and equal to
K units of capital services and L units of labor services. But the demand for
capital and labor services depends on the announced r^ and w: Let the potential
pure pro…t of the representative …rm be denoted : If r^ and w are so that < 0;
the …rm declares K d = 0 and Ld = 0: If on the other hand at the announced r^
and w; = 0 (as when r^ = r(k) + and w = w(k)); the desired capital-labor
ratio is given as k d = f 0 1 (^
r) from (3.20), but the …rm is indi¤erent w.r.t. the
absolute level of the factor inputs. In this situation the auctioneer tells the …rm
to declare Ld = L (recall L is the given labor supply) and K d = k d Ld which is
certainly acceptable for the …rm. Finally, if > 0; the …rm is tempted to declare
in…nite factor demands, but to avoid that, the auctioneer imposes the rule that
the maximum allowed demands for capital and labor are 2K and 2L; respectively:
Within these constraints the factor demands will be uniquely determined by r^
and w and we have

= r; w; 1) = F (K d ; Ld )
(^ r^K d wLd : (3.46)

The owners of both the capital stock K and the representative …rm must be
those who saved in the previous period, namely the currently old. These elderly
will together declare the consumption c2 L 1 = (1 + r^ )K + and the net
investment K (which amounts to disinvestment). The young will declare the
e e
consumption c1 L = wL s(w; r+1 )L and the net investment sL = s(w; r+1 )L: So
e
aggregate declared consumption will be C = (1 + r^ )K + + wL s(w; r+1 )L
e
and aggregate net investment I K = s(w; r+1 )L K: It follows that C + I
= wL + r^K + : The aggregate declared supply of output is Y s = F (K d ; Ld ):
The values of excess demands in the three markets now add to

Z(^
r; w; 1) w(Ld L) + r^(K d K) + C + I Y s
= wLd wL + r^K d r^K + wL + r^K + F (K d ; Ld )
= wLd + r^K d + F (K d ; Ld ) = 0;

by (3.46).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.9. Appendix 113

This is a manifestation of Walras’law for each period: whatever the announced

price vector for the period is, the aggregate value of excess demands in the period
is zero. The reason is the following. When each household satis…es its budget
constraint and each …rm pays out its ex ante pro…t,24 then the economy as a
whole has to satisfy an aggregate budget constraint for the period considered.
The budget constraints, demands, and supplies operating in this thought ex-
periment (and in Walras’law in general) are the Walrasian budget constraints,
demands, and supplies. Outside equilibrium these are somewhat arti…cial con-
structs. A Walrasian budget constraint is based on the assumption that the
desired actions can be realized. This assumption will be wrong unless r^ and w
are already at their equilibrium levels. But the assumption that desired actions
can be realized is never falsi…ed because the thought experiment does not allow
trades to take place outside Walrasian equilibrium. Similarly, the Walrasian con-
sumption demand by the worker is rather hypothetical outside equilibrium. This
demand is based on the income the worker would get if fully employed at the
announced real wage, not on the actual employment (or unemployment) at that
real wage.
These ambiguities notwithstanding, the important message of Walras’ law
goes through, namely that when two of the three markets clear (in the sense of
the Walrasian excess demands being nil), so does the third.

D. Proof of (i) and (ii) of Proposition 2

For convenience we repeat the fundamental di¤erence equation characterizing an

equilibrium path:
s (w (kt ) ; r (kt+1 ))
kt+1 = ;
1+n
where w(k) f (k) f 0 (k)k > 0 for all k > 0 and r(k) f 0 (k) > 1 for all
k 0: The key to the proof of Proposition 2 about existence of an equilibrium
path is the following lemma.
LEMMA D1 Suppose the No Fast Assumption (A1) applies and let w > 0 and
n > 1 be given. Then the equation

s (w; r (k))
= 1 + n: (3.47)
k

has at least one solution k > 0:

24
By ex ante pro…t is meant the hypothetical pro…t calculated on the basis of …rms’desired
supply evaluated at the announced price vector, (^
r; w; 1).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

114 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

Proof. Note that 1 + n > 0: From Lemma 1 in Section 3.3 follows that for all
possible values of r(k); 0 < s(w; r(k)) < w: Hence, for any k > 0;
s (w; r (k)) w
0< < :
k k
Letting k ! 1 we then have s (w; r (k)) =k ! 0 since s (w; r (k)) =k is squeezed
between 0 and 0 (as indicated in the two graphs in Fig. 3.9).

Figure 3.9: Existence of a solution to equation (3.47).

Next we consider k ! 0: There are two cases.

Case 1: limk!0 s (w; r (k)) > 0:25 Then obviously limk!0 s (w; r (k)) =k = 1:
Case 2: limk!0 s (w; r (k)) = 0:26 In this case we have
lim r (k) = 1: (3.48)
k!0

Indeed, since f 0 (k) rises monotonically as k ! 0; the only alternative would be

that limk!0 r (k) exists and is < 1; then, by Lemma 1 in Section 3.3, we would
be in case 1 rather than case 2. By the second-period budget constraint, with
r = r(k); consumption as old is c2 = s (w; r (k)) (1 + r(k)) c(w; k) > 0 so that
s (w; r (k)) c(w; k)
= :
k [1 + r(k)] k
The right-hand side of this equation goes to 1 for k ! 0 since limk!0 [1 + r(k)] k =
0 by Technical Remark in Section 3.4 and limk!0 c(w; k) = 1; this latter fact
follows from the …rst-order condition (3.8), which can be written
u0 (w s(w; r(k)) u0 (w)
0 u0 (c(w; k)) = (1 + ) (1 + ) :
1 + r(k) 1 + r(k)
25
If the limit does not exist, the proof applies to the limit inferior of s (w; r (k)) for k ! 0:
1
The limit inferior for i ! 1 of a sequence fxi gi=0 is de…ned as limi!1 inf fxj j j = i; i+1; . . . g ;
where inf of a set Si = fxj j j = i; i + 1; . . . g is de…ned as the greatest lower bound for Si .
26
If the limit does not exist, the proof applies to the limit inferior of s (w; r (k)) for k ! 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.9. Appendix 115

Taking limits on both sides gives

0 u0 (w s (w; r (k))) u0 (w)

lim u (c(w; k)) = (1 + ) lim = (1 + ) lim = 0;
k!0 k!0 1 + r(k) k!0 1 + r(k)

where the second equality comes from the fact that we are in case 2 and the
third comes from (3.48). But since u0 (c) > 0 and u00 (c) < 0 for all c > 0;
limk!0 u0 (c(w; k)) = 0 requires limk!0 c(w; k) = 1, as was to be shown:
In both Case 1 and Case 2 we thus have that k ! 0 implies s (w; r (k)) =k !
1. Since s (w; r (k)) =k is a continuous function of k; there must be at least one
k > 0 such that (3.47) holds (as illustrated by the two graphs in Fig. 3.14).
Now, to prove (i) of Proposition 2, consider an arbitrary kt > 0: We have
w(kt ) > 0: In (3.47), let w = w(kt ). By Lemma C1, (3.47) has a solution k > 0.
Set kt+1 = k: Starting with t = 0; from a given k0 > 0 we thus …nd a k1 > 0 and
letting t = 1; from the now given k1 we …nd a k2 and so on. The resulting in…nite
sequence fkt g1t=0 is an equilibrium path. In this way we have proved existence of
an equilibrium path if k0 > 0: Thereby (i) of Proposition 2 is proved:
But what if k0 = 0? Then, if f (0) = 0; no temporary equilibrium is possible in
period 0, in view of (ii) of Proposition 1; hence there can be no equilibrium path.
Suppose f (0) > 0: Then w(k0 ) = w(0) = f (0) > 0; as explained in Technical
Remark in Section 3.4. Let w in equation (3.47) be equal to f (0): By Lemma
C1 this equation has a solution k > 0. Set k1 = k: Letting period 1 be the new
initial period, we are back in the case with initial capital positive. This proves
(ii) of Proposition 2.

E. Su¢ cient conditions for certain properties of the transition curve

Positive slope everywhere For convenience we repeat here the condition
(3.36):
1 1
> ; (*)
1 1 + (1 + ) (1 + f 0 (k) ) 1
where we have substituted 1= : In Section 3.5.3 we claimed that in the
CRRA-CES case this condition is su¢ cient for the transition curve to be posi-
tively sloped everywhere. We here prove the claim.
Consider an arbitrary kt > 0 and let w w(kt ) > 0: Knowing that w0 (kt ) > 0
for all kt > 0; we can regard kt+1 as directly linked to w: With k representing
kt+1 , k must satisfy the equation k = s(w; r(k))=(1 + n): A su¢ cient condition
for this equation to implicitly de…ne k as an increasing function of w is also a
su¢ cient condition for the transition curve to be positively sloped for all kt > 0:
When u(c) belongs to the CRRA class, by (3.15) with 1= ; we have
1
s(w; r(k)) = [1 + (1 + ) (1 + r(k))1 ] w: The equation k = s(w; r(k))=(1 + n)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

116 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

then implies
w
= k 1 + (1 + ) R(k)1 h(k); (3.49)
1+n
where R(k) 1 + r(k) 1 + f 0 (k) > 0 for all k > 0: It remains to provide a
0
su¢ cient condition for obtaining h (k) > 0 for all k > 0: We have

h0 (k) = 1 + (1 + ) R(k)1 [1 (1 ) (k)] ; (3.50)

since (k) kR0 (k)=R(k) > 0; the sign being due to R0 (k) = f 00 (k) < 0: So
h (k) > 0 if and only if 1 (1 ) (k) > (1+ ) R(k) 1 ; a condition equivalent
0

to
1 1
> : (3.51)
(k) 1 + (1 + ) R(k) 1
To make this condition more concrete, consider the CES production function

f (k) = A( k + 1 ); A > 0; 0 < < 1; < 1: (3.52)

Then f 0 (k) = A (f (k)=k)1 and de…ning (k) f 0 (k)k=f (k) we …nd

(1 (k))f 0 (k)
(k) = (1 ) (1 )(1 (k)) < 1 ; (3.53)
1 + f 0 (k)
where the …rst inequality is due to 0 1 and the second to 0 < (k) < 1;
which is an implication of strict concavity of f combined with f (0) 0: Thus,
(k) 1 > (1 ) 1 so that if (*) holds for all k > 0; then so does (3.51), i.e.,
h0 (k) > 0 for all k > 0: We have hereby shown that (*) is su¢ cient for the
transition curve to be positively sloped everywhere.

Transition curve steep for k small Here we specialize further and consider
the CRRA-Cobb-Douglas case: u(c) = (c1 1)=(1 ); > 0; and f (k) = Ak ,
A > 0; 0 < < 1. In the prelude to Proposition 4 in Section 3.5 it was claimed
that if this combined utility and technology condition holds at least for small k;
then (ii) of (A3) is satis…ed. We now show this.
Letting ! 0 in (3.52) gives the Cobb-Douglas function f (k) = Ak (this
is proved in the appendix to Chapter 4). With = 0; clearly (1 ) 1 = 1
1
>1 ; where > 0: This inequality implies that (*) above holds and
so the transition curve is positively sloped everywhere. As an implication there
is a transition function, '; such that kt+1 = '(kt ); '0 (kt ) > 0: Moreover, since
f (0) = 0; we have, by Lemma 5, limkt !0 '(kt ) = 0:
Given the imposed CRRA utility, the fundamental di¤erence equation of the
model is
w(kt )
kt+1 = (3.54)
(1 + n) [1 + (1 + ) R(kt+1 )1 ]

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.9. Appendix 117

or, equivalently,
w(kt )
h(kt+1 ) = ;
1+n
where h(kt+t ) is de…ned as in (3.49). By implicit di¤erentiation we …nd h0 (kt+1 )'0 (kt )
= w0 (kt )=(1 + n); i.e.,
w0 (kt )
'0 (kt ) = > 0:
(1 + n)h0 (kt+1 )
If k > 0 is a steady-state value of kt ; (3.54) implies
w(k )
1 + (1 + ) R(k )1 = ; (3.55)
(1 + n)k
and the slope of the transition curve at the steady state will be
w0 (k )
'0 (k ) = > 0: (3.56)
(1 + n)h0 (k )
If we can show that such a k > 0 exists, is unique, and implies '0 (k ) < 1; then
the transition curve crosses the 45 line from above, and so (ii) of (A3) follows in
view of limkt !0 = 0.
De…ning x(k) f (k)=k = Ak 1 ; where x0 (k) = ( 1)Ak 2 < 0; and using
that f (k) = Ak ; we have R(k) = 1 + x(k) and w(k)=k = (1 )x(k):
Hence, (3.55) can be written
1
1 + (1 + ) (1 + x )1 = x; (3.57)
1+n
where x = x(k ): It is easy to show graphically that this equation has a unique
solution x > 0 whether < 1; = 1; or > 1: Then k = (x =A)1=( 1) > 0 is
also unique.
By (3.50) and (3.57),
1 1
h0 (k ) = 1 + ( x 1) [1 (1 ) (k )] > 1 + ( x 1)(1 (k ))
1+n 1+n
1
1+( x 1) ;
1+n
where the …rst inequality is due to > 0 and the second to the fact that (k)
1 in view of (3.53) with = 0 and (k) = : Substituting this together with
w0 (k ) = (1 ) x into (3.56) gives
x
0 < '0 (k ) < < 1; (3.58)
1+n+ x
as was to be shown.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

118 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

The CRRA-Cobb-Douglas case is well-behaved For the case of CRRA

utility and Cobb-Douglas technology with CRS, existence and uniqueness of a
steady state has just been proved. Asymptotic stability follows from (3.58). So
the CRRA-Cobb-Douglas case is well-behaved.

3.10 Exercises
3.1 The dynamic accounting relation for a closed economy is
Kt+1 = Kt + S N (*)
where Kt is the aggregate capital stock and StN is aggregate net saving. In the
Diamond model, let S1t be aggregate net saving of the young in period t and
S2t aggregate net saving of the old in the same period. On the basis of (*)
give a direct proof that the link between two successive periods takes the form
kt+1 = st =(1+n); where st is the saving of each young, n is the population growth
rate, and kt+1 is the capital/labor ratio at the beginning of period t + 1. Hint:
by de…nition, the increase in …nancial wealth is the same as net saving (ignoring
gifts).
3.2 Suppose the production function in Diamond’s OLG model is Y = A( K +
(1 )L )1= ; A > 0; 0 < < 1; < 0; and A 1= < 1+n. a) Given k K=L; …nd
the equilibrium real wage, w(k): b) Show that w(k) < (1 + n)k for all k > 0: Hint:
consider the roof. c) Comment on the implication for the long-run evolution of
the economy. Hint: consider the ceiling.
3.3 (multiple temporary equilibria with self-ful…lling expectations) Fig. 3.10
shows the transition curve for a Diamond OLG model with u(c) = c1 =(1 );
p 1=p
= 8; = 0:4; n = 0:2; = 0:6, f (k) = A(bk + 1 b) ; A = 7; b = 0:33;
p = 0:4:
a) Let t = 0: For a given k0 slightly below 1, how many temporary equilibria
with self-ful…lling expectations are there?
b) Suppose the young in period 0 expect the real interest rate on their saving
to be relatively low. Describe by words the resulting equilibrium path in
this case. Comment (what is the economic intuition behind the path?).
c) In the …rst sentence under b), replace “low”by “high”. How is the answer
to b) a¤ected? What kind of di¢ culty arises?
3.4 (plotting the transition curve by MATLAB) This exercise requires compu-
tation on a computer. You may use MATLAB OLG program.27
27
Made by Marc P. B. Klemp and available at the address:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.10. Exercises 119

Figure 3.10: Transition curve for Diamond’s OLG model in the case described in Ex-
ercise 3.3.

a) Enter the model speci…cation from Exercise 3.3 and plot the transition
curve.

b) Plot examples for two other values of the substitution parameter: p = 1:0
and p = 0:5: Comment.

c) Find the approximate largest lower bound for p such that higher values of
p eliminates multiple equilibria.

d) In continuation of c), what is the corresponding elasticity of factor substi-

tution, ? Hint: as shown in §4.4, the formula is = 1=(1 p):

e) The empirical evidence for industrialized countries suggests that 0:4 < <
1:0: Is your from d) empirically realistic? Comment.

3.5 (one stable and one unstable steady state) Consider the following Diamond
model: u(c) = ln c; = 2:3; n = 2:097; = 1:0; f (k) = A(bk p + 1 b)1=p ; A = 20;
b = 0:5; p = 1:0:
http://www.econ.ku.dk/okocg/Computation/main.htm.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

120 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

a) Plot the transition curve of the model. Hint: you may use either a program
like MATLAB OLG Program (available on the course website) or …rst a
little algebra and then Excel (or similar simple software).

b) Comment on the result you get. Will there exist a poverty trap? Why or
why not?

c) At the stable steady state calculate numerically the output-capital ratio,

the aggregate saving-income ratio, the real interest rate, and the capital
income share of gross national income.

d) Brie‡y discuss how your results in c) comply with your knowledge of cor-
responding empirical magnitudes in industrialized Western countries?

e) There is one feature which this model, as a long-run model, ought to incor-
porate, but does not. Extend the model, taking this feature into account,
and write down the fundamental di¤erence equation for the extended model
in algebraic form.

f) Plot the new transition curve. Hint: given the model speci…cation, this
should be straightforward if you use Excel (or similar); and if you use MAT-
LAB OLG Program, note that by a simple “trick”you can transform your
new model into the “old”form.

g) The current version of the MATLAB OLG Program is not adapted to this
question. So at least here you need another approach, for instance based on
a little algebra and then Excel (or similar simple software). Given k0 = 10;
calculate numerically the time path of kt and plot the time pro…le of kt ; i.e.,
the graph (t; kt ) in the tk-plane. Next, do the same for k0 = 1: Comment.

3.6 (dynamics under myopic foresight)

(incomplete) Show the possibility of a chaotic trajectory.
3.7 Given the period utility function is CRRA, derive the saving function of the
young in Diamond’s OLG model. Hint: substitute the period budget constraints
into the Euler equation.
3.8 Short questions a) A steady-state capital-labor ratio can be in the “dy-
namically e¢ cient” region or in the “dynamically ine¢ cient” region. How are
the two mentioned regions de…ned? b) Give a simple characterization of the two
regions. c) The First Welfare Theorem states that, given certain conditions, any
competitive equilibrium ( Walrasian equilibrium) is Pareto optimal. Give a list
of circumstances that each tend to obstruct Pareto optimality of a competitive
equilibrium.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

3.10. Exercises 121

3.9 Consider a Diamond OLG model for a closed economy. Let the utility
discount rate be denoted and let the period utility function be speci…ed as
u (c) = ln c:

a) Derive the saving function of the young. Comment.

b) Let the aggregate production function be a neoclassical production function

with CRS and ignore technological progress. Let Lt denote the number of
young in period t. Derive the fundamental di¤erence equation of the model.

From now, assume that the production function is Y = L + KL=(K + L);

where > 0 and > 0 (as in Problem 2.4).

c) Draw a transition diagram illustrating the dynamics of the economy. Make

sure that you draw the diagram so as to exhibit consistency with the pro-
duction function.

d) Given the above information, can we be sure that there exists a unique and
globally asymptotically stable steady state? Why or why not?

e) Suppose the economy is in a steady state up to and including period t0 > 0:

Then, at the shift from period t0 to period t0 + 1; a negative technology
shock occurs such that the technology level in period t0 + 1 is below that of
period t0 : Illustrate by a transition diagram the evolution of the economy
from period t0 onward. Comment.

f) Let k K=L: In the (t; ln k) plane, draw a graph of ln kt such that the
qualitative features of the time path of ln k before and after the shock,
including the long run, are exhibited.

g) How, if at all, is the real interest rate in the long run a¤ected by the shock?

h) How, if at all, is the real wage in the long run a¤ected by the shock?

i) How, if at all, is the labor income share of national income in the long run
a¤ected by the shock?

j) Explain by words the economic intuition behind your results in h) and i).

3.10

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

122 CHAPTER 3. THE BASIC OLG MODEL: DIAMOND

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 4

A growing economy

In the previous chapter we ignored technological progress. An incontestable fact

of real life in industrialized countries is, however, the presence of a persistent rise
in GDP per capita on average between 1.5 and 2.5 percent per year since 1870
in many developed economies. In regard to UK, USA, and Japan, see Fig. 4.1;
and in regard to Denmark, see Fig. 4.2. In spite of the somewhat dubious quality
of the data from before the Second World War, this observation should be taken
into account in a model which, like the Diamond model, aims at dealing with
long-run issues. For example, in relation to the question of dynamic ine¢ ciency,
cf. Chapter 3, the cut-o¤ value of the steady-state interest rate is the steady-state
GDP growth rate of the economy and this growth rate increases one-to-one with
the rate of technological progress. We shall therefore now introduce technological
progress:

On the basis of a summary of “stylized facts” about growth, Section 4.1

motivates the assumption that technological progress at the aggregate level takes
the Harrod-neutral form. In Section 4.2 we extend the Diamond OLG model by
incorporating this form of technological progress. Section 4.3 extends the concept
of the golden rule to allow for the existence of technological progress. In Section
4.4 what is known as the marginal productivity theory of factor income shares is
addressed. In this connection an expedient analytical tool, the elasticity of factor
substitution, is presented. Section 4.5 goes into detail with the special case of a
constant elasticity of factor substitution (the CES production function). Finally,
Section 4.6 concludes.

123
 124
GDP per capita in United States, United KingdomCHAPTER 4. A GROWING ECONOMY
and Japan (1870-2010)

Figure 4.1: GDP per capita in USA, UK, and Japan 1870-2010. Source: Bolt and van
Zanden
Sources: Bolt,(2013).
J. and J. L. van Zanden (2013): The First Update of the Maddison Project; Re-Estimating
Growth Before 1820. Maddison Project Working Paper 4.
4.1 Harrod-neutrality and Kaldor’s stylized facts
Suppose the technology changes over time in such a way that we can write the
aggregate production function as

Yt = F (Kt ; Tt Lt ); (4.1)

where the level of technology is represented by the factor Tt which is growing over
time, and where Yt ; Kt ; and Lt stand for output, capital input, and labor input,
respectively. When technological change takes this purely “labor-augmenting”
form, it is known as Harrod-neutral technological progress.

Kaldor’s stylized facts

The reason that macroeconomists often assume that technological change at the
aggregate level takes the Harrod-neutral form as in (4.1) and not for example
the form Yt = F (Xt Kt ; Tt Lt ) (where both X and T are changing over time), is
the following. You want the long-run properties of the model to comply with
Kaldor’s list of “stylized facts” (Kaldor 1961) concerning the long-run evolution
of industrialized economies. Abstracting from short-run ‡uctuations, Kaldor’s
“stylized facts”are:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 4.1. Harrod-neutrality and Kaldor’s stylized facts 125
GDP and GDP per capita in Denmark (1870-2010)

Figure 4.2: GDP and GDP per capita. Denmark 1870-2006. Sources: Bolt and van
Zanden (2013); Maddison (2010); The Conference Board Total Economy Database
Sources: Bolt, J. and J. L. van Zanden (2013): The First Update of the Maddison Project; Re-Estimating
(2013).
Growth Before 1820. Maddison Project Working Paper 4, Maddison (2010): Statistics on World Population,
GDP and Per Capita GDP, 1-2008 AD, and The Conference Board Total Economy Database (2013).
1. the growth rates in K=L and Y =L are roughly constant;

2. the output-capital ratio, Y =K; the income share of labor, wL=Y; and the
average rate of return, (Y wL K)=K;1 are roughly constant;

3. the growth rate of Y =L can vary substantially across countries for quite
long time.

Ignoring the conceptual di¤erence between the path of Y =L and that of Y

per capita (a di¤erence not so important in this context), the …gures 4.1 and
4.2 illustrate Kaldor’s “fact 1” about the long-run property of the Y =L path for
the more developed countries: Japan had an extraordinarily high growth rate
for a couple of decades after World War II, usually explained by fast technology
transfer from the most developed countries (the catching-up process which can
only last until the technology gap is eliminated). Fig. 4.3 gives rough support
1
In this formula w is the real wage and is the capital depreciation rate. Land (and/or
similar natural resources) is ignored. For countries where land is a quantitatively important
production factor, the denominator should be replaced by K + pJ J, where pJ is the real price
of land, J:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

126 CHAPTER 4. A GROWING ECONOMY

for that part of Kaldor’s “fact 2” which claims long-run constancy of the labor
income share. The third fact is a fact well documented empirically.2

1
Denmark
USA
0.8
Labor’s share of income

0.6

0.4

0.2

0
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Figure 4.3: Labor’s share of GDP in USA (1950-2011) and Denmark (1970-2011).
Source: Feenstra, Inklaar and Timmer (2013), www.ggdc.net/pwt.

It is fair to add, however, that the claimed regularities 1 and 2 do not …t

all developed countries equally well. While Solow’s growth model (Solow, 1956)
can be seen as the …rst successful attempt at building a model consistent with
Kaldor’s “stylized facts”, Solow himself once remarked about them: “There is no
doubt that they are stylized, though it is possible to question whether they are
facts” (Solow, 1970). Recently, several empiricists have questioned the methods
which standard national income accounting applies to separate the income of
entrepreneurs, sole proprietors, and unincorporated businesses into labor and
capital income. It is claimed that these methods obscure a tendency in recent
decades of the labor income share to fall.
Notwithstanding these ambiguities, it is de…nitely a fact that many long-run
models are constructed so a to comply with Kaldor’s stylized facts. Let us brie‡y
take a look at the Solow model (in discrete time) and check its consistency with
Kaldor’s “stylized facts”. The point of departure of the Solow model, and many
other growth models, is the aggregate dynamic resource constraint for a closed
economy:

Kt+1 Kt = It Kt = St Kt Yt Ct Kt , K0 > 0 given, (4.2)

2
For a summary, see Pritchett (1997).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.1. Harrod-neutrality and Kaldor’s stylized facts 127

where It is gross investment, which in a closed economy equals gross saving, St

Yt Ct ; is a constant capital depreciation rate, 0 1:3

The Solow model and Kaldor’s stylized facts

As is well-known, the Solow model postulates a constant aggregate saving-income

ratio, s^; so that St = s^Yt ; 0 < s^ < 1.4 Further, the model assumes that the aggre-
gate production function is neoclassical and features Harrod-neutral technological
progress. So, let F in (4.1) be Solow’s production function. To this Solow adds
assumptions of CRS and exogenous geometric growth in both the technology
level T and the labour force L; i.e., Tt = T0 (1 + g)t , g 0; and Lt = L0 (1 + n)t ;
n > 1. In view of CRS, we have Y = F (K; AL) = T LF (k; ~ 1) T Lf (k);
~ where
~ 0
k K=(T L) is the e¤ective capital-labor ratio while f > 0 and f < 0: 00

Substituting St = s^Yt into Kt+1 Kt = St Kt ; dividing through by Tt (1 +

g)Lt (1 + n) and rearranging gives the “law of motion”of the Solow economy:

s^f (k~t ) + (1 )k~t

k~t+1 = '(k~t ): (4.3)
(1 + g)(1 + n)

De…ning G (1 + g)(1 + n); we have '0 (k) ~ = (^ ~ +1

sf 0 (k) ~
)=G > 0 and '00 (k)
00 ~ 0 ~
= s^f (k)=G < 0: If G > 1 and f satis…es the Inada conditions limk!0
~ f (k)
= 1 and limk!1 ~
~ = 0; there is a unique and globally asymptotically stable
f 0 (k)
steady state k~ > 0: The transition diagram looks entirely as in Fig. 3.4 of the
previous chapter (ignoring the tildes).5 The convergence of k~ to k~ implies that
in the long run we have K=L = k~ T and Y =L = f (k~ )T: Both K=L and Y =L are
consequently growing at the same constant rate as T; the rate g. And constancy of
k~ implies that Y =K = f (k)=~ k~ is constant and so is the labor income share, wL=Y
~ kf
= (f (k) ~ 0 (k))=f
~ ~ and hence also the net rate of return, (1 wL=Y )Y =K :
(k);
It follows that the Solow model complies with the stylized facts 1 and 2 above.
Many di¤erent models do that. What these models must then have in common
is a capability of generating balanced growth.
3
In both (4.1) and (4.2) it is implicitly assumed, as is usual in simple macroeconomic models,
that technological progress is disembodied rather than embodied, a distinction described in
Section 2.2 of Chapter 2.
4
Note that s^ is a ratio while the s in the Diamond model stands for the saving per young.
5
What makes the Solow model so easily tractable compared to the Diamond OLG model
is the constant saving-income ratio which makes the transition function essentially dependent
only on the production function in intensive form. Owing to dimishing marginal productivity
of capital, this is a strictly concave function. Anyway, the Solow model emerges as a special
case of the Diamond model, see Exercise IV.??.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

128 CHAPTER 4. A GROWING ECONOMY

Balanced growth
With Kt , Yt , and Ct denoting aggregate capital, output, and consumption as
above, we de…ne a balanced growth path the following way:
DEFINITION 1 A balanced growth path is a path f(Kt ; Yt ; Ct )g1t=0 along which the
variables Kt ; Yt ; and Ct are positive and grow at constant rates (not necessarily
positive).
At least for a closed economy there is a general equivalence relationship be-
tween balanced growth and constancy of certain key ratios like Y =K and C=Y .
This relationship is an implication of accounting based on the above aggregate
dynamic resource constraint (4.2).
For an arbitrary variable xt 2 R++ ; we de…ne xt xt xt 1 : Whenever
xt 1 > 0; the growth rate of x from t 1 to t; denoted gx (t); is de…ned by gx (t)
xt =xt 1 . When there is no risk of confusion, we suppress the explicit dating
and write gx x=x:
PROPOSITION 1 (the balanced growth equivalence theorem). Let f(Kt ; Yt ; Ct )g1 t=0
be a path along which Kt , Yt ; Ct ; and St ( Yt Ct ) are positive for all t =
0; 1; 2; : : : : Then, given the dynamic resource constraint (4.2), the following holds:
(i) if there is balanced growth, then gY = gK = gC and so the ratios Y =K and
C=Y are constant;
(ii) if Y =K and C=Y are constant, then Y; K; and C grow at the same constant
rate, i.e., not only is there balanced growth but the growth rates of Y; K; and C
are the same.
Proof Consider a path f(Kt ; Yt ; Ct )g1 t=0 along which K, Y; C; and St Y Ct
are positive for all t = 0; 1; 2; : : : :
(i) Suppose the path is a balanced growth path. Then, by de…nition, gY ; gK ;
and gC are constant. Hence, by (4.2), S=K = gK + must be constant, implying6

gS = gK : (*)

By (4.2), Y C + S; and so

Y C S C S C S
gY = = + = gC + gS = gC + gK (by (*))
Y Y Y Y Y Y Y
C Y C C
= gC + gK = (gC gK ) + gK : (**)
Y Y Y
6
The ratio between two positive variables is constant if and only if the variables have the
same growth rate (not necessarily constant or positive). For this and similar simple growth-
arithmetic rules, see Appendix A.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.1. Harrod-neutrality and Kaldor’s stylized facts 129

Let us provisionally assume that gC 6= gK : Then (**) gives

C gY gK
= ; (***)
Y gC gK
a constant since gY ; gK ; and gC are constant. Constancy of C=Y requires gC = gY ;
hence, by (***), C=Y = 1, i.e., C = Y: In view of Y C + S, however, this
implication contradicts the given condition that S > 0: Hence, our provisional
assumption and its implication (***) are falsi…ed. Instead we have gC = gK : By
(**), this implies gY = gK = gC ; but now without the condition C=Y = 1 being
implied. It follows that Y =K and C=Y are constant.
(ii) Suppose Y =K and C=Y are positive constants. Applying that the ratio
between two variables is constant if and only if the variables have the same (not
necessarily constant or positive) growth rate, we can conclude that gY = gK = gC .
By constancy of C=Y follows that S=Y 1 C=Y is constant. So gS = gY = gK ;
which in turn implies that S=K is constant. By (4.2),
S K+ K
= = gK + ;
K K
so that also gK is constant. This, together with constancy of Y =K and C=Y;
implies that also gY and gC are constant.
Remark. It is part (i) of the proposition which requires the assumption S > 0 for
all t 0: If S = 0; we would have gK = and C Y S = Y; hence gC = gY for
all t 0: Then there would be balanced growth if the common value of gC and gY
had a constant growth rate. This growth rate, however, could easily di¤er from
that of K: Suppose Y = AK L1 ; 0 < < 1; gA = and gL = n; where and
n are constants. We would then have 1 + gC = 1 + gY = (1 + )(1 ) (1 + n)1 ;
which could easily be larger than 1 and thereby di¤erent from 1 + gK = 1 1
so that (i) no longer holds.
It is part (ii) of the proposition which requires the assumption of a closed
economy. In an open economy we do not necessarily have I = S; hence constancy
of S=K no longer implies constancy of gK = I=K :
For many long-run closed-economy models, including the Diamond OLG model,
it holds that if and only if the dynamic system implied by the model is in a steady
state, will the economy feature balanced growth, cf. Proposition 4 below. There
exist cases, however, where this equivalence between steady state and balanced
growth does not hold (some open economy models and some models with em-
bodied technological change). Hence, we shall maintain a distinction between the
two concepts.
Note that Proposition 1 pertains to any model for which (4.2) is valid. No
assumption about market form and economic agents’behavior are involved. And

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

130 CHAPTER 4. A GROWING ECONOMY

except for the assumed constancy of the capital depreciation rate , no assumption
about the technology is involved, not even that constant returns to scale is present.
Proposition 1 suggests that if one accepts Kaldor’s stylized facts as a rough
description of more than a century’s growth experience and therefore wants the
model to be consistent with them, one should construct the model so that it can
generate balanced growth.

Balanced growth requires Harrod-neutrality

Our next proposition states that for a model to be capable of generating balanced
growth, technological progress must take the Harrod-neutral form (i.e., be labor-
augmenting). Also this proposition holds in a fairly general setting, but not as
general as that of Proposition 1. Constant returns to scale and a constant growth
rate in the labor force, two aspects about which Proposition 1 is silent, will now
have a role to play.7
Consider an aggregate production function

@ F~
Yt = F~ (Kt ; ALt ; t); A > 0; > 0; (4.4)
@t
where F~ is homogeneous of degree one w.r.t. the …rst two arguments (CRS) and
A is a constant that depends on measurement units. The third argument, t;
represents technological progress: as time proceeds, unchanged inputs of capital
and labor result in more and more output. Let the labor force grow at a constant
rate n;
Lt = L0 (1 + n)t ; n > 1; (4.5)
where L0 > 0. The Japanese economist Hirofumi Uzawa (1928-) is famous for
several contributions, not least his balanced growth theorem (Uzawa 1961), which
we here state in a modernized form.
PROPOSITION 2 (Uzawa’s balanced growth theorem). Let f(Kt ; Yt ; Ct )g1 t=0 be a
path along which Yt ; Kt , Ct , and St Yt Ct are positive for all t = 0; 1; 2;. . . ,
and satisfy the dynamic resource constraint (4.2), given the production function
(4.4) and the labor force (4.5). Assume (1 + g)(1 + n) > 1 . Then:
(i) a necessary condition for this path to be a balanced growth path is that along
the path it holds that
Yt = F~ (Kt ; Tt Lt ; 0); (4.6)
where Tt = A(1 + g)t with 1 + g (1 + gY )=(1 + n); gY being the constant growth
rate of output along the balanced growth path;
7
On the other hand we do not imply that CRS is always necessary for a balanced growth
path (see Exercise 4.??).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.1. Harrod-neutrality and Kaldor’s stylized facts 131

(ii) for any g 0 such that there is a q > (1 + g)(1 + n) (1 ) with the
property that the production function F~ in (4.4) allows an output-capital ratio
equal to q at t = 0 (i.e., F~ (1; k~ 1 ; 0) = q for some real number k~ > 0), a su¢ cient
condition for F~ to be consistent with a balanced growth path with output-capital
ratio equal to q is that F~ can be written as in (4.6) with Tt = A(1 + g)t .
Proof (i) Suppose the given path f(Kt ; Yt ; Ct )g1t=0 is a balanced growth path.
By de…nition, gK and gY are then constant so that Kt = K0 (1 + gK )t and Yt
= Y0 (1 + gY )t : With t = 0 in (4.4) we then have

Yt (1 + gY ) t
= Y0 = F~ (K0 ; AL0 ; 0) = F~ (Kt (1 + gK ) t ; ALt (1 + n) t ; 0): (4.7)

In view of the assumption that St Yt Ct > 0; we know from (i) of Proposition

1, that Y =K is constant so that gY = gK . By CRS, (4.7) then implies

Yt = F~ (Kt ; A(1 + gY )t (1 + n) t Lt ; 0):

We see that (4.6) holds for Tt = A(1 + g)t with 1 + g (1 + gY )=(1 + n):
(ii) See Appendix B.
The form (4.6) indicates that along a balanced growth path (BGP from
now), technological progress must be purely labor augmenting, that is, Harrod-
neutral. Moreover, by de…ning a new CRS production function F by F (Kt ; Tt Lt )
F~ (Kt ; Tt Lt ; 0); we see that (i) of the proposition implies that at least along the
BGP, we can rewrite the original production function this way:

Yt = F~ (Kt ; ALt ; t) = F~ (Kt ; Tt Lt ; 0) F (Kt ; Tt Lt ): (4.8)

where T0 = A and Tt = T0 (1 + g)t with 1 + g (1 + gY )=(1 + n).

As emphasized also in Chapter 2, presence of Harrod-neutrality says nothing
about what the source of technological progress is. Harrod-neutrality does not
mean that technological change emanates speci…cally from the labor input. It
only means that technical innovations predominantly are such that not only do
labor and capital in combination become more productive, but this happens to
manifest itself such that we can rewrite the aggregate production function as in
(4.8). (Often introductions to economic growth theory focus on the case where
the production function F is Cobb-Douglas. In this case but only in this case
Harrod-neutrality is equivalent to both Hicks-neutrality and Solow-neutrality.)
What is the intuition behind the Uzawa result that for balanced growth to be
possible, technological progress must at the aggregate level have the purely labor-
augmenting form? First, notice that there is an asymmetry between capital and
labor. Capital is an accumulated amount of non-consumed output. In contrast,
labor is a non-produced production factor which in the present context grows in

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

132 CHAPTER 4. A GROWING ECONOMY

an exogenous way. Second, because of CRS, the original production function,

(4.4), implies that
Kt Lt
1 = F~ ( ; ; t): (4.9)
Yt Yt

Now, since capital is accumulated non-consumed output, it tends to inherit the

trend in output such that Kt =Yt must be constant along a BGP (this is what
Proposition 1 is about). Labor does not inherit the trend in output; indeed, the
ratio Lt =Yt is free to adjust as t proceeds. When there is technological progress
(@ F~ =@t > 0) along a BGP, this progress must manifest itself in the form of a
changing Lt =Yt in (4.9) as t proceeds, precisely because Kt =Yt must be constant
along the path. In the “normal” case where @ F~ =@L > 0; the needed change in
L(t)=Y (t) is a fall (i.e., rise in Y (t)=L(t)): This is what (4.9) shows. Indeed, the
fall in Lt =Yt must exactly o¤set the e¤ect on F~ of the rising t; when there is a
…xed capital-output ratio. It follows that along the BGP, Yt =Lt is an increasing
implicit function of t: If we denote this function Tt , we end up with (4.8).
The generality of Uzawa’s theorem is noteworthy. Like Proposition 1, Uzawa’s
theorem is about technically feasible paths, while economic institutions, market
forms, and agents’ behavior are not involved. The theorem presupposes CRS,
but does not need that the technology has neoclassical properties not to speak of
satisfying the Inada conditions. And the theorem holds for exogenous as well as
endogenous technological progress.
A simple implication of the theorem is the following. Let yt denote “labor
productivity”in the sense of Yt =Lt , kt denote the capital-labor ratio, Kt =Lt ; and
ct the consumption-labor ratio, Ct =Lt : We have:

COROLLARY Along a BGP with positive gross saving and the technology level
T growing at a constant rate g 0; output grows at the rate (1 + g)(1 + n) 1
( g + n for g and n “small”) while labor productivity, y; capital-labor ratio, k;
and consumption-labor ratio, c; all grow at the rate g:

Proof That gY = (1 + g)(1 + n) 1 follows from (i) of Proposition 2. As to gy

we have
Yt Y0 (1 + gY )t
yt = = y0 (1 + g)t ;
Lt L0 (1 + n)t

showing that y grows at the rate g: Moreover, y=k = Y =K; which is constant
along a BGP, by (i) of Proposition 1. Hence k grows at the same rate as y:
Finally, also c=y C=Y is constant along a BGP, implying that also c grows at
the same rate as y.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.1. Harrod-neutrality and Kaldor’s stylized facts 133

Factor income shares

There is one facet of Kaldor’s stylized facts which we have not yet related to
Harrod-neutral technological progress, namely the claimed long-run “approxi-
mate” constancy of the income share of labor and the rate of return on capital.
It turns out that, if we assume (a) neoclassical technology, (b) pro…t maximiz-
ing …rms, and (c) perfect competition in the output and factor markets, then
these constancies are inherent in the combination of constant returns to scale
and balanced growth.
To see this, let the aggregate production function be Yt = F (Kt ; Tt Lt ) where
F is neoclassical and has CRS. With wt denoting the real wage at time t; in
equilibrium under perfect competition the labor income share will be
@Yt
wt Lt L
@Lt t F2 (Kt ; Tt Lt )Tt Lt
= = : (4.10)
Yt Yt Yt
When the capital good is nothing but non-consumed output, the rate of return
on capital at time t can be written
Yt wt Lt Kt Yt wt Lt Yt
rt = = : (4.11)
Kt Yt Kt
Since land as a production factor is ignored, gross capital income equals non-
labor income, Yt wt Lt : Denoting the gross capital income share by t ; we thus
have
Yt wt Lt F (Kt ; Tt Lt ) F2 (Kt ; Tt Lt )Tt Lt
t = =
Yt Yt
@Yt
F1 (Kt ; Tt Lt )Kt Kt Kt
= = @Kt = (rt + ) ; (4.12)
Yt Yt Yt
where the third equality comes from Euler’s theorem8 and the last from (4.11.
PROPOSITION 3 (factor income shares) Suppose a given path f(Kt ; Yt ; Ct )g1 t=0
is a BGP with positive saving in this competitive economy. Then t = ; a
constant 2 (0; 1). The labor income share will be 1 and the rate of return on
capital q ; where q is the constant output-capital ratio along the BGP.
Proof We have Yt = F (Kt ; Tt Lt ) = Tt Lt F (k~t ; 1) Tt Lt f (k~t ): From Proposition
1 follows that along the given BGP, Yt =Kt is some constant, q. Since Yt =Kt =
f (k~t )=k~t and f 00 < 0; this implies k~t constant, say equal to k~ : Along the BGP,
@Yt =@Kt (= f 0 (k~t )) thus equals the constant f 0 (k~ ). From (4.12) then follows
8
Indeed, from Euler’s theorem follows that F1 K+F2 T L = F (K; T L); when F is homogeneous
of degree one:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

134 CHAPTER 4. A GROWING ECONOMY

that t = f 0 (k~ )=q : Moreover, 0 < < 1; since 0 < is implied by f 0 > 0;
and < 1 is implied by the fact that q = Y =K = f (k~ )=k~ > f 0 (k~ ); in view of
f 00 < 0 and f (0) 0: So, by the …rst equality in (4.12), the labor income share
can be written wt Lt =Yt = 1 t = 1 : Consequently, by (4.11), the rate of
return on capital is rt = (1 wt Lt =Yt )Yt =Kt = q :
Although this proposition implies constancy of the factor income shares under
balanced growth, it does not determine them. The proposition expresses the
factor income shares in terms of the unknown constants and q: These constants
will generally depend on the e¤ective capital-labor ratio in steady state, k~ ; which
will generally be an unknown as long as we have not formulated a theory of saving.
This takes us back to Diamond’s OLG model which provides such a theory.

4.2 The Diamond OLG model with Harrod-neutral

technological progress
Recall from the previous chapter that in the Diamond OLG model people live in
two periods, as young and as old. Only the young work and each young supplies
one unit of labor inelastically. The period utility function, u(c); satis…es the
No Fast Assumption. The saving function of the young is st = s(wt ; rt+1 ). We
now include Harrod-neutral technological progress in the aggregate production
function of the Diamond model:

Yt = F (Kt ; Tt Lt ); (4.13)

where F is neoclassical with CRS and Tt represents the level of technology in

period t: We assume that Tt grows at a constant exogenous rate, that is,

Tt = T0 (1 + g)t ; g 0: (4.14)

The initial level of technology, T0 ; is historically given. Employment equals Lt

which is the number of young, growing at the constant exogenous rate n > 1:
Suppressing for a while the explicit dating of the variables, in view of CRS
w.r.t. K and T L, we have

Y K ~ 1) ~
y~ = F( ; 1) = F (k; f (k); f 0 > 0; f 00 < 0;
TL TL

where T L is labor input in e¢ ciency units and k~ K=(T L) is known as the

e¤ective or technology-corrected capital-labor ratio - also sometimes called the
e¤ective capital intensity. There is perfect competition in all markets. In each

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.2. The Diamond OLG model with Harrod-neutral technological progress 135

period the representative …rm maximizes pro…t, = F (K; T L) r^K wL: With
respect to capital this leads to the …rst-order condition
h i
@ T Lf ( ~
k)
@Y ~ =r+ ;
= = f 0 (k) (4.15)
@K @K
where is a constant capital depreciation rate, 0 1: With respect to labor
we get the …rst-order condition
h i
@ T Lf ( ~
k) h i
@Y ~ ~ k~ T = w:
= = f (k) f 0 (k) (4.16)
@L @L
In view of f 00 < 0; a k~ satisfying (4.15) is unique. Let us denote its value in
period t; k~td : Assuming equilibrium in the factor markets, this desired e¤ective
capital-labor ratio equals the e¤ective capital-labor ratio from the supply side,
k~t Kt =(Tt Lt ) kt =Tt ; which is predetermined in every period. The equilibrium
interest rate and real wage in period t are thus given by

rt = f 0 (k~t ) r(k~t ); where r0 (k~t ) = f 00 (k~t ) < 0; (4.17)

h i
wt = f (k~t ) f 0 (k~t )k~ Tt w(~ k~t )Tt ; where w~ 0 (k~t ) = k~t f 00 (k~t ) > 0: (4.18)

~ k~t ) = wt =Tt is known as the technology-corrected real wage.

Here, w(

The equilibrium path

The aggregate capital stock at the beginning of period t+1 must still be owned by
the old generation in that period and thus equal the aggregate saving these people
did as young in the previous period. Hence, as before, Kt+1 = st Lt = s(wt ; rt+1 )Lt :
In view of Kt+1 k~t+1 Tt+1 Lt+1 = k~t+1 Tt (1 + g)Lt (1 + n); together with (4.17)
and (4.18), we get
~ k~t )Tt ; r(k~t+1 ))
s(w(
k~t+1 = : (4.19)
Tt (1 + g)(1 + n)
This is the general version of the law of motion of the Diamond OLG model with
Harrod-neutral technological progress.
For the model to comply with Kaldor’s “stylized facts”, the model should be
capable of generating balanced growth. Essentially, this capability is equivalent
to being able to generate a steady state. In the presence of technological progress
this latter capability requires a restriction on the lifetime utility function, U: In-
deed, we see from (4.19) that the model is consistent with existence of a steady
state only if the time-dependent technology level, Tt , in the numerator and de-
nominator cancels out. This requires that the saving function is homogeneous of

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

136 CHAPTER 4. A GROWING ECONOMY

~ k~t )Tt ; r(k~t+1 )) = s(w(

degree one in its …rst argument such that s(w( ~ k~t ); r(k~t+1 ))Tt :
In turn this is so if and only if the lifetime utility function of the young is ho-
mothetic. So, in addition to the No Fast Assumption from Chapter 3, we impose
the Homotheticity Assumption:

the lifetime utility function U is homothetic. (A4)

This property entails that if the value of the “endowment”, here the human wealth
wt ; is multiplied by a > 0, then the chosen c1t and c2t+1 are also multiplied by
this factor (see Appendix C); it then follows that st is multiplied by as well.
Letting = 1=(w( ~ k~t )Tt ); (A4) thus allows us to write

st = s(1; r(k~t+1 ))w(

~ k~t )Tt s^(r(k~t+1 ))w(
~ k~t )Tt ; (4.20)

where s^(r(k~t+1 )) is the saving-wealth ratio of the young. The distinctive feature
is that this saving-wealth ratio is independent of wealth (but in general it depends
on the interest rate). By (4.19), the law of motion of the economy reduces to

s^(r(k~t+1 ))
k~t+1 = ~ k~t ):
w( (4.21)
(1 + g)(1 + n)
The equilibrium path of the economy can be analyzed in a similar way as in
the case of no technological progress. In the assumptions (A2) and (A3) from
Chapter 3 we replace k by k~ and 1 + n by (1 + g)(1 + n). As a generalization
of Proposition 4 from Chapter 3, these generalized versions of (A2) and (A3),
together with the No Fast Assumption (A1) and the Homotheticity Assumption
(A4), guarantee that there exists at least one locally asymptotically stable steady
state k~ > 0: That is, given these assumptions, we have k~t ! k~ for t ! 1 and so
the system will sooner or later settle down in a steady state. The convergence of
k~ implies convergence of many key variables, for instance the equilibrium factor
prices given in (4.17) and (4.18). We see that, for t ! 1;

rt = f 0 (k~t ) ! f 0 (k~ ) r ; and

h i
wt = f (k~t ) k~t f 0 (k~t ) Tt ! [f (k ) k f 0 (k )] Tt w~ Tt = w~ T0 (1 + g)t :

The prediction of the model is now that the economy will in the long run
behave in accordance with Kaldor’s stylized facts. Indeed, in many models, in-
cluding the present one, convergence toward a steady state is equivalent to saying
that the time path of the economy converges toward a BGP. In the present case,
with perfect competition, the implication is that in the long run the economy will
be consistent with Kaldor’s stylized facts.
The claimed equivalence follows from:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.2. The Diamond OLG model with Harrod-neutral technological progress 137

PROPOSITION 4 Consider a Diamond economy with Harrod-neutral techno-

logical progress at the constant rate g 0 and positive gross saving for all t.
Then:
(i) if the economy features balanced growth, then it is in a steady state;
(ii) if the economy is in a steady state, then it features balanced growth.
Proof (i) Suppose the considered economy features balanced growth. Then, by
Proposition 1, Y =K is constant. As Y =K = y~=k~ = f (k)= ~ k;
~ also k~ is constant.
Thereby the economy is in a steady state. (ii) Suppose the considered economy
is in a steady state, i.e., given (4.21), k~t = k~t+1 = k~ for some k~ > 0. The
constancy of k~ K=(T L) and y~ Y =(T L) = f (k) ~ implies that both gK and gY
equal gT L = (1 + g)(1 + n) 1 > 0: As K and Y thus grow at the same rate, Y =K
is constant. With S Y C; constancy of S=K = ( K + K)=K = gK + ,
implies constancy of S=K so that S also grows at the rate gK and thereby at the
same rate as output. Hence S=Y is constant. Because C=Y 1 S=Y; also C
grows at the constant rate gY : All criteria for a balanced growth path are thus
satis…ed.

Figure 4.4: Transition curve for a well-behaved Diamond OLG model with Harrod-
neutral technical progress.

Let us portray the dynamics by a transition diagram. Fig. 4.4 shows a “well-
behaved” case in the sense that there is only one steady state. In the …gure the
initial e¤ective capital-labor ratio, k~0 ; is assumed to be relatively large. This
need not be interpreted as if the economy is highly developed and has a high

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

138 CHAPTER 4. A GROWING ECONOMY

initial capital-labor ratio, K0 =L0 : Indeed, the reason that k~0 K0 =(T0 L0 ) is
large relative to its steady-steady value may be that the economy is “backward”
in the sense of having a relatively low initial level of technology. Growing at a
given rate g; the technology will in this situation grow faster than the capital-
labor ratio, K=L; so that the e¤ective capital-labor ratio declines over time. The
process continues until the steady state is essentially reached with a real interest
rate r = f 0 (k~ ) : This is to remind the reader that from an empirical point of
view, the adjustment towards a steady state can be from above as well as from
below.
The output growth rate in steady state, (1 + g)(1 + n) 1; is sometimes called
the “natural rate of growth”. Since (1 + g)(1 + n) 1 = g + n + gn g + n for
g and n “small”, the natural rate of growth approximately equals the sum of the
rate of technological progress and the growth rate of the labor force. Warning:
When measured on an annual basis, the growth rates of technology and labor
force, g and n; do indeed tend to be “small”, say g = 0:02 and n = 0:005; so
that g + n + gn = 0:0251 0:0250 = g + n: But in the context of models like
Diamond’s, the period length is, say, 30 years. Then the corresponding g and n
will satisfy the equations 1 + g = (1 + g)30 = 1:0230 = 1:8114 and 1 + n = (1 + n)30
= 1:00530 = 1:1614, respectively. We get g + n = 0:973, which is about 10 per
cent smaller than the true output growth rate over 30 years, which is g + n + gn
= 1:104:
We end our account of Diamond’s OLG model with some remarks on a popular
special case of a homothetic utility function.

An example: CRRA period utility

An example of a homothetic lifetime utility function is obtained by letting the
period utility function take the CRRA form introduced in the previous chapter.
Then
c11 1 1
1 c2 1
U (c1 ; c2 ) = + (1 + ) ; > 0: (4.22)
1 1
Recall that the CRRA utility function with parameter has the property that
the (absolute) elasticity of marginal utility of consumption equals the constant
> 0 for all c > 0. Up to a positive linear transformation it is, in fact, the only
period utility function with this property. A proof that the utility function (4.22)
is indeed homothetic is given in Appendix C.
One of the reasons that the CRRA function is popular in macroeconomics is
that in representative agent models, the period utility function must have this
form to obtain consistency with balanced growth and Kaldor’s stylized facts (this
is shown in Chapter 7). In contrast, a model with heterogeneous agents, like the
Diamond model, does not need CRRA utility to comply with the Kaldor facts.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.3. The golden rule under Harrod-neutral technological progress 139

CRRA utility is just a convenient special case leading to homothetic lifetime

utility. And this is what is needed for a BGP to exist and thereby for compatibility
with Kaldor’s stylized facts.
Given the CRRA assumption in (4.22), the saving-wealth ratio of the young
becomes
1
s^(r) = ( 1)=
: (4.23)
1+r
1 + (1 + ) 1+

It follows that s^0 (r) R 0 for Q 1:

When = 1 (the case u(c) = ln c), s^(r) = 1=(2 + ) s^; a constant, and the
law of motion (4.21) thus simpli…es to

1
k~t+1 = ~ k~t ):
w(
(1 + g)(1 + n)(2 + )

We see that in the = 1 case, whatever the production function, k~t+1 enters
only at the left-hand side of the fundamental di¤erence equation, which thereby
~ > 0; the transition curve
reduces to a simple transition function. Since w~ 0 (k)
is positively sloped everywhere. If the production function is Cobb-Douglas, Yt
= Kt (Tt Lt )1 ; then w(~ k~t ) = (1 )k~t : Combining this with = 1 yields a
“well-behaved” Diamond model (thus having a unique and globally asymptoti-
cally stable steady state), cf. Fig. 4.4 above. In fact, as noted in Chapter 3,
in combination with Cobb-Douglas technology, CRRA utility results in “well-
behavedness”whatever the value of > 0.

4.3 The golden rule under Harrod-neutral tech-

nological progress
Given that there is technological progress, consumption per unit of labor is likely
to grow over time. Therefore the de…nition of the golden-rule capital-labor ratio
from Chapter 3 has to be extended to cover the case of growing consumption per
unit of labor. To allow existence of steady states and balanced growth paths, we
maintain the assumption that technological progress is Harrod-neutral, that is,
we maintain (4.13) where the technology level, T; grows at a constant rate g > 0:
DEFINITION 2 The golden-rule capital intensity, k~GR ; is that level of k~
K=(T L) which gives the highest sustainable path for consumption per unit of
labor in the economy.
As before, we let time be discrete but allow the period length to be arbitrary,

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

140 CHAPTER 4. A GROWING ECONOMY

possibly one year for instance. Consumption per unit of labor is

Ct F (Kt ; Tt Lt ) St f (k~t )Tt Lt (Kt+1 Kt + K t )

ct = =
Lt Lt Lt
= f (k~t )Tt (1 + g)Tt (1 + n)k~t+1 + (1 )T k~
h i t t
= f (k~t ) + (1 )k~t (1 + g)(1 + n)k~t+1 Tt :

In a steady state we have k~t+1 = k~t = k~ and therefore

h i
~ + (1
ct = f (k) )k~ (1 + g)(1 + n)k~ Tt ~ t;
c~(k)T

~ is the “technology-corrected” level of consumption per unit of labor

where c~(k)
in steady state. We see that in steady state, consumption per unit of labor will
~ + ln T0 + t ln(1 + g):
grow at the same rate as the technology. Thus, ln ct = ln c~(k)
Fig. 4.5 illustrates.
Since the evolution of technology, parameterized by T0 and g; is exogenous, the
~ This gives the …rst-order
highest possible path of ct is found by maximizing c~(k):
condition
~ = f 0 (k)
c~0 (k) ~ + (1 ) (1 + g)(1 + n) = 0: (4.24)
Assuming, for example, n 0; we have (1 + g)(1 + n) (1 ) > 0 since g > 0:
Then, by continuity the equation (4.24) necessarily has a unique solution in k~ > 0;
if the production function satis…es the condition
~ > (1 + g)(1 + n)
lim f 0 (k) (1 ~
) > lim f 0 (k);
~
k!0 ~
k!1

which is a milder condition than the Inada conditions. Considering the second-
~ = f 00 (k)
order condition c~00 (k) ~ < 0; the k~ satisfying (4.24) does indeed maximize
~ By de…nition, this k~ is the golden-rule capital intensity, k~GR : Thus
c~(k):

f 0 (k~GR ) = (1 + g)(1 + n) 1 g + n; (4.25)

where the right-hand side is the “natural rate of growth”. This says that the
golden-rule capital intensity is that level of the capital intensity at which the net
marginal productivity of capital equals the output growth rate in steady state.

Is dynamic ine¢ ciency a problem in practice? As in the Diamond model

without technological progress, it is theoretically possible that the economy ends
up in a steady state with k~ > k~GR :9 If this happens, the economy is dynamically
9
The proof is analogue to that in Chapter 3 for the case g = 0.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.3. The golden rule under Harrod-neutral technological progress 141

Figure 4.5: The highest sustainable path of consumption is where k~ = k~GR .

ine¢ cient and r < (1 + g)(1 + n) 1 g + n. To check whether dynamic

ine¢ ciency is a realistic outcome in an industrialized economy or not, we should
compare the observed average GDP growth rate over a long stretch of time to the
average real interest rate or rate of return in the economy. For the period after the
Second World War the average GDP growth rate ( g + n) in Western countries
is typically about 3 per cent per year. But what interest rate should one choose?
In simple macro models, like the Diamond model, there is no uncertainty and no
need for money to carry out trades. In such models all assets earn the same rate
of return, r; in equilibrium. In the real world there is a spectrum of interest rates,
re‡ecting the di¤erent risk and liquidity properties of the di¤erent assets. The
expected real rate of return on a short-term government bond is typically less
than 3 per cent per year (a relatively safe and liquid asset). This is much lower
than the expected real rate of return on corporate stock, 7-9 per cent per year.
Our model cannot tell which rate of return we should choose, but the conclusion
hinges on that choice.

Abel et al. (1989) study the problem on the basis of a model with uncertainty.
They show that a su¢ cient condition for dynamic e¢ ciency is that gross invest-
ment, I; does not exceed the gross capital income in the long run, that is I
Y wL: They …nd that for the U.S. and six other major OECD nations this seems
to hold. Indeed, for the period 1929-85 the U.S. has, on average, I=Y = 0:15 and
(Y wL)=Y = 0:29: A similar di¤erence is found for other industrialized coun-
tries, suggesting that they are dynamically e¢ cient. At least in these countries,
therefore, the potential coordination failure laid bare by OLG models does not
seem to have been operative in practice.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

142 CHAPTER 4. A GROWING ECONOMY

4.4 The functional distribution of income

.....Text to be inserted

The neoclassical theory of factor income shares

.....Text to be inserted

How the labor income share depends on the capital-labor ratio

To begin with we ignore technological progress and write aggregate output as
Y = F (K; L); where F is neoclassical with CRS. From Euler’s theorem follows
that F (K; L) = F1 K + F2 L = f 0 (k)K + (f (k) kf 0 (k))L; where k K=L: In
equilibrium under perfect competition we have

Y = r^K + wL;

where r^ = r + = f 0 (k) r^(k) is the cost per unit of capital input and w
= f (k) kf 0 (k) w(k) is the real wage, i.e., the cost per unit of labor input.
The labor income share is
w=^
r
wL f (k) kf 0 (k) w(k) wL k
= SL(k) = = ; (4.26)
Y f (k) f (k) r^K + wL 1 + w=^
k
r

where the function SL( ) is the income share of labor function, w=^ r is the factor
price ratio, and (w=^ r)=k = w=(^rk) is the factor income ratio. As r^ (k) = f 00 (k) < 0
0
0 00
and w (k) = kf (k) > 0; the factor price ratio, w=^ r, is an increasing function
of k:
Suppose that capital tends to grow faster than labor so that k rises over time.
Unless the production function is Cobb-Douglas, this will under perfect competi-
tion a¤ect the labor income share. But apriori it is not obvious in what direction.
By (4.26) we see that the labor income share moves in the same direction as the
factor income ratio, (w=^ r)=k: The latter goes up (down) depending on whether
the percentage rise in the factor price ratio w=^ r is greater (smaller) than the
percentage rise in k. So, if we let E`x g(x) denote the elasticity of a function g(x)
w.r.t. x; we can only say that
w
SL0 (k) R 0 for E`k R 1; (4.27)
r^
respectively. In words: if the production function is such that the ratio of the
marginal productivities of the two production factors is strongly (weakly) sensitive
to the capital-labor ratio, then the labor income share rises (falls) along with a
rise in K=L:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.4. The functional distribution of income 143

Usually, however, the inverse elasticity is considered, namely E`w=^r k (= 1=E`k wr^ ):
This elasticity indicates how sensitive the cost minimizing capital-labor ratio, k;
is to a given factor price ratio w=^r: Under perfect competition E`w=^r k coincides
with what is known as the elasticity of factor substitution (for a general de…n-
ition, see below). The latter is often denoted : In the CRS case, will be a
function of only k so that we can write E`w=^r k = (k): By (4.27), we therefore
have
SL0 (k) R 0 for (k) Q 1;
respectively.
If F is Cobb-Douglas, i.e., Y = K L1 ; 0 < < 1; we have (k) 1; as
shown in Section 4.5. In this case variation in k does not change the labor income
share under perfect competition. Empirically there is not agreement about the
“normal”size of the elasticity of factor substitution for industrialized economies,
but the bulk of studies seems to conclude with (k) < 1 (see below).

Adding Harrod-neutral technical progress We now add Harrod-neutral

technical progress. We write aggregate output as Y = F (K; T L); where F is
neoclassical with CRS, and T = Tt = T0 (1 + g)t . Then the labor income share is

wL w=T w~
= :
Y Y =(T L) y~

The above formulas still hold if we replace k by k~ K=(T L) and w by w~ w=T:

We get
~ R 0 for (k)
SL0 (k) ~ Q 1;

respectively. We see that if (k) ~ < 1 in the relevant range for k;~ then market
forces tend to increase the income share of the factor that is becoming relatively
more scarce, which is e¢ ciency-adjusted labor, T L; if k~ is increasing. And if
instead (k)~ > 1 in the relevant range for k;
~ then market forces tend to decrease
the income share of the factor that is becoming relatively more scarce.
While k empirically is clearly growing, k~ k=T is not necessarily so because
also T is increasing. Indeed, according to Kaldor’s “stylized facts”, apart from
short- and medium-term ‡uctuations, k~ and therefore also r^ and the labor
income share tend to be more or less constant over time. This can happen
whatever the sign of (k~ ) 1; where k~ is the long-run value of the e¤ective
~ Given CRS and the production function f; the elasticity
capital-labor ratio k.
of substitution between capital and labor does not depend on whether g = 0 or
g > 0, but only on the function f itself and the level of K=(T L).
As alluded to earlier, there are empiricists who reject Kaldor’s “facts” as a
general tendency. For instance Piketty (2014) essentially claims that in the very

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

144 CHAPTER 4. A GROWING ECONOMY

long run the e¤ective capital-labor ratio k~ has an upward trend, temporarily
braked by two world wars and the Great Depression in the 1930s. If so, the sign
~
of (k) 1 becomes decisive for in what direction wL=Y will move. Piketty
~ > 1; which means there
interprets the econometric literature as favoring (k)
should be downward pressure on wL=Y . This particular source behind a falling
wL=Y can be questioned, however. Indeed, (k) ~ > 1 contradicts the more general
10
empirical view referred to above.

Immigration
~ Con-
Here is another example that illustrates the importance of the size of (k):
sider an economy with perfect competition and a given aggregate capital stock K
and technology level T (entering the production function in the labor-augmenting
way as above). Suppose that for some reason, immigration, say, aggregate labor
supply, L; shifts up and full employment is maintained by the needed real wage
adjustment. Given the present model, in what direction will aggregate labor in-
come wL = w( ~ L then change? The e¤ect of the larger L is to some extent
~ k)T
o¤set by a lower w brought about by the lower e¤ective capital-labor ratio. In-
~ k~ = kf
deed, in view of dw=d ~ 00 (k)
~ > 0; we have k~ # implies w # for …xed T: So
we cannot apriori sign the change in wL: The following relationship can be shown
(Exercise 4.??), however:

@(wL) ~
(k)
= (1 )w R 0 for (k)
~ Q (k);
~ (4.28)
@L ~
(k)

~
respectively, where a(k) ~ 0 (k)=f
kf ~ (k)~ is the output elasticity w.r.t. capital
which under perfect competition equals the gross capital income share. It follows
that the larger L will not be fully o¤set by the lower w as long as the elasticity
~ exceeds the gross capital income share, (k).
of factor substitution, (k); ~ This
condition seems con…rmed by most of the empirical evidence (see Section 4.5).

The elasticity of factor substitution*

We shall here discuss the concept of elasticity of factor substitution at a more
general level. Fig. 4.6 depicts an isoquant, F (K; L) = Y ; for a given neoclassical
production function, F (K; L); which need not have CRS. Let M RS denote the
marginal rate of substitution of K for L; i.e., M RS = FL (K; L)=FK (K; L):11 At
10
According to Summers (2014), Piketty’s interpretation relies on con‡ating gross and net
returns to capital.
11
When there is no risk of confusion as to what is up and what is down, we use M RS as a
shorthand for the more precise expression M RSKL .

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.4. The functional distribution of income 145

a given point (K; L) on the isoquant curve, M RS is given by the absolute value
of the slope of the tangent to the isoquant at that point. This tangent coincides
with that isocost line which, given the factor prices, has minimal intercept with
the vertical axis while at the same time touching the isoquant. In view of F ( )
being neoclassical, the isoquants are by de…nition strictly convex to the origin.
Consequently, M RS is rising along the curve when L decreases and thereby K
increases. Conversely, we can let M RS be the independent variable and consider
the corresponding point on the indi¤erence curve, and thereby the ratio K=L, as a
function of M RS. If we let M RS rise along the given isoquant, the corresponding
value of the ratio K=L will also rise.

Figure 4.6: Substitution of capital for labor as the marginal rate of substitution in-
creases from M RS to M RS 0 .

The elasticity of substitution between capital and labor is de…ned as the elas-
ticity of the ratio K=L with respect to M RS when we move along a given isoquant,
evaluated at the point (K; L). Let this elasticity be denoted ~ (K; L): Thus,

d(K=L)
M RS d(K=L) K=L
~ (K; L) = = dM RS
: (4.29)
K=L dM RS jY =Y M RS jY =Y

Although the elasticity of factor substitution is a characteristic of the tech-

nology as such and is here de…ned without reference to markets and factor prices,
it helps the intuition to refer to factor prices. At a cost-minimizing point, M RS
equals the factor price ratio w=^ r: Thus, the elasticity of factor substitution will
under cost minimization coincide with the percentage increase in the ratio of the
cost-minimizing factor ratio induced by a one percentage increase in the inverse

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

146 CHAPTER 4. A GROWING ECONOMY

factor price ratio, holding the output level unchanged.12 The elasticity of factor
substitution is thus a positive number and re‡ects how sensitive the capital-labor
ratio K=L is under cost minimization to an increase in the factor price ratio w=^ r
for a given output level: The less curvature the isoquant has, the greater is the
elasticity of factor substitution. In an analogue way, in consumer theory one con-
siders the elasticity of substitution between two consumption goods or between
consumption today and consumption tomorrow, cf. Chapter 3. In that context
the role of the given isoquant is taken over by an indi¤erence curve. That is also
the case when we consider the intertemporal elasticity of substitution in labor
supply, cf. the next chapter.
Calculating the elasticity of substitution between K and L at the point (K; L),
we get
FK FL (FK K + FL L)
~ (K; L) = ; (4.30)
KL [(FL )2 FKK 2FK FL FKL + (FK )2 FLL ]
where all the derivatives are evaluated at the point (K; L): When F (K; L) has
CRS, the formula (4.30) simpli…es to

FK (K; L)FL (K; L) f 0 (k) (f (k) f 0 (k)k)

~ (K; L) = = (k); (4.31)
FKL (K; L)F (K; L) f 00 (k)kf (k)

where k K=L:13 We see that under CRS, the elasticity of substitution depends
only on the capital-labor ratio k, not on the output level. We will now consider the
case where the elasticity of substitution is independent also of the capital-labor
ratio:

4.5 The CES production function*

It can be shown14 that if a neoclassical production function with CRS has a
constant elasticity of factor substitution di¤erent from one, it must be of the
form 1
Y = A K + (1 )L ; (4.32)
where A; ; and are parameters satisfying A > 0, 0 < < 1; and < 1;
6= 0: This function has been used intensively in empirical studies and is called
a CES production function (CES for Constant Elasticity of Substitution). For a
given choice of measurement units, the parameter A re‡ects e¢ ciency (or what
12
This characterization is equivalent to interpreting the elasticity of substitution as the per-
centage decrease in the factor ratio (when moving along a given isoquant) induced by a one-
percentage increase in the corresponding factor price ratio.
13
The formulas (4.30) and (4.31) are derived in Appendix D.
14
See, e.g., Arrow et al. (1961).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.5. The CES production function* 147

is known as total factor productivity) and is thus called the e¢ ciency parameter.
The parameters and are called the distribution parameter and the substitution
parameter, respectively. The restriction < 1 ensures that the isoquants are
strictly convex to the origin. Note that if < 0; the right-hand side of (4.32)
is not de…ned when either K or L (or both) equal 0: We can circumvent this
problem by extending the domain of the CES function and assign the function
value 0 to these points when < 0. Continuity is maintained in the extended
domain (see Appendix E).
By taking partial derivatives in (4.32) and substituting back we get
1 1
@Y Y @Y Y
= A and = (1 )A ; (4.33)
@K K @L L
1 1
where Y =K = A + (1 )k and Y =L = A k +1 : The marginal
rate of substitution of K for L therefore is
@Y =@L 1
M RS = = k1 > 0:
@Y =@K

Consequently,
dM RS 1
= (1 )k ;
dk
where the inverse of the right-hand side is the value of dk=dM RS: Substituting
these expressions into (4.29) gives

1
~ (K; L) = ; (4.34)
1

con…rming the constancy of the elasticity of substitution. Since < 1; > 0

always: A higher substitution parameter, ; results in a higher elasticity of factor
substitution, : And 7 1 for 7 0; respectively.
Since = 0 is not allowed in (4.32), at …rst sight we cannot get = 1 from
this formula. Yet, = 1 can be introduced as the limiting case of (4.32) when
! 0; which turns out to be the Cobb-Douglas function. Indeed, one can show15
that, for …xed K and L;
1
A K + (1 )L ! AK L1 ; for ! 0:

By a similar procedure as above we …nd that a Cobb-Douglas function always

has elasticity of substitution equal to 1; this is exactly the value taken by in
15
Proofs of this and the further claims below are in Appendix E.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

148 CHAPTER 4. A GROWING ECONOMY

(4.34) when = 0. In addition, the Cobb-Douglas function is the only production

function that has unit elasticity of substitution whatever the capital-labor ratio.
Another interesting limiting case of the CES function appears when, for …xed
K and L; we let ! 1 so that ! 0: We get
1
A K + (1 )L ! A min(K; L); for ! 1: (4.35)
So in this case the CES function approaches a Leontief production function, the
isoquants of which form a right angle, cf. Fig. 4.7. In the limit there is no
possibility of substitution between capital and labor. In accordance with this the
elasticity of substitution calculated from (4.34) approaches zero when goes to
1:
Finally, let us consider the “opposite” transition. For …xed K and L we let
the substitution parameter rise towards 1 and get
1
A K + (1 )L ! A [ K + (1 )L] ; for ! 1:
Here the elasticity of substitution calculated from (4.34) tends to 1 and the
isoquants tend to straight lines with slope (1 )= : In the limit, the production
function thus becomes linear and capital and labor become perfect substitutes.
Fig. 4.7 depicts isoquants for alternative CES production functions and their
limiting cases. In the Cobb-Douglas case, = 1; the horizontal and vertical
asymptotes of the isoquant coincide with the coordinate axes. When < 1; the
horizontal and vertical asymptotes of the isoquant belong to the interior of the
positive quadrant. This implies that both capital and labor are essential inputs.
When > 1; the isoquant terminates in points on the coordinate axes. Then
neither capital, nor labor are essential inputs. Empirically there is not complete
agreement about the “normal” size of the elasticity of factor substitution for
industrialized economies. The elasticity also di¤ers across the production sectors.
A thorough econometric study (Antràs, 2004) of U.S. data indicate the aggregate
elasticity of substitution to be in the interval (0:5; 1:0). The survey by Chirinko
(2008) concludes with the interval (0:4; 0:6): Starting from micro data, a recent
study by Ober…eld and Raval (2014) …nds that the elasticity of factor substitution
for the US manufacturing sector as a whole has been stable since 1970 at about
0:7.

The CES production function in intensive form

Dividing through by L on both sides of (4.32), we obtain the CES production
function in intensive form,
Y 1
y = A( k + 1 ) ; (4.36)
L
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
4.5. The CES production function* 149

2
Aα
3

2
σ=∞
2 σ=0
A
1
σ = 0.5

σ = 1.5 σ=1
L
0 1 2
2 2
3 4 5 6 7
A A(1−α)

Figure 4.7: Isoquants for the CES function for alternative values of (A = 1:5, Y = 2;
and = 0:42).

where k K=L. The marginal productivity of capital can be written

dy 1 y 1
MP K = = A + (1 )k = A ;
dk k
which of course equals @Y =@K in (4.33). We see that the CES function violates
either the lower or the upper Inada condition for M P K, depending on the sign
of : Indeed, when < 0 (i.e., < 1); then for k ! 0 both y=k and dy=dk
approach an upper bound equal to A 1= < 1; thus violating the lower Inada
condition for M P K (see the right-hand panel of Fig. 2.3 of Chapter 2). It is also
noteworthy that in this case, for k ! 1, y approaches an upper bound equal to
A(1 )1= < 1. These features re‡ect the low degree of substitutability when
< 0:
When instead > 0; there is a high degree of substitutability ( > 1). Then,
for k ! 1 both y=k and dy=dk ! A 1= > 0; thus violating the upper Inada
condition for M P K (see right panel of Fig. 4.8). It is also noteworthy that for
k ! 0, y approaches a positive lower bound equal to A(1 )1= > 0. Thus, in
this case capital is not essential. At the same time dy=dk ! 1 for k ! 0 (so the
lower Inada condition for the marginal productivity of capital holds). Details are

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

150 CHAPTER 4. A GROWING ECONOMY

in Appendix E.
The marginal productivity of labor is
@Y
MP L = = (1 )A y 1 = (1 )A( k + 1 )(1 )=
w(k);
@L
from (4.33).
Since (4.32) is symmetric in K and L; we get a series of symmetric results by
1=
considering output per unit of capital as x Y =K = A + (1 )(L=K) :
In total, therefore, when there is low substitutability ( < 0); for …xed input
of either of the production factors, there is an upper bound for how much an
unlimited input of the other production factor can increase output. And when
there is high substitutability ( > 0); there is no such bound and an unlimited
input of either production factor take output to in…nity.
The Cobb-Douglas case, i.e., the limiting case for ! 0; constitutes in several
respects an intermediate case in that all four Inada conditions are satis…ed and
we have y ! 0 for k ! 0; and y ! 1 for k ! 1:
y y

1
∆x · Aα β
1 ∆x
5 ∆x · Aα β 5

1 ∆x
A (1 − α) β

1
k A (1 − α) β k
0 5 10 0 5 10

a) The case of σ < 1. a) The case of σ > 1.

Figure 4.8: The CES production function in intensive form, = 1=(1 ); < 1:

Generalizations
The CES production function considered above has CRS. By adding an elasticity
of scale parameter, , we get the generalized form

Y =A K + (1 )L ; > 0: (4.37)
In this form the CES function is homogeneous of degree : For 0 < < 1; there are
DRS, for = 1 CRS, and for > 1 IRS. If 6= 1; it may be convenient to consider
1=
Q Y 1= = A1= K + (1 )L and q Q=L = A1= ( k + 1 )1= :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.6. Concluding remarks 151

The elasticity of substitution between K and L is = 1=(1 ) whatever the

value of : So including the limiting cases as well as non-constant returns to scale
in the “family” of production functions with constant elasticity of substitution,
we have the simple classi…cation displayed in Table 4.1.

Table 4.1 The family of production functions

with constant elasticity of substitution.
=0 0< <1 =1 >1
Leontief CES Cobb-Douglas CES

Note that only for 1 is (4.37) a neoclassical production function. This

is because, when > 1; the conditions FKK < 0 and FN N < 0 do not hold
everywhere.
We may generalize further by assuming there are n inputs, in the amounts
X1 ; X2 ; :::; Xn : Then the CES production function takes the form
X
Y = A 1 X1 + 2 X2 + ::: n Xn ; i > 0 for all i; i = 1; > 0:
i
(4.38)
In analogy with (4.29), for an n-factor production function the partial elasticity
of substitution between factor i and factor j is de…ned as

M RSij d(Xi =Xj )

ij = ;
Xi =Xj dM RSij jY =Y

where it is understood that not only the output level but also all Xk , k 6= i, j,
are kept constant. Note that ji = ij : In the CES case considered in (4.38), all
the partial elasticities of substitution take the same value, 1=(1 ):

4.6 Concluding remarks

(incomplete)
When speaking of “sustained growth” in variables like K; Y; and C, we do
not mean growth in a narrow physical sense. Given limited natural resources
(matter and energy), exponential growth in a physical sense is of course not
possible. But sustained growth in terms of economic value is not ruled out. We
should for instance understand K broadly as “produced means of production”
of rising quality and falling material intensity (think of the rising e¢ ciency of
microprocessors). Similarly, C must be seen as a composite of consumer goods
and services with declining material intensity over time. This accords with the

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

152 CHAPTER 4. A GROWING ECONOMY

empirical fact that as income rises, the share of consumption expenditures devoted
to agricultural and industrial products declines and the share devoted to services,
hobbies, and amusement increases. Although “economic development”is perhaps
a more appropriate term (suggesting qualitative and structural change), we retain
standard terminology and speak of “economic growth”.

4.7 Literature notes and discussion

1. We introduced the assumption that at the macroeconomic level the “direc-
tion”of technological progress tends to be Harrod-neutral. Otherwise the model
will not be consistent with Kaldor’s stylized facts. The Harrod-neutrality of the
“direction” of technological progress is in the present model just an exogenous
feature. This raises the question whether there are mechanisms tending to gen-
erate Harrod-neutrality. Fortunately new growth theory provides clues as to the
sources of the speed as well as the direction of technological change. A facet
of this theory is that the direction of technological change is linked to the same
economic forces as the speed, namely pro…t incentives. See Acemoglu (2003) and
Jones (2006).
2. The literature discussing Kaldor’s “stylized facts” includes Att…eld and
Temple (2010), Rognlie (2015), Gollin (2002), Elsby et al. (2013), and Karabar-
bounis and Neiman (2014). The latter three references conclude with serious
scepticism.
3. In Section 4.2 we claimed that from an empirical point of view, the adjust-
ment towards a steady state can be from above as well as from below. Indeed,
Cho and Graham (1996) …nd that “on average, countries with a lower income per
adult are above their steady-state positions, while countries with a higher income
are below their steady-state positions”.
As to the assessment of the role of uncertainty for the condition that dynamic
e¢ ciency is satis…ed, in addition to Abel et al. (1989) other useful sources include
Ball et al. (1998) and Blanchard and Weil (2001).
4. In the Diamond OLG model as well as in many other models, a steady
state and a balanced growth path imply each other. Indeed, they are two sides
of the same process. There exist cases, however, where this equivalence does not
hold (some open economy models and some models with embodied technological
change, see Groth et al., 2010). Therefore, it is recommendable always to maintain
a terminological distinction between the two concepts.
5. Based on time-series econometrics, Att…eld and Temple (2010) and others
…nd support for the Kaldor “facts”for the US and UK and thereby for an evolu-
tion roughly obeying balanced growth in terms of aggregate variables. This does
not rule out structural change. A changing sectorial composition of the economy

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.7. Literature notes and discussion 153

is under certain conditions compatible with balanced growth (in a generalized

sense) at the aggregate level, cf. the “Kuznets facts” (see Kongsamut et al.,
2001, and Acemoglu, 2009).
6. Cases where the equivalence between steady state and balanced growth
does not hold include some open economy models and some models with embodied
technological change, see, e.g., Groth et al. (2010).
7. La Grandville (2009) contains a lot about analytical aspects linked to the
CES production function and the concept of elasticity of factor substitution.
8. On the declining material intensity of consumer goods and services as
technology develops, see Fagnart and Germain (2011).
From here incomplete:
Piketty (2014), Zucman ( ).
Demange and Laroque (1999, 2000) extend Diamond’s OLG model to uncer-
tain environments.
Keeping-up-with-the-Jones externalities. Do we work too much?
Blanchard, O., (2004) The Economic Future of Europe, J. Economic Perspec-
tives, vol. 18 (4), 3-26.
Prescott, E. (2003), Why do Americans work so much more than Europeans?
Federal Reserve Bank of Minneapolis Research Department Sta¤ Report No. 321.
I Ch. 5?
Chari, V. V., and P. J. Kehoe (2006), Modern macroeconomics in practice:
How theory is shaping policy, J. of Economic Perspectives, vol. 20 (4), 3-28.
For expositions in depth of OLG modeling and dynamics in discrete time, see
Azariadis (1993), de la Croix and Michel (2002), and Bewley (2007).
Dynamic ine¢ ciency, see also Burmeister (1980).
Two-sector OLG: Galor (1992). Galor’s book??
Bewley (2007).
Uzawa’s theorem: Uzawa (1961), Schlicht (2006).
The way the intuition behind the Uzawa theorem was presented in Section
4.1 draws upon Jones and Scrimgeour (2008).
La Grandville’s normalization of the CES function.
For more general and ‡exible production functions applied in econometric
work, see, e.g., Nadiri (1982).
Other aspects of life cycle behavior: education. OLG where people live three
periods.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

154 CHAPTER 4. A GROWING ECONOMY

4.8 Appendix
A. Growth and interest arithmetic in discrete time
Let t = 0; 1; 2; : : : ; and consider the variables zt ; xt ; and yt ; assumed positive
for all t. De…ne zt = zt zt 1 and xt and yt similarly. These ’s need not
be positive. The growth rate of xt from period t 1 to period t is de…ned as the
relative rate of increase in x; i.e., xt =xt 1 xt =xt 1 : And the growth factor for
xt from period t 1 to period t is de…ned as 1 + xt =xt 1 :
As we are here interested not in the evolution of growth rates, we simplify
notation by suppressing the t’s. So we write the growth rate of x as gx x=x 1
and similarly for y and z:
PRODUCT RULE If z = xy; then 1 + gz = (1 + gx )(1 + gy ) and gz gx + gy ;
when gx and gy are “small”.
Proof. By de…nition, z = xy; which implies z 1 + z = (x 1 + x)(y 1 +
y): Dividing by z 1 = x 1 y 1 gives 1 + z=z 1 = (1 + x=x 1 )(1 + y=y 1 )
as claimed. By carrying out the multiplication on the right-hand side of this
equation, we get 1 + z=z 1 = 1 + x=x 1 + y=y 1 + ( x=x 1 )( y=y 1 )
1 + x=x 1 + y=y1 when x=x 1 and y=y 1 are “small”. Subtracting 1 on
both sides gives the stated approximation.
So the product of two positive variables will grow at a rate approximately
equal to the sum of the growth rates of the two variables.
1+gx
FRACTION RULE If z = xy ; then 1 + gz = 1+gy
and gz gx gy ; when gx and
gy are “small”.
Proof. By interchanging z and x in Product Rule and rearranging, we get 1 +
z=z 1 = 1+ x=x 1
1+ y=y 1
; as stated in the …rst part of the claim. Subtracting 1 on
both sides gives z=z 1 = x=x 1 y=y 1
1+ y=y 1
x=x 1 y=y 1 ; when x=x 1
and y=y 1 are “small”. This proves the stated approximation.
So the ratio between two positive variables will grow at a rate approximately
equal to the excess of the growth rate of the numerator over that of the denomina-
tor. An implication of the …rst part of Claim 2 is: the ratio between two positive
variables is constant if and only if the variables have the same growth rate (not
necessarily constant or positive).
POWER FUNCTION RULE If z = x ; then 1 + gz = (1 + gx ) :
Proof. 1 + gz z=z 1 = (x=x 1 ) (1 + gx ) .
Given a time series x0 ; x1 ; :::; xn , by the average growth rate per period, or
more precisely, the average compound growth rate, is meant a g which satis…es
xn = x0 (1 + g)n : The solution for g is g = (xn =x0 )1=n 1:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.8. Appendix 155

Finally, the following approximation may be useful (for intuition) if used with
caution:
THE GROWTH FACTOR With n denoting a positive integer above 1 but “not
too large”, the growth factor (1 + g)n can be approximated by 1 + ng when g is
“small”. For g 6= 0; the approximation error is larger the larger is n:
Proof. (i) We prove the claim by induction. Suppose the claim holds for a …xed
n 2; i.e., (1 + g)n 1 + ng for g “small”: Then (1 + g)n+1 = (1 + g)n (1 + g)
(1 + ng)(1 + g) = 1 + ng + g + ng 2 1 + (n + 1)g since g “small” implies g 2
“very small”and therefore ng 2 “small”if n is not “too”large. So the claim holds
also for n + 1: Since (1 + g)2 = 1 + 2g + g 2 1 + 2g; for g “small”, the claim does
indeed hold for n = 2:
THE EFFECTIVE ANNUAL RATE OF INTEREST Suppose interest on a
loan is charged n times a year at the rate r=n per year. Then the e¤ective annual
interest rate is (1 + r=n)n 1:

B. Proof of the su¢ ciency part of Uzawa’s theorem

For convenience we restate the full theorem here:
PROPOSITION 2 (Uzawa’s balanced growth theorem). Let f(Kt ; Yt ; Ct )g1 t=0 be a
path along which Yt ; Kt , Ct , and St Yt Ct are positive for all t = 0; 1; 2; : : : ,
and satisfy the dynamic resource constraint (4.2), given the production function
(4.4) and the labor force (4.5). Assume (1 + g)(1 + n) > 1 . Then:
(i) a necessary condition for this path to be a balanced growth path is that along
the path it holds that
Yt = F~ (Kt ; Tt Lt ; 0); (*)
where Tt = A(1 + g)t with 1 + g (1 + gY )=(1 + n); gY being the constant growth
rate of output along the balanced growth path;
(ii) for any g 0 such that there is a q > (1 + g)(1 + n) (1 ) with the
~
property that the production function F in (4.4) allows an output-capital ratio
equal to q at t = 0 (i.e., F~ (1; k~ 1 ; 0) = q for some real number k~ > 0), a su¢ cient
condition for F~ to be consistent with a balanced growth path with output-capital
ratio equal to q is that F~ can be written as in (*) with Tt = A(1 + g)t .
Proof (i) See Section 4.1. (ii) Suppose (*) holds with Tt = A(1 + g)t : Let g 0
be given such that there is a q > (1 + g)(1 + n) (1 ) > 0 with the property
that
F~ (1; k~ 1 ; 0) = q (**)
for some constant k~ > 0: Our strategy is to prove the claim by construction of
a path P = (Yt ; Kt ; Ct )1
t=0 which satis…es it. We let P be such that the saving-
income ratio is a constant s^ [(1 + g)(1 + n) (1 )] =q 2 (0; 1), i.e., Yt Ct

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

156 CHAPTER 4. A GROWING ECONOMY

St = s^Yt for all t = 0; 1; 2; : : : : Inserting this, together with Yt = f (k~t )Tt Lt ,

where f (k~t ) F~ (k~t ; 1; 0) and k~t Kt =(Tt Lt ); into (4.2), rearranging gives the
Solow equation (4.3), which we may rewrite as

s^f (k~t ) [(1 + g)(1 + n) (1 )] k~t

k~t+1 k~t = :
(1 + g)(1 + n)

We see that k~t is constant if and only if k~t satis…es the equation f (k~t )=k~t =
[(1 + g)(1 + n) (1 )] =^
s q: By (**) and the de…nition of f; the required
~ ~
value of kt is k; which is thus the steady state for the constructed Solow model.
Letting K0 satisfy K0 = kAL ~ 0 ; where A = T0 ; we thus have k~0 = K0 =(T0 L0 ) = k:
~
So that the initial value of k~t equals the steady-state value. It follows that k~t = k~
for all t = 0; 1; 2; : : : ; and so Yt =Kt = f (k~t )=k~t = f (k)=
~ k~ = q for all t 0: In
addition, Ct = (1 s^)Yt ; so that Ct =Yt is constant along the path P: As both Y =K
and C=Y are thus constant along the path P , by (ii) of Proposition 1 follows that
P is a balanced growth path.
It is noteworthy that the proof of the su¢ ciency part of the theorem is con-
structive. It provides a method for constructing a balanced growth path with a
given technology growth rate and a given output-capital ratio.

C. Homothetic utility functions

Generalities A set C in Rn is called a cone if x 2 C and > 0 implies x 2 C:
A function f (x) = f (x1 ;. . . ,xn ) is homothetic in the cone C if for all x; y 2 C
and all > 0; f (x) = f (y) implies f ( x) = f ( y):
Consider the continuous utility function U (x1 ; x2 ); de…ned in R2+ : Suppose U
is homothetic and that the consumption bundles (x1 ; x2 ) and (y1 ; y2 ) are on the
same indi¤erence curve, i.e., U (x1 ; x2 ) = U (y1 ; y2 ): Then for any > 0 we have
U ( x1 ; x2 ) = U ( y1 ; y2 ) so that the bundles ( x1 ; x2 ) and ( y1 ; y2 ) are also
on the same indi¤erence curve.
For a continuous utility function U (x1 ; x2 ); de…ned in R2+ and increasing in
each of its arguments (as is our life time utility function in the Diamond model),
one can show that U is homothetic if and only if U can be written U (x1 ; x2 )
F (f (x1 ; x2 )) where the function f is homogeneous of degree one and F is an
increasing function. The “if”part is easily shown. Indeed, if U (x1 ; x2 ) = U (y1 ; y2 ),
then F (f (x1 ; x2 )) = F (f (y1 ; y2 )): Since F is increasing, this implies f (x1 ; x2 )
= f (y1 ; y2 ). Because f is homogeneous of degree one, if > 0; then f ( x1 ; x2 )
= f (x1 ; x2 ) and f ( y1 ; y2 ) = f (y1 ; y2 ) so that U ( x1 ; x2 ) = F (f ( x1 ; x2 ))
= F (f ( y1 ; y2 )) = U ( y1 ; y2 ); which shows that U is homothetic. As to the
“only if”part, see Sydsaeter et al. (2002).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.8. Appendix 157

Using di¤erentiability of our homothetic time utility function U (x1 ; x2 )

F (f (x1 ; x2 )), we …nd the marginal rate of substitution of good 2 for good 1 to be

dx2 @U=@x1 F 0 f1 (x1 ; x2 ) f1 (1; xx21 )

M RS = = = 0 = : (4.39)
dx1 jU =U @U=@x2 F f2 (x1 ; x2 ) f2 (1; xx21 )

The last equality is due to Euler’s theorem saying that when f is homogeneous of
degree 1, then the …rst-order partial derivatives of f are homogeneous of degree
0. Now, (4.39) implies that for a given M RS; in optimum re‡ecting a given
relative price of the two goods, the same consumption ratio, x2 =x1 ; will be chosen
whatever the budget. For a given relative price, a rising budget (rising wealth)
will change the position of the budget line, but not its slope. So M RS will not
change, which implies that the chosen pair (x1 ; x2 ) will move outward along a
given ray in R2+ : Indeed, as the intercepts with the axes rise proportionately with
the budget (the wealth), so will x1 and x2 .

Proof that the utility function in (4.22) is homothetic In Section 4.2 we

claimed that (4.22) is a homothetic utility function. This can be proved in the
following way. There are two cases to consider. Case 1: > 0; 6= 1: We rewrite
(4.22) as
1 1 1+
U (c1 ; c2 ) = (c11 + c21 )1=(1 ) ;
1 1
where (1 + ) 1 . The function x = g(c1 ; c2 ) (c11 + c21 )1=(1 ) is
homogeneous of degree one and the function G(x) (1=(1 ))x1 (1 +
)=(1 ) is an increasing function, given > 0; 6= 1. Case 2: = 1: Here we
start from U (c1 ; c2 ) = ln c1 + ln c2 : This can be written
h i
U (c1 ; c2 ) = (1 + ) ln (c1 c2 )1=(1+ ) ;

where x = g(c1 ; c2 ) = (c1 c2 )1=(1+ ) is homogeneous of degree one and G(x)

(1 + ) ln x is an increasing function.

D. General formulas for the elasticity of factor substitution

We here prove (4.30) and (4.31). Given the neoclassical production function
F (K; L); the slope of the isoquant F (K; L) = Y at the point (K; L) is

dK FL (K; L)
= M RS = : (4.40)
dL jY =Y FK (K; L)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

158 CHAPTER 4. A GROWING ECONOMY

We consider this slope as a function of the value of k K=L as we move along

the isoquant. The derivative of this function is

dM RS dM RS dL
=
dk jY =Y dL jY =Y dk jY =Y
(FL )2 FKK 2FK FL FKL + (FK )2 FLL dL
= (4.41)
FK3 dk jY =Y

by (2.53) of Chapter 2. In view of L K=k we have

dL k dK
dk jY =Y
K k dK
dL jY =Y
dL
dk jY =Y
K kM RS dL
dk jY =Y
K
= = = :
dk jY =Y k2 k2 k2

From this we …nd

dL K
= ;
dk jY =Y (k + M RS)k

to be substituted into (4.41). Finally, we substitute the inverse of (4.41) together

with (4.40) into the de…nition of the elasticity of factor substitution:

M RS dk
(K; L)
k dM RS jY =Y
FL =FK (k + FL =FK )k FK3
=
k K [(FL )2 FKK 2FK FL FKL + (FK )2 FLL ]
FK FL (FK K + FL L)
= ;
KL [(FL ) FKK 2FK FL FKL + (FK )2 FLL ]
2

which is the same as (4.30).

Under CRS, this reduces to

FK FL F (K; L)
(K; L) = (from (2.54) with h = 1)
KL [(FL )2 FKK 2FK FL FKL + (FK )2 FLL ]
FK FL F (K; L)
= (from (2.60))
KLFKL [ (FL )2 L=K 2FK FL (FK )2 K=L]
FK FL F (K; L) FK FL
= 2
= ; (using (2.54) with h = 1)
FKL (FL L + FK K) FKL F (K; L)

which proves the …rst part of (4.31). The second part is an implication of rewriting
the formula in terms of the production in intensive form.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.8. Appendix 159

E. Properties of the CES production function

The generalized CES production function is

Y =A K + (1 )L ; (4.42)
where A; ; and are parameters satisfying A > 0, 0 < < 1; and < 1;
6= 0; > 0: If < 1; there is DRS, if = 1; CRS, and if > 1, IRS. The
elasticity of substitution is always = 1=(1 ): Throughout below, k means
K=L:

The limiting functional forms We claimed in the text that, for …xed K > 0
and L > 0, (4.42) implies:
lim Y = A(K L1 ) = AL k ; (*)
!0
lim Y = A min(K ; L ) = AL min(k ; 1): (**)
! 1

=
Proof. Let q Y =(AL ): Then q = ( k + 1 ) so that
ln( k + 1 ) m( )
ln q = ; (4.43)

where
k ln k ln k
m0 ( ) = = : (4.44)
k +1 + (1 )k
Hence, by L’Hôpital’s rule for “0/0”,
m0 ( )
lim ln q = lim = ln k = ln k ;
!0 !0 1
so that lim !0 q = k ; which proves (*). As to (**), note that
8
1 < 0 for k > 1;
lim k = lim ! 1 for k = 1;
! 1 ! 1 k :
1 for k < 1:
Hence, by (4.43),
0 for k 1;
lim ln q = m0 ( )
! 1 lim ! 1 1
= ln k = ln k for k < 1;
where the result for k < 1 is based on L’Hôpital’s rule for “1/-1”. Consequently,
1 for k 1;
lim q =
! 1 k for k < 1;
which proves (**).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

160 CHAPTER 4. A GROWING ECONOMY

Properties of the isoquants of the CES function The absolute value of

the slope of an isoquant for (4.42) in the (L; K) plane is
@Y =@L 1 0 for k ! 0;
M RSKL = = k1 ! (*)
@Y =@K 1 for k ! 1:
This holds whether < 0 or 0 < < 1:
Concerning the asymptotes and terminal points, if any, of the isoquant Y = Y
we have from (4.42) Y = = A K + (1 )L : Hence,
!1
Y 1
K = L ;
A
!1
Y
L = K :
A(1 ) 1
From these two equations follows, when < 0 (i.e., 0 < < 1); that
1 1
K ! (A ) Y for L ! 1;
1 1
L ! [A(1 )] Y for K ! 1:
When instead > 0 (i.e., > 1), the same limiting formulas obtain for L ! 0
and K ! 0; respectively.

Properties of the CES function on intensive form Given = 1, i.e., CRS,

we have y Y =L = A( k + 1 )1= from (4.42). Then
dy 1 1 1
=A ( k +1 ) 1 k 1=A + (1 )k :
dk
Hence, when < 0 (i.e., 0 < < 1);
A 0 for k ! 0;
y = !
(ak + 1 ) 1= A(1 )1= for k ! 1:
dy A 1=
A for k ! 0;
= !
dk [ + (1 )k ] ( 1)= 0 for k ! 1:
If instead > 0 (i.e., > 1),
A(1 )1= for k ! 0;
y !
1 for k ! 1:
dy 1 for k ! 0;
! 1=
dk A for k ! 1:
1
The output-capital ratio is y=k = A + (1 )k and has the same limiting
values as dy=dk; when > 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

4.9. Exercises 161

Continuity at the boundary of R2+ When 0 < < 1; the right-hand side of
(4.42) is de…ned and continuous also on the boundary of R2+ : Indeed, we get
(
A K for L ! 0;
Y = F (K; L) = A K + (1 )L !
A(1 ) L for K ! 0:

When < 0; however, the right-hand side is not de…ned on the boundary. We
circumvent this problem by rede…ning the CES function in the following way
when < 0:
(
Y = F (K; L) = A K + (1 )L when K > 0 and L > 0; (4.45)
0 when either K or L equals 0.

We now show that continuity holds in the extended domain. When K > 0 and
L > 0, we have

Y =A K + (1 )L A G(K; L): (4.46)

Let < 0 and (K; L) ! (0; 0): Then, G(K; L) ! 1; and so Y = ! 1: Since
= < 0; this implies Y ! 0 = F (0; 0); where the equality follows from the
de…nition in (4.45): Next, consider a …xed L > 0 and rewrite (4.46) as
1 1 1 1 1
Y = A K + (1 )L = A L( k + 1 )
1
A L
= 1=
! 0 for k ! 0;
(ak + 1 )

when < 0: Since 1= > 0; this implies Y ! 0 = F (0; L); from (4.45): Finally,
consider a …xed K > 0 and let L=K ! 0: Then, by an analogue argument we get
Y ! 0 = F (K; 0); (4.45): So continuity is maintained in the extended domain.

4.9 Exercises
4.1 (the aggregate saving rate in steady state)

a) In a well-behaved Diamond OLG model let n be the rate of population

growth and k the steady state capital-labor ratio (until further notice, we
ignore technological progress). Derive a formula for the long-run aggregate
net saving rate, S N =Y; in terms of n and k . Hint: use that for a closed
economy S N = Kt+1 Kt :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

162 CHAPTER 4. A GROWING ECONOMY

b) In the Solow growth model without technological change a similar relation

holds, but with a di¤erent interpretation of the causality. Explain.

c) Compare your result in a) with the formula for S N =Y in steady state one
gets in any model with the same CRS-production function and no techno-
logical change. Comment.

d) Assume that n = 0. What does the formula from a) tell you about the level
of net aggregate savings in this case? Give the intuition behind the result in
terms of the aggregate saving by any generation in two consecutive periods.
One might think that people’s rate of impatience (in Diamond’s model the
rate of time preference ) a¤ect S N =Y in steady state. Does it in this case?
Why or why not?

e) Suppose there is Harrod-neutral technological progress at the constant rate

g > 0: Derive a formula for the aggregate net saving rate in the long run in
a well-behaved Diamond model in this case.

f) Answer d) with “from a)”replaced by “from e)”. Comment.

g) Consider the statement: “In Diamond’s OLG model any generation saves
as much when young as it dissaves when old.”True or false? Why?

4.2 (increasing returns to scale and balanced growth)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 6

Long-run aspects of …scal policy

and public debt

We consider an economy with a government providing public goods and services.

It …nances its spending by taxation and borrowing. The term …scal policy refers
to the government’s decisions about spending and the …nancing of this spending,
be it by taxes or debt issue. The government’s choice concerning the level and
composition of its spending and how to …nance it, may aim at:

1 a¤ecting resource allocation (provide public goods that would otherwise not
be supplied in a su¢ cient amount, correct externalities and other markets
failures, prevent monopoly ine¢ ciencies, provide social insurance);

2 a¤ecting income distribution, be it a) within generations or b) between

generations;

3 contribute to macroeconomic stabilization (dampening of business cycle

‡uctuations through aggregate demand policies).

The design of …scal policy with regard to the aims 1 and 2 at a disaggregate
level is a major theme within the …eld of public economics. Macroeconomics
studies ways of dealing with aim 3 as well as big-picture aspects of 1 and 2, like
overall policies to maintain and promote sustainable prosperity.
In this chapter we address …scal sustainability and long-run implications of
debt …nance. This relates to one of the conditions that constrain public …nancing
instruments. To see the issue of …scal sustainability in a broader context, Section
6.1 provides an overview of conditions and factors that constrain public …nanc-
ing instruments. Section 6.2 introduces the basics of government budgeting and
Section 6.3 de…nes the concepts of government solvency and …scal sustainability.
In Section 6.4 the analytics of debt dynamics is presented. As an example, the

203
 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
204 AND PUBLIC DEBT

Stability and Growth Pact of the EMU (the Economic and Monetary Union of
the European Union) is discussed. Section 6.5 looks more closely at the link be-
tween government solvency and the government’s No-Ponzi-Game condition and
intertemporal budget constraint. In Section 6.6 we widen public sector accounting
by introducing separate operating and capital budgets so as to allow for proper
accounting of public investment. A theoretical claim, known as the Ricardian
equivalence proposition, is studied in Section 6.7. The question “is Ricardian
equivalence likely to be a good approximation to reality?”is addressed, applying
the Diamond OLG framework extended with a public sector.

6.1 An overview of government spending and

…nancing issues
Before entering the more specialized sections, it is useful to have a general idea
about circumstances that condition public spending and …nancing. These cir-
cumstances include:

(i) …nancing by debt issue is constrained by the need to remain solvent and
avoid catastrophic debt dynamics;

(ii) …nancing by taxes is limited by problems arising from:

(a) distortionary supply-side e¤ects of many kinds of taxes;

(b) tax evasion (cf. the rise of the shadow economy, tax havens used by
multinationals, etc.).

(iii) time lags in spending as well as taxing may interfere with attempts to
stabilize the economy (recognition lag, decision lag, implementation lag,
and e¤ect lag);

(iv) credibility problems due to time-inconsistency;

(v) conditions imposed by political processes, bureaucratic self-interest, lobby-

ing, and rent seeking.

Point (i) is the main focus of sections 6.2-6.6. Point (ii) is brie‡y considered
in Section 6.4.1 in connection with the so-called La¤er curve. In Section 6.6 point
(iii) is brie‡y commented on. The remaining points, (iv) - (v), are not addressed
speci…cally in this chapter. They should always be kept in mind, however, when
discussing …scal policy. Hence some remarks at the end of the chapter.
Now to the speci…cs of government budget accounting and debt …nancing.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.2. The government budget 205

6.2 The government budget

We generally perceive the public sector (or the nation state) as consisting of the
national government and a central bank. In economics the term “government”
does not generally refer to the particular administration in o¢ ce at a point in
time. The term is rather used in a broad sense, encompassing both legislation
and administration. The aspects of legislation and administration in focus in
macroeconomics are the rules and decisions concerning spending on public con-
sumption, public investment, transfers, and subsidies on the expenditure side and
on levying taxes and incurring debts on the …nancing side. Within certain limits
the government has usually delegated the management of the nation’s currency
to the central bank, also called the monetary authority. Our accounting treats
“government budgeting” as covering the public sector as a whole, that is, the
consolidated government (including local government) and central bank. Gov-
ernment bonds held by the central bank are thus excluded from what we call
“government debt”. So the terms government debt, public debt, and state debt
are used synonymously.

The basics of government budget accounting cannot be described without

including money, nominal prices, and in‡ation. Elementary aspects of money and
in‡ation will therefore be included in this section. We shall not, however, consider
money and in‡ation in any systematic way until later chapters. Whether the
economy considered is a closed or open economy will generally not be important
in this chapter.

Table 6.1 lists key variables of government budgeting.

Table 6.1. List of main variable symbols

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
206 AND PUBLIC DEBT

Symbol Meaning
Yt real GDP (= real GNP if the economy is closed)
Ctg public consumption
Itg public …xed capital investment
Gt Ctg + Itg real public spending on goods and services
Xt real transfer payments
T~t real gross tax revenue
Tt T~t Xt real net tax revenue
Mt the monetary base (currency and bank reserves in the central bank)
Pt price level (in money) for goods and services (the GDP de‡ator)
Dt nominal net public debt
Dt
Bt Pt 1
real net public debt
Bt
bt Yt
government debt-to-income ratio
it nominal short-term interest rate
xt = xt xt 1 (where x is some arbitrary variable)
Pt Pt Pt 1
t Pt 1 Pt 1
in‡ation rate
Pt 1 (1+it ) 1+it
1 + rt Pt 1+ t
real short-term interest rate

Note that Yt ; Gt ; and Tt are quantities de…ned per period, or more generally,
per time unit, and are thus ‡ow variables. On the other hand, Mt ; Dt ; and Bt
are stock variables, that is, quantities de…ned at a given point in time, here at
the beginning of period t. We measure Dt and Bt net of …nancial claims held
by the government. Almost all countries have positive government net debt, but
in principle Dt < 0 is possible.1 The monetary base, Mt ; is currency plus fully
liquid deposits in the central bank held by the private sector at the beginning of
period t; Mt is by de…nition nonnegative.
We shall in this chapter most of the time ignore uncertainty and risk of default.
Then the nominal interest rate on government bonds must be the same as that on
other interest-bearing assets in the economy. For ease of exposition we imagine
that all government bonds are one-period bonds. That is, each government bond
promises a payout equal to one unit of account at the end of the period and
then the bond expires. Given the interest rate, it ; the market value of a bond at
the start of period t is vt = 1=(1 + it ): If the number of outstanding bonds (the
quantity of bonds) in period t is qt ; the government debt has face value (value at
maturity) equal to qt . The market value at the start of period t of this quantity
of bonds will be Dt = qt =(1 + it ). The nominal expenditure to be made at the
1
If Dt < 0, the government has positive net …nancial claims on the private sector and earns
interest on these claims which is then an additional source of government revenue besides
taxation.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.2. The government budget 207

end of the period to redeem the outstanding debt can then be written

qt = Dt (1 + it ): (6.1)

This is the usual way of writing the expenditure to be made, namely as if the
government debt were like a given bank loan of size Dt with a variable rate of
interest. We should not forget, however, that given the quantity, qt ; of the bonds,
the value, Dt , of the government debt at the issue date depends negatively on it :
Anyway, the total nominal government expenditure in period t can be written

Pt (Gt + Xt ) + Dt (1 + it ):

It is common to refer to this expression as expenditure “in period t”. Yet, in a

discrete time model (with a period length of a year or a quarter corresponding
to typical macroeconomic data) one has to imagine that the payment for goods
and services delivered in the period occurs either at the beginning or the end of
the period. We follow the latter interpretation and so the nominal price level Pt
for period-t goods and services refers to payment occurring at the end of period
t: As an implication, the real value, Bt ; of government debt at the beginning of
period t (= end of period t 1) is Dt =Pt 1 : This may look a little awkward but
is nevertheless meaningful. Indeed, Dt is a stock of liabilities at the beginning
of period t while Pt 1 is a price referring to a ‡ow paid for at the end of period
t 1 which is essentially the same point in time as the beginning of period t:
Anyway, whatever timing convention is chosen, some kind of awkwardness will
always arise in discrete time analysis. This is because the discrete time approach
arti…cially treats the continuous ‡ow of time as a sequence of discrete points in
time.2
The government expenditure is …nanced by a combination of taxes, bonds
issue, and increase in the monetary base:

Pt T~t + Dt+1 + Mt+1 = Pt (Gt + Xt ) + Dt (1 + it ): (6.2)

By rearranging we have

Dt+1 + Mt+1 = Pt (Gt + Xt T~t ) + it Dt : (6.3)

In standard government budget accounting the nominal government budget

de…cit, GBD; is de…ned as the excess of total government spending over govern-
ment revenue, P T~. That is, according to this de…nition the right-hand side of
(6.3) is the nominal budget de…cit in period t; GBDt : The …rst term on the right-
hand side, Pt (Gt + Xt T~t ); is named the primary budget de…cit (non-interest
2
In a theoretical model this kind of problems is avoided when government budgeting is
formulated in continuous time, cf. Chapter 13.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
208 AND PUBLIC DEBT

spending less taxes). The second term, it Dt ; is called the debt service. Simi-
larly, Pt (T~t Xt Gt ) is called the primary budget surplus. A negative value of
a “de…cit” thus amounts to a positive value of a corresponding “surplus”, and
a negative value of a “surplus” amounts to a positive value of a corresponding
“de…cit”.
We immediately see that this accounting deviates from “normal” principles.
Business companies typically have sharply separated capital and operating bud-
gets. In contrast, the budget de…cit de…ned above treats that part of G which
represents government net investment as parallel to government consumption.
Government net investment is attributed as an expense in a single year’s ac-
count; according to “normal” principles it is only the depreciation on the public
capital that should …gure as an expense. Likewise, the above accounting does
not consider that a part of D (or perhaps more than D) may be backed by the
value of public physical capital. And if the government sells a physical asset to
the private sector, the sale will appear as a reduction of the government budget
de…cit while in reality it is merely a conversion of an asset from a physical form
to a …nancial form. So the cost and asset aspects of government net investment
are not properly dealt with in the standard public accounting.3
With the exception of Section 6.6 we will nevertheless stick to the traditional
vocabulary. Where this might create logical di¢ culties, it helps to imagine that:
(a) all of G is public consumption, i.e., Gt = Ctg for all t;
(b) there is no public physical capital.
Now, from (6.2) and the de…nition Tt T~t Xt (net tax revenue) follows that
real government debt at the beginning of period t + 1 is:
Dt+1 Dt Mt+1
Bt+1 = Gt + Xt T~t + (1 + it )
Pt Pt Pt
Dt =Pt 1 Mt+1 1 + it Mt+1
= Gt Tt + (1 + it ) = Gt Tt + Bt
Pt =Pt 1 Pt 1+ t Pt
Mt+1
(1 + rt )Bt + Gt Tt : (6.4)
Pt
We see from the second line that, everything else equal, in‡ation curtails the real
value of the debt and interest payments. Hence, sometimes not only the actual
nominal budget de…cit is recorded but also a measure where t Dt is subtracted.
3
Another anomaly is related to the fact that some countries, for instance Denmark, have
large implicit government assets due to deferred taxes on the part of personal income invested
in pension funds. If the government then decides to reverse the deferred taxation (as the Danish
government did 2012 and 2014 to comply better with the 3%-de…cit rule of the Stability and
Growth Pact of the EMU), the o¢ cial budget de…cit is reduced, but essentially it is just a
matter of replacing one government asset by another.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.2. The government budget 209

The last term, Mt+1 =Pt ; in (6.4) is seigniorage, i.e., public sector revenue
obtained by issuing base money (ignoring the diminutive cost of printing money).
To get a sense of this variable, suppose real output grows at the constant rate gY
so that Yt+1 = (1 + gY )Yt : Then the public debt-to-income ratio can be written
Bt+1 1 + rt Gt Tt Mt+1
bt+1 = bt + : (6.5)
Yt+1 1 + gY (1 + gY )Yt Pt (1 + gY )Yt
Apart from the growth-correcting factor, (1+gY ) 1 ; the last term is the seigniorage-
income ratio,
Mt+1 Mt+1 Mt
= :
P t Yt Mt Pt Yt
If in the long run the base money growth rate, Mt+1 =Mt , as well as the nominal
interest rate (i.e., the opportunity cost of holding money) are constant, then the
velocity of money and its inverse, the money-nominal income ratio, Mt =(Pt Yt ),
are also likely to be roughly constant. So is, therefore, the seigniorage-income
ratio.4 For the more developed countries this ratio tends to be a fairly small
number although not immaterial. For emerging economies with poor institutions
for collecting taxes seigniorage matters more.5
The U.S. has a single monetary authority, the central bank, and a single
…scal authority, the treasury. The seigniorage created is immediately transferred
from the …rst to the latter. The Eurozone has a single monetary authority but
multiple …scal authorities, namely the treasuries of the member countries. The
seigniorage created by the ECB is every year shared by the national central banks
of the Eurozone countries in proportion to their equity share in the ECB. And
the national central banks then transfer their share to the national treasuries.
This makes up a Mt+1 term for the consolidated public sector of the individual
Eurozone countries.
In monetary unions and countries with their own currency, government budget
de…cits are thus generally …nanced both by debt creation and money creation, as
envisioned in the above equations. Nonetheless, from now on, for simplicity, in
this chapter we will predominantly ignore the seigniorage term in (6.5) and only
occasionally refer to the modi…cations implied by taking it into account.
4
A reasonable money demand function is Mtd = Pt Yt e i ; > 0; where i is the nominal
interest rate. With clearing in the money market, we thus have Mt =(Pt Yt ) = e i . In view of
1 + i (1 + r)(1 + ); when r and are constant, so is i and, thereby, Mt =(Pt Yt ):
5
In the U.S. over the period 1909-1950s seigniorage ‡uctuated a lot and peaked 4 % of GDP
in the 1930s and 3 % of GDP at the end of WW II. But over the period from the late 1960s
to 1986 seigniorage ‡uctuated less around an average close to 0.5 %.of GDP (Walsh, 2003, p.
177). In Denmark seigniorage was around 0.2 % of GDP during the 1990s (Kvartalsoversigt 4.
kvartal 2000, Danmarks Nationalbank). In Bolivia, up to the event of hyperin‡ation 1984-85,
seigniorage reached 5 % of GDP and more than 50 % of government revenue (Sachs and Larrain,
1993).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
210 AND PUBLIC DEBT

We thus proceed with the simple government accounting equation:

Bt+1 Bt = rt Bt + Gt Tt ; (DGBC)

where the right-hand side is the real budget de…cit. This equation is in macro-
economics often called the dynamic government budget constraint (or DGBC for
short). It is in fact just an accounting identity conditional on M = 0. It says
that if the real budget de…cit is positive and there is essentially no …nancing
by money creation, then the real public debt grows. We come closer to a con-
straint when combining (DGBC) with the requirement that the government stays
solvent.

6.3 Government solvency and …scal sustainabil-

ity
To be solvent means being able to meet the …nancial commitments as they fall
due. In practice this concept is closely related to the government’s No-Ponzi-
Game condition and intertemporal budget constraint (to which we return in Sec-
tion 6.5), but at the theoretical level it is more fundamental.
We may view the public sector as an in…nitely-lived agent in the sense that
there is no last date where all public debt has to be repaid. Nevertheless, as we
shall see, there tends to be stringent constraints on government debt creation in
the long run.

6.3.1 The critical role of the growth-corrected interest

factor
Very much depends on whether the real interest rate in the long-run is higher
than the growth rate of GDP or not.
To see this, suppose the country considered has positive government debt at
time 0 and that the government levies taxes equal to its non-interest spending:

T~t = Gt + Xt or Tt T~t Xt = Gt for all t 0: (6.6)

So taxes cover only the primary expenses while interest payments (and debt
repayments when necessary) are …nanced by issuing new debt. That is, the
government attempts a permanent roll-over of the debt including the interest
due for payment. In view of (DGBC), this implies that Bt+1 = (1 + rt )Bt ; saying
that the debt grows at the rate rt . Assuming, for simplicity, that rt = r (a

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.3. Government solvency and …scal sustainability 211

constant), the law of motion for the public debt-to-income ratio is

Bt+1 1 + r Bt 1+r
bt+1 = bt ; b0 > 0;
Yt+1 1 + gY Yt 1 + gY
where we have maintained the assumption of a constant output growth rate, gY .
The solution to this linear di¤erence equation then becomes
1+r t
bt = b0 ( );
1 + gY
where we consider both r and gY as exogenous. We see that the growth-corrected
1+r
interest rate, 1+g Y
1 r gY (for gY and r “small”) plays a key role. There
are contrasting cases to discuss.
Case 1: r > gY : In this case, bt ! 1 for t ! 1: Owing to compound interest,
the debt grows so large in the long run that the government will be unable to …nd
buyers for all the debt. Permanent debt roll-over is thus not feasible. Imagine for
example an economy described by the Diamond OLG model. Here the buyers of
the debt are the young who place part of their saving in government bonds. But
if the stock of these bonds grows at a higher rate than income, the saving of the
young cannot in the long run keep track with the fast-growing government debt.
In this situation the private sector will understand that bankruptcy is threatening
and nobody will buy government bonds except at a low price, which means a high
interest rate. The high interest rate only aggravates the problem. That is, the
…scal policy (6.6) breaks down. Either the government defaults on the debt or T
must be increased or G decreased (or both) until the growth rate of the debt is
no longer higher than gY :
If the debt is denominated in the country’s own currency, an alternative way
out is of course a shift to money …nancing of the budget de…cit, that is, seignior-
age. When capacity utilization is high, this leads to rising in‡ation and thus
the real value of the debt is eroded. Bond holders will then demand a higher
nominal interest rate, thus aggravating the …scal di¢ culties. The economic and
social chaos of hyperin‡ation threatens.6 The hyperin‡ation in Germany 1922-23
peaked in Nov. 1923 at 29,525% per month; it eroded the real value of the huge
government debt of Germany after WW I by 95 percent.
Case 2: r = gY : If r = gY ; we get bt = b0 for all t 0: Since the debt, increas-
ing at the rate r; does not increase faster than national income, the government
has no problem …nding buyers of its newly issued bonds the government stays
6
In economists’ standard terminology “hyperin‡ation” is present when the in‡ation rate
exceeds 50 percent per month. As we shall see in Chapter 18, the monetary …nancing route comes
to a dead end if the needed seigniorage reaches the backward-bending part of the “seigniorage
La¤er curve”.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
212 AND PUBLIC DEBT

%
30
Real short-term DK interest rate (ex post)
Real annual DK GDP growth rate
20

10

−10

−20
Average interest rate (1875-2003): 2.9%
Compound annual GDP growth rate (1875-2002):2.7%

−30
1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
%
30
Real short-term US interest rate (ex post)
Real annual US GDP growth rate
20

10

−10

−20
Average interest rate (1875-2003): 2.4%
Compound annual GDP growth rate (1875-2002):3.4%

−30
1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000

Figure 6.1: Real short-term interest rate and annual growth rate of real GDP in Den-
mark and the US since 1875. The real short-term interest rate is calculated as the
money market rate minus the contemporaneous rate of consumer price in‡ation. Source:
Abildgren (2005) and Maddison (2003).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.3. Government solvency and …scal sustainability 213

solvent. Thereby the government is able to …nance its interest payments simply
by issuing new debt. The growing debt is passed on to ever new generations with
higher income and saving and the debt roll-over implied by (6.6) can continue
forever.
Case 3: r < gY : Here we get bt ! 0 for t ! 1; and the same conclusion
holds a fortiori.
In Case 2 as well as Case 3, where the interest rate is not higher than the
growth rate of the economy, the government can thus pursue a permanent debt
roll-over policy as implied by (6.6) and still remain solvent. But in Case 1,
permanent debt roll-over is impossible and sooner or later the interest payments
must be tax …nanced.
Which of the cases is relevant in real life? Fig. 6.1 shows for Denmark (upper
panel) and the US (lower panel) the time paths of the real short-term interest
rate and the GDP growth rate, both on an annual basis. Overall, the levels of
the two are more or less the same, although on average the interest rate is in
Denmark slightly higher but in the US somewhat lower than the growth rate.
(Note that the interest rates referred to are not the average rate of return in the
economy but a proxy for the lower interest rate on government bonds.)
Nevertheless, many macroeconomists believe there is good reason for paying
attention to the case r > gY , also for a country like the US. This is because we live
in a world of uncertainty, with many di¤erent interest rates, and imperfect credit
markets, aspects the above line of reasoning has not incorporated. The prudent
debt policy needed whenever, under certainty, r > gY can be shown to apply
to a larger range of circumstances when uncertainty is present (see Literature
notes). To give a ‡avor we may say that a prudent debt policy is needed when
the average interest rate on the public debt exceeds gY " for some “small”but
positive ":7 On the other hand there is a di¤erent feature which draws the matter
in the opposite direction. This is the possibility that a tax, 2 (0; 1); on interest
income is in force so that the net interest rate on the government debt is (1 )r
rather than r:

6.3.2 Sustainable …scal policy

The concept of sustainable …scal policy is closely related to the concept of gov-
ernment solvency. As already noted, to be solvent means being able to meet the
…nancial commitments as they fall due. A given …scal policy is called sustainable
if by applying its spending and tax rules forever, the government stays solvent.
“Sustainable” conveys the intuitive meaning. The issue is: can the current tax
and spending rules continue forever?
7
This is only a “rough” characterization, see, e.g., Blanchard and Weil (2001).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
214 AND PUBLIC DEBT

To be more speci…c, suppose Gt and Tt are determined by …scal policy rules

represented by the functions
Gt = G(x1t ; :::; xnt ; t); and Tt = T (x1t ; :::; xnt ; t);
where t = 0; 1; 2; : : : ; and x1t ,..., xnt are key macroeconomic and demographic
variables (like national income, old-age dependency ratio, rate of unemployment,
extraction of natural resources, say oil from the North Sea, etc.). In this way a
given …scal policy is characterized by the rules G( ) and T ( ): Suppose further
that we have an economic model, M, of how the economy functions.
DEFINITION Let the current period be period 0 and let the public debt at
the beginning of period 0 be given. Then, given a forecast of the evolution
of the demographic and foreign economic environment in the future and given
the economic model M, the …scal policy (G( ); T ( )) is said to be sustainable
relative to this model if the forecast calculated on the basis of M is that the
government stays solvent under this policy. The …scal policy (G( ); T ( )) is called
unsustainable, if it is not sustainable.
This de…nition of …scal sustainability is silent about the presence of uncer-
tainty. Without going into detail about this di¢ cult issue, suppose the model
M is stochastic and let " be a “small” positive number. Then we may say that
the …scal policy (G( ); T ( )) with 100-" percent probability is sustainable relative
to the model M if the forecast calculated on the basis of M is that with 100-"
percent probability the government stays solvent under this policy.
Governments, rating agencies, and other institutions evaluate sustainability
of …scal policy on the basis of simulations of giant macroeconometric models.
Essentially, the operational criterion for sustainability is whether the …scal policy
can be deemed compatible with upward boundedness of the public debt-to-income
ratio. Normally, the income measure applied here is GDP. Other measures are
conceivable such as GNP, taxable income, or after-tax income. Moreover, even
if a debt spiral is not (yet) underway in a given country, a high level of the
debt-income ratio may in itself be worrisome. This is because a high level of
debt under certain conditions may trigger a spiral of self-ful…lling expectations of
default. We come back to this in the section to follow.
Owing to the increasing pressure on public …nances caused by factors such
as reduced birth rates, increased life expectancy, and a fast-growing demand for
medical care, many industrialized countries have for a long time been assessed
to be in a situation where their …scal policy is not sustainable (Elmendorf and
Mankiw 1999). The implication is that sooner or later one or more expenditure
rules and/or tax rules (in a broad sense) will probably have to be changed.
Two major kinds of strategies have been suggested. One kind of strategy is
the pre-funding strategy. The idea is to prevent sharp future tax increases by

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 215

ensuring a …scal consolidation prior to the expected future demographic changes.

Another strategy (alternative or complementary to the former) is to attempt a
gradual increase in the labor force by letting the age limits for retirement and
pension increase along with expected lifetime this is the indexed retirement
strategy. The …rst strategy implies that current generations bear a large part
of the adjustment cost. In the second strategy the costs are shared by current
and future generations in a way more similar to the way the bene…ts in the
form of increasing life expectancy are shared. We shall not go into detail about
these matters here, but refer the reader to a large literature about securing …scal
sustainability in the ageing society, see Literature notes.

6.4 Debt arithmetic

A key tool for evaluating …scal sustainability is debt arithmetic, i.e., the ana-
lytics of debt dynamics. The previous section described the important role of
the growth-corrected interest rate. The next subsection considers the minimum
primary budget surplus required for …scal sustainability in di¤erent situations.

6.4.1 The required primary budget surplus

Ignoring the seigniorage term Mt+1 =Pt in the dynamic government budget iden-
tity (6.4), we have:
Bt+1 = (1 + r)Bt (Tt Gt ); (DGBC)
where Tt Gt is the primary surplus in real terms. Suppose aggregate income,
Yt ; grows at a given constant rate rate, gY : Let the spending-to-income ratio,
Gt =Yt , and the (net) tax revenue-to-income ratio, Tt =Yt ; be constants, and ;
respectively. We assume that interest income on government bonds is not taxed.
It follows that the public debt-to-income ratio bt Bt =Yt (from now just denoted
debt-income ratio) changes over time according to

Bt+1 1+r
bt+1 = bt ; (6.7)
Yt+1 1 + gY 1 + gY

where we have assumed a constant interest rate, r. There are (again) three cases
to consider.
Case 1: r > gY : As emphasized above this case is generally considered the one
of most practical relevance. And it is in this case that latent debt instability is
present and the government has to pay attention to the danger of runaway debt
dynamics. To see this, note that the solution of the linear di¤erence equation

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
216 AND PUBLIC DEBT

(6.7) is
t
1+r
bt = (b0 b) +b ; where (6.8)
1 + gY
1
1+r s
b = 1 = ; (6.9)
1 + gY 1 + gY r gY r gY

where s is the primary surplus as a share of GDP. Here b0 is historically given. But
the steady-state debt-income ratio, b ; depends on …scal policy. The important
feature is that the growth-corrected interest factor is in this case higher than 1
and has the exponent t. Therefore, if …scal policy is such that b < b0 ; the debt-
income ratio exhibits geometric growth. The solid curve in the topmost panel in
Fig. 6.2 shows a case where …scal policy is such that < (r gY )b0 whereby we
get b < b0 when r > gY ; so that the debt-income ratio, bt ; grows without bound.
This re‡ects that with r > gY ; compound interest is stronger than compound
growth. The sequence of discrete points implied by our discrete-time model is in
the …gure smoothed out as a continuous curve.
The American economist and Nobel Prize laureate George Akerlof (2004, p.
6) came up with this analogy:

“It takes some time after running o¤ the cli¤ before you begin to fall.
But the law of gravity works, and that fall is a certainty”.

Somewhat surprisingly, perhaps, when r > gY , there can be debt explosion in

the long run even if > , namely if 0 < < (r gY )b0 : Debt explosion can
also arise if b0 < 0; namely if < (r gY )b0 < 0:
The only way to avoid the snowball e¤ects of compound interest when the
growth-corrected interest rate is positive is to ensure a primary budget surplus as
a share of GDP, ; high enough such that b b0 : So the minimum primary
surplus as a share of GDP, s^; required for …scal sustainability is the one implying
b = b0 ; i.e., by (6.9),
s^ = (r gY )b0 : (6.10)
If by adjusting and/or , the government obtains = s^; then b = b0
whereby bt = b0 for all t 0 according to (6.8), cf. the second from the top panel
in Fig. 6.2. The di¤erence between s^ and the actual primary surplus as a share
of GDP is named the primary surplus gap or the sustainability gap.
Note that s^ will be larger:
- the higher is the initial level of debt, b0 ; and,
- when b0 > 0; the higher is the growth-corrected interest rate, r gY .

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 217

Figure 6.2: Evolution of the debt-income ratio, depending on the sign of b0 b ; in the
cases r > gY (the three upper panels) and r < gY (the two lower panels), respectively.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
218 AND PUBLIC DEBT

Delaying the adjustment increases the size of the needed policy action, since
the debt-income ratio, and thereby s^; will become higher in the meantime.
For …xed spending-income ratio ; the minimum tax-to-income ratio needed
for …scal sustainability is
^ = + (r gY )b0 : (6.11)
Given b0 and ; this tax-to-income ratio is sometimes called the sustainable tax
rate. The di¤erence between this rate and the actual tax rate, ; indicates the
size of the needed tax adjustment, were it to take place at time 0, assuming a
given :
Suppose that the debt build-up can be and is prevented already at time
0 by ensuring that the primary surplus as a share of income, ; at least equals
s^ so that b b0 : The solid curve in the midmost panel in Fig. 6.2 illustrates the
resulting evolution of the debt-income ratio if b is at the level corresponding to
the hatched horizontal line while b0 is unchanged compared with the top panel.
Presumably, the government would in such a state of a¤airs relax its …scal policy
after a while in order not to accumulate large government …nancial net wealth.
Yet, the pre-funding strategy vis-a-vis the …scal challenge of population ageing
(referred to above) is in fact based on accumulating some positive public …nancial
net wealth as a bu¤er before the substantial e¤ects of population ageing set in. In
this context, the higher the growth-corrected interest rate, the shorter the time
needed to reach a given positive net wealth position.
Case 2: r = gY : In this knife-edge case there is still a danger of runaway dy-
namics, but of a less explosive form. The formula (6.8) is no longer valid. Instead
the solution of (6.7) is bt = b0 + [( )=(1 + gY )] t = b0 [( )=(1 + gY )] t:
Here, a non-negative primary surplus is both necessary and su¢ cient to avoid
bt ! 1 for t ! 1.
Case 3: r < gY : This is the case of stable debt dynamics. The formula (6.8)
is again valid, but now implying that the debt-income ratio is non-explosive.
Indeed, bt ! b for t ! 1; whatever the level of the initial debt-income ratio
and whatever the sign of the budget surplus. Moreover, when r < gY ;

b = S 0 for T 0: (*)
r gY
So, if there is a forever positive primary surplus, the result is a negative long-run
debt, i.e., a positive government …nancial net wealth in the long run. And if there
is a forever negative primary surplus, the result is not debt explosion but just
convergence toward some positive long-run debt-income ratio. The second from
bottom panel in Fig. 6.2 illustrates this case for a situation where b0 > b and
b > 0; i.e., < 0; by (*). When the GDP growth rate continues to exceed
the interest rate on government debt, a large debt-income ratio can be brought

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 219

down quite fast, as witnessed by the evolution of both UK and US government

debt in the …rst three decades after the second world war. Indeed, if the growth-
corrected interest rate remains negative, permanent debt roll-over can handle the
…nancing, and taxes need never be levied.8
Finally, the bottom panel in Fig. 6.2 shows the case where, with a large
primary de…cit ( < 0 but large in absolute value), excess of output growth
over the interest rate still implies convergence towards a constant debt-income
ratio, albeit a high one.
In this discussion we have treated r as exogenous. But r may to some extent
be dependent on prolonged budget de…cits. Indeed, in Chapter 13 we shall see
that with prolonged budget de…cits, r tends to become higher than otherwise.
Everything else equal, this reduces the likelihood of Case 2 and Case 3.

La¤er curve*

We return to Case 1 because we have ignored supply-side e¤ects of taxation, and

such e¤ects could be important in Case 1.
A La¤er curve (so named after the American economist Arthur La¤er, 1940-)
refers to a hump-shaped relationship between the income tax rate and the tax
revenue. For simplicity, suppose the tax revenue equals taxable income times
a given average tax rate. A 0% tax rate and most likely also a 100% tax rate
generate no tax revenue. As the tax rate increases from a low initial level, a rising
tax revenue is obtained. But after a certain point some people may begin to work
less (in the legal economy), stop reporting all their income, and stop investing.
So it is reasonable to think of a tax rate above which the tax revenue begins to
decline.
While La¤er was wrong about where USA was “on the curve” (see, e.g.,
Fullerton 2008), and while, strictly speaking, there is no such thing as the La¤er
curve and the tax rate,9 La¤er’s intuition is hardly controversial. Ignoring, for
simplicity, transfers, we therefore now assume that for a given tax system there
is a gross tax-income ratio, L ; above which the tax revenue declines. Then, if
the presumed sustainable tax-income ratio, ^; in (6.11) exceeds L , it can not be
realized.
To see what the value of L could be, suppose aggregate taxable income before
8
On the other hand, we should not forget that this analysis presupposes absence of uncer-
tainty. As touched on in Section 6.3.1, in the presence of uncertainty and therefore existence of
many interest rates, the issue becomes more complicated.
9
A lot of contingencies are involved: income taxes are typically progressive (i.e., average tax
rates rise with income); it matters whether a part of tax revenue is spent to reduce tax evasion,
etc.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
220 AND PUBLIC DEBT

tax is a function, f; of the net-of-tax share 1 : Then tax revenue is

R( ) = f (1 );

which we assume is a hump-shaped function of in the interval [0; 1] : Taking

logs and di¤erentiating w.r.t. gives the …rst-order condition R0 ( )=R( ) =
1= f 0 (1 )=f (1 ) = 0; which holds for = L ; the tax-income ratio that
maximizes R: It follows that 1= L = f 0 (1 L )=f (1 L ); hence

1 L 1 L
= f 0 (1 L) E`1 f (1 L ):
L f (1 L)

Rearranging gives
1
L = :
1 + E`1 f (1 L)

If the elasticity of income w.r.t. 1 is given as 0.4,10 we get L = 5=7 0:7:

Thus, if the required tax-income ratio, ^; calculated on the basis of (6.11) (under
the simplifying assumption of no transfers), exceeds 0.7, …scal sustainability can
not be obtained by just raising taxation.

The level of the debt-income ratio and self-ful…lling expectations of

default
We again consider Case 1: r > gY : The incumbent chief economist at the IMF,
Olivier Blanchard remarked in the midst of the 2010-2012 debt crisis in the Eu-
rozone:

“The higher the level of debt, the smaller is the distance between
solvency and default”.11

The background for this remark is the following. There is likely to be an upper
bound for the tax-income ratio deemed politically or economically feasible by the
government as well as the market participants. Similarly, a lower bound for the
spending-income ratio is likely to exist, be it for economic or political reasons. In
the present framework we therefore let the government face the constraints
and ; where is the least upper bound for the tax-income ratio and is
the greatest lower bound for the spending-income ratio. Then the actual primary
surplus, s; can at most equal s :
10
As suggested for the U.S. by Gruber and Saez, 2002.
11
Blanchard (2011).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 221

Suppose that at …rst the situation in the considered country is as in the second
from the top panel in Fig. 6.2. That is, initially,

s= = s^ = (r gY )b0 s ; (6.12)

with b0 > 0: De…ne r to be the value of r satisfying

s
(r gY )b0 = s; i.e., r = + gY : (6.13)
b0
Thereby r is the maximum level of the interest rate consistent with absence of
an explosive debt-income ratio.
According to (6.12), fundamentals (tax- and spending-income ratios, growth-
corrected interest rate, and initial debt) are consistent with absence of an explo-
sive debt-income ratio as long as r is unchanged. Nevertheless …nancial investors
may be worried about default if b0 is high. Investors are aware that a rise in the
actual interest rate, r; can always happen and that if it does, a situation with
r > r is looming, in particular if the country has high debt. The larger is b0 ; the
lower is the critical interest rate, r; as witnessed by (6.13).
The worrying scenario is that the fear of default triggers a risk premium, and
if the resulting level of the interest rate on the debt, say r0 ; exceeds r, unpleasant
debt dynamics like that in the top panel of Fig. 6.2 set in. To r0 corresponds a
new value of the primary surplus, say s^0 ; de…ned by s^0 = (r0 gY )b0 : So s^0 is the
minimum primary surplus (as a share of GDP) required for a non-accelerating
debt-income ratio in the new situation. With b0 > 0 and r0 > r, we get

s^0 = (r0 gY )b0 > (r gY )b0 = s;

where s is given in (6.12). The government could possibly increase its primary
surplus, s; but at most up to s; and this will not be enough since the required
primary surplus, s^0 ; exceeds s: The situation would be as illustrated in the top
panel of Fig. 6. 2 with b given as s=(r0 gY ) < b0 :
That is, if the actual interest rate should rise above the critical interest rate,
r; runaway debt dynamics would take o¤ and debt default thereby be threatening.
A fear that it may happen may be enough to trigger a fall in the market price of
government bonds which means a rise in the actual interest rate, r. So …nancial
investors’fear can be a self-ful…lling prophesy. Moreover, as we saw in connection
with (6.13), the risk that r becomes greater than r is larger the larger is b0 .
It is not so that across countries there is a common threshold value for a
“too large” public debt-to-income ratio. This is because variables like ; ; r;
and gY , as well as the net foreign debt position and the current account de…cit
(not in focus in this chapter), di¤er across countries. Late 2010 Greece had

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
222 AND PUBLIC DEBT

(gross) government debt of 148 percent of GDP and the interest rate on 10-year
government bonds skyrocketed. Conversely Japan had (gross) government debt
of more than 200 percent of GDP while the interest rate on 10-year government
bonds remained very low.

Finer shades

1. As we have just seen, even when in a longer-run perspective a solvency problem

is unlikely, self-ful…lling expectations can here and now lead to default. Such a
situation is known as a liquidity crisis rather than a true solvency crisis. In a
liquidity crisis there is an acute problem of insu¢ cient cash to pay the next bill
on time (“cash-‡ow insolvency”) because lending is di¢ cult due to actual and
potential creditors’fear of default. A liquidity crisis can be braked by the central
bank stepping in and acting as a “lender of last resort” by printing money. In a
country with its own currency, the central bank can do so and thereby prevent a
bad self-ful…lling expectations equilibrium to unfold.12
2. In the above analysis we simpli…ed by assuming that several variables,
including ; ; and r; are constants. The upward trend in the old-age dependency
ratio, due to a decreased birth rate and rising life expectancy, together with a
rising request for medical care is likely to generate upward pressure on . Thereby
a high initial debt-income ratio becomes more challenging.
3. On the other hand, rBt is income to the private sector and can be taxed at
the same average tax rate as factor income, Yt : Then the benign inequality is
no longer r gY but (1 )r gY ; which is more likely to hold. Taxing interest
income is thus supportive of …scal sustainability (cf. Exercise B.28).
4. Having ignored seigniorage, there is an upward bias in our measure (6.10)
of the minimum primary surplus as a share of GDP, s^; required for …scal sustain-
ability when r > gY . Imposing stationarity of the debt-income ratio at the level b
into the general debt-accumulation formula (6.5), multiplying through by 1 + gY ;
and cancelling out, we …nd

Mt+1 Mt+1 Mt
s^ = (r gY )b = (r gY )b :
Pt Y t Mt P t Yt
12
In a monetary union which is not also a …scal union (think of the eurozone), the situation
is more complicated. A single member country with large government debt (or large debt in
commercial banks for that matter) may …nd itself in an acute liquidity crisis without its own
means to solve it. Indeed, the elevation of interest rates on government bonds in the Southern
part of the eurozone in 2010-2012 can be seen as a manifestation of investors’fear of payment
di¢ culties. The elevation was not reversed until the European Central Bank in September 2012
declared its willingness to e¤ectively act as a “lender of last resort” (on a conditional basis),
see Box 6.2 in Section 6.4.2.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 223

With r = 0:04; gY = 0:03; and b = 0:60; we get (r gY )b = 0:006: With a

seigniorage-income ratio even as small as 0:003, the “true”required primary sur-
plus is 0:003 rather than 0:006. As long as the seigniorage-income ratio is approx-
imately constant, our original formula, given in (6.10), for the required primary
surplus as a share of GDP is in fact valid if we interpret as the (tax+seigniorage)-
income ratio.
5. Having assumed a constant gY ; we have ignored business cycle ‡uctuations.
Allowing for booms and recessions, the timing of …scal consolidation in a country
with a structural primary surplus gap (^ s s > 0) becomes a crucial issue. The
case study in the next section will be an opportunity to touch upon this issue.

6.4.2 Case study: The Stability and Growth Pact of the

EMU
The European Union (EU) is approaching its aim of establishing a “single mar-
ket”(unrestricted movement of goods and services, workers, and …nancial capital)
across the territory of its member countries, 28 sovereign nations. Nineteen of
these have joined the common currency, the euro. They constitute what is known
as the Eurozone with the European Central Bank (ECB) as supranational institu-
tion responsible for conducting monetary policy in the Eurozone. The Eurozone
countries as well as the nine EU countries outside the Eurozone (including UK,
Denmark, Sweden, and Poland) are, with minor exceptions, required to abide
with a set of …scal rules, …rst formulated already in the Treaty of Maastrict from
1992. In that year a group of European countries decided a road map leading to
the establishment of the euro in 1999 and a set of criteria for countries to join.
These …scal rules included a de…cit rule as well as a debt rule. The de…cit rule
says that the annual nominal government budget de…cit must not be above 3
percent of nominal GDP. The debt rule says that the government debt should not
be above 60 percent of GDP. The …scal rules were upheld and in minor respects
tightened in the Stability and Growth Pact (SGP) which was implemented in 1997
as the key …scal constituent of the Economic and Monetary Union (EMU). The
latter name is a popular umbrella term for the …scal and monetary legislation of
the EU. The EU member countries that have adopted the euro are often referred
to as “the full members of the EMU”.
Some of the EU member states (Belgium, Italy, and Greece) had debt-income
ratios above 100 percent since the early 1990s and still have. Committing to
the requirement of a gradual reduction of their debt-income ratios, they became
full members of the EMU essentially from the beginning (that is, 1999 except
Greece, 2001). The 60 percent debt rule of the SGP is to be understood as a
long-run ceiling that, by the stock nature of debt, can not be accomplished here

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
224 AND PUBLIC DEBT

and now if the country is highly indebted.

The de…cit and debt rules (with associated detailed contingencies and arrange-
ments including ultimate pecuniary …nes for de…ance) are meant as discipline de-
vices aiming at “sound budgetary policy”, alternatively called “…scal prudence”.
The motivation is protection of the ECB against political demands to loosen mon-
etary policy in situations of …scal distress. A …scal crisis in one or more of the
Eurozone countries, perhaps “too big to fail”, could set in and entail a state of
a¤airs approaching default on government debt and chaos in the banking sector
with rising interest rates spreading to neighboring member countries (a negative
externality). This could lead to open or concealed political pressure on the ECB
to in‡ate away the real value of the debt, thus challenging the ECB’s one and
only concern with “price stability”.13 Or a …scal crisis might at least result in
demands on the ECB to curb soaring interest rates by purchasing government
bonds from the country in trouble. In fact, such a scenario is close to what we
have seen in southern Europe in the wake of the Great Recession triggered by
the …nancial crisis starting 2007. Such “bailing out” could give governments in-
centives to be relaxed about de…cits and debts (a “moral hazard”problem). And
the lid on de…cit spending imposed by the SGP should help to prevent needs for
“bailing out”to arise.

The link between the de…cit and the debt rule

Whatever the virtues or vices of the design of the de…cit and debt rules, one may
ask the plain question: what is the arithmetical relationship, if any, between the
3 percent and 60 percent tenets?
First a remark about measurement. The measure of government debt, called
the EMU debt, used in the SGP criterion is based on the book value of the
…nancial liabilities rather than the market value. In addition, the EMU debt is
more of a gross nature than the theoretical net debt measure represented by our
D. The EMU debt measure allows fewer of the government …nancial assets to
be subtracted from the government …nancial liabilities.14 In our calculation and
subsequent discussion we ignore these complications.
Consider a de…cit rule saying that the (total) nominal budget de…cit must
never be above 100 percent of nominal GDP. By (6.3) with Mt+1 “small”
enough to be ignored, this de…cit rule is equivalent to the requirement
Dt+1 Dt = GBDt = it Dt + Pt (Gt Tt ) P t Yt : (6.14)
13
The ECB interprets “price stability” as a consumer price in‡ation rate “below, but close
to, 2 percent per year over the medium term”.
14
For Denmark the di¤erence between the EMU and the net debt is substantial. In 2013 the
Danish EMU debt was 44.6% of GDP while the government net debt was 5.5% of GDP (Danish
Ministry of Finance, 2014).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 225

In the SGP, = 0:03. Here we consider the general case: > 0: To see the
implication for the (public) debt-to-income ratio in the long run, let us …rst
imagine a situation where the de…cit ceiling, ; is always binding for the economy
we look at. Then Dt+1 = Dt + Pt Yt and so
Bt+1 Dt+1 Dt
bt+1 = + ;
Yt+1 Pt Yt+1 (1 + )Pt 1 (1 + gY )Yt 1 + gY
assuming constant output growth rate, gY ; and in‡ation rate : This reduces to
1
bt+1 = bt + : (6.15)
(1 + )(1 + gY ) 1 + gY
Assuming that (1 + )(1 + gY ) > 1 (as is normal over the medium run), this linear
di¤erence equation has the stable solution
t
1
bt = (b0 b) + b ! b for t ! 1; (6.16)
(1 + )(1 + gY )
where
(1 + )
b = : (6.17)
(1 + )(1 + gY ) 1
Consequently, if the de…cit rule (6.14) is always binding, the debt-income ratio
tends in the long run to be proportional to the de…cit bound : The factor of
proportionality is a decreasing function of the long-run growth rate of real GDP
and the in‡ation rate. This result con…rms the general tenet that if there is
economic growth, perpetual budget de…cits need not lead to …scal problems.
If on the other hand the de…cit rule is not always binding, then the budget
de…cit is on average smaller than above so that the debt-income ratio will in the
long run be smaller than b .
The conclusion is the following. With one year as the time unit, suppose the
de…cit rule is = 0:03 and that gY = 0:03 and = 0:02 (the upper end of
the in‡ation interval aimed at by the ECB). Suppose further the de…cit rule is
never violated. Then in the long run the debt-income ratio will be at most b
= 1:02 0:03=(1:02 1:03 1) 0:60: This is in agreement with the debt rule
of the SGP according to which the maximum value allowed for the debt-income
ratio is 60%.
Although there is nothing sacred about either of the numbers 0:60 or 0:03;
they are mutually consistent, given = 0:02 and gY = 0:03.
We observe that the de…cit rule (6.14) implies that:
The upper bound, b ; on the long-run debt income ratio is lower the higher
is in‡ation. The reason is that the growth factor [(1 + ) (1 + gY )] 1
for bt in (6.15) depends negatively on the in‡ation rate, . So does therefore
b since, by (6.16), b (1 + gY ) 1 (1 ) 1:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
226 AND PUBLIC DEBT

For a given ; the upper bound on the long-run debt income ratio is inde-
pendent of both the nominal and real interest rate (this follows from the
indicated formula for the growth factor for bt and the fact that (1+i)(1+r) 1
= 1 + ):

The debate about the design of the SGP

In addition to the aimed long-run implications, by its design the SGP has short-
run implications for the economy. Hence an evaluation of the SGP cannot ignore
the way the economy functions in the short run. How changes in government
spending and taxation a¤ects the economy depends on the “state of the business
cycle”: is the economy in a boom with full capacity utilization or in a slump with
slack aggregate demand?
Much of the debate about the SGP has centered around the consequences
of the de…cit rule in an economic recession triggered by a collapse of aggregate
demand (for instance due to private deleveraging in the wake of a banking crisis).
Although the Eurozone countries are economically quite di¤erent, they are sub-
ject to the same one-size-…ts-all monetary policy. Facing dissimilar shocks, the
single member countries in need of aggregate demand stimulation in a recession
have by joining the euro renounced on both interest rate policy and currency de-
preciation.15 The only policy tool left for demand stimulation is therefore …scal
policy. Instead of a supranational …scal authority responsible for handling the
problem, it is up to the individual member countries to act and to do so within
the constraints of the SGP.
On this background, the critiques of the de…cit rule of the SGP include the fol-
lowing points. (It may here be useful to have at the back of one’s mind the simple
Keynesian income-expenditure model, where output is demand-determined and
below capacity while the general price level is sticky.)

Critiques 1. When considering the need for …scal stimuli in a recession, a

ceiling at 0:03 is too low unless the country has almost no government debt in
advance. Such a de…cit rule gives too little scope for counter-cyclical …scal policy,
including the free working of the automatic …scal stabilizers (i.e., the provisions,
through tax and transfer codes, in the government budget that automatically
cause tax revenues to fall and spending to rise when GDP falls).16 As an econ-
omy moves towards recession, the de…cit rule may, bizarrely, force the government
to tighten …scal policy although the situation calls for stimulation of aggregate
15
Denmark is in a similar situation. In spite of not joining the euro after the referendum in
2000, the Danish krone has been linked to the euro through a …xed exchange rate since 1999.
16
Over the …rst 13 years of existence of the euro even Germany violated the 3 percent rule
…ve of the years.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 227

demand. The pact has therefore sometimes been called the “Instability and De-
pression Pact” it imposes a wrong timing of …scal consolidation.17
2. Since what really matters is long-run …scal sustainability, a de…cit rule
should be designed in a more ‡exible way than the 3% rule of the SGP. A mean-
ingful de…cit rule would relate the de…cit to the trend nominal GDP, which we
may denote (P Y ) . Such a criterion would imply

GBD (P Y ) : (6.18)

Then
GBD (P Y )
:
PY PY
In recessions the ratio (P Y ) =(P Y ) is high, in booms it is low. This has the
advantage of allowing more room for budget de…cits when they are needed
without interfering with the long-run aim of stabilizing government debt below
some speci…ed ceiling.
3. A further step in this direction is a rule directly in terms of the structural
or cyclically adjusted budget de…cit rather than the actual year-by-year de…cit.
The cyclically adjusted budget de…cit in a given year is de…ned as the value the
de…cit would take in case actual output were equal to trend output in that year.
Denoting the cyclically adjusted budget de…cit GBD ; the rule would be

GBD
:
(P Y )

In fact, in its original version as of 1997 the SGP contained an additional rule
like that, but in the very strict form of 0: This requirement was implicit in
the directive that the cyclically adjusted budget “should be close to balance or
in surplus”. By this requirement it is imposed that the debt-income ratio should
be close to zero in the long run. Many EMU countries certainly had and have
larger cyclically adjusted de…cits. Taking steps to comply with such a low
structural de…cit ceiling may be hard and endanger national welfare by getting in
the way of key tasks of the public sector. The minor reform of the SGP endorsed
in March 2005 allowed more contingencies, also concerning this structural bound.
By the more recent reform in 2012, the Fiscal Pact, the lid on the cyclically
17
The SGP has an exemption clause referring to “exceptional”circumstances. These circum-
stances were originally de…ned as “severe economic recession”, interpreted as an annual fall
in real GDP of at least 1-2%. By the reform of the SGP in March 2005, the interpretation
was changed into simply “negative growth”. Owing to the international economic crisis that
broke out in 2008, the de…cit rule was thus suspended in 2009 and 2010 for most of the EMU
countries. But the European Commission brought the rule into e¤ect again from 2011, which
according to many critics was much too early, given the circumstances.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
228 AND PUBLIC DEBT

adjusted de…cit-income ratio was raised to 0.5% and to 1.0% for members with a
debt-income ratio “signi…cantly below 60%”. These are still quite small numbers.
Abiding by the 0.5% or 1.0% rule implies a long-run debt-income ratio of at most
10% or 20%, respectively, given structural in‡ation and structural GDP growth
at 2% and 3% per year, respectively.18
4. Regarding the composition of government expenditure, critics have argued
that the SGP pact entails a problematic disincentive for public investment. The
view is that a …scal rule should be based on a proper accounting of public invest-
ment instead of simply ignoring the composition of government expenditure. We
consider this issue in Section 6.6 below.
5. At a more general level critics have contended that policy rules and sur-
veillance procedures imposed on sovereign nations will hardly be able to do their
job unless they encompass stronger incentive-compatible elements. Enforcement
mechanisms are bound to be week. The SGP’s threat of pecuniary …nes to a
country which during a recession has di¢ culties to reduce its budget de…cit seems
absurd (and has not been made use of so far). Moreover, abiding by the …scal
rules of the SGP prior to the Great Recession was certainly no guarantee of not
ending up in a …scal crisis in the wake of a crisis in the banking sector, as wit-
nessed by Ireland and Spain. A seemingly strong …scal position can vaporize fast,
particularly if banks, “too big to fail”, need be bailed out.

Counter-arguments Among the counter-arguments raised against the criti-

cisms of the SGP has been that the potential bene…ts of the proposed alternative
rules are more than o¤set by the costs in terms of reduced simplicity, measurabil-
ity, and transparency. The lack of ‡exibility may even be a good thing because it
helps “tying the hands of elected policy makers”. Tight rules are needed because
of a “de…cit bias”arising from short-sighted policy makers’temptation to promise
spending without ensuring the needed …nancing, especially before an upcoming
election. These points are sometimes linked to the view that market economies
are generally self-regulating. Keynesian stabilization policy is not needed and
may do more harm than good.

Box 6.1. The 2010-2012 debt crisis in the Eurozone

What began as a banking crisis became a deep economic recession combined with a
government debt crisis.
At the end of 2009, in the aftermath of the global economic downturn, it became
evident that Greece faced an acute debt crisis driven by three factors: high government
debt, low ability to collect taxes, and lack of competitiveness due to cost in‡ation.
Anxiety broke out about the debt crisis spilling over to Spain, Portugal, Italy, and
18
Again apply (6.17).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.4. Debt arithmetic 229

Ireland, thus widening bond yield spreads in these countries vis-a-vis Germany in the
midst of a serious economic recession. Moreover, the solvency of big German banks
that were among the prime creditors of Greece was endangered. The major Eurozone
governments and the International Monetary Fund (IMF) reached an agreement to
help Greece (and indirectly its creditors) with loans and guarantees for loans, condi-
tional on the government of Greece imposing yet another round of harsh …scal austerity
measures. The elevated bond interest rates of Greece, Italy, and Spain were not con-
vincingly curbed, however, until in August-September 2012 the president of the ECB,
Mario Draghi, launched the “Outright Monetary Transactions” (OMT) program ac-
cording to which, under certain conditions, the ECB will buy government bonds in
secondary bond markets with the aim of “safeguarding an appropriate monetary policy
transmission and the singleness of the monetary policy” and with “no ex ante quan-
titative limits”. Considerably reduced government bond spreads followed and so the
sheer announcement of the program seemed e¤ective in its own right. Doubts raised by
the German Constitutional Court about its legality vis-à-vis Treaties of the European
Union were …nally repudiated by the European Court of Justice mid-June 2015. At
the time of writing (late June 2015) the OMT program has not been used in practice.
Early 2015, a di¤erent massive program for purchases of government bonds, including
long-term bonds, in the secondary market as well as private asset-backed bonds was
decided and implemented by the ECB. The declared aim was to brake threatening de-
‡ation and return to “price stability”, by which is meant in‡ation close to 2 percent
per year.
So much about the monetary policy response. What about …scal policy? On the
basis of the SGP, the EU Commission imposed “…scal consolidation” initiatives to be
carried out in most EU countries in the period 2011-2013 (some of the countries were
required to start already in 2010). With what consequences? By many observers, partly
including the research department of IMF, the initiatives were judged self-defeating.
When at the same time comprehensive deleveraging in the private sector is going on,
“austerity” policy deteriorates aggregate demand further and raises unemployment.
Thereby, instead of budget de…cits being decreased, the numerator in the debt-income
ratio, D=(P Y ); is decreased. Fiscal multipliers are judged to be large (“in the 0.9 to
1.7 range since the Great Recession”, IMF, World Economic Outlook, Oct. 2012) in
a situation of idle resources, monetary policy aiming at low interest rates, and nega-
tive spillover e¤ects through trade linkages when “…scal consolidation”is synchronized
across countries. The unemployment rate in the Eurozone countries was elevated from
7.5 percent in 2008 to 12 percent in 2013. The British economists, Holland and Portes
(2012), concluded: “It is ironic that, given that the EU was set up in part to avoid
coordination failures in economic policy, it should deliver the exact opposite”.
The whole crisis has pointed to a basic di¢ culty faced by the Eurozone. In spite
of the member countries being economically very di¤erent sovereign nations, they are

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
230 AND PUBLIC DEBT

subordinate to the same one-size-…ts-all monetary policy without sharing a federal

government ready to use …scal instruments to mitigate regional consequences of country-
speci…c shocks. Adverse demand shocks may lead to sharply rising budget de…cits in
some countries, and …nancial investors may loose con…dence and so elevate government
bond interest rates. A liquidity crisis may arise, thereby amplifying adverse shocks.
Even when a common negative demand shock hits all the member countries in a similar
way, and a general relaxation of both monetary and …scal policy is called for, there is
the problem that the individual countries, in fear of boosting their budget de…cit and
facing the risk of exceeding the de…cit or debt limit, may wait for the others to initiate
a …scal expansion. The possible consequence of this “free rider” problem is general
under-stimulation of the economies.
The dismal experience regarding the ability of the Eurozone to handle the Great
Recession has incited proposals along two dimensions. One dimension is about allowing
the ECB greater scope for acting as a “lender of last resort”. The other dimension is
about centralizing a larger part of the national budgets into a common union budget
(see, e.g., De Grauwe, 2014). (END OF BOX)

6.5 Solvency, the NPG condition, and the in-

tertemporal government budget constraint
Up to now we have considered the issue of government solvency from the per-
spective of dynamics of the government debt-to-income ratio. It is sometimes
useful to view government solvency from another angle the intertemporal bud-
get constraint (GIBC). Under a certain condition stated below, the intertemporal
budget constraint is as relevant for a government as for private agents. A simple
condition closely linked to whether the government’s intertemporal budget con-
straint is satis…ed or not is what is known as the government’s No-Ponzi-Game
(NPG) condition. It is convenient to …rst focus on this condition. We concentrate
on government net debt measured in real terms and ignore seigniorage.

6.5.1 When is the NPG condition necessary for solvency?

Consider a situation with a constant interest rate, r: Suppose taxes are lump sum
or at least that there is no tax on interest income from owning government bonds.
Then the government’s NPG condition is that the present discounted value of the
public debt in the far future is not positive, i.e.,

t
lim Bt (1 + r) 0: (NPG)
t!1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.5. Solvency, the NPG condition, and the intertemporal government budget
constraint 231

This condition says that government debt is not allowed to grow in the long
run at a rate as high as (or even higher than) the interest rate.19 That is, a
…scal policy satisfying the NPG condition rules out a permanent debt rollover.
Indeed, as we saw in Section 6.3.1, with B0 > 0; a permanent debt rollover
policy (…nancing all interest payments and perhaps even also part of the primary
government spending) by debt issue leads to Bt B0 (1 + r)t for t = 0; 1; 2; : : : :
Substituting into (NPG) gives limt!1 Bt B0 (1 + r)t (1 + r) t = B0 > 0; thus
violating (NPG).
The designation No-Ponzi-Game condition refers to a guy from Boston, Charles
Ponzi, who in the 1920s made a fortune out of an investment scam based on the
chain-letter principle. The principle was to pay o¤ old investors with money from
new investors. Ponzi was sentenced to many years in prison for his transactions;
he died poor and without friends!
To our knowledge, this kind of …nancing behavior is nowhere forbidden for
the government as it generally is for private agents. But under “normal”circum-
stances a government has to plan its expenditures and taxation so as to comply
with its NPG condition since otherwise not enough lenders will be forthcoming.
As the state is in principle in…nitely-lived, however, there is no …nal date where
all government debt should be over and done with. Indeed, the NPG condition
does not even require that the debt has ultimately to be non-increasing. The
NPG condition “only” says that the debtor, here the government, can not let
the debt grow forever at a rate as high as (or higher than) the interest rate. For
instance the U.K. as well as the U.S. governments have had positive debt for
centuries and high debt after both WW I and WW II.
Suppose Y (GDP) grows at the constant rate gY (actually, for most of the
following results it is enough that limt!1 Yt+1 =Yt = 1 + gY ). We have:
PROPOSITION 1 Let bt Bt =Yt and interpret “solvency”as absence of an for
ever accelerating debt-income ratio. Then:

(i) if r > gY ; solvency requires (NPG) satis…ed;

(ii) if r gY ; the government can remain solvent without (NPG) being satis…ed.

Proof. When bt 6= 0;

bt+1 Bt+1 =Yt+1 Bt+1 =Bt Bt+1 =Bt

lim lim = lim = lim : (6.19)
t!1 bt t!1 Bt =Yt t!1 Yt+1 =Yt t!1 1 + gY
19
If there is e¤ective taxation of interest income at the rate r 2 (0; 1); then the after-
tax interest rate, (1 r )r; is the relevant discount rate, and the NPG condition would read
t
limt!1 Bt [1 + (1 r )r] 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
232 AND PUBLIC DEBT

Case (i): r > gY : If limt!1 Bt 0; then (NPG) is trivially satis…ed. As-

sume limt!1 Bt > 0: For this situation we prove the statement by contradic-
tion. Suppose (NPG) is not satis…ed. Then, limt!1 Bt (1 + r) t > 0; implying
that limt!1 Bt+1 =Bt 1 + r: In view of (6.19) this implies that limt!1 bt+1 =bt
(1 + r)=(1 + gY ) > 1: Thus, bt ! 1; which violates solvency. By contradiction,
this proves that solvency implies (NPG) when r > gY .
Case (ii): r gY : Consider the permanent debt roll-over policy Tt = Gt for
all t 0; and assume B0 > 0. By (DGBC) of Section 6.2 this policy yields
Bt+1 =Bt = 1 + r; hence, in view of (6.19), lim t!1 bt+1 =bt = (1 + r)=(1 + gY )
1: The policy consequently implies solvency. On the other hand the solution
of the di¤erence equation Bt+1 = (1 + r)Bt is Bt = B0 (1 + r)t : Thus Bt (1 + r) t
= B0 > 0 for all t; thus violating (NPG).
Hence imposition of the NPG condition on the government relies on the in-
terest rate being in the long run higher than the growth rate of GDP. If instead
r gY ; the government can cut taxes, run a budget de…cit, and postpone the
tax burden inde…nitely. In that case the government can thus run a Ponzi Game
and still stay solvent. Nevertheless, as alluded to earlier, if uncertainty is added
to the picture, there will be many di¤erent interest rates, matters become more
complicated, and quali…cations to Proposition 1 are needed (Blanchard and Weil,
2001). The prevalent view among macroeconomists is that imposition of the NPG
condition on the government is generally warranted.
While in the case r > gY ; the NPG condition is necessary for solvency, it is
not su¢ cient. Indeed, we could have
1 + gY < lim Bt+1 =Bt < 1 + r: (6.20)
t!1

Here, by the upper inequality, (NPG) is satis…ed, yet, by the lower inequality
together with (6.19), we have limt!1 bt+1 =bt > 1 so that the debt-income ratio
explodes.
EXAMPLE 1 Let GDP = Y; a constant, and r > 0; so r > gY = 0: Let the
budget de…cit in real terms equal "Bt + ; where 0 " < r and > 0: Assuming
no money-…nancing of the de…cit, government debt evolves according to Bt+1 Bt
= "Bt + which implies a simple linear di¤erence equation:
Bt+1 = (1 + ")Bt + : (*)
Case 1: " = 0: Then the solution of (*) is
Bt = B0 + t; (**)
B0 being historically given. Then Bt (1 + r) t = B0 (1 + r) t + t(1 + r) t ! 0 for
t ! 1: So, (NPG) is satis…ed. Yet the debt-GDP ratio, Bt =Y; goes to in…nity

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.5. Solvency, the NPG condition, and the intertemporal government budget
constraint 233

for t ! 1: That is, in spite of (NPG) being satis…ed, solvency is not present. For
" = 0 we thus get the insolvency result even though the lower strict inequality in
(6.20) is not satis…ed. Indeed, (**) implies Bt+1 =Bt = 1 + =Bt ! 1 for t ! 1
and 1 + gY = 1:
Case 2: 0 < " < r: Then the solution of (*) is

Bt = (B0 + )(1 + ")t ! 1 for t ! 1;

" "
if B0 > =": So Bt =Y ! 1 for t ! 1 and solvency is violated. Nevertheless
Bt (1 + r) t ! 0 for t ! 1 so that (NPG) holds.
The example of this case fully complies with both strict inequalities in (6.20)
because Bt+1 =Bt = 1 + " + =Bt ! 1 + " for t ! 1:
An approach to …scal budgeting that ensures debt stabilization and thereby
solvency is the following. First impose that the cyclically adjusted primary budget
surplus as a share of GDP equals a constant, s. Next adjust taxes and/or spending
such that s s^ = (r gY )b0 , ignoring short-run di¤erences between Yt+1 =Yt and
1 + gY and between rt and its long-run value, r; as in (6.10), s^ is the minimum
primary surplus as a share of GDP required to obtain bt+1 =bt 1 for all t 0.
This s^ is a measure of the burden that the government debt imposes on tax payers.
If the policy steps needed to realize at least s^ are not taken, the debt-income ratio
will grow, thus worsening the …scal position in the future by increasing s^.

6.5.2 Equivalence of NPG and GIBC

The condition under which the NPG condition is necessary for solvency is also
the condition under which the government’s intertemporal budget constraint is
necessary. To show this we let t denote the current period and t + i denote a
period in the future. As above, we ignore seigniorage. Debt accumulation is then
described by

Bt+1 = (1 + r)Bt + Gt + Xt T~t ; where Bt is given. (6.21)

The government intertemporal budget constraint (GIBC), as seen from the begin-
ning of period t; is the requirement
X
1 X
1
(Gt+i + Xt+i )(1 + r) (i+1)
T~t+i (1 + r) (i+1)
Bt : (GIBC)
i=0 i=0

This condition requires that the present value (PV) of current and expected
future government spending does not exceed the government’s net wealth. The
latter equals the PV of current and expected future tax revenue minus existing

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
234 AND PUBLIC DEBT

P PI
government debt. By the symbol 1 i=0 x i we mean limI!1 i=0 xi : Until further
notice we assume this limit exists.
What connection is there between the dynamic accounting relationship (6.21)
and the intertemporal budget constraint, (GIBC)? To …nd out, we rearrange
(6.21) and use forward substitution to get

Bt = (1 + r) 1 (T~t Xt Gt ) + (1 + r) 1 Bt+1
j
X
= (1 + r) (i+1) (T~t+i Xt+i Gt+i ) + (1 + r) (j+1)
Bt+j+1
i=0
X
1
= (1 + r) (i+1)
(T~t+i Xt+i Gt+i ) + lim (1 + r) (j+1)
Bt+j+1
j!1
i=0
X
1
(1 + r) (i+1)
(T~t+i Xt+i Gt+i ); (6.22)
i=0

if and only if the government debt ultimately grows at a rate less than r so that
(j+1)
lim (1 + r) Bt+j+1 0: (6.23)
j!1

This latter condition is exactly the NPG condition above (replace t in (6.23) by
0 and j by t 1). And the condition (6.22) is just a rewriting of (GIBC). We
conclude:
PROPOSITION 2 Given the book-keeping relation (6.21), then:

(i) (NPG) is satis…ed if and only if (GIBC) is satis…ed;

(ii) there is strict equality in (NPG) if and only if there is strict equality in
(GIBC).

We know from Proposition 1 that in the “normal case”where r > gY ; (NPG) is

needed for government solvency. The message of (i) of Proposition 2 is then that
also (GIBC) need be satis…ed. Given r > gY ; to appear solvent a government has
to realistically plan taxation and spending pro…les such that the PV of current and
expected future primary budget surpluses matches the current debt, cf. (6.22).
Otherwise debt default is looming and forward-looking investors will refuse to
buy government bonds or only buy them at a reduced price, thereby aggravating
the …scal conditions.20
20
Government debt defaults have their own economic as well as political costs, including loss
of credibility. Yet, they occur now and then. Recent examples include Russia in 1998 and
Argentina in 2001-2002. During 2010-12, Greece was on the brink of debt default. At the time
of writing (June 2015) such a situation has turned up again for Greece.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.5. Solvency, the NPG condition, and the intertemporal government budget
constraint 235

In view of the remarks around the inequalities in (6.20), however, satisfying

the condition (6.22) is only a necessary condition (if r > gY ), not in itself a
su¢ cient condition for solvency. A simple condition under which satisfying the
condition (6.22) is su¢ cient for solvency is that both Gt and Tt are proportional
to Yt ; cf. Example 2.
EXAMPLE 2 Consider a small open economy facing an exogenous constant
real interest rate r: Suppose that at time t government debt is Bt > 0; GDP is
growing at the constant rate gY ; and r > gY : Assume Gt = Yt and Tt T~t Xt
= Yt ; where and are positive constants. What is the minimum size of the
primary budget surplus as a share of GDP required for satisfying the government’s
intertemporal budget constraint asPseen from time t? Inserting into the formula
(6.22), with strict equality, yields 1i=0 (1 + r)
(i+1)
( )Yt+i = Bt : This gives
P1 1+gY (i+1)
Y
1+gY t i=0 1+r
= r gY Yt = Bt ; where we have used the rule for the
sum of an in…nite geometric series. Rearranging, we conclude that the required
primary surplus as a share of GDP is

Bt
= (r gY ) :
Yt

This is the same result as in (6.10) above if we substitute s^ = and t = 0.

Thus, maintaining Gt =Yt and Tt =Yt constant while satisfying the government’s
intertemporal budget constraint ensures a constant debt-income ratio and thereby
government solvency.
On the other hand, if r gY ; it follows from propositions 1 and 2 together that
the government can remain solvent without satisfying its intertemporal budget
constraint (at least as long as we ignore uncertainty). The background for this
fact may become more apparent when we recognize how the condition r gY
a¤ects the constraint (GIBC). Indeed, to the extent that the tax revenue tends
to grow at the same rate as national income, we have T~t+i = T~t (1 + gY )i . Then

X
1
T~t X
1
1 + gY
(i+1)
T~t+i (1 + r) (i+1)
= ;
i=0
1 + gY i=0 1+r

which is clearly in…nite if r gY : The PV of expected future tax revenues is thus

unbounded in this case. Suppose that also government spending, Gt+i + Xt+i ;
grows at the rate gY . Then the evolution of the primary surplus is described by
T~t+i Xt+i Gt+i = (T~t (Gt + Xt ))(1 + gY )i ; i = 1; 2; : : : . Although in this
case also the PV of future government spending is in…nite, (6.22) shows that any
positive initial primary budget surplus, T~t (Gt + Xt ); ever so small can repay
any level of initial debt in …nite time.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
236 AND PUBLIC DEBT

In (GIBC) and (6.23) we allow strict inequalities to obtain. What is the

interpretation of a strict inequality here? The answer is:
COROLLARY OF PROPOSITION 2 Given the book-keeping relation (6.21),
then strict inequality in (GIBC) is equivalent to the government in the long run
accumulating positive net …nancial wealth.
Proof. Strict inequality in (GIBC) is equivalent to strict inequality in (6.22),
which in turn, by (ii) of Proposition 2, is equivalent to strict inequality in (6.23),
which is equivalent to limj!1 (1 + r) (j+1) Bt+j+1 < 0: This latter inequality is
equivalent to limj!1 Bt+j+1 < 0, that is, positive net …nancial wealth in the long
run. Indeed, by de…nition, r > 1; hence limj!1 (1 + r) (j+1) 0:
It is common to consider as the regular case the case where the government
does not attempt to accumulate positive net …nancial wealth in the long run
and thereby become a net creditor vis-à-vis the private sector. Returning to
the assumption r > gY ; in the regular case …scal solvency thus amounts to the
requirement
X
1 X
1
T~t+i (1 + r) (i+1)
= (Gt+i + Xt+i )(1 + r) (i+1)
+ Bt ; (GIBC’)
i=0 i=0

which is obtained by rearranging (GIBC) and replacing weak inequality with strict
equality. It is certainly not required that the budget is balanced all the time. The
point is “only” that for a given planned expenditure path, a government should
plan realistically a stream of future tax revenues the PV of which matches the
PV of planned expenditure plus the current debt. If an unplanned budget de…cit
is run so that the public debt rises during a recession, say then higher taxes
than otherwise must be levied in the future.
We may rewrite (GIBC’) as

X
1
T~t+i (Gt+i + Xt+i ) (1 + r) (i+1)
= Bt : (GIBC”)
i=0

This expresses the basic principle that when r > gY ; solvency requires that the
present value of planned future primary surpluses equals the initial debt. If debt
is positive today, then the government has to run a positive primary surplus for
a su¢ ciently long time in the future.

Finer shades
1. If the real interest rate varies over time, all the above formulas remain valid if
(1 + r) (i+1) is replaced by ij=0 (1 + rt+j ) 1 :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.6. A proper accounting of public investment* 237

2. We have essentially ignored seigniorage. Under “normal” circumstances

seigniorage is present and this relaxes (GIBC”) somewhat. Indeed, as noted in
Section 6.2, the money-nominal income ratio, M=P Y; tend to be roughly constant
over time, re‡ecting that money and nominal income tend to grow at the same
rate. So a rough indicator of gM is the sum + gY . Seigniorage is S M=P
= gM M=P = sY; where s is the seigniorage-income ratio. Taking seigniorage into
account amounts to subtracting the present value of expected future seigniorage,
PV(S), from the right-hand side of (GIBC”). With s constant and Y growing at
the constant rate gY < r, PV(S) can be written
X
1 X
1
sYt X
1
1 + gY
(i+1)
(i+1) (i+1)
PV(S) = St+i (1 + r) =s Yt+i (1 + r) =
i=0 i=0
1 + gY i=0 1+r
sYt 1 + gY 1 sYt
= 1+gY = ;
1 + gY 1 + r 1 1+r
r gY
where the second to last equality comes from the rule for the sum of an in…nite
geometric series. So the right-hand side of (GIBC”) becomes Bt sYt =(r gY )
[bt s=(r gY )] Yt :21
3. Should a public de…cit rule not make a distinction between public con-
sumption and public investment? This question is taken up in the next section.

6.6 A proper accounting of public investment*

Public investment as a share of GDP has been falling in the EMU countries since
the middle of the 1970s, in particular since the run-up to the euro 1993-97. This
later development is seen as in part induced by the de…cit rule of the Maastrict
Treaty and the Stability and Growth Pact (SGP) which, like the standard gov-
ernment budget accounting we have considered up to now, attributes government
gross investment as an expense in a single year’s operating account instead of just
the depreciation of the public capital. Already Musgrave (1939) recommended
applying separate capital and operating budgets. Thereby government net in-
vestment will be excluded from the de…nition of the public “budget de…cit”. And
more meaningful de…cit rules can be devised.
To see the gist of this, we partition G into public consumption, C g ; and public
investment, I g ; that is, G = C g + I g : Public investment produces public capital
(infrastructure etc.). Denoting the public capital K g we may write
Kg = Ig Kg; (6.24)
21
In a recession where the economy is in a liquidity trap the non-conventional monetary policy
called Quantitative Easing may partly take the form of seigniorage. This is taken up in Chapter
24.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
238 AND PUBLIC DEBT

where is a (constant) capital depreciation rate. Let the annual (direct) …nancial
return per unit of public capital be rg : This is the sum of user fees and the
like. Net government revenue, T 0 ; now consists of net tax revenue, T; plus the
direct …nancial return rg K g :22 In that now only interest payments and the capital
depreciation, K g ; along with C g ; enter the operating account as “true”expenses,
the “true”budget de…cit is rB + C g + K g T 0 ; where T 0 = T + rg K g :
We impose a rule requiring balancing the “true structural budget”in the sense
that on average over the business cycle

T 0 = rB + C g + K g (6.25)

should hold. The spending on public investment of course enters the debt accu-
mulation equation which now takes the form

B = rB + C g + I g T 0:

Substituting (13.68) into this, we get

B = Ig Kg = Kg; (6.26)

by (13.67). So the balanced “true structural budget” implies that public net
investment is …nanced by an increase in public debt. Other public spending is
tax …nanced.
Suppose that public capital keeps pace with trend GDP, Yt ; thereby growing
at the same constant rate gY > 0: So K g =K g = gY and the ratio K g =Y remains
positive constant at some level, say h: Then (13.69) implies
g
Bt+1 Bt = Kt+1 Ktg = gY Ktg = gY hYt : (6.27)

What is the implication for the evolution of the debt-to-trend-income ratio, ^bt
Bt =Yt , over time? By (6.27) together with Yt+1 = (1 + gY )Yt follows

^bt+1 Bt+1 Bt gY h 1 ^ gY h
= + bt + :
Yt+1 (1 + gY )Yt 1 + gY 1 + gY 1 + gY
This linear …rst-order di¤erence equation has the solution

^bt = (^b0 ^b )(1 + gY ) 1 ^ gY h

t
+ ^b ; where ^b = b + = h;
1 + gY 1 + gY
22
There is also an indirect …nancial return deriving from the fact that better infrastructure
may raise e¢ ciency in the supply of public services and increase productivity in the private
sector and thereby the tax base. While such expected e¤ects matter for a cost-bene…t analysis
of a public investment project, from an accounting point of view they will be included in the
net tax revenue, T; in the future.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.6. A proper accounting of public investment* 239

assuming gY > 0: Then ^bt ! h for t ! 1: Run-away debt dynamics is pre-

cluded.23 Moreover, the ratio Bt =Ktg ; which equals ^bt =h; approaches 1. Eventu-
ally the public debt is in relative terms thus backed by the accumulated public
capital.
Fiscal sustainability is here ensured in spite of a positive “budget de…cit”in
the traditional sense of Section 6.2 and given by B in (13.69). This result holds
even when rg < r; which is perhaps the usual case. Still, the public investment
may be worthwhile in view of indirect …nancial returns as well as non-…nancial
returns in the form of the utility contribution of public goods.

Additional remarks

1. The de…cit rule described says only that the “true structural budget” should
be balanced “on average”over the business cycle. This invites de…cits in slumps
and surpluses in booms. Indeed, in economic slumps government borrowing is
usually cheap. As Harvard economist Lawrence Summers put it: “Idle workers
+ Low interest rates = Time to rebuild infrastructure”(Summers, 2014).
2. When separating government consumption and investment in budget ac-
counting, a practical as well as theoretical issue arises: where to draw the border
between the two? A sizeable part of what is investment in an economic sense is in
standard public sector accounting categorized as “public consumption”: spending
on education, research, and health are obvious examples. Distinguishing between
such categories and public consumption in a narrower sense (administration, ju-
dicial system, police, defence) may be important when economic growth policy is
on the agenda. Apart from noting the issue, we shall not pursue the matter here.
3. That time lags, cf. point (iii) in Section 6.1, are a constraining factor
for …scal policy is especially important for macroeconomic stabilization policy
aiming at dampening business cycle ‡uctuations. If the lags are ignored, there is
a risk that government intervention comes too late and ends up amplifying the
‡uctuations instead of dampening them. In particular the monetarists, lead by
Milton Friedman (1912-2006), warned against this risk. Other economists …nd
awareness of this potential problem relevant but point to ways to circumvent the
problem. During a recession there is for instance the option of reimbursing a
part of last year’s taxes, a policy that can be quickly implemented. At a more
structural level, legislation concerning taxation, transfers, and other spending can
be designed with the aim of strengthening the automatic …scal stabilizers.

23
This also holds if gY = 0: Indeed, in this case, (6.27) implies Bt+1 = Bt = B0 .

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
240 AND PUBLIC DEBT

6.7 Ricardian equivalence?

Having so far concentrated on the issue of …scal sustainability, we shall now
consider how budget policy a¤ects resource allocation and intergenerational dis-
tribution. The role of budget policy for economic activity within a time horizon
corresponding to the business cycle is not the issue here. The focus is on the
longer run: does it matter for aggregate consumption and aggregate saving in
an economy with full capacity utilization whether the government …nances its
current spending by (lump-sum) taxes or borrowing?
There are two opposite answers in the literature to this question. Some macro-
economists tend to answer the question in the negative. This is the debt neutral-
ity view, also called the Ricardian equivalence view. The in‡uential American
economist Robert Barro is in this camp. Other macroeconomists tend to answer
the question in the positive. This is the debt non-neutrality view or absence of
Ricardian equivalence view. The in‡uential French-American economist Olivier
Blanchard is in this camp.
The two di¤erent views rest on two di¤erent models of the economic reality.
The two models have a common point of departure, though, namely a state of
a¤airs where:

1) r > gY ;

2) …scal policy satis…es the intertemporal budget constraint with strict equal-
ity:
X1 X1
~
Tt (1 + r) (t+1)
= (Gt + Xt )(1 + r) (t+1) + B0 ; (6.28)
t=0 t=0

where the initial debt, B0 ; and the planned path of Gt + Xt are given;

3) agents have rational (model consistent) expectations;

4) at least some of the taxes are lump sum and only these are varied in the
thought experiment to be considered;

5) no money …nancing;

6) credit market imperfections are absent.

For a given planned time path of Gt + Xt ; equation (6.28) implies that a tax
cut in any period has to be met by an increase in future taxes of the same present
discounted value as the tax cut.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 241

6.7.1 Two di¤ering views

Ricardian equivalence

The Ricardian equivalence view is the conception that government debt is neutral
in the sense that for a given time path of government spending, aggregate private
consumption is una¤ected by a temporary tax cut. The temporary tax cut does
not make the households feel richer because they expect that the ensuing rise
in government debt will lead to higher taxes in the future. The essential claim
is that the timing of (lump-sum) taxes does not matter. The name Ricardian
equivalence comes from a seemingly false association of this view with the
early nineteenth-century British economist David Ricardo. It is true that Ricardo
articulated the possible logic behind debt neutrality. But he suggested several
reasons that debt neutrality would not hold in practice and in fact he warned
against high public debt levels (Ricardo, 1969, pp. 161-164). Therefore it is
doubtful whether Ricardo was a Ricardian.
Debt neutrality was rejuvenated, however, by Robert Barro in a paper entitled
“Are government bonds net wealth [of the private sector]?”, a question which
Barro answered in the negative (Barro 1974). Barro’s debt neutrality view rests
on a representative agent model, that is, a model where the household sector
is described as consisting of a …xed number of in…nitely-lived forward-looking
“dynasties”. With perfect …nancial markets, a change in the timing of taxes
does not change the PV of the in…nite stream of taxes imposed on the individual
dynasty. A cut in current taxes is o¤set by the expected higher future taxes.
Though current government saving (T G rB) goes down, private saving and
bequests left to the members of the next generation go up equally much.
More precisely, the logic of the debt neutrality view is as follows. Suppose, for
simplicity, that the government waits only 1 period to increase taxes and then does
so in one stroke. Then, for each unit of account current taxes are reduced, taxes
next period are increased by (1+r) units of account. The PV as seen from the end
of the current period of this future tax increase is (1+r)=(1+r) = 1: As 1 1 = 0;
the change in the time pro…le of taxation will make the dynasty feel neither richer
nor poorer. Consequently, its current and planned future consumption will be
una¤ected. That is, its current saving goes up just as much as its current taxation
is reduced. In this way the altruistic parents make sure that the next generation
is fully compensated for the higher future taxes. Current private consumption in
society is thus una¤ected and aggregate saving stays the same.24

24
The complete Barro model is presented in Chapter 7.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
242 AND PUBLIC DEBT

Absence of Ricardian equivalence

Other economists dissociate themselves from such representative agent models
because of their unrealistic description of the household sector. Instead attention
is drawn to overlapping generations models which emphasize …nite lifetime and
life-cycle behavior of human beings and lead to a refutation of Ricardian equiva-
lence. The essential point is that those individuals who bene…t from lower taxes
today will at most be a fraction of those who bear the higher tax burden in the
future. As taxes levied at di¤erent times are thereby levied at partly di¤erent
sets of agents, the timing of taxes generally matters. The current tax cut makes
current tax payers feel wealthier and so they increase their consumption and de-
crease their saving. The present generations bene…t and future tax payers (partly
future generations) bear the cost in the form of access to less national wealth
than otherwise.
The next subsection provides an example showing in detail how a change
in the timing of taxes a¤ects aggregate private consumption in an overlapping
generations framework.

6.7.2 A small open OLG economy with a temporary bud-

get de…cit
We consider a Diamond-style overlapping generations (OLG) model of a small
open economy (henceforth named SOE) with a government sector. The rela-
tionship between SOE and international markets is described by the same four
assumptions as in Section 5.3 of Chapter 5:

(a) There is perfect mobility of goods and …nancial capital across borders.

(b) There is no uncertainty and domestic and foreign …nancial claims are perfect
substitutes.

(c) The need for means of payment is ignored; hence so is the need for a foreign
exchange market.

(d) There is no labor mobility across borders.

The assumptions (a) and (b) imply real interest rate equality. That is, in
equilibrium the real interest rate in SOE must equal the real interest rate in
the world …nancial market, r. And by saying that SOE is “small” we mean
it is small enough to not a¤ect the world market interest rate as well as other
world market factors. We imagine that all countries trade one and the same

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 243

homogeneous good. International trade will then be only intertemporal trade,

i.e., international borrowing and lending of this good.
We assume that r is constant over time and that r > n 0. As earlier
we let Lt denote the size of the young generation and Lt = L0 (1 + n)t . Each
young supplies one unit of labor inelastically, hence Lt is aggregate labor supply.
Assuming full employment, gross domestic product, GDP, is Yt = F (Kt ; Lt ):

Some national accounting for the open economy

Gross national saving is
S t = Yt rN F Dt Ct Gt = Yt rN F Dt (c1t Lt + c2t Lt 1 ) Gt ; (6.29)
where N F Dt is (net) foreign debt (also called external debt) at the beginning
of period t; Gt is government consumption in period t; and c1t and c2t are con-
sumption by a young and an old in period t; respectively. In the open economy,
generally, gross investment, It ; di¤ers from gross saving. If N F Dt > 0; the in-
terpretation is that some of the capital stock, Kt ; is directly or indirectly owned
by foreigners. On the other hand, if N F Dt < 0; SOE has positive net claims on
resources in the rest of the world.
National wealth, Vt ; of SOE at the beginning of period t is, by de…nition,
national assets minus national liabilities,
Vt Kt N F Dt :
National wealth is also, by de…nition, the sum of private …nancial (net) wealth,
At ; and government …nancial (net) wealth, Bt : We assume the government has
no physical assets and Bt is government (net) debt. Thus,
Vt At + ( Bt ): (6.30)
We may also view national wealth from the perspective of national saving.
First, when the young save, they accumulate private …nancial wealth. The private
…nancial wealth at the start of period t+1 must in our Diamond framework equal
N
the (net) saving by the young in the previous period, S1t ; and the latter must
N
equal minus the (net) saving by the old in the next period, S2t+1 :
N N
At+1 = st Lt S1t = S2t+1 : (6.31)
Next, the increase in national wealth equals by de…nition net national saving,
StN ; which in turn equals the sum of net saving by the private sector, S1t
N N
+ S2t ;
N
and the net saving by the public sector, Sgt : So
Vt+1 Vt = St Kt = StN S1t
N N
+ S2t N
+ Sgt = At+1 + ( At ) + ( GBDt )
= At+1 At (Bt+1 Bt );

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
244 AND PUBLIC DEBT

N
where the second to last equality comes from (6.31) and the identity Sgt
GBDt ; while the last equality re‡ects the maintained assumption that bud-
get de…cits are fully …nanced by debt issue.

Firms’behavior
GDP is produced by an aggregate neoclassical production function with CRS:

Yt = F (Kt ; Lt ) = Lt F (kt ; 1) Lt f (kt );

where Kt and Lt are input of capital and labor, respectively, and kt Kt =Lt .
Technological change is ignored. Imposing perfect competition in all markets,
markets clear so that Lt can be interpreted as both employment and labor supply
(exogenous). Pro…t maximization leads to f 0 (kt ) = r + ; where is a constant
capital depreciation rate, 0 1. When f satis…es the condition limk!0 f 0 (k)
0
> r + > limk!1 f (k), there is always a solution in k to this equation and it is
unique (since f 00 < 0) and constant over time (as long as r and are constant):
Thus,
kt = f 0 1 (r + ) k; for all t: (6.32)
The stock of capital, Kt ; is determined by the equation Kt = kLt :
In view of …rms’pro…t maximization, the equilibrium real wage before tax is

@Yt
wt = = f (k) f 0 (k)k w; (6.33)
@Lt
a constant. GDP will evolve according to

Yt = f (k)Lt = f (k)L0 (1 + n)t = Y0 (1 + n)t :

The growth rate of Y thus equals the growth rate of the labor force, i.e., gY = n.

Government and household behavior

We assume that the role of the government sector is to deliver some public good
or service in the amount Gt in period t. Think of a non-rival good like “rule of
law”, TV-transmitted theatre, or another public service free of charge. Suppose

Gt = G0 (1 + n)t ;

where 0 < G0 < F (K0 ; L0 ): It is assumed that the production of Gt uses the
same technology and therefore involves the same unit production costs as the
other components of GDP. As the focus is not on distortionary e¤ects of taxation,

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 245

taxes are assumed to be lump sum, i.e., levied on individuals irrespective of their
economic behavior.
To get explicit solutions, we specify the period utility function to be CRRA:
u(c) = (c1 1)=(1 ); where > 0: To keep things simple, the utility of
the public good enters the life-time utility additively so that it does not a¤ect
marginal utilities of private consumption. In addition we assume that the public
good does not a¤ect productivity in the private sector. There is a tax on the
young as well as the old in period t, 1 and 2 ; respectively. Until further notice
these taxes are time-independent. Possibly, 1 or 2 is negative, in which case
there is a transfer to either the young or the old.
The consumption-saving decision of the young will be the solution to the
following problem:
c11t 1 1 c12t+1 1
max U (c1t ; c2t+1 ) = + v(Gt ) + (1 + ) + v(Gt+1 ) s.t.
1 1
c1t + st = w 1;
c2t+1 = (1 + r)st 2;
c1t 0; c1t+1 0;
where the function v represents the utility contribution of the public good. The
implied Euler equation can be written
1=
c2t+1 1+r
= :
c1t 1+
Inserting the two budget constraints and solving for st , we get
1+ 1=
w 1 + 1+r 2
st = ( 1)=
s(w; r; 1; 2 ): (6.34)
1+r
1 + (1 + ) 1+

Consumption in the …rst and the second period then is

c1t = w 1 st = c^1 (r)ht (6.35)
and
c2t+1 = c^2 (r)ht ; (6.36)
respectively, where
1+
c^1 (r) (1 )=
2 (0; 1) and (6.37)
1+r
1+ + 1+
1=
1+r 1+r
c^2 (r) = c^1 (r) = ( 1)=
(6.38)
1+ 1+r
1 + (1 + ) 1+

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
246 AND PUBLIC DEBT

are the marginal (= average) propensities to consume out of wealth, and where
ht is the after-tax human wealth of the young, i.e., the present value, evaluated
at the end of period t; of disposable lifetime income (the “endowment”). Thus,

2
ht = w 1 h: (6.39)
1+r
Under the given conditions human wealth is thus time-independent. We assume
1 and 2 are such that h > 0: Given r; individual consumption in the …rst as well
as the second period of life is thus proportional to individual human wealth. This
is as expected in view of the homothetic life time utility function. If = r; then
c^1 (r) = c^2 (r) = (1 + r)=(2 + r); that is, there is complete consumption smoothing
as also the Euler equation indicates when = r.25
The tax revenue in period t is Tt = 1 Lt + 2 Lt 1 = ( 1 + 2 =(1 + n))Lt : Let
B0 = 0 and let the “reference path” be a path along which the budget is and
remains balanced for all t; i.e., Tt = Gt = G0 (1 + n)t : In the reference path the
tax code ( 1 ; 2 ) thus satis…es

2
1 + L0 = G0 :
1+n

Consistency with h > 0 in (6.39) requires a “not too large”G0 :

Along the reference path, aggregate private consumption grows at the same
constant rate as GDP and public consumption, the rate n. Indeed,
c2t c2t
Ct = c1t Lt + Lt = (c1t + )L0 (1 + n)t = C0 (1 + n)t :
1+n 1+n

A one-o¤ tax cut

As an alternative to the reference path, consider the case where an unexpected
one-o¤ cut in taxation by x takes place in period 0 for every individual, whether
young or old. Given 0 < x < 1 ; what are the consequences of this? The tax
cut amounts to creating a budget de…cit in period 0 equal to (L0 + L 1 )x: At
the start of period 1 there is thus a government debt B10 = (L0 + L 1 )x, while in
the reference path, B1 = 0. Since we assume r > n = gY ; government solvency
requires that the present value of future taxes, as seen from the beginning of
period 1, rises by (L0 + L 1 )x. This may be accomplished by, for instance,
raising the tax on all individuals from period 1 onward by m. Suppose this way
of addressing the arisen debt is already in period 0 credibly announced by the
25
By calculating backwards from (6.38) to (6.37) to (6.34), the reader may check whether the
calculated st ; c1t and c2t+1 are consistent.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 247

government to be followed. The required value of m will satisfy

X
1
(L0 + L 1 )(1 + n)t m(1 + r) t
= (L0 + L 1 )x:
t=1

This gives
X
1
1+n
t
m = x:
t=1
1+r
As r > n; from the rule for the sum of an in…nite geometric series follows that
r n
m= x m: (6.40)
1+n
The needed rise in future taxes is thus higher the higher is the interest rate r.
This is because the interest burden of the debt will be higher. On the other
hand, a higher population growth rate, n; reduces the needed rise in future taxes.
This is because the interest burden per capita is mitigated by population growth.
Finally, a greater tax cut, x; in the …rst period implies greater tax rises in future
periods.
Let the value of the variables along this alternative path be marked with a
prime. In period 0 the tax cut unambiguously bene…ts the old whose increase in
consumption equals the saved tax:

c020 c20 = x > 0: (6.41)

The young in period 0 know that per capita taxes next period will be increased
by m: In view of the tax cut in period 0, the young nevertheless experiences an
increase in after-tax human wealth equal to

+m 2 2
h00 h0 = w 1 +x w 1
1+r 1+r
r n
= 1 x (by (6.40))
(1 + r)(1 + n)
1 + (2 + r)n
= x > 0:
(1 + r)(1 + n)

Consequently, through the wealth e¤ect this generation enjoys increases in con-
sumption through life equal to

c010 c10 = c^1 (r)(h00 h0 ) > 0; (6.42)

c021 c21 = c^2 (r)(h00 h0 ) > 0; (6.43)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
248 AND PUBLIC DEBT

by (6.35) and (6.36), respectively. So the two generations alive in period 0 gain
from the temporary budget de…cit. But all future generations are worse o¤. These
generations do not bene…t from the tax relief in period 0, but they have to bear
the future cost of the tax relief by a reduction in individual after-tax human
wealth. Indeed, for t = 1; 2; : : : ;

2+m 2
h0t ht = h01 h=w 1 m w 1
1+r 1+r
m
= m+ < 0: (6.44)
1+r

All things considered, since both the young and the old in period 0 increase
their consumption, aggregate consumption in period 0 rises. Ricardian equiva-
lence thus fails.

National saving and wealth accumulation*

The direct impact on national wealth of the temporary tax cut How
N N
does aggregate private net saving, S10 + S20 ; respond to the temporary tax cut?
In both the reference path and the alternative path, the old enter period 0 with
the …nancial wealth A0 and leave the period with zero …nancial wealth. So their
N
net saving is S20 = A0 in both …scal regimes. Although the young in period 0
increase their consumption in response to the temporary tax cut, they increase
their period 0-saving as well. The increased saving by the young is revealed by
the fact that they in period 1, as old, can a¤ord to increase their consumption
in spite of the tax increase of size m in that period. Indeed, from (6.43) and the
period budget constraint as old follows

0 < c021 c21 = (1 + r)s00 ( 2 + m) ((1 + r)s0 2)

= (1 + r)(s00 s0 ) m < (1 + r)(s00 s0 );

thus implying s00 s0 > 0: Since A01 =L0 = s00 > s0 = A1 =L0 ; also aggregate private
…nancial wealth per old at the beginning of period 1 is larger than it would have
been without the temporary tax cut. This might seem paradoxical in view of the
higher aggregate private consumption in period 0. The explanation lies in the
fact that the lower taxation in period 0 means higher disposable income, allowing
both higher private consumption and higher private saving in period 0.
Nevertheless, gross national saving, cf. (6.29), is lower than in the reference
path. Indeed, C00 > C0 implies

S00 = F (K0 ; L0 ) rN F D0 C00 G0 < F (K0 ; L0 ) rN F D0 C0 G0 = S0 :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 249

A counterpart of the increased private saving is the public dissaving, re‡ecting the
budget de…cit created one-to-one by the reduction in taxation. As the increased
disposable income resulting from the latter partly goes to increased private saving
and partly to increased private consumption, the rise in private saving is smaller
than the public dissaving. Consequently, gross national saving ends up lower
than in the reference path.
Net national saving in the reference path is S0N = S0 K0 : The public
dissaving in the alternative path reduces net national saving by the amount

S0N S0N 0 = C00 C0 = c010 L0 + c020 L 1 (c10 L0 + c20 L 1 )

= (c010 c10 )L0 + (c020 c20 )L 1 = c^1 (r)(h00 h0 )L0 + xL 1
1 + (2 + r)n 1
= c^1 (r) xL0 + x L0
(1 + r)(1 + n) 1+n
1 + (2 + r)n 1
= c^1 (r) +1 L0 x > 0: (6.45)
1+r 1+n

For our national income accounting to be consistent, national wealth should

decrease by the same amount as net national saving. Let us check. By the
de…nition (6.30) follows

1 1
V10 = A01 B10 = s00 L0 (1 + )L0 x = (w ( 1 x) c010 x)L0 L0 x
1+n 1+n
1
= (w 1 c^1 (r)h00 ) L0 L0 x
1+n
2+m 1
= w 1 c^1 (r) w 1+x L0 L0 x
1+r 1+n
2 m 1
= w 1 c^1 (r) w 1 L0 c^1 (r) x L0 L0 x
1+r 1+r 1+n
r n 1
= s0 L 0 c^1 (r) 1 L0 x L0 x
(1 + r)(1 + n) 1+n
1 + (2 + r)n 1
= s0 L 0 c^1 (r) +1 L0 x < s0 L0 = V1 : (6.46)
1+r 1+n

We see that national wealth has decreased by an amount equal to the decrease
in net national saving in (6.45), as it should.

Later consequences As revealed by (6.44), all future generations (those born

in period 1; 2; : : : ) are worse o¤ along the alternative path. One might think that
also aggregate private …nancial wealth per old along the alternative path would

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
250 AND PUBLIC DEBT

necessarily be lower. But this is not so. As of period t = 2; 3;. . . , aggregate

private …nancial wealth per old along the alternative path is

A0t =Lt 1 = s0t 1 =w ( 1 + m) c01t 1 =w ( 1 + m) c^1 (r)h0t

2+m
= w ( 1 + m) c^1 (r) w 1 m
1+r
2 m
= w 1 m c^1 (r)(w 1) + c^1 (r)m + c^1 (r) + c^1 (r)
1+r 1+r
2 m
= w 1 c^1 (r)(w 1 ) m + c^1 (r)m + c^1 (r)
1+r 1+r
2 1
= w 1 c^1 (r) w 1 1 c^1 (r) 1 + m
1+r 1+r
2+r r n
= st 1 1 c^1 (r) x: (6.47)
1+r 1+n

Thus, for t = 2; 3; : : : ;

A0t At
Q holds for s0t 1 Q st 1 ; respectively, which in turn holds for
Lt 1 Lt 1
1+r
c^1 (r) Q ; respectively. (6.48)
2+r

In the benchmark case = 1, (6.37) gives c^1 (r) = (1 + r)=(2 + ): In combination

with (6.48), this implies that aggregate private …nancial wealth per old along the
alternative path is lower than, equal to, or higher than that along the reference
path if R r; respectively (in the benchmark case = 1). The reason that it
may be higher is that the saving by the young, which next period constitutes the
private …nancial wealth, has to cover not only the consumption as old but also
the taxes as old which have been increased. In view of st = (c2t+1 + 2 )=(1 + r);
a rise in 2 thus gives scope for a rise in st at the same time as c2t+1 decreases.
For certain, however, national wealth as of period t = 2; 3; : : : ; is smaller
along the alternative path. In Exercise 6.? the reader is asked to show that for
t = 1; 2; : : : ; we have Bt0 = [(2 + n)=(1 + n)] L1 (1 + n)t 2 x: With this evolution of
public debt, the evolution in national wealth per old along the alternative path

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.7. Ricardian equivalence? 251

as of period t = 2; 3; : : : , is
Vt0 A0t Bt 2+n
= s0t 1 x
Lt 1 Lt 1 Lt 1 1+n
2+r 1
= st 1 1 c^1 (r) (r n) + 2 + n x (by (6.47))
1+r 1+n
1 + (2 + r)n 1 1+r
= s0 c^1 (r) +1 x (1 c^1 (r)) x
1+r 1+n 1+n
1 + (2 + r)n 1
< s0 c^1 (r) +1 x = V10 =L0 (by (6.46))
1+r 1+n
A1 At Vt
< s0 = = = ;
L0 Lt 1 Lt 1
where the second to last inequality is due to c^1 (r) < 1; cf. (6.37), while the two
…rst equalities in the last line are due to the constancy of “per old” variables
along the reference path. The last equality is due to the absence of government
debt along that path. So, like period 1, also the subsequent periods experience a
reduction in national wealth as a consequence of the temporary tax cut in period
0.
Period 1 is special, though. While there is a per capita tax increase by m like
in the subsequent periods, period 1’s old generation still bene…ts from the higher
disposable income in period 0. Hence, in period 2 national wealth per old is even
lower than in period 1 but remains constant henceforth.

A closed economy Also in a closed economy would the future generations be

worse o¤ as a result of a temporary tax cut. Indeed, national wealth (which in the
closed economy equals K) would, in view of the reduced national saving in period
0, in period 1 be smaller than otherwise. As of period 2 national wealth would be
even smaller than in period 1, in view of the further reduction in national saving
that occurs in period 1.

Perspectives on the debt neutrality issue

The fundamental point underlined by OLG models is that there is a di¤erence
between the public sector’s future tax base, including the resources of individuals
yet to be born, and the future tax base emanating from individuals alive today.
This may be called the composition-of-tax-base argument for a tendency to non-
neutrality of shifting the timing of (lump-sum) taxation.26
26
In Exercise 6.?? the reader is asked how the burden of the public debt is distributed across
generations if the debt should be completely wiped out through a tax increase in only periods
1 and 2.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
252 AND PUBLIC DEBT

The conclusion that under full capacity utilization budget de…cits imply a
burden for future generations may be seen in a somewhat di¤erent light if per-
sistent technological progress is included in the model. In that case, everything
else equal, future generations will generally be better o¤ than current generations.
Then it might seem less unfair if the former carry some public debt forward to the
latter. In particular this is so if a part of Gt represents spending on infrastructure,
education, research, health, and environmental protection. As future generations
directly bene…t from such investment, it seems fair that they also contribute to
the …nancing. This is the “bene…ts received principle”known from public …nance
theory.
A further concern is whether the economy is a state of full capacity utiliza-
tion or serious unemployment. The above analysis assumes the …rst. What if the
economy in period 0 is in economic depression with high unemployment due to
insu¢ cient aggregate demand? Some economists maintain that also in this situa-
tion is a cut in (lump-sum) taxes to stimulate aggregate demand futile because it
has no real e¤ect. The argument is again that foreseeing the higher taxes needed
in the future, people will save more to prepare themselves (or their descendants
through higher bequests) for paying the higher taxes in the future. The opposite
view is, …rst, that the composition-of-tax-base argument speaks against this as
usual. Second, there is in a depression an additional and quantitatively impor-
tant factor. The “…rst-round”increase in consumption due to the temporary tax
cut raises aggregate demand. Thereby production and income is stimulated and
a further (but smaller) rise in consumption occurs in the “second round”and so
on (the Keynesian multiplier process).
This Keynesian mechanism is important for the debate about e¤ects of budget
de…cits because there are limits to how large deviations from Ricardian equiva-
lence the composition-of-tax-base argument alone can deliver. Indeed, taking into
account the sizeable life expectancy of the average citizen, Poterba and Summers
(1987) point out that the composition-of-tax-base argument by itself delivers only
modest deviations if the issue is timing of taxes over the business cycle.
Another concern is that in the real world, taxes tend to be distortionary and
not lump sum. On the one hand, this should not be seen as an argument against
the possible theoretical validity of the Ricardian equivalence proposition. The
reason is that Ricardian equivalence (in its strict meaning) claims absence of
allocational e¤ects of changes in the timing of lump-sum taxes.
On the other hand, in a wider perspective the interesting question is, of course,
how changes in the timing of distortionary taxes is likely to a¤ect resource allo-
cation. Consider …rst income taxes. When taxes are proportional to income or
progressive (average tax rate rising in income), they provide insurance through re-
ducing the volatility of after-tax income. The fall in taxes in a recession thus helps

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.8. Concluding remarks 253

stimulating consumption through reduced precautionary saving (the phenomenon

that current saving tends to rise in response to increased uncertainty, cf. Chapter
??). In this way, replacing lump-sum taxation by income taxation underpins the
positive wealth e¤ect on consumption, arising from the composition-of-tax-base
channel, of a debt-…nanced tax-cut in an economic recession.
What about consumption taxes? A debt-…nanced temporary cut in consump-
tion taxes stimulates consumption through a positive wealth e¤ect, arising from
the composition-of-tax-base channel. On top of this comes a positive intertempo-
ral substitution e¤ect on current consumption caused by the changed consumer
price time pro…le.
The question whether Ricardian non-equivalence is important from a quan-
titative and empirical point of view pops up in many contexts within macroeco-
nomics. We shall therefore return to the issue several times later in this book.

6.8 Concluding remarks

(incomplete)
Point (iv) in Section 6.1 hints at the fact that when outcomes depend on
forward-looking expectations in the private sector, governments may face a time-
inconsistency problem. In this context time inconsistency refers to the possible
temptation of the government to deviate from its previously announced course
of action once the private sector has acted. An example: With the purpose
of stimulating private saving, the government announces that it will not tax
…nancial wealth. Nevertheless, when …nancial wealth has reached a certain level,
it constitutes a tempting base for taxation and so a tax on wealth might be levied.
To the extent the private sector anticipates this, the attempt to a¤ect private
saving in the …rst place fails. This raises issues of commitment and credibility.
We return to this kind of problems in later chapters.
Finally, point (v) in Section 6.1 alludes to the fact that political processes,
bureaucratic self-interest, rent seeking, and lobbying by powerful interest groups
interferes with …scal policy.27 This is a theme in the branch of economics called
political economy and is outside the focus of this chapter.

6.9 Literature notes

(incomplete)
27
Rent seeking refers to attempts to gain by increasing one’s share of existing wealth, instead
of trying to produce wealth.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
254 AND PUBLIC DEBT

Sargent and Wallace (1981) study consequences of and limits to a shift

from debt …nancing to money …nancing of sustained government budget de…cits
in response to threatening increases in the government debt-income ratio.
How the condition r > gY , for prudent debt policy to be necessary, is modi…ed
when the assumption of no uncertainty is dropped is dealt with in Abel et al.
(1989), Bohn (1995), Ball et al. (1998), and Blanchard and Weil (2001).
Readers wanting to go more into detail with the debate about the design of the
EMU and the Stability and Growth Pact is referred to the discussions in for exam-
ple Buiter (2003), Buiter and Grafe (2004), Fogel and Saxena (2004), Schuknecht
(2005), and Wyplosz (2005). As to discussions of the actual functioning of mone-
tary and …scal policy in the Eurozone in response to the Great Recession, see for
instance the opposing views by De Grauwe and Ji (2013) and Buti and Carnot
(2013). Blanchard and Giavazzi (2004) discuss how proper accounting of public
investment would modify the de…cit and debt rules of the EMU. Beetsma and
Giuliodori (2010) survey recent research of costs and bene…ts of the EMU.
On the theory of optimal currency areas, see Krugman, Obstfeld, and Melitz
(2012).
In addition to the hampering of Keynesian stabilization policy discussed in
Section 6.4.2, also demographic staggering (due to baby booms succeeded by
baby busts) may make rigid de…cit rules problematic. In Denmark for instance
demographic staggering is prognosticated to generate considerable budget de…cits
during several decades after 2030 where younger and smaller generations will suc-
ceed older and larger ones in the labor market. This is prognosticated to take
place, however, without challenging the long-run sustainability of current …scal
policy as assessed by the Danish Economic Council (see the English Summary in
De Økonomiske Råd, 2014). This phenomenon is in Danish known as “hængekø-
jeproblemet”(the “hammock problem”).
Sources for last part of Section 6.7 ....

6.10 Exercises
6.xx In the OLG model of Section 6.6.2, derive (6.37) and (6.38).

6.? In the OLG model of Section 6.6.2, show that for t = 1; 2; 3; : : : ; public debt
along the “alternative path” evolves according to Bt0 = [(2 + n)=(1 + n)] L1 (1 +
n)t 2 x; where x is the temporary per capita tax cut in period 0. Hint: given
the information in Section 6.6.2 you may start by deriving a …rst-order di¤erence
equation in bt Bt =Yt with constant coe¢ cients. The information that the
“reference path" has a balanced budget for all t should be taken into account. In

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

6.10. Exercises 255

addition, you should explain - and apply - that the initial condition is b1 = B1 =Y1
= (2 + n)x= [f (k)(1 + n)2 ] :

6.?? Consider the OLG model of Section 6.6.2. a) Show that if the temporary
per capita tax cut, x; is su¢ ciently small, the debt can be completely wiped out
through a per capita tax increase in only periods 1 and 2. b) Investigate how in
this case the burden of the debt is distributed across generations. Compare with
the alternative debt policy described in the text.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 6. LONG-RUN ASPECTS OF FISCAL POLICY
256 AND PUBLIC DEBT

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 9

The intertemporal
consumption-saving problem in
discrete and continuous time

In the next two chapters we shall discuss and apply the continuous-time ver-
sion of the basic representative agent model, the Ramsey model. As a preparation
for this, the present chapter gives an account of the transition from discrete time
to continuous time analysis and of the application of optimal control theory to
formalize and solve the household’s consumption/saving problem in continuous
time.
There are many …elds in economics where a setup in continuous time is prefer-
able to one in discrete time. One reason is that continuous time formulations
expose the important distinction in dynamic theory between stock and ‡ows in
a much clearer way. A second reason is that continuous time opens up for appli-
cation of the mathematical apparatus of di¤erential equations; this apparatus is
more powerful than the corresponding apparatus of di¤erence equations. Simi-
larly, optimal control theory is more developed and potent in its continuous time
version than in its discrete time version, considered in Chapter 8. In addition,
many formulas in continuous time are simpler than the corresponding ones in
discrete time (cf. the growth formulas in Appendix A).
As a vehicle for comparing continuous time modeling with discrete time mod-
eling we consider a standard household consumption/saving problem. How does
the household assess the choice between consumption today and consumption in
the future? In contrast to the preceding chapters we allow for an arbitrary num-
ber of periods within the time horizon of the household. The period length may
thus be much shorter than in the previous models. This opens up for capturing
additional aspects of economic behavior and for undertaking the transition to

343
 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
344 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

continuous time in a smooth way.

We …rst specify the market environment in which the optimizing household
operates.

9.1 Market conditions

In the Diamond OLG model no loan market was active and wealth e¤ects on
consumption or saving through changes in the interest rate were absent. It is
di¤erent in a setup where agents live for many periods and realistically have a
hump-shaped income pro…le through life. This motivates a look at the …nancial
market and more re…ned notions related to intertemporal choice.

A perfect loan market Consider a given household or, more generally, a

given contractor. Suppose the contractor at a given date t wants to take a loan or
provide loans to others at the going interest rate, it , measured in money terms. So
two contractors are involved, a borrower and a lender. Let the market conditions
satisfy the following four criteria:

(a) the contractors face the same interest rate whether borrowing or lending
(that is, monitoring, administration, and other transaction costs are ab-
sent);

(b) there are many contractors on each side and none of them believe to be able
to in‡uence the interest rate (the contractors are price takers in the loan
market);

(c) there are no borrowing restrictions other than the requirement on the part
of the borrower to comply with her …nancial commitments;

(d) the lender faces no default risk (the borrower can somehow cost-less be
forced to repay the debt with interest on the conditions speci…ed in the
contract).

A loan market satisfying these idealized conditions is called a perfect loan

market. In such a market,

1. various payment streams can be subject to comparison in a simple way; if

they have the same present value (PV for short), they are equivalent;

2. any payment stream can be converted into another one with the same
present value;

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.1. Market conditions 345

3. payment streams can be compared with the value of stocks.

Consider a payment stream fxt gTt=01 over T periods, where xt is the payment
in currency at the end of period t. Period t runs from time t to time t + 1 for t
= 0; 1; :::; T 1: We ignore uncertainty and so it is the interest rate on a riskless
loan from time t to time t + 1: Then the present value, P V0 , as seen from the
beginning of period 0, of the payment stream is de…ned as1
x0 x1 xT 1
P V0 = + + + : (9.1)
1 + i0 (1 + i0 )(1 + i1 ) (1 + i0 )(1 + i1 ) (1 + iT 1)

If Ms. Jones is entitled to the income stream fxt gTt=01 and at time 0 wishes
to buy a durable consumption good of value P V0 , she can borrow this amount
and use a part of the income stream fxt gTt=01 to repay the debt with interest over
the periods t = 0; 1; 2; :::; T 1. In general, when Jones wishes to have a time
pro…le on the payment stream di¤erent from the income stream, she can attain
this through appropriate transactions in the loan market, leaving her with any
stream of payments of the same present value as the given income stream.

Real versus nominal rate of return In this chapter we maintain the as-
sumption of perfect competition in all markets, i.e., households take all prices as
given from the markets. In the absence of uncertainty, the various assets (real
capital, stocks, loans etc.) in which households invest give the same rate of return
in equilibrium. The good which is traded in the loan market can be interpreted
as a (riskless) bond. The borrower issues bonds and the lender buys them. In
this chapter all bonds are assumed to be short-term, i.e., one-period bonds. For
every unit of account borrowed at the end of period t 1, the borrower pays back
with certainty (1 + short-term interest rate) units of account at the end of period
t: If a borrower wishes to maintain debt through several periods, new bonds are
issued at the end of the current period and the obtained loans are spent rolling
over the older loans at the going market interest rate. For the lender, who lends
in several periods, this is equivalent to o¤ering a variable-rate demand deposit
like in a bank.2
Our analysis will be in real terms, that is, in‡ation-corrected terms. In prin-
ciple the unit of account is a …xed bundle of consumption goods. In the simple
macroeconomic models to be studied in this and most subsequent chapters, such
1
We use “present value” as synonymous with “present discounted value”. As usual our
timing convention is such that P V0 denotes the time-0 value of the payment stream, including
the discounted value of the payment (or dividend) indexed by 0.
2
Unless otherwise speci…ed, this chapter uses terms like “loan market” and “bond market”
interchangeably. As uncertainty is ignored, this is legitimate.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
346 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

a bundle is reduced to one consumption good. The models simply assume there
is only one consumption good in the economy. In fact, there will only be one
produced good, “the” output good, which can be used for both consumption and
capital investment. Whether our unit of account is seen as the consumption good
or the output good is thus immaterial.
The real (net) rate of return on an investment is the rate of return in units
of the output good. More precisely, the real rate of return in period t; rt ; is the
(proportionate) rate at which the real value of an investment, made at the end
of period t 1; has grown after one period.
The link between this rate of return and the more commonplace concept of a
nominal rate of return is the following. Imagine that at the end of period t 1
you make a bank deposit of value Vt euro. The real value of the deposit when
you invest is then Vt =Pt 1 ; where Pt 1 is the price in euro of the output good at
the end of period t 1: If the nominal short-term interest rate is it ; the deposit is
worth Vt+1 = Vt (1 + it ) euro at the end of period t: By de…nition of rt ; the factor
by which the deposit in real terms has expanded is

Vt+1 =Pt Vt+1 =Vt 1 + it

1 + rt = = = ; (9.2)
Vt =Pt 1 Pt =Pt 1 1+ t

where t (Pt Pt 1 )=Pt 1 is the in‡ation rate in period t: So the real (net)
rate of return on the investment is rt = (it t )=(1 + t ) it t for it and t
“small”. The number 1 + rt is called the real interest factor and measures the
rate at which current units of output can be traded for units of output one period
later.
In the remainder of this chapter we will think in terms of real values and
completely ignore monetary aspects of the economy.

9.2 Maximizing discounted utility in discrete time

As mentioned, the consumption/saving problem faced by the household is as-
sumed to involve only one consumption good. The composition of consumption
in each period is not part of the problem. What remains is the question how to
distribute consumption over time.

The intertemporal utility function

A plan for consumption in the periods 0; 1; :::; T 1 is denoted fct gTt=01 , where ct
is the consumption in period t. We say the plan has time horizon T: Period 0
(“the initial period”) need not refer to the “birth” of the household but is just
an arbitrary period within the lifetime of the household.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.2. Maximizing discounted utility in discrete time 347

We assume the preferences of the household can be represented by a time-

separable intertemporal utility function with a constant utility discount rate and
no utility from leisure. The latter assumption implies that the labor supply
of the household in each period is inelastic. The time-separability itself just
means that the intertemporal utility function is additive, i.e., U (c0 ; c1 ;. . . ; cT 1 )
= u(0) (c0 ) + u(1) (c1 )+ . . . +u(T 1) (cT 1 ); where u(t) (ct ) is the utility contribution
from period-t consumption, t = 0; 1;. . . ; T 1: In addition we assume geometric
utility discounting, meaning that utility obtained t periods ahead is converted
into a present equivalent by multiplying by the discount factor (1 + ) t ; where
is a constant utility discount rate. So u(t) (ct ) = u(ct )(1 + ) t ; where u(c)
is a time-independent period utility function. Together, these two assumptions
amount to
u(c1 ) u(cT 1 ) X u(ct )
T 1
U (c0 ; c1 ; ; cT 1 ) = u(c0 ) + + ::: + = : (9.3)
1+ (1 + )T 1 t=0
(1 + )t

The period utility function is assumed to satisfy u0 (c) > 0 and u00 (c) < 0. As
explained in Box 9.1, only linear positive transformations of the period utility
function are admissible.
As (9.3) indicates, the number 1 + tells how many units of utility in the next
period the household insists on “in return”for a decrease of one unit of utility in
the current period. So, a > 0 will re‡ect that if the chosen level of consumption
is the same in two periods, then the individual always appreciates a marginal
unit of consumption higher if it arrives in the earlier period. This explains why
is named the rate of time preference or, even more to the point, the rate of
impatience. The utility discount factor, 1=(1 + )t , indicates how many units of
utility the household is at most willing to give up in period 0 to get one additional
unit of utility in period t.3
It is generally believed that human beings are impatient and that should
therefore be assumed positive.4 There is, however, a growing body of evidence
suggesting that the utility discount rate is typically not constant, but declining
with the time distance from the current period to the future periods within the
horizon. This phenomenon is referred to as “present bias” or, with a more tech-
nical term, “hyperbolic discounting”. Macroeconomics often, as a …rst approach,
3
Multiplying through in (9.3) by (1 + ) 1 would make the objective function appear in a
way similar to (9.1) in the sense that also the …rst term in the sum becomes discounted. At the
same time the ranking of all possible alternative consumption paths would remain una¤ected.
For ease of notation, however, we use the form (9.3) which is more standard. Economically,
there is no di¤erence.
4
If uncertainty were included in the model, (1 + ) 1 might be interpreted as (roughly)
re‡ecting the probability of surviving to the next period. In this perspective, > 0 is de…nitely
a plausible assumption.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
348 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

ignores the problem and assumes a constant to keep things simple. We will
generally follow this practice.
For many issues the size of is immaterial. Except when needed, we shall
therefore not impose any other constraint on than the de…nitional requirement
in discrete time that > 1:

Box 9.1. Admissible transformations of the period utility function

When preferences, as assumed here, can be represented by discounted utility, the

concept of utility appears at two levels. The function U in (9.3) is de…ned on
the set of alternative feasible consumption paths and corresponds to an ordinary
utility function in general microeconomic theory. That is, U will express the
same ranking between alternative consumption paths as any increasing transfor-
mation of U . The period utility function, u, de…ned on the consumption in
a single period, is a less general concept, requiring that reference to “utility
units” is legitimate. That is, the size of the di¤erence in terms of period utility
between two outcomes has signi…cance for choices. Indeed, the essence of the
discounted utility hypothesis is that we have, for example,

u(c0 ) u(c00 ) > 0:95 u(c01 ) u(c1 ) , (c0 ; c1 ) (c00 ; c01 );

meaning that the household, having a utility discount factor 1=(1 + ) = 0:95,
0
strictly prefers consuming (c0 ; c1 ) to (c0 ; c01 ) in the …rst two periods, if and only
if the utility di¤erences satisfy the indicated inequality. (The notation x y
means that x is strictly preferred to y:)
Only a linear positive transformation of the utility function u; that is,
v(c) = au(c) + b, where a > 0, leaves the ranking of all possible alternative
T 1
consumption paths, fct gt=0 , unchanged. This is because a linear positive
transformation does not a¤ect the ratios of marginal utilities (the marginal
rates of substitution across time).

The saving problem in discrete time

Suppose the household considered has income from two sources: work and …-
nancial wealth. Let at denote the real value of (net) …nancial wealth held by
the household at the beginning of period t (a for “assets”). We treat at as pre-
determined at time t and in this respect similar to a variable-interest deposit with
a bank. The initial …nancial wealth, a0 ; is thus given, independently of what in
interest rate is formed in the loan market. And a0 can be positive as well as
negative (in the latter case the household is initially in debt).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.2. Maximizing discounted utility in discrete time 349

The labor income of the household in period t is denoted wt 0 and may

follow a typical life-cycle pattern, …rst rising, then more or less stationary, and
…nally vanishing due to retirement. Thus, in contrast to previous chapters where
wt denoted the real wage per unit of labor, here a broader interpretation of wt
is allowed. Whatever the time pro…le of the amount of labor delivered by the
household through life, in this chapter, where the focus is on individual saving,
we regard this time pro…le, as well as the hourly wage as exogenous. The present
interpretation of wt will coincide with the one in the other chapters if we imagine
that the household in each period delivers one unit of labor.
To avoid corner solutions we impose the No Fast Assumption limc!0 u0 (c) =
1: Since uncertainty is by assumption ruled out, the problem is to choose a plan
(c0 ; c1 ;. . . ; cT 1 ) so as to maximize
X
T 1
t
U = u(ct )(1 + ) s.t. (9.4)
t=0
ct 0; (9.5)
at+1 = (1 + rt )at + wt ct ; a0 given; (9.6)
aT 0; (9.7)
where rt is the interest rate. The control region (9.5) re‡ects the de…nitional
non-negativity of the control variable, consumption. The dynamic equation (9.6)
is an accounting relation telling how …nancial wealth moves over time. Indeed,
income in period t is rt at + wt and saving is then rt at + wt ct . Since saving is by
de…nition the same as the increase in …nancial wealth, at+1 at ; we obtain (9.6).
Finally, the terminal condition (9.7) is a solvency requirement that no …nancial
debt be left over at the terminal date, T . We shall refer to this decision problem
as the standard discounted utility maximization problem without uncertainty.

Solving the problem

To solve the problem, let us use the substitution method.5 From (9.6) we have ct
= (1 + rt )at + wt at+1 ; for t = 0; 1;. . . ; T 1: Substituting this into (9.4), we
obtain a function of a1 ; a2 ;. . . ; aT : Since u0 > 0; saturation is impossible and so an
optimal solution cannot have aT > 0: Hence we can put aT = 0 and the problem
is reduced to an essentially unconstrained problem of maximizing a function U~
w.r.t. a1 ; a2 ;. . . ; aT 1 : Thereby we indirectly choose c0 ; c1 ;. . . ; cT 2 : Given aT 1 ;
consumption in the last period is trivially given as
cT 1 = (1 + rT 1 )aT 1 + wT 1;
5
Alternative methods include the Maximum Principle as described in the previous chapter
or Dynamic Programming as described in Math Tools.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
350 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

ensuring
aT = 0; (9.8)
the terminal optimality condition, necessary when u0 (c) > 0 for all c 0 (satu-
ration impossible).
To obtain …rst-order conditions we put the partial derivatives of U~ w.r.t. at+1 ,
t = 0; 1;. . . ; T 2; equal to 0:

@ U~ t
= (1 + ) u0 (ct ) ( 1) + (1 + ) 1 u0 (ct+1 )(1 + rt+1 ) = 0:
@at+1
Reordering gives the Euler equations describing the trade-o¤ between consump-
tion in two succeeding periods,

u0 (ct ) = (1 + ) 1 u0 (ct+1 )(1 + rt+1 ); t = 0; 1; 2; :::; T 2: (9.9)

One of the implications of this condition is that

S rt+1 causes u0 (ct ) T u0 (ct+1 ); i.e., ct S ct+1 (9.10)

in the optimal plan (due to u00 < 0): Absent uncertainty the optimal plan entails
either increasing, constant, or decreasing consumption over time depending on
whether the rate of time preference is below, equal to, or above the rate of return
on saving.

Interpretation The interpretation of (9.9) is as follows. Let the consumption

path (c0 ; c1 ;. . . ; cT 1 ) be our “reference path”. Imagine an alternative path which
coincides with the reference path except for the periods t and t + 1. If it is
possible to obtain a higher total discounted utility than in the reference path
by varying ct and ct+1 within the constraints (9.5), (9.6), and (9.7), at the same
time as consumption in the other periods is kept unchanged, then the reference
path cannot be optimal. That is, “local optimality” is a necessary condition for
“global optimality”. So the optimal plan must be such that the current utility
loss by decreasing consumption ct by one unit equals the discounted expected
utility gain next period by having 1 + rt+1 extra units available for consumption,
namely the gross return on saving one more unit in the current period.
A more concrete interpretation, avoiding the notion of “utility units”, is ob-
tained by rewriting (9.9) as
u0 (ct )
= 1 + rt+1 : (9.11)
(1 + ) 1 u0 (ct+1 )
The left-hand side indicates the marginal rate of substitution, MRS, of period-
(t+1) consumption for period-t consumption, namely the increase in period-(t+1)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.2. Maximizing discounted utility in discrete time 351

consumption needed to compensate for a one-unit marginal decrease in period-t

consumption:
dct+1 u0 (ct )
M RSt+1;t = jU =U = : (9.12)
dct (1 + ) 1 u0 (ct+1 )
And the right-hand side of (9.11) indicates the marginal rate of transformation,
MRT, which is the rate at which the loan market allows the household to shift
consumption from period t to period t + 1:
So, in an optimal plan MRS must equal MRT. This has implications for the
time pro…le of optimal consumption as indicated by the relationship in (9.10).
The Euler equations, (9.9), can also be seen in a comparative perspective. Con-
sider two alternative values of rt+1 : The higher interest rate will induce a negative
substitution e¤ect on current consumption, ct . There is also an income e¤ect,
however, and this goes in the opposite direction. The higher interest rate makes
the present value of a given consumption plan lower. This allows more consump-
tion in all periods for a given total wealth. Moreover, there is generally a third
e¤ect of the rise in the interest rate, a wealth e¤ect. As indicated by the in-
tertemporal budget constraint in (9.20) below, total wealth includes the present
value of expected future after-tax labor earnings and this present value depends
negatively on the interest rate, cf. (9.15) below.
From the formula (9.12) we see one of the reasons that the assumption of a
constant utility discount rate is convenient (but also restrictive). The marginal
rate of substitution between consumption this period and consumption next pe-
riod is independent of the level of consumption as long as this level is the same
in the two periods.
The formula for MRS between consumption this period and consumption two
periods ahead is
dct+2 u0 (ct )
M RSt+2;t = jU =U = :
dct (1 + ) 2 u0 (ct+2 )
This displays one of the reasons that the time-separability of the intertemporal
utility function is a strong assumption. It implies that the trade-o¤ between
consumption this period and consumption two periods ahead is independent of
consumption in the interim.

Deriving the consumption function when utility is CRRA The …rst-

order conditions (9.9) tell us about the relative consumption levels over time,
not the absolute level. The latter is determined by the condition that initial
consumption, c0 ; must be highest possible, given that the …rst-order conditions
and the constraints (9.6) and (9.7) must be satis…ed.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
352 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

To …nd an explicit solution we have to specify the period utility function. As

an example we choose the CRRA function u(c) = c1 =(1 ); where > 0:6
Moreover we simplify by assuming rt = r; a constant > 1. Then the Euler
equations take the form (ct+1 =ct ) = (1 + r)(1 + ) 1 so that
1=
ct+1 1+r
= ; (9.13)
ct 1+

and thereby ct = t c0 ; t = 0; 1;. . . ; T 1: Substituting into the accounting equa-

t
tion (9.6), we thus have at+1 = (1 + r)at + wt c0 : By backward substitution
we …nd the solution of this di¤erence equation to be
" #
X
t 1
at = (1 + r)t a0 + (1 + r) (i+1) (wi i
c0 ) :
i=0

Optimality requires that the left-hand side of this equation vanishes for t = T .
So we can solve for c0 :
" #
1+r X
T 1
1+r
c0 = PT 1 i a0 + (1 + r) (i+1) wi = PT 1 i (a0 + h0 ); (9.14)
i=0 1+r i=0 i=0 1+r

where we have inserted the human wealth of the household (present value of
expected lifetime labor income) as seen from time zero:

X
T 1
(i+1)
h0 = (1 + r) wi : (9.15)
i=0

Thus (9.14) says that initial consumption is proportional to initial total wealth,
the sum of …nancial wealth and human wealth at time 0: To allow for positive
consumption we need a0 + h0 > 0:
In (9.14) is not one of the original parameters, but a derived parameter. To
express the consumption function only in terms of the original parameters, not
that, by (9.14), the propensity to consume out of total wealth depends on:
( T
X
T 1 i 1 ( 1+r )
= 1 1+r
when 6= 1 + r; (9.16)
i=0
1+r T when = 1 + r;
6
In later sections of this chapter we let the time horizon of the decision maker go to in…nity.
To ease convergence of an in…nite sum of discounted utilities, it is an advantage not to have to
bother with additive constants in the period utilities and therefore we write the CRRA function
as c1 =(1 ) instead of the form, (c1 1)=(1 ); introduced in Chapter 3. As implied by
Box 9.1, the two forms represent the same preferences.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.2. Maximizing discounted utility in discrete time 353

where the result for 6= 1 + r follows from the formula for the sum of a …nite
geometric series. Inserting this together with (9.13) into (9.14), we end up with
the expression
8
1=
>
< (1+r)[1 (1+T )= (1+r)
1= (1 )=
] 1+r
1 (1+ ) (1+r) (1 )T = (a 0 + h0 ) when 1+
6= 1 + r;
c0 = 1= (9.17)
>
: 1+r (a0 + h0 ) when 1+r
= 1 + r:
T 1+

This, together with (9.14), thus says:

Result 1 : Consumption is proportional to total wealth, and the factor of
proportionality, often called the marginal propensity to consume out of wealth,
depends on the interest rate r; the time horizon T; and the preference parame-
ters and ; that is, the impatience rate and the strength of the preference for
consumption smoothing, respectively.
For the subsequent periods we have from (9.13) that
!t
1=
1+r
ct = c0 ; t = 1; : : : ; T 1: (9.18)
1+

EXAMPLE 1 Consider the special case = 1 (i.e., u(c) = ln c) together with

> 0: The upper case in (9.17) is here the relevant one and period-0 consumption
will be
(1 + r)(1 (1 + ) 1 )
c0 = (a0 + h0 ) for = 1:
1 (1 + ) T
We see that c0 ! (1 + r) (1 + ) 1 (a0 + h0 ) for T ! 1; assuming the right-hand
side of (9.15) converges for T ! 1:
We have assumed that payment for consumption occurs at the end of the
period at the price 1 per consumption unit. To compare with the corresponding
result in continuous time with continuous compounding (see Section 9.4), we
might want to have initial consumption in the same present value terms as a0
and h0 : That is, we consider c~0 c0 (1 + r) 1 = (1 + ) 1 (a0 + h0 ) for T ! 1:

So far the expression (9.17) is only a candidate consumption function. But

in view of strict concavity of the objective function, (9.17) is indeed the unique
optimal solution when a0 + h0 > 0.
The conclusion from (9.17) and (9.18) is that consumers look beyond current
income. More precisely:
Result 2 : Under the idealized conditions assumed, including a perfect loan
market and perfect foresight, and given the marginal propensity to consume out

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
354 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

of total wealth shown in (9.17), the time pro…le of consumption is determined

by the total wealth and the interest rate (relative to impatience corrected for
the preference for consumption smoothing). The time pro…le of income does not
matter because consumption can be smoothed over time by drawing on the loan
market.
EXAMPLE 2 Consider the special case = r > 0: Again the upper case in (9.17)
is the relevant one and period-0 consumption will be
r
c0 = T
(a0 + h0 ):
1 (1 + r)

We see that c0 ! r(a0 + h0 ) for T ! 1; assuming the right-hand side of (9.15)

converges for T ! 1. So, with an in…nite time horizon current consumption
equals the interest on total current wealth. By consuming this the individual
or household maintains total wealth intact. This consumption function provides
an interpretation of Milton Friedman’s permanent income hypothesis. Friedman
de…ned “permanent income”as “the amount a consumer unit could consume (or
believes it could) while maintaining its wealth intact” (Friedman, 1957). The
key point of Friedman’s theory was the idea that a random change in current
income only a¤ects current consumption to the extent that it a¤ects “perma-
nent income”. Replacing Friedman’s awkward term “permanent income” by the
straightforward “total wealth”, this feature is a general aspect of all consump-
tion functions considered in this chapter. In contrast to this chapter, however,
Friedman emphasized credit market imperfections and thought of a “subjective
income discount rate”of as much as 33% per year. His interpretation of the em-
pirics was that households adopt a much shorter “horizon” than the remainder
of their expected lifetimes (Friedman, 1963, Carroll 2001).
(i+1)
If the real interest rate varies over time, the discount factor (1 + r) for
a payment made at the end of period i is replaced by ij=0 (1 + rj ) 1 :

Alternative approach based on the intertemporal budget constraint

There is another approach to the household’s saving problem. With its choice of
consumption plan the household must act in conformity with its intertemporal
budget constraint (IBC for short). The present value of the consumption plan
(c1 ; :::; cT 1 ), as seen from time zero, is

X
T 1
ct
P V (c0 ; c1 ; :::; cT 1) t
: (9.19)
t=0 =0 (1 + r )

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.2. Maximizing discounted utility in discrete time 355

This value cannot exceed the household’s total initial wealth, a0 + h0 : So the
household’s intertemporal budget constraint is

X
T 1
ct
t
a0 + h0 : (9.20)
t=0 =0 (1 + r )

In this setting the household’s problem is to choose its consumption plan so as

to maximize U in (9.4) subject to this budget constraint.
This way of stating the problem is equivalent to the approach above based
on the dynamic budget condition (9.6) and the solvency condition (9.7). Indeed,
given the accounting equation (9.6), the consumption plan of the household will
satisfy the intertemporal budget constraint (9.20) if and only if it satis…es the
solvency condition (9.7). And there will be strict equality in the intertemporal
budget constraint if and only if there is strict equality in the solvency condition
(the proof is similar to that of a similar claim relating to the government sector
in Chapter 6.2).
Moreover, since in our speci…c saving problem saturation is impossible, an
optimal solution must imply strict equality in (9.20). So it is straightforward to
apply the substitution method also within the IBC approach. Alternatively one
can introduce
P the Lagrange function associated with the problem of maximizing
U = Tt=01 (1 + ) t u(ct ) s.t. (9.20) with strict equality.

In…nite time horizon In the Ramsey model of the next chapter the idea is
used that households may have an in…nite time horizon. One interpretation of
this is that parents care about their children’s future welfare and leave bequests
accordingly. This gives rise to a series of intergenerational links. The household
is then seen as a family dynasty with a time horizon beyond the lifetime of
the current members of the family. Barro’s bequest model in Chapter 7 is an
application of this idea. Given a su¢ ciently large rate of time preference, it is
ensured that the sum of achievable discounted utilities over an in…nite horizon is
bounded from above.
One could say, of course, that in…nity is a long time. The sun will eventually,
in some billion years, burn out and life on earth become extinct. Nonetheless,
there are several reasons that an in…nite time horizon may provide a convenient
substitute for …nite but remote horizons. First, in many cases the solution to
an optimization problem for T “large” is in a major part of the time horizon
close to the solution for T ! 1.7 Second, an in…nite time horizon tends to ease
aggregation because at any future point in time, remaining time is still in…nite.
Third, an in…nite time horizon may be a convenient notion when in any given
7
The turnpike proposition in Chapter 8 exempli…es this.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
356 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

period there is a always a positive probability that there will be a next period to
be concerned about. This probability may be low, but this can be re‡ected in a
high e¤ective utility discount rate. This idea will be applied in chapters 12 and
13.
We may perform the transition to in…nite horizon by letting T ! 1 in the ob-
jective function, (9.4) and the intertemporal budget constraint, (9.20). On might
think that, in analogy of (9.8) for the case of …nite T; the terminal optimality
condition for the case of in…nite horizon is limT !1 aT = 0: This is generally not
so, however. The reason is that with in…nite horizon there is no …nal date where
all debt must be settled. The terminal optimality condition in the present prob-
lem is simply that the intertemporal budget constraint should hold with strict
equality.
As with …nite time horizon, the saving problem with in…nite time horizon
may alternatively be framed in terms of a series of dynamic period-by-period
budget identities, in the form (9.6), together with the borrowing limit known as
the No-Ponzi-Game condition:
t 1 1
lim at i=0 (1 + ri ) 0:
t!1

As we saw in Section 6.5.2 of Chapter 6, such a “‡ow”formulation of the prob-

lem is equivalent to the formulation based on the intertemporal budget constraint.
We also recall from Chapter 6 that the name Ponzi refers to a guy, Charles Ponzi,
who in Boston in the 1920s temporarily became very rich by a loan arrangement
based on the chain letter principle. The fact that debts grow without bounds is
irrelevant for the lender if the borrower can always …nd new lenders and use their
outlay to pay o¤ old lenders with the contracted interest. In the real world, en-
deavours to establish this sort of …nancial eternity machine sooner or later break
down because the ‡ow of new lenders dries up. Such …nancial arrangements,
in everyday speech known as pyramid companies, are universally illegal.8 It is
exactly such arrangements the No-Ponzi-Game condition precludes.
The terminal optimality condition, known as a transversality condition, can
be shown9 to be
lim (1 + ) (t 1) u0 (ct 1 )at = 0:
t!1
8
A related Danish instance, though on a modest scale, could be read in the Danish newpaper
Politiken on the 21st of August 1992. “A twenty-year-old female student from Tylstrup in
Northern Jutland is charged with fraud. In an ad she o¤ered to tell the reader, for 200 DKK,
how to make easy money. Some hundred people responded and received the reply: do like me”.
A more serious present day example is the Wall Street stockbroker, Bernard Mado¤, who
admitted a Ponzi scheme that is considered to be the largest …nancial fraud in U.S. history. In
2009 Mado¤ was sentenced to 150 years in prison. Other examples of large-scale Ponzi games
appeared in Albania 1995-97 and Ukraine 2008.
9
The proof is similar to that given in Chapter 8, Appendix C.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.3. Transition to continuous time analysis 357

9.3 Transition to continuous time analysis

In the discrete time framework the run of time is divided into successive periods
of equal length, taken as the time-unit. Let us here index the periods by i =
0; 1; 2; :::. Thus …nancial wealth accumulates according to

ai+1 ai = s i ; a0 given,

where si is (net) saving in period i:

Multiple compounding per year

With time ‡owing continuously, we let a(t) refer to …nancial wealth at time t:
Similarly, a(t + t) refers to …nancial wealth at time t + t: To begin with, let
t equal one time unit. Then a(i t) equals a(i) and is of the same value as ai :
Consider the forward …rst di¤erence in a; a(t) a(t+ t) a(t): It makes sense
to consider this change in a in relation to the length of the time interval involved,
that is, to consider the ratio a(t)= t: As long as t = 1; with t = i t we have
a(t)= t = (ai+1 ai )=1 = ai+1 ai :
Now, keep the time unit unchanged, but let the length of the time interval
[t; t + t) approach zero, i.e., let t ! 0. When a is a di¤erentiable function of
t, we have
a(t) a(t + t) a(t) da(t)
lim = lim = ;
t!0 t t!0 t dt
where da(t)=dt; often written a(t);
_ is known as the derivative of a at the point t:
Wealth accumulation in continuous time can then be written

a(t)
_ = s(t); a(0) = a0 given, (9.21)

where s(t) is the saving ‡ow (saving intensity) at time t. For t “small”we have
the approximation a(t) a(t) _ t = s(t) t: In particular, for t = 1 we have
a(t) = a(t + 1) a(t) s(t):
As time unit choose one year. Going back to discrete time, if wealth grows at
a constant rate g per year, then after i periods of length one year, with annual
compounding, we have

ai = a0 (1 + g)i ; i = 0; 1; 2; ::: . (9.22)

If instead compounding (adding saving to the principal) occurs n times a year,

then after i periods of length 1=n year and a growth rate of g=n per such period,
we have
g
ai = a0 (1 + )i : (9.23)
n
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
358 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

With t still denoting time measured in years passed since date 0, we have i = nt
periods. Substituting into (9.23) gives
gt
g 1 n
a(t) = ant = a0 (1 + )nt = a0 (1 + )m ; where m :
n m g

We keep g and t …xed, but let n ! 1: Thus m ! 1: In the limit there is

continuous compounding and we get

a(t) = a0 egt ; (9.24)

where e is a mathematical constant called the base of the natural logarithm and
de…ned as e limm!1 (1 + 1=m)m ' 2.7182818285....
The formula (9.24) is the continuous-time analogue to the discrete time for-
mula (9.22) with annual compounding. A geometric growth factor is replaced by
an exponential growth factor, egt ; and this growth factor is valid for any t in the
time interval ( 1 ; 2 ) for which the growth rate of a equals the constant g ( 1
and 2 being some positive real numbers).
We can also view the formulas (9.22) and (9.24) as the solutions to a di¤erence
equation and a di¤erential equation, respectively. Thus, (9.22) is the solution to
the linear di¤erence equation ai+1 = (1 + g)ai , given the initial value a0 : And
(9.24) is the solution to the linear di¤erential equation a(t)
_ = ga(t); given the
initial condition a(0) = a0 : Now consider a time-dependent growth rate, g(t); a
continuous function of t: The corresponding di¤erential equation is a(t)
_ = g(t)a(t)
and it has the solution Rt
a(t) = a(0)e 0 g( )d ; (9.25)
Rt
where the exponent, 0 g( )d , is the de…nite integral of the function g( ) from 0
to t: The result
Rt
(9.25) is called the accumulation formula in continuous time and
the factor e 0 )d is called the growth factor or the accumulation factor.10
g(

Compound interest and discounting in continuous time

Let r(t) denote the short-term real interest rate in continuous time at time t.
To clarify what is meant by this, consider a deposit of V (t) euro in a bank at
time t. If the general price level in the economy at time t is P (t) euro, the real
value of the deposit is a(t) = V (t)=P (t) at time t: By de…nition the real rate of
return on the deposit in continuous time (with continuous compounding) at time
t is the (proportionate) instantaneous rate at which the real value of the deposit
expands per time unit when there is no withdrawal from the account. Thus, if
10
Sometimes the accumulation factor with time-dependent growth rate is written in a di¤erent
way, see Appendix B.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.3. Transition to continuous time analysis 359

the instantaneous nominal interest rate is i(t); we have V_ (t)=V (t) = i(t) and so,
by the fraction rule in continuous time (cf. Appendix A),
a(t)
_ V_ (t) P_ (t)
r(t) = = = i(t) (t); (9.26)
a(t) V (t) P (t)
where (t) P_ (t)=P (t) is the instantaneous in‡ation rate. In contrast to the
corresponding formula in discrete time, this formula is exact. Sometimes i(t) and
r(t) are referred to as the nominal and real force of interest.
Calculating the terminal value of the deposit at time t1 > t0 ; given its value at
time t0 and assuming no withdrawal in the time interval [t0 ; t1 ], the accumulation
formula (9.25) immediately yields
R t1
r(t)dt
a(t1 ) = a(t0 )e t0
:
When calculating present values in continuous time, we use compound dis-
counting. We reverse the accumulation formula and go from the compounded or
terminal value to the present value, a(t0 ). Similarly, given a consumption plan
(c(t))tt=t
1
0
, the present value of this plan as seen from time t0 is
Z t1
PV = c(t) e rt dt; (9.27)
t0

presupposing a constant interest rate, r. Instead of the geometric discount factor,

1=(1+r)t ; from discrete time analysis, we have here an exponential discount factor,
1=(ert ) = e rt ; and instead of a sum, an integral. When the interest rate varies
over time, (9.27) is replaced by
Z t1 Rt
r( )d
PV = c(t) e t0 dt:
t0

In (9.27) c(t) is discounted by e rt (1 + r) t for r “small”. This might not

seem analogue to the discrete-time discounting in (9.19) where it is ct 1 that is
discounted by (1 + r) t ; assuming a constant interest rate. When taking into
account the timing convention that payment for ct 1 in period t 1 occurs at
the end of the period (= time t); there is no discrepancy, however, since the
continuous-time analogue to this payment is c(t).

The range for particular parameter values

The allowed range for parameters may change when we go from discrete time to
continuous time with continuous compounding. For example, the usual equation
for aggregate capital accumulation in continuous time is
_
K(t) = I(t) K(t); K(0) = K0 given, (9.28)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
360 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

where K(t) is the capital stock, I(t) is the gross investment at time t and 0
is the (physical) capital depreciation rate. Unlike in discrete time, here > 1 is
conceptually allowed. Indeed, suppose for simplicity that I(t) = 0 for all t 0;
then (9.28) gives K(t) = K0 e t . This formula is meaningful for any 0:
Usually, the time unit used in continuous time macro models is one year (or, in
business cycle theory, rather a quarter of a year) and then a realistic value of
is of course < 1 (say, between 0.05 and 0.10). However, if the time unit applied
to the model is large (think of a Diamond-style OLG model), say 30 years, then
> 1 may …t better, empirically, if the model is converted into continuous time
with the same time unit. Suppose, for example, that physical capital has a half-
life of 10 years. With 30 years as our time unit, inserting into the formula 1=2
= e =3 gives = (ln 2) 3 ' 2:
In many simple macromodels, where the level of aggregation is high, the
relative price of a unit of physical capital in terms of the consumption good is
1 and thus constant. More generally, if we let the relative price of the capital
good in terms of the consumption good at time t be p(t) and allow p(t)
_ 6= 0; then
we have to distinguish between the physical depreciation of capital, ; and the
economic depreciation, that is, the loss in economic value of a machine per time
unit. The economic depreciation will be d(t) = p(t) p(t);
_ namely the economic
value of the physical wear and tear (and technological obsolescence, say) minus
the capital gain (positive or negative) on the machine.
Other variables and parameters that by de…nition are bounded from below
in discrete time analysis, but not so in continuous time analysis, include rates of
return and discount rates in general.

Stocks and ‡ows

An advantage of continuous time analysis is that it forces the analyst to make
a clear distinction between stocks (say wealth) and ‡ows (say consumption or
saving). Recall, a stock variable is a variable measured as a quantity at a given
point in time. The variables a(t) and K(t) considered above are stock variables.
A ‡ow variable is a variable measured as quantity per time unit at a given point
_
in time. The variables s(t); K(t) and I(t) are ‡ow variables.
One can not add a stock and a ‡ow, because they have di¤erent denomina-
tions. What is meant by this? The elementary measurement units in economics
are quantity units (so many machines of a certain kind or so many liters of oil
or so many units of payment, for instance) and time units (months, quarters,
years). On the basis of these elementary units we can form composite mea-
surement units. Thus, the capital stock, K; has the denomination “quantity of
machines”, whereas investment, I; has the denomination “quantity of machines
per time unit” or, shorter, “quantity/time”. A growth rate or interest rate has

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.3. Transition to continuous time analysis 361

Figure 9.1: With t small the integral of s(t) from t0 to t0 + t the hatched area.

the denomination “(quantity/time)/quantity”= “time 1 ”. If we change our time

unit, say from quarters to years, the value of a ‡ow variable as well as a growth
rate is changed, in this case quadrupled (presupposing annual compounding).
In continuous time analysis expressions like K(t) + I(t) or K(t) + K(t) _ are
thus illegitimate. But one can write K(t + t) K(t) + (I(t) K(t)) t; or
_
K(t) t (I(t) K(t)) t: In the same way, suppose a bath tub at time t
contains 50 liters of water and that the tap pours 12 liter per second into the
tub for some time. Then a sum like 50 ` + 21 (`/sec) does not make sense. But
the amount of water in the tub after one minute is meaningful. This amount
would be 50 ` + 12 60 ((`/sec) sec) = 80 `. In analogy, economic ‡ow variables
in continuous time should be seen as intensities de…ned for every t in the time
interval considered, say the time interval [0, T ) or perhaps [0, 1). For example,
when we say that I(t) is “investment” at time t, this is really a short-hand
for “investment intensity” at time t. The actual investment in a time interval
[t ; t + t) ; i.e., the invested amount during this time interval, is the integral,
R 0t0 +0 t
t0
I(t)dt I(t0 ) t: Similarly, the ‡ow of individual saving, s(t); should be
interpreted as the saving intensity (or saving density), at time t: The actual saving
in a time interval [t0 ; t0 + t) ; i.e., the saved (or accumulated) amount during
Rt + t
this time interval, is the integral, t00 s(t)dt: If t is “small”, this integral is
approximately equal to the product s(t0 ) t, cf. the hatched area in Fig. 9.1.
The notation commonly used in discrete time analysis blurs the distinction
between stocks and ‡ows. Expressions like ai+1 = ai + si ; without further com-
ment, are usual. Seemingly, here a stock, wealth, and a ‡ow, saving, are added.
In fact, however, it is wealth at the beginning of period i and the saved amount
during period i that are added: ai+1 = ai + si t. The tacit condition is that
the period length, t; is the time unit, so that t = 1. But suppose that, for
example in a business cycle model, the period length is one quarter, but the time
unit is one year. Then saving in quarter i is si = (ai+1 ai ) 4 per year.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
362 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

The choice between discrete and continuous time formulation

In empirical economics, data typically come in discrete time form and data for
‡ow variables typically refer to periods of constant length. One could argue that
this discrete form of the data speaks for discrete time rather than continuous
time modelling. And the fact that economic actors often think, decide, and plan
in period terms, may seem a good reason for putting at least microeconomic
analysis in period terms. Nonetheless real time is continuous. Moreover, as for
instance Allen (1967) argued, it can hardly be said that the mass of economic
actors think and decide with the same time distance between successive decisions
and actions. In macroeconomics we consider the sum of the actions. In this
perspective the continuous time approach has the advantage of allowing variation
within the usually arti…cial periods in which the data are chopped up. In addition,
centralized asset markets equilibrate very fast and respond almost immediately
to new information. For such markets a formulation in continuous time seems a
good approximation.
There is also a risk that a discrete time model may generate arti…cial oscil-
lations over time. Suppose the “true” model of some mechanism is given by the
di¤erential equation
x_ = x; < 1: (9.29)
The solution is x(t) = x(0)e t which converges in a monotonic way toward 0 for
t ! 1: However, the analyst takes a discrete time approach and sets up the
seemingly “corresponding”discrete time model
xt+1 xt = x t :
This yields the di¤erence equation xt+1 = (1+ )xt , where 1+ < 0: The solution
is xt = (1+ )t x0 ; t = 0; 1; 2; : : : : As (1+ )t is positive when t is even and negative
when t is odd, oscillations arise (together with divergence if < 2) in spite of
the “true” model generating monotonous convergence towards the steady state
x = 0.
This potential problem can always be avoided, however, by choosing a su¢ -
ciently short period length in the discrete time model. The solution to a di¤eren-
tial equation can always be obtained as the limit of the solution to a corresponding
di¤erence equation for the period length approaching zero. In the case of (9.29),
the approximating di¤erence equation is xi+1 = (1 + t)xi ; where t is the
period length, i = t= t, and xi = x(i t): By choosing t small enough, the
solution comes arbitrarily close to the solution of (9.29). It is generally more
di¢ cult to go in the opposite direction and …nd a di¤erential equation that ap-
proximates a given di¤erence equation. But the problem is solved as soon as a
di¤erential equation has been found that has the initial di¤erence equation as an
approximating di¤erence equation.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 363

From the point of view of the economic contents, the choice between discrete
time and continuous time may be a matter of taste. Yet, everything else equal, the
clearer distinction between stocks and ‡ows in continuous time than in discrete
time speaks for the former. From the point of view of mathematical convenience,
the continuous time formulation, which has worked so well in the natural sciences,
is preferable. At least this is so in the absence of uncertainty. For problems where
uncertainty is important, discrete time formulations are easier to work with unless
one is familiar with stochastic calculus.11

9.4 Maximizing discounted utility in continuous

time
9.4.1 The saving problem in continuous time
In continuous time the analogue to the intertemporal utility function, (9.3), is
RT
U0 = 0 u(c(t))e t dt: (9.30)
In this context it is common to name the utility ‡ow, u; the instantaneous utility
function. We still assume that u0 > 0 and u00 < 0: The analogue in continuous
time to the intertemporal budget constraint (9.20) is
RT Rt
0
c(t)e 0 r( )d dt a0 + h0 ; (9.31)
where, as before, a0 is the historically given initial …nancial wealth, while h0 is
the given human wealth,
RT Rt
h0 = 0 w(t)e 0 r( )d dt: (9.32)
The household’s problem is then to choose a consumption plan (c(t))Tt=0 so as
to maximize discounted utility, U0 ; subject to the budget constraint (9.31).

In…nite time horizon Transition to in…nite horizon is performed by letting

T ! 1 in (9.30), (9.31), and (9.32). In the limit the household’s, or dynasty’s,
problem becomes one of choosing a plan, (c(t))1
t=0 , which maximizes
Z 1
U0 = u(c(t))e t dt s.t. (9.33)
0
Z 1 Rt
c(t)e 0 r( )d dt a0 + h0 ; (IBC)
0
11
In the latter case, the arguments by Nobel laureate Robert C. Merton in favor of a contin-
uous time formulation are worth consideration.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
364 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

where h0 emerges by letting T in (9.32) approach 1. With an in…nite horizon

there may exist technically feasible paths along which the integrals in (9.30),
(9.31), and (9.32) go to 1 for T ! 1: In that case maximization is not well-
de…ned. However, the assumptions we are going to make when working with
in…nite horizon will guarantee that the integrals converge as T ! 1 (or at least
that some feasible paths have 1 < U0 < 1; while the remainder have U0
= 1 and are thus clearly inferior). The essence of the matter is that the rate
of time preference, ; must be assumed su¢ ciently high.
Generally we de…ne a person as solvent if she is able to meet her …nancial
obligations as they fall due. Each person is considered “small” relative to the
economy as a whole. As long as all agents in an economy with a perfect loan
market remain “small”, they will in general equilibrium remain solvent if and
only if their gross debt does not exceed their gross assets. The “gross assets”
should be understood as including the present value of the expected future labor
income. Considering the net debt d0 gross debt gross assets, the solvency
requirement becomes
Z 1 Rt
d0 (w(t) c(t))e 0 r( )d dt;
0

where the right-hand side of the inequality is the present value of the expected
future primary saving.12 By the de…nition in (9.32), we see that this requirement
is identical to the intertemporal budget constraint (IBC) which consequently
expresses solvency.

The budget constraint in ‡ow terms

The method which is particularly apt for solving intertemporal decision problems
in continuous time is based on the mathematical discipline optimal control theory.
To apply the method, we have to convert the household’s budget constraint from
the present-value formulation considered above into ‡ow terms.
By mere accounting, in every short time interval (t, t + t) the household’s
consumption plus saving equals the household’s total income, that is,
(c(t) + a(t))
_ t = (r(t)a(t) + w(t)) t:
Here, a(t)
_ da(t)=dt is the increase per time unit in …nancial wealth, and thereby
the saving intensity, at time t (assuming no robbery). If we divide through by
t and rearrange, we get for all t 0
a(t)
_ = r(t)a(t) + w(t) c(t); a(0) = a0 given. (9.34)
12
By primary saving is meant the di¤erence between current earned income and current
consumption, where earned income means income before interest transfers.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 365

This equation in itself is just a dynamic budget identity. It tells how much
and in which direction the …nancial wealth is changing due to the di¤erence
between current income and current consumption. The equation per se does
not impose any restriction on consumption over time. If this equation were the
only “restriction”, one could increase consumption inde…nitely by incurring an
increasing debt without limits. It is not until we add the requirement of solvency
that we get a constraint. When T < 1, the relevant solvency requirement is
a(T ) 0 (that is, no debt is left over at the terminal date). This is equivalent to
satisfying the intertemporal budget constraint (9.31).
When T = 1; the relevant solvency requirement is the No-Ponzi-Game con-
dition Rt
lim a(t)e 0 r( )d 0: (NPG)
t!1

This condition says that the present value of debts, measured as a(t); in…nitely
far out in the future, is not permitted to be positive. We have the following
equivalency:
PROPOSITION 1 (equivalence of NPG condition and intertemporal budget con-
straint) Let the time horizon be in…nite and assume that the integral (9.32)
remains …nite for T ! 1. Then, given the accounting relation (9.34), we have:
(i) the requirement (NPG) is satis…ed if and only if the intertemporal budget
constraint, (IBC), is satis…ed; and
(ii) there is strict equality in (NPG) if and only if there is strict equality in (IBC).
Proof. See Appendix C.
The condition (NPG) does not preclude that the household, or family dynasty,
can remain in debt. This would also be an unnatural requirement as the dynasty
is in…nitely-lived. The condition does imply, however, that there is an upper
bound for the speed whereby debt can increase in the long term. The NPG
condition says that in the long term, debts are not allowed to grow at a rate as
high as (or higher than) the interest rate.
To understand the implication, consider the case with a constant interest rate
r > 0. Assume that the household at time t has net debt d(t) > 0, i.e., a(t)
d(t) < 0. If d(t) were persistently growing at a rate equal to or greater than
the interest rate, (NPG) would be violated.13 Equivalently, one can interpret
(NPG) as an assertion that lenders will only issue loans if the borrowers in the
long run cover their interest payments by other means than by taking up new
_
loans. In this way, it is avoided that d(t) rd(t) in the long run. In brief, the
borrowers are not allowed to run a Ponzi Game.
13 _
Starting from a given initial positive debt, d0 ; when d(t)=d(t) r > 0; we have d(t) d0 ert
so that d(t)e rt d0 > 0 for all t 0. Consequently, a(t)e rt = d(t)e rt d0 < 0 for
all t 0; that is, lim t!1 a(t)e rt < 0; which violates (NPG).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
366 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

9.4.2 Solving the saving problem

The household’s consumption/saving problem is one of choosing a path for the
control variable c(t) so as to maximize a criterion function, in the form of an in-
tegral, subject to constraints that include a …rst-order di¤erential equation where
the control variable enters, namely (9.34). Choosing a time path for the con-
trol variable, this equation determines the evolution of the state variable, a(t):
Optimal control theory, which in Chapter 8 was applied to a related discrete
time problem, o¤ers a well-suited apparatus for solving this kind of optimization
problems. We will make use of a special case of Pontryagin’s Maximum Principle
(the basic tool of optimal control theory) in its continuous time version. We shall
consider both the …nite and the in…nite horizon case. The only regularity con-
dition required is that the exogenous variables, here r(t) and w(t); are piecewise
continuous and that the control variable, here c(t); is piecewise continuous and
take values within some given set C R, called the control region.
For T < 1 the problem is: choose a plan (c(t))Tt=0 that maximizes
Z T
U0 = u(c(t))e t dt s.t. (9.35)
0
c(t) 0; (control region) (9.36)
a(t)
_ = r(t)a(t) + w(t) c(t); a(0) = a0 given, (9.37)
a(T ) 0: (9.38)
With an in…nite time horizon, T in (9.35) is interpreted as 1 and the solvency
condition (9.38) is replaced by
Rt
r( )d
lim a(t)e 0 0: (NPG)
t!1

Let I denote the time interval [0; T ] if T < 1 and the time interval [0; 1)
if T = 1: If c(t) and the corresponding evolution of a(t) ful…l (9.36) and (9.37)
for all t 2 I as well as the relevant solvency condition, we call (a(t); c(t))Tt=0 an
admissible path. If a given admissible path (a(t); c(t))Tt=0 solves the problem, it is
referred to as an optimal path.14 We assume that w(t) > 0 for all t: No condition
on the impatience parameter is imposed (in this chapter).

First-order conditions
The solution procedure for this problem is as follows:15
14
The term “path”, sometimes “trajectory”, is common in the natural sciences for a solution
to a di¤erential equation because one may think of this solution as the path of a particle moving
in two- or three-dimensional space.
15
The four-step solution procedure below is applicable to a large class of dynamic optimization
problems in continuous time, see Math tools.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 367

1. We set up the current-value Hamiltonian function (often just called the

current-value Hamiltonian):

H(a; c; ; t) u(c) + (ra + w c);

where is the adjoint variable (also called the co-state variable) associated
with the dynamic constraint (9.37).16 That is, is an auxiliary variable
which is a function of t and is analogous to the Lagrange multiplier in
static optimization.
2. At every point in time, we maximize the Hamiltonian w.r.t. the control
variable. Focusing on an interior optimal path,17 we calculate
@H
= u0 (c) = 0:
@c
For every t 2 I we thus have the condition

u0 (c(t)) = (t): (9.39)

3. We calculate the partial derivative of H with respect to the state variable

and put it equal to minus the time derivative of plus the discount rate
(as it appears in the integrand of the criterion function) multiplied by :
@H _+
= r= :
@a
This says that, for all t 2 I, the adjoint variable should ful…l the di¤er-
ential equation
_ (t) = ( r(t)) (t): (9.40)

4. We now apply the Maximum Principle which applied to this problem says:
an interior optimal path (a(t); c(t))Tt=0 will satisfy that there exits a contin-
uous function = (t) such that for all t 2 I; (9.39) and (9.40) hold along
the path, and the transversality condition,

a(T ) (T ) = 0; if T < 1; and

lim a(t) (t)e t = 0; if T = 1; (TVC)
t!1

is satis…ed.
16
The explicit dating of the time-dependent variables a, c; and is omitted where not needed
for clarity.
17
A path, (at ; ct )Tt=0 ; is an interior path if for no t 2 [0; T ) ; (at ; ct ) is at a boundary point of
the set of admissible values. In the present case where at is not constrained, except at t = T;
(at ; ct )Tt=0 ; is an interior path if ct > 0 for all t 2 [0; T ) :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
368 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Let us provide some interpretation of these optimality conditions. Overall,

the Maximum Principle characterizes an optimal path as a path that for every
t maximizes the Hamiltonian associated with the problem. The intuition is that
the Hamiltonian weighs the direct contribution of the marginal unit of the con-
trol variable to the criterion function in the “right” way relative to the indirect
contribution, which comes from the generated change in the state variable (here
…nancial wealth); “right”means in accordance with the opportunities o¤ered by
the rate of return vis-a-vis the time preference rate, : The optimality condition
(9.39) can be seen as a M C = M B condition in utility terms: on the margin
one unit of account (here the consumption good) must be equally valuable in its
two uses: consumption and wealth accumulation. Together with the optimality
condition (9.40) this signi…es that the adjoint variable can be interpreted as
the shadow price (measured in units of current utility) of …nancial wealth along
the optimal path.18
Reordering the di¤erential equation (9.40) gives

r +_
= : (9.41)

This can be interpreted as a no-arbitrage condition. The left-hand side gives the
actual rate of return, measured in utility units, on the marginal unit of saving.
Indeed, r can be seen as a dividend and _ as a capital gain. The right-hand side
is the required rate of return in utility units, : Along an optimal path the two
must coincide. The household is willing to save the marginal unit of income only
up to the point where the actual return on saving equals the required return.
We may alternatively write the no-arbitrage condition as

_
r= : (9.42)

On the left-hand-side appears the actual real rate of return on saving and on
the right-hand-side the required real rate of return. The intuition behind this
condition can be seen in the following way. Suppose Mr. Jones makes a deposit
of V utility units in a “bank”that o¤ers a proportionate rate of expansion of the
utility value of the deposit equal to i (assuming no withdrawal occurs), i.e.,

V_
= i:
V
18
Recall, a shadow price (measured in some unit of account) of a good is, from the point of
view of the buyer, the maximum number of units of account that the optimizing buyer is willing
to o¤er for one extra unit of the good.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 369

This is the actual utility rate of return, a kind of “nominal interest rate”. To
calculate the corresponding “real interest rate”let the “nominal price”of a con-
sumption good be utility units. Dividing the number of invested utility units,
V; by ; we get the real value, m = V = ; of the deposit at time t. The actual real
rate of return on the deposit is therefore

m
_ V_ _ _
r= = =i : (9.43)
m V
Mr. Jones is just willing to save the marginal unit of income if this actual
real rate of return on saving equals the required real rate, that is, the right-hand
side of (9.42); in turn this necessitates that the “nominal interest rate”, i; in
(9.43) equals the required nominal rate, : The formula (9.43) is analogue to the
discrete-time formula (9.2) except that the unit of account in (9.43) is current
utility while in (9.2) it is currency.
The transversality condition (TVC) is a terminal optimality condition. We
could, for the case T < 1; have expressed it on the equivalent form
T
a(T ) (T )e = 0;

since e T > 0 always. This form has the advantage of being “parallel” to the
transversality condition for the case T = 1: More importantly, the transversal-
ity condition has a¢ nity with the principle of complementary slackness in linear
and nonlinear programming. Let us spell out in general terms. Consider the case
T < 1. Interpret the solvency condition a(T ) 0 as just an example of a general
terminal constraint a(T ) aT , where a(T ) is the terminal value of some general
state variable with a nonnegative shadow price (T ); besides, aT is an arbitrary
real number. Continuing this line of thought, interpret (9.35) as an abstract cri-
terion function and c(t) as an abstract control variable with control region R and
with the property that a higher value of c(t) makes a(t)
_ smaller. Then “comple-
mentary slackness”is the principle that given the terminal constraint a(T ) aT ,
the terminal optimality condition must be (a(T ) aT ) (T ) = 0: The intuition
is that if the shadow price (T ) > 0 (a “slackness”), then optimality requires
a(T ) = aT : Indeed, in this case a(T ) > aT has an avoidable positive opportunity
cost. On the other hand, if a(T ) > aT is optimal (another “slackness”), then the
shadow price must be nil, i.e., (T ) = 0: There is “complementary slackness” in
the sense that at most one of the weak inequalities a(T ) aT and (T ) 0 can
be strict in optimum.
Anyway, returning to the household’s saving problem, the transversality con-
dition becomes more concrete if we insert (9.39). For the case T < 1; we then
have
a(T )u0 (c(T ))e T = 0: (9.44)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
370 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Since u0 (c(T ))e T is always positive, an optimal plan obviously must satisfy a(T )
= aT = 0. The reason is that, given the solvency requirement a(T ) 0, the only
alternative to a(T ) = 0 is a(T ) > 0: But this would imply that the level of the
consumption path could be raised, and U0 thereby be increased, by allowing a
decrease in a(T ) without violating the solvency requirement.
Now, write the solvency requirement as a(T )e T 0 and let T ! 1: Then
in the limit the solvency requirement takes the form of (NPG) above (replace T
by t), and (9.44) is replaced by
lim a(T )u0 (c(T ))e T
= 0: (9.45)
T !1

This says the same as (TVC) above. Intuitively, a plan that violates this condition
by having “>”instead “=”indicates scope for improvement and thus cannot be
optimal. There would be “purchasing power left for eternity”. This purchasing
power could be transferred to consumption on earth at an earlier date.
Generally, care must be taken when extending a necessary transversality con-
dition from a …nite to an in…nite horizon. But for the present problem, the
extension is valid. To see this, note that by Proposition 1, strict inequality in
the (NPG) condition is (by Proposition1) equivalent to strict inequality in the in-
tertemporal budget constraint (IBC). Such a path can always be improved upon
by raising c(t) a little in some time interval without decreasing c(t) in any other
time interval and without violating the (NPG) and (IBC). Hence, an optimal
plan must have strict equality in both NPG and IBC. This amounts to requiring
that none of these two conditions is “over-satis…ed”. And this requirement can
be shown to be equivalent to the condition (TVC) above. Indeed:
PROPOSITION 2 (the household’s necessary transversality condition with in-
…nite time horizon) Let T ! 1 in the criterion function (9.35) and assume
the human wealth integral (9.32) converges (and thereby remains bounded) for
T ! 1. Provided the adjoint variable, (t), satis…es the …rst-order conditions
(9.39) and (9.40), (TVC) holds if and only if (NPG) holds with strict equality.
Proof. See Appendix D.
In view of this proposition, we can write the transversality condition for T !
1 as the NPG condition with strict equality:
Rt
r( )d
lim a(t)e 0 = 0: (TVC’)
t!1

In view of the equivalence of the NPG condition with strict equality and the IBC
with strict equality, established in Proposition 1, the transversality condition for
T ! 1 can also be written
Z 1 Rt
c(t)e 0 r( )d dt = a0 + h0 : (IBC’)
0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 371

The current-value Hamiltonian versus the present-value Hamiltonian

The pre…x “current-value” is used to distinguish the current-value Hamiltonian
from what is known as the present-value Hamiltonian. The latter is de…ned as
H^ He t with e t substituted by ; which is the associated (discounted)
adjoint variable. The solution procedure is similar except that step 3 is replaced
^
by @ H=@a = _ and (t)e t in the transversality condition is replaced by (t):
The two methods are equivalent (and if the discount rate is nil, the formulas for
the optimality conditions coincide). But for many economic problems the current-
value Hamiltonian has the advantage that it makes both the calculations and the
interpretation slightly simpler. The adjoint variable, (t), which as mentioned
acts as a shadow price of the state variable, becomes a current price along with
the other prices in the problem, w(t) and r(t): This is in contrast to (t) which
is a discounted price.

9.4.3 The Keynes-Ramsey rule

The …rst-order conditions have interesting implications. Di¤erentiate both sides
of (9.39) w.r.t. t to get u00 (c)c_ = _ . This equation can be written u00 (c)c=u
_ 0 (c) =
_ = by drawing on (9.39) again. Applying (9.40) now gives

c(t)
_ 1
= (r(t) ); (9.46)
c(t) (c(t))
where (c) is the (absolute) elasticity of marginal utility w.r.t. consumption,
c 00
(c) u (c) > 0: (9.47)
u0 (c)
As in discrete time, (c) indicates the strength of the consumer’s preference for
consumption smoothing. The inverse of (c) measures the instantaneous in-
tertemporal elasticity of substitution in consumption, which in turn indicates the
willingness to accept variation in consumption over time when the interest rate
changes, see Appendix F.
The result (9.46) says that an optimal consumption plan is characterized in
the following way. The household will completely smooth i.e., even out
consumption over time if the rate of time preference equals the real interest rate.
The household will choose an upward-sloping time path for consumption if and
only if the rate of time preference is less than the real interest rate. In this case
the household will have to accept a relatively low level of current consumption
with the purpose of enjoying higher consumption in the future. The higher the
real interest rate relative to the rate of time preference, the more favorable is
it to defer consumption everything else equal. The proviso is important. In-
deed, in addition to the negative substitution e¤ect on current consumption of a

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
372 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Figure 9.2: Optimal consumption paths for a low and a high constant , given a constant
r> .

higher interest rate there is a positive income e¤ect due to the present value of
a given intertemporal consumption plan being reduced by a higher interest rate
(see (IBC)). On top of this comes a negative wealth e¤ect due to a higher interest
rate causing a lower present value of expected future labor earnings (again see
(IBC)). The special case of a CRRA utility function provides a convenient agenda
for sorting these details out, see Example 1 in Section 9.5.
By (9.46) we also see that the greater the elasticity of marginal utility (that
is, the greater the curvature of the utility function), the greater the incentive to
smooth consumption for a given value of r(t) . The reason for this is that a
strong curvature means that the marginal utility will drop sharply if consumption
increases, and will rise sharply if consumption decreases. Fig. 9.2 illustrates this
in the CRRA case where (c) = ; a positive constant. For a given constant
r > ; the consumption path chosen when is high has lower slope, but starts
from a higher level, than when is low.
The condition (9.46), which holds for all t within the time horizon whether this
is …nite or in…nite, is referred to as the Keynes-Ramsey rule. The name springs
from the English mathematician Frank Ramsey who derived the rule in 1928,
while his mentor, John Maynard Keynes, suggested a simple and intuitive way
of presenting it. The rule is the continuous-time counterpart to the consumption
Euler equation in discrete time.
The Keynes-Ramsey rule re‡ects the general microeconomic principle that
the consumer equates the marginal rate of substitution between any two goods to
the corresponding price ratio. In the present context the principle is applied to a
situation where the “two goods”refer to the same consumption good delivered at
two di¤erent dates. In Section 9.2 we used the principle to interpret the optimal
saving behavior in discrete time. How can the principle be translated into a
continuous time setting?

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.4. Maximizing discounted utility in continuous time 373

Local optimality in continuous time* Let (t, t+ t) and (t+ t, t+2 t) be

two short successive time intervals. The marginal rate of substitution, M RSt+ t;t ,
of consumption in the second time interval for consumption in the …rst, is19
dc(t + t) u0 (c(t))
M RSt+ t;t jU =U = t u0 (c(t +
; (9.48)
dc(t) e t))
approximately. On the other hand, by saving c(t) more per time unit (where
c(t) < 0) in the short time interval (t, t+ t), one can, via the market, transform
c(t) t units of consumption in this time interval into
R t+ t
r( )d
c(t + t) t c(t) t e t (9.49)
units of consumption in the time interval (t + t, t + 2 t). The marginal rate of
transformation is therefore
dc(t + t)
M RTt+ t;t jU =U
dc(t)
R t+ t
r( )d
= e t :
In the optimal plan we must have M RSt+ t;t = M RTt+ t;t which gives
u0 (c(t)) R t+ t
r( )d
t u0 (c(t +
=e t ; (9.50)
e t))
approximately. When t = 1 and and r(t) are small, this relation can be
approximated by (9.11) from discrete time (generally, by a …rst-order Taylor
approximation, we have ex 1 + x; when x is close to 0).
Taking logs on both sides of (9.50), dividing through by t; inserting (9.49),
and letting t ! 0; we get (see Appendix E)
u00 (c(t))
c(t)
_ = r(t): (9.51)
u0 (c(t))
With the de…nition of (c) in (9.47), this is exactly the same as the Keynes-
Ramsey rule (9.46) which, therefore, is merely an expression of the general op-
timality condition M RS = M RT: When c(t) _ > 0; the household is willing to
sacri…ce some consumption today for more consumption tomorrow only if it is
compensated by an interest rate su¢ ciently above : Naturally, the required com-
pensation is higher, the faster marginal utility declines with rising consumption,
i.e., the larger is ( u00 =u0 )c_ already. Indeed, a higher ct in the future than today
implies a lower marginal utility of consumption in the future than of consumption
today. Saving of the marginal unit of income today is thus only warranted if the
rate of return is su¢ ciently above ; and this is what (9.51) indicates.
19
The underlying analytical steps can be found in Appendix E.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
374 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

9.4.4 Mangasarian’s su¢ cient conditions

For dynamic optimization problems with one state variable, the Maximum Prin-
ciple delivers a set of …rst-order conditions and suggests a terminal optimality
condition, the transversality condition. The …rst-order conditions are necessary
conditions for an interior path to be optimal, while, with in…nite horizon, the
necessity of the suggested transversality condition in principle requires a veri…ca-
tion in each case; in the present case the veri…cation is implied by Proposition 2.
So, up to this point we have only shown that if the consumption/saving problem
has an interior solution, then this solution satis…es the Keynes-Ramsey rule and
a transversality condition, (TVC’).
But are these conditions also su¢ cient? The answer is yes in the present case.
This follows from Mangasarian’s su¢ ciency theorem (see Math tools) which, ap-
plied to the present problem, tells us that if the Hamiltonian is jointly concave
in (a; c) for every t within the time horizon, then the listed …rst-order conditions,
together with the transversality condition, are also su¢ cient. Because the in-
stantaneous utility function (the …rst term in the Hamiltonian) is here strictly
concave in c and the second term is linear in (a; c), the Hamiltonian is jointly
concave in (a; c):
To sum up: if we have found a path satisfying the Keynes-Ramsey rule and
(TVC’), we have a candidate solution. Applying the Mangasarian theorem, we
check whether our candidate is an optimal solution. In the present case it is. In
fact the strict concavity of the Hamiltonian with respect to the control variable
in this problem ensures that the optimal solution is unique (Exercise 9.?).

9.5 The consumption function

We have not yet fully solved the saving problem. The Keynes-Ramsey rule gives
only the optimal rate of change of consumption over time. It says nothing about
the level of consumption at any given time. In order to determine, for instance,
the level c(0), we implicate the solvency condition which limits the amount the
household can borrow in the long term. Among the in…nitely many consumption
paths satisfying the Keynes-Ramsey rule, the household will choose the “highest”
one that also ful…ls the solvency requirement (NPG). Thus, the household acts
so that strict equality in (NPG) obtains. As we saw in Proposition 2, this is
equivalent to the transversality condition being satis…ed.
To avoid misunderstanding: The examples below should not be interpreted
such that for any evolution of wages and interest rates there exists a solution to
the household’s maximization problem with in…nite horizon. There is generally
no guarantee that integrals converge and thus have an upper bound for T ! 1.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.5. The consumption function 375

The evolution of wages and interest rates which prevails in general equilibrium
is not arbitrary, however. It is determined by the requirement of equilibrium.
In turn, of course existence of an equilibrium imposes restrictions on the utility
discount rate relative to the potential growth in instantaneous utility. We shall
return to these aspects in the next chapter.
EXAMPLE 1 (constant elasticity of marginal utility; in…nite time horizon). In
the problem in Section 9.4.2 with T = 1; we consider the case where the elasticity
of marginal utility (c); as de…ned in (9.47), is a constant > 0. From Appendix
A of Chapter 3 we know that this requirement implies that up to a positive linear
transformation the utility function must be of the form:
c1
; when > 0; 6= 1;
u(c) = 1 (9.52)
ln c; when = 1:
This is our familiar CRRA utility function. In this case the Keynes-Ramsey rule
_ = 1 (r(t)
implies c(t) )c(t): Solving this linear di¤erential equation yields
1
Rt
c(t) = c(0)e 0 (r( ) )d
; (9.53)
cf. the general accumulation formula, (9.25).
We know from Proposition 2 that the transversality condition is equivalent
to the NPG condition being satis…ed with strict equality, and from Proposition 1
we know that this condition is equivalent to the intertemporal budget constraint
being satis…ed with strict equality, i.e.,
Z 1 Rt
c(t)e 0 r( )d dt = a0 + h0 ; (IBC’)
0

where h0 is the human wealth,

Z 1 Rt
r( )d
h0 = w(t)e 0 dt: (9.54)
0

This result can be used to determine c(0):20 Substituting (9.53) into (IBC’) gives
Z 1 R
t 1
c(0) e 0 [ (r( ) ) r( )]d dt = a0 + h0 :
0

The consumption function is thus

c(0) = 0 (a0 + h0 ); where
1 1
0 R1 Rt 1
= R1 1
Rt (9.55)
e 0[ (r( ) ) r( )]d
dt e 0 [(1 )r( ) ]d
dt
0 0
20
The method also applies if instead of T = 1; we have T < 1:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
376 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

is the marginal propensity to consume out of wealth. We have here assumed

that these improper integrals over an in…nite horizon are bounded from above for
all admissible paths. We see that consumption is proportional to total wealth.
The factor of proportionality, often called the marginal propensity to consume
out of wealth, depends on the expected future interest rates and on the preference
parameters and ; that is, the impatience rate and the strength of the preference
for consumption smoothing, respectively.
Generally, an increase in the interest rate level, for given total wealth, a0 + h0 ,
can e¤ect c(0) both positively and negatively.21 On the one hand, such an increase
makes future consumption cheaper in present value terms. This change in the
trade-o¤ between current and future consumption entails a negative substitution
e¤ect on c(0). On the other hand, the increase in the interest rates decreases the
present value of a given consumption plan, allowing for higher consumption both
today and in the future, for given total wealth, cf. (IBC’). This entails a positive
pure income e¤ect on consumption today as consumption is a normal good. If
< 1 (small curvature of the utility function), the substitution e¤ect will dominate
the pure income e¤ect, and if > 1 (large curvature), the reverse will hold. This
is because the larger is , the stronger is the propensity to smooth consumption
over time.
In the intermediate case = 1 (the logarithmic case) we get from (9.55) that
0 = , hence
c(0) = (a0 + h0 ): (9.56)

In this special case the marginal propensity to consume is time independent and
equal to the rate of time preference. For a given total wealth, a0 + h0 , current
consumption is thus independent of the expected path of the interest rate. That
is, in the logarithmic case the substitution and pure income e¤ects on current
consumption exactly o¤set each other. Yet, on top of this comes the negative
wealth e¤ect on current consumption of an increase in the interest rate level.
The present value of future wage incomes becomes lower (similarly with expected
future dividends on shares and future rents in the housing market in a more
general setup). Because of this, h0 (and so a0 + h0 ) becomes lower, which adds
to the negative substitution e¤ect. Thus, even in the logarithmic case, and a
fortiori when < 1; the total e¤ect of an increase in the interest rate level is
unambiguously negative on c(0):

21
By an increase in the interest rate level we mean an upward shift in the time-pro…le of the
interest rate. That is, there is at least one time interval within [0; 1) where the interest rate is
higher than in the original situation and no time interval within [0; 1) where the interest rate
is lower.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.5. The consumption function 377

If, for example, r(t) = r and w(t) = w (positive constants), we get

0 = [( 1)r + ]= ;
a0 + h0 = a0 + w=r:

When = 1; the negative e¤ect of a higher r on h0 is decisive. When < 1;

a higher r reduces both 0 and h0 ; hence the total e¤ect on c(0) is even “more
negative”: When > 1, a higher r implies a higher 0 which more or less o¤sets
the lower h0 , so that the total e¤ect on c(0) becomes ambiguous. As referred to
in Chapter 3, available empirical studies generally suggest a value of somewhat
above 1.
A remark on …xed-rate loans and positive net debt is appropriate here. Sup-
pose a0 < 0 and assume that this net debt is not in the form of a variable-rate
loan (as hitherto assumed), but for instance a …xed-rate mortgage loan. Then
a rise in the interest rate level implies a lowering of the present value of the
debt and thereby raises …nancial wealth and possibly total wealth. If so, the rise
in the interest rate level implies a positive wealth e¤ect on current consumption,
thereby “joining”the positive pure income e¤ect in counterbalancing the negative
substitution e¤ect.
EXAMPLE 2 (constant absolute semi-elasticity of marginal utility; in…nite time
horizon). In the problem in Section 9.4.2 with T = 1; we consider the case
where the sensitivity of marginal utility, measured by the absolute value of the
semi-elasticity of marginal utility, u00 (c)=u0 (c) ( u0 =u0 )= c, is a positive
constant, : The utility function must then, up to a positive linear transformation,
be of the form,
1
u(c) = e c ; > 0: (9.57)
This is known as the CARA utility function (where the name CARA comes from
“Constant Absolute Risk Aversion”). The Keynes-Ramsey rule now becomes
_ = 1 (r(t)
c(t) ): When the interest rate is a constant r > 0, we …nd, through
(IBC’) and partial integration, c(0) = r(a0 + h0 ) (r )=( r), presupposing
r and a0 + h0 > (r )=(ar2 ):
This hypothesis of a “constant absolute variability aversion”implies that the
degree of relative variability aversion is (c) = c and thus greater, the larger is
c. The CARA function is popular in the theory of behavior under uncertainty.
One of the theorems of expected utility theory is that the degree of absolute risk
aversion, u00 (c)=u0 (c), is proportional to the risk premium which the economic
agent will require to be willing to exchange a speci…ed amount of consumption
received with certainty for an uncertain amount having the same mean value.
Empirically this risk premium seems to be a decreasing function of the level of

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
378 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

consumption. Therefore the CARA function is generally considered less realistic

than the CRRA function of the previous example.
EXAMPLE 3 (logarithmic utility; …nite time horizon; retirement). We consider
a life-cycle saving problem. A worker enters the labor market at time 0 with
a …nancial wealth of 0, has …nite lifetime T (assumed known), retires at time
t1 2 (0; T ] ; and does not wish to pass on bequests. For simplicity we assume that
rt = r > 0 for all t 2 [0; T ] and labor income is w(t) = w > 0 for t 2 [0; t1 ], while
w(t) = 0 for t > t1 . The decision problem is
Z T
max U0 = (ln c(t))e t dt s.t.
(c(t))T
t=0 0
c(t) 0;
a(t)
_ = ra(t) + w(t) c(t); a(0) = 0;
a(T ) 0:

The Keynes-Ramsey rule becomes c_t =ct = r . A solution to the problem

will thus ful…l
c(t) = c(0)e(r )t : (9.58)
Inserting this into the di¤erential equation for a; we get a …rst-order linear dif-
ferential equation the solution of which (for a(0) = 0) can be reduced to

w c0
a(t) = ert (1 e rz
) (1 e t
) ; (9.59)
r

where z = t if t t1 , and z = t1 if t > t1 . We need to determine c(0). The

transversality condition implies a(T ) = 0. Having t = T , z = t1 and aT = 0 in
(9.59), we get
c(0) = ( w=r)(1 e rt1 )=(1 e T ): (9.60)
Substituting this into (9.58) gives the optimal consumption plan.22
If r = , consumption is constant over time at the level given by (9.60). If, in
addition, t1 < T , this consumption level is less than the wage income per year up
to t1 (in order to save for retirement); in the last years the level of consumption
is maintained although there is no wage income; the retired person uses up both
the return on …nancial wealth and this wealth itself.
The examples illustrate the importance of forward-looking expectations, here
expectations about future wage income and interest rates. The expectations
a¤ect c(0) both through their impact on the marginal propensity to consume
22
For t1 = T and T ! 1 we get in the limit c(0) = w=r h0 ; which is also what (9.55)
gives when a(0) = 0 and = 1:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.6. Concluding remarks 379

(cf. 0 in Example 1) and through their impact on the present value of expected
future labor income (or of expected future dividends on shares or imputed rental
income on owner-occupied houses in a more general setup).23

9.6 Concluding remarks

(incomplete)
...
The examples above and the consumption theory in this chapter in gen-
eral should only be seen as a …rst, crude approximation to actual consump-
tion/saving behavior. Real world factors such as uncertainty and narrow credit
constraints (absence of perfect loan and insurance markets) also a¤ect the behav-
ior. When these factors are included, current income and expected income in the
near future tend to become important co-determinants of current consumption,
at least for a large fraction of the population with little …nancial wealth. We
return to this in connection with short- and medium-run macro models later in
this book.

9.7 Literature notes

(incomplete)
In Chapter 6, where the borrower was a “large” agent with …scal and mon-
etary policy mandates, namely the public sector, satisfying the intertemporal
budget constraint was a necessary condition for solvency (when the interest rate
exceeds the growth rate of income), but not a su¢ cient condition. When the
modelled borrowers are “small” private agents as in this chapter, the situation
is di¤erent. Neoclassical models with perfect markets then usually contain equi-
librium mechanisms such that the agents’ compliance with their intertemporal
budget constraint is su¢ cient for lenders’ willingness and ability to supply the
demanded …nance. See ...
Present-bias and time-inconsistency. Strots (1956). Laibson, QJE 1997: 1,
; 2 ; :::
Loewenstein and Thaler (1989) survey the evidence suggesting that the utility
discount rate is generally not constant, but declining with the time distance from
the current period to the future periods within the horizon. This is known as
hyperbolic discounting.
23
There exist cases where, due to new information, a shift in expectations occurs so that
a discontinuity in a responding endogenous variable results. How to deal with such cases is
treated in Chapter 11.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
380 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

The assumptions regarding the underlying intertemporal preferences which

allow them to be represented by the present value of period utilities discounted
at a constant rate are dealt with by Koopmans (1960), Fishburn and Rubinstein
(1982), and in summary form by Heal (1998).
Borovika, WP 2013, Recursive preferences, separation of risk aversion and
IES.
Deaton, A., Understanding Consumption, OUP 1992.
On continuous-time …nance, see for instance Merton (1990).
Goldberg (1958).
Allen (1967).
To Math Tools: Rigorous and more general presentations of the Maximum
Principle in continuous time applied in economic analysis are available in, e.g.,
Seierstad and Sydsæter (1987), Sydsæter et al. (2008) and Seierstad and Sydsæter
(Optimization Letters, 2009, 3, 507-12).

9.8 Appendix
A. Growth arithmetic in continuous time
Let the variables z; x; and y be di¤erentiable functions of time t: Suppose z(t);
x(t); and y(t) are positive for all t: Then:
PRODUCT RULE z(t) = x(t)y(t) ) z(t)=z(t)
_ = x(t)=x(t)
_ + y(t)=y(t):
_
Proof. Taking logs on both sides of the equation z(t) = x(t)y(t) gives ln z(t) =
ln x(t)+ln y(t). Di¤erentiation w.r.t. t, using the chain rule, gives the conclusion.

The procedure applied in this proof is called logarithmic di¤erentiation w.r.t.

t:
FRACTION RULE z(t) = x(t)=y(t) ) z(t)=z(t)
_ = x(t)=x(t)
_ y(t)=y(t):
_
The proof is similar.
POWER FUNCTION RULE z(t) = x(t) ) z(t)=z(t)
_ = x(t)=x(t):
_
The proof is similar.
In continuous time these simple formulas are exactly true. In discrete time
the analogue formulas are only approximately true and the approximation can
be quite bad unless the growth rates of x and y are small, cf. Appendix A to
Chapter 4.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.8. Appendix 381

B. Average growth and interest rates

Sometimes in the literature the accumulation formula in continuous time,
Rt
g( )d
a(t) = a(0)e 0 ;

is expressed in terms of the arithmetic average, also called the arithmetic mean,
of the growth rates in the time interval [0; t]. This average is de…ned as g0;t
Rt
= (1=t) 0 g( )d : So we can write

a(t) = a(0)eg0;t t ; (9.61)

which has form similar to (9.24). Similarly, let r0;t denote the arithmetic average
Rt
of the (short-term) interest rates from time 0 to time t; i.e., r0;t = (1=t) 0 r( )d :
Then we can write the present value of the consumption stream (c(t))Tt=0 as P V
RT
= 0 c(t)e r0;t t dt:
The arithmetic average growth rate, g0;t ; coincides with the average compound
growth rate from time 0 to time t; that is, the number g satisfying

a(t) = a(0)egt ; (9.62)

for the same a(0) and a(t) as in (9.61).

There is no similar concordance within discrete time modeling. To see this,
suppose that the period-by-period observations, a0 ; a1 : : : ; an ; are available. Let
g^0;n be the average compound growth rate from period 0 to period n; that is,
the number x satisfying an = a0 (1 + x)n : We …nd 1 + g^0;n = 1 + x = (an =a0 )1=n :
This compound growth factor is the geometric mean, mg ; of the period-by-period
growth factors since
1=n
a1 a2 an an 1=n
mg ::: =( ) :
a0 a1 an 1 a0

The arithmetic mean, Ma , of the period-by-period growth factors is

1 a1 a2 an
ma + + + mg ; (9.63)
n a0 a1 an 1

where strict inequality holds unless all the n growth factors are identical. Indeed,
when the growth factors are not identical, we have, by Jensen’s inequality,

Xn X
n
'( w i xi ) > wi '(xi );
i=1 i=1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
382 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Pn
when ' is strictly concave and i=1 wi = 1; wi 0; i = 1; 2; : : : ; n: So, by (9.63),

Xn
1 ai Xn
1 ai 1X
n
ai
ln ma = ln > ln = ln = ln mg ;
i=1
n ai 1 i=1
n ai 1 n i=1 ai 1

since ln is a strictly concave function. This inequality implies ma > mg since ln

is also an increasing function. Consequently, unless the period-by-period growth

rate is a constant, multiplying the initial value a0 with the arithmetic mean of
the growth factors results in a number larger than an :

Discrete versus continuous compounding Suppose the period length is

one year so that the given observations, a0 ; a1 : : : ; an ; are annual data. There
are two alternative ways of calculating an average compound growth rate (often
just called the “average growth rate”) for the data. We may apply the geometric
growth formula,

an = a0 (1 + G)n ; (9.64)
which is natural if the compounding behind the data is discrete and occurs annu-
ally. If the compounding is much more frequent, it is in principle better to apply
the exponential growth formula,

an = a0 egn ; (9.65)

corresponding to continuous compounding. Unless an = a0 ; the resulting g will

be smaller than the average compound growth rate G calculated from a geometric
growth formula (discrete time) for the same data. Indeed,

ln aan0
g= = ln(1 + G) / G
n
for G “small”, where “/”means “close to”(by a …rst-order Taylor approximation
about G = 0) but “less than”except if G = 0. The intuitive reason for “less than”
is that a given growth force is more powerful when compounding is continuous.
To put it di¤erently: rewriting (1 + G)n into exponential form gives (1 + G)n
= (eln(1+G) )n = egn < eGn ; as ln(1 + G) < G for all G 6= 0.
Anyway, the di¤erence between G and g is usually unimportant. If for example
G refers to the annual GDP growth rate, it will be a small number, and the
di¤erence between G and g immaterial. For example, to G = 0:040 corresponds g
0:039: Even if G = 0:10, the corresponding g is 0:0953. But if G stands for the
in‡ation rate and there is high in‡ation, the di¤erence between G and g will be

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.8. Appendix 383

substantial. During hyperin‡ation the monthly in‡ation rate may be, say, G =
100%, but the corresponding g will be only 69%.24

C. Proof of Proposition 1 (about equivalence between the No-Ponzi-

Game condition and the intertemporal budget constraint)
We consider the book-keeping relation

a(t)
_ = r(t)a(t) + w(t) c(t); (9.66)

where a(0) = a0 (given), and the solvency requirement

Rt
r( )d
lim a(t)e 0 0: (NPG)
t!1

Technical remark. The expression

Rt
in (NPG) should be understood to include
r( )d
the possibility that a(t)e 0 ! 1 for t ! 1: Moreover, if full generality
were aimed at, we should allow for in…nitely ‡uctuating paths in both the (NPG)
and (TVC) and therefore replace “limt!1 ” by “lim inf t!1 ”, i.e., the limit in-
ferior. The limit inferior for t ! 1 of a function f (t) on [0; 1) is de…ned as
limt!1 inf ff (s)j s tg.25 As noted in Appendix E of the previous chapter, how-
ever, undamped in…nitely ‡uctuating paths never turn up in “normal”economic
optimization problems, whether in discrete or continuous time. Hence, we apply
the simpler concept “lim”rather than “lim inf”.
On the background of (9.66), Proposition 1 in the text claimed that (NPG)
is equivalent to the intertemporal budget constraint,
Z 1 Rt
c(t)e 0 r( )d dt h0 + a0 ; (IBC)
0

being satis…ed, where h0 is de…ned as in (9.54) and is assumed to be a …nite

number. In addition, Proposition 1 in Section 9.4 claimed that there is strict
equality in (IBC) if and only there is strict equality in (NPG). A plain proof goes
as follows.
Rt
Proof. Isolate c(t) in (9.66) and multiply through by e 0 r( )d to obtain
Rt Rt Rt
r( )d r( )d
c(t)e 0 = w(t)e 0 (a(t)
_ r(t)a(t))e 0 r( )d :
24
Apart from the discrete compounding instead of continuous compounding, a geometric
growth factor is equivalent to a “corresponding” exponential growth factor. Indeed, we can
rewrite the growth factor (1+g)t ; t = 0; 1; 2; : : : , into exponential form since (1+g)t = (eln(1+g) )t
= e[ln(1+g)]t : Moreover, if g is “small”, we have e[ln(1+g)]t egt .
25
By “inf” is meant in…mum of the set, that is, the largest number less than or equal to all
numbers in the set.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
384 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

RT Rt
r( )d
Integrate from 0 to T > 0 to get 0
c(t)e 0 dt
Z T Rt Z T Rt Z T Rt
r( )d r( )d r( )d
= w(t)e 0 dt a(t)e
_ 0 dt + r(t)a(t)e 0 dt
0 0 0
Z Rt Rt Z T Rt !
T T
r( )d r( )d r( )d
= w(t)e 0 dt a(t)e 0 a(t)e 0 ( r(t))dt
0 0 0
Z T Rt
r( )d
+ r(t)a(t)e 0 dt
0
Z T Rt RT
r( )d r( )d
= w(t)e 0 dt (a(T )e 0 a(0));
0

where the second equality follows from integration by parts. If we let T ! 1 and
use the de…nition of h0 and the initial condition a(0) = a0 , we get (IBC) if and
only if (NPG) holds. It follows that when (NPG) is satis…ed with strict equality,
so is (IBC), and vice versa.
An alternative proof is obtained by using the general solution to a linear
inhomogeneous …rst-order di¤erential equation and then let T ! 1. Since this
is a more generally applicable approach, we will show how it works and use it
for Claim 1 below (an extended version of Proposition 1) and for the proof of
Proposition 2 in the text. Claim 1 will for example prove useful in Exercise 9.1
and in the next chapter.
CLAIM 1 Let f (t) and g(t) be given continuous functions of time, t: Consider
the di¤erential equation
x(t)
_ = g(t)x(t) + f (t); (9.67)
with x(t0 ) = xt0 ; a given initial value: Then the inequality
Rt
g(s)ds
lim x(t)e t0
0 (9.68)
t!1

is equivalent to Z 1 R
g(s)ds
f ( )e t0
d xt0 : (9.69)
t0

Moreover, if and only if (9.68) is satis…ed with strict equality, then (9.69) is
satis…ed with strict equality.
Proof. The linear di¤erential equation, (9.67), has the solution
Rt Z t Rt
g(s)ds g(s)ds
x(t) = x(t0 )e t0
+ f ( )e d . (9.70)
t0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.8. Appendix 385

Rt
g(s)ds
Multiplying through by e t0
yields
Rt Z t R
g(s)ds g(s)ds
x(t)e t0
= x(t0 ) + f ( )e t0
d :
t0

By letting t ! 1; it can be seen that if and only if (9.68) is true, we have

Z 1 R
g(s)ds
x(t0 ) + f ( )e t0 d 0:
t0

Since x(t0 ) = xt0 ; this is the same as (9.69). We also see that if and only if (9.68)
holds with strict equality, then (9.69) also holds with strict equality.
COROLLARY Let n be a given constant and let
Z 1 R
(r(s) n)ds
ht0 w( )e t0 d ; (9.71)
t0

which we assume is a …nite number. Then, given

a(t)
_ = (r(t) n)a(t) + w(t) c(t); where a(t0 ) = at0 , (9.72)

it holds that
Rt Z 1 R
t0 (r(s) n)ds t0 (r(s) n)ds
lim a(t)e 0, c( )e d at0 + ht0 ; (9.73)
t!1 t0

where a strict equality on the left-hand side of “,” implies a strict equality on
the right-hand side, and vice versa.
Proof. In (9.67), (9.68) and (9.69), let x(t) = a(t); g(t) = r(t) n and f (t) =
w(t) c(t). Then the conclusion follows from Claim 1.

By setting t0 = 0 in the corollary and replacing by t and n by 0, we have

hereby provided an alternative proof of Proposition 1.

D. Proof of Proposition 2 (about the transversality condition with an

in…nite time horizon)
In the di¤erential equation (9.67) we let x(t) = (t); g(t) = (r(t) ); and
f (t) = 0: This gives the linear di¤erential equation _ (t) = ( r(t)) (t); which
is identical to the …rst-order condition (9.40) in Section 9.4. The solution is
Rt
t0 (r(s) )ds
(t) = (t0 )e :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
386 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Substituting this into (TVC) in Section 9.4 yields

Rt
t0 (r(s) n)ds
(t0 ) lim a(t)e = 0: (9.74)
t!1

From the …rst-order condition (9.39) in Section 9.4 we have (t0 ) = u0 (c(t0 )) > 0
so that (t0 ) in (9.74) can be ignored. Thus, (TVC) in Section 9.4 is equivalent
to the condition that (NPG) in that section is satis…ed with strict equality (let
t0 = 0 = n). This proves Proposition 2 in the text.

E. Intertemporal consumption smoothing

We claimed in Section 9.4 that equation (9.48) gives approximately the marginal
rate of substitution of consumption in the time interval (t + t, t + 2 t) for
consumption in (t, t+ t): This can be seen in the following way. To save notation
we shall write our time-dependent variables as ct ; rt , etc., even though they are
continuous functions of time. The contribution from the two time intervals to the
criterion function is
Z t+2 t Z t+ t Z t+2 t
t ( t)
u(c )e d e u(ct )e d + u(ct+ t )e ( t) d
t t t+ t
!
( t) t+ t ( t) t+2 t
e e
= e t u(ct ) + u(ct+ t )
t t+ t
t t
e (1 e ) t
= u(ct ) + u(ct+ t )e :

Requiring unchanged utility integral U0 = U0 is thus approximately the same as

t
requiring [u(ct ) + u(ct+ t )e ] = 0; which by carrying through the di¤erenti-
ation and rearranging gives (9.48).
The instantaneous local optimality condition, equation (9.51), can be inter-
preted on the basis of (9.50). Take logs on both sides of (9.50) to get
Z t+ t
0 0
ln u (ct ) + t ln u (ct+ t ) = r d :
t

Dividing by t, substituting (9.49), and letting t ! 0 we get

ln u0 (ct+ t ) ln u0 (ct ) Rt+ t Rt
lim = lim ; (9.75)
t!0 t t!0 t
where Rt is the antiderivative of rt . By the de…nition of a time derivative, (9.75)
can be written
d ln u0 (ct ) dRt
= :
dt dt
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
9.8. Appendix 387

Carrying out the di¤erentiation, we get

1
u00 (ct )c_t = rt ;
u0 (c t)

which was to be shown.

F. Elasticity of intertemporal substitution in continuous time

The relationship between the elasticity of marginal utility and the concept of
instantaneous elasticity of intertemporal substitution in consumption can be ex-
posed in the following way: consider an indi¤erence curve for consumption in the
non-overlapping time intervals (t, t+ t) and (s, s+ t). The indi¤erence curve is
depicted in Fig. 9.3. The consumption path outside the two time intervals is kept
unchanged. At a given point (ct t; cs t) on the indi¤erence curve, the marginal
rate of substitution of s-consumption for t-consumption, M RSst , is given by the
absolute slope of the tangent to the indi¤erence curve at that point. In view of
u00 (c) < 0, M RSst is rising along the curve when ct decreases (and thereby cs
increases).
Conversely, we can consider the ratio cs =ct as a function of M RSst along the
given indi¤erence curve. The elasticity of this consumption ratio w.r.t. M RSst
as we move along the given indi¤erence curve then indicates the elasticity of
substitution between consumption in the time interval (t, t+ t) and consumption
in the time interval (s, s + t): Denoting this elasticity by (ct ; cs ); we thus have:
(cs =ct )
M RSst d(cs =ct ) cs =ct
(ct ; cs ) = M RSst
:
cs =ct dM RSst M RSst

At an optimum point, M RSst equals the ratio of the discounted prices of

good t and good s: Thus, the elasticity of substitution can be interpreted as
approximately equal to the percentage increase in the ratio of the chosen goods,
cs =ct , generated by a one percentage increase in the inverse price ratio, holding
the utility level and the amount of other goods unchanged. If s = t + t and the
interest rate from date t to date s is r; then (with continuous compounding) this
price ratio is er t ; cf. (9.50). Inserting M RSst from (9.48) with t + t replaced
by s, we get

u0 (ct )=[e (s t) u0 (cs )] d(cs =ct )

(ct ; cs ) =
cs =ct dfu (ct )=[e (s t) u0 (cs )]g
0

u0 (ct )=u0 (cs ) d(cs =ct )

= ; (9.76)
cs =ct d(u (ct )=u0 (cs ))
0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
388 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

Figure 9.3: Substitution of s-consumption for t-consumption as M RSst increases to

0
M RSst .

since the factor e (t s) cancels out.

We now interpret the d’s in (9.76) as di¤erentials (recall, the di¤erential of a
di¤erentiable function y = f (x) is denoted dy and de…ned as dy = f 0 (x)dx where
dx is some arbitrary real number). Calculating the di¤erentials we get

u0 (ct )=u0 (cs ) (ct dcs cs dct )=c2t

(ct ; cs ) :
cs =ct [u0 (cs )u00 (ct )dct u0 (ct )u00 (cs )dcs ]=u0 (cs )2

Hence, for s ! t we get cs ! ct and

ct (dcs dct )=c2t u0 (ct )

(ct ; cs ) ! 0 = ~ (ct ):
u (ct )u00 (ct )(dct dcs )=u0 (ct )2 ct u00 (ct )

This limiting value is known as the instantaneous elasticity of intertemporal sub-

stitution of consumption. It re‡ects the opposite of the preference for consump-
tion smoothing. Indeed, we see that ~ (ct ) = 1= (ct ), where (ct ) is the elasticity
of marginal utility at the consumption level c(t).

9.9 Exercises
9.1 We look at a household (or dynasty) with in…nite time horizon. The house-
hold’s labor supply is inelastic and grows at the constant rate n > 0: The house-
hold has a constant rate of time preference > n and the individual instantaneous
utility function is u(c) = c1 =(1 ); where is a positive constant. There is
no uncertainty. The household maximizes the integral of per capita utility dis-
counted at the rate n. Set up the household’s optimization problem. Show

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

9.9. Exercises 389

that the optimal consumption plan satis…es

c(0) = 0 (a0 + h0 ); where

1
0 = R1 Rt (1 )r( )
; and
( +n)d
e 0 dt
Z0 1 Rt
h0 = w(t)e 0 (r( ) n)d
dt;
0

where w(t) is the real wage per unit of labor and otherwise the same notation as
in this chapter is used. Hint: apply the corollary to Claim 1 in Appendix C and
the method of Example 1 in Section 9.5. As to h0 ; start by considering
Z 1 Rt
H0 h0 L0 = w(t)Lt e 0 (r( ) n)d dt
0

and apply that L(t) = L0 ent :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 9. THE INTERTEMPORAL CONSUMPTION-
390 SAVING PROBLEM IN DISCRETE AND CONTINUOUS TIME

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 10

The basic representative agent

model: Ramsey

As early as 1928 a sophisticated model of a society’s optimal saving was pub-

lished by the British mathematician and economist Frank Ramsey (1903-1930).
Ramsey’s contribution was mathematically demanding and did not experience
much response at the time. Three decades had to pass until his contribution was
taken up seriously (Samuelson and Solow, 1956). His model was merged with the
growth model by Solow (1956) and became a cornerstone in neoclassical growth
theory from the mid 1960s. The version of the model which we present below was
completed by the work of Cass (1965) and Koopmans (1965). Hence the model
is also known as the Ramsey-Cass-Koopmans model.
The model is one of the basic workhorse models in macroeconomics. As
we conclude in Section 10.6, the model can be seen as placed at one end of a
line segment. At the other end appears another basic workhorse model, namely
Diamond’s overlapping generations model considered in chapters 3 and 4. While
in the Diamond model there is an unbounded number of agents (since in every new
period a new generation enters the economy) and these have a …nite time horizon,
in the Ramsey model there is a …nite number of agents with an unbounded time
horizon. The agents in the Ramsey model are completely alike. The model is thus
an example of a representative agent model. In contrast, the Diamond model has
heterogeneous agents, young versus old, interacting in every period. There are
important economic questions where these di¤erences in the setup lead to salient
di¤erences in the answers.
The purpose of this chapter is to describe and analyze the continuous-time
version of the Ramsey framework. In the main sections we consider the case of a
perfectly competitive market economy. In this context we shall see, for example,
that the Solow growth model can be interpreted as a special case of the Ramsey

391
 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
392 MODEL: RAMSEY

model. Towards the end of the chapter we consider the Ramsey framework in a
setting with an “all-knowing and all-powerful”social planner.

10.1 Preliminaries
We consider a closed economy. Time is continuous. We assume households own
the capital goods and hire them out to …rms at a market rental rate, r^. This
is just to have something concrete in mind. If instead the capital goods were
owned by the …rms using them in production and the capital investment by these
…rms were …nanced by issuing shares and bonds, the conclusions would remain
the same as long as we ignore uncertainty.
The variables in the model are considered as (piecewise) continuous and dif-
ferentiable functions of time, t: Yet, to save notation, we shall write them as wt ;
r^t ; etc. instead of w(t); r^(t), etc. In every short time interval (t, t + t); the in-
dividual …rm employs labor at the market wage wt and rents capital goods at the
rental rate r^t . The combination of labor and capital produces the homogeneous
output good. This good can be used for consumption as well as investment. So
in every short time interval there are at least three active markets, one for the
“all-purpose”output good, one for labor, and one for capital services (the rental
market for capital goods). It may be convenient to imagine that there is also a
market for loans. As all households are alike, however, the loan market will not
be active in general equilibrium.
There is perfect competition in all markets, that is, households and …rms are
price takers. Any need for means of payment money is abstracted away.
Prices are measured in units of the output good.
There are no stochastic elements in the model. We assume households under-
stand exactly how the economy works and can predict the future path of wages
and interest rates. In other words, we assume “rational expectations”. In our
non-stochastic setting this amounts to perfect foresight. The results that emerge
from the model are thereby the outcome of economic mechanisms in isolation
from expectational errors.
Uncertainty being absent, rates of return on alternative assets are in equilib-
rium the same. In spite of the not active loan market, it is usual to speak of this
common rate of return as the real interest rate of the economy. Denoting this
rate rt ; for a given rental rate, r^t ; we have
r^t Kt Kt
rt = = r^t ; (10.1)
Kt
where the right-hand side is the rate of return on holding Kt capital goods,
( 0) being a constant rate of capital depreciation. This relationship may be

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.2. The agents 393

considered a no-arbitrage condition between investing in the loan market and in

capital goods.
We describe, …rst, the households’ behavior and next the …rms’ behavior.
Thereafter the interaction between households and …rms in general equilibrium
and the resulting dynamics will be analyzed.

10.2 The agents

10.2.1 Households
There is a …xed number, N; of identical households with an in…nite time horizon.
This feature makes aggregation very simple: we just have to multiply the behavior
of a single household with the number of households (we later normalize N to
equal 1). Every household has Lt (adult) members and Lt changes over time at
a constant rate, n :
Lt = L0 ent ; L0 > 0: (10.2)
Indivisibility is ignored.
Each household member supplies inelastically one unit of labor per time unit.
Equation (10.2) therefore describes the growth of the population as well as the
labor force. Since there is only one consumption good, the only decision problem
is how to distribute current income between consumption and saving.

Intertemporal utility function

The household’s preferences can be represented by an additive intertemporal util-
ity function with a constant rate of time preference, . Seen from time 0; the
intertemporal utility function is
Z 1
U0 = u(ct )Lt e t dt;
0

where ct Ct =Lt is consumption per family member. The instantaneous utility

function, u(c); has u0 (c) > 0 and u00 (c) < 0; i.e., positive but diminishing marginal
utility of consumption. The utility contribution from consumption per family
member is weighted by the number of family members, Lt : So it is the sum of the
family members’utility that counts. Such a utility function is called a utilitarian
utility function (with discounting).
The household is seen as an in…nitely-lived family, a family dynasty. The
current members of the dynasty act in unity and are concerned about the utility
from own consumption as well as the utility of the future generations within the

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
394 MODEL: RAMSEY

dynasty.1 Births (into adult life) do not amount to emergence of new economic
agents with independent interests. Births and population growth are seen as
just an expansion of the size of the already existing families. In contrast, in the
Diamond OLG model births imply entrance of new economic decision makers
whose preferences no-one cared about in advance.
In view of (10.2), U0 can be written as
Z 1
U0 = u(ct )e ( n)t dt; (10.3)
0

where the inconsequential positive factor L0 has been eliminated. Here n is

known as the e¤ective rate of time preference while is the pure rate of time
preference. We later introduce a restriction on n to ensure boundedness from
above of the utility integral in general equilibrium.
The household chooses a consumption-saving plan which maximizes U0 subject
to its budget constraint. Let At at Lt be the household’s (net) …nancial wealth
in real terms at time t: We have
dAt
A_ t = rt At + wt Lt ct L t ; A0 given. (10.4)
dt
This equation is a book-keeping relation telling how …nancial wealth or debt ( A)
changes over time depending on how consumption relates to current income. The
equation merely says that the increase in …nancial wealth per time unit equals
saving which equals income minus consumption. Income is the sum of the net
return on …nancial wealth, rt At ; and labor income, wt Lt ; where wt is the real
wage.2 Saving can be negative. In that case the household “dissaves” and does
so simply by selling a part of its stock of capital goods or by taking loans in the
loan market. The market prices, wt and rt ; faced by the household are assumed
to be piecewise continuous functions of time.
When the dynamic budget identity (10.4) is combined with a requirement of
solvency, we have a budget constraint. The relevant solvency requirement is the
No-Ponzi-Game condition (NPG for short):
Rt
rs ds
lim At e 0 0: (10.5)
t!1

This condition says that …nancial wealth far out in the future cannot have a
negative present value. That is, in the long run, debt is at most allowed to rise
1
The descrete-time Barro model of Chapter 7 articulated such an altruistic bequest motive.
In that chapter we also discussed some of the conceptual di¢ culties of the dynasty setup.
2
Since the technology exhibits constant returns to scale, in competitive equilibrium the …rms
make no (pure) pro…t to pay out to their owners.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.2. The agents 395

at a rate less than the real interest rate r: The NPG condition thus precludes
permanent …nancing of the interest payments by new loans.3
The decision problem is: choose a plan (ct )1
t=0 so as to achieve a maximum
of U0 subject to non-negativity of the control variable, c; and the constraints
(10.4) and (10.5). The problem is a slight generalization of the problem studied
in Section 9.4 of the previous chapter.
To solve the problem we shall apply the Maximum Principle. This method can
be applied directly to the problem as stated above or to an equivalent problem
with constraints expressed in per capita terms. Let us follow the latter approach.
From the de…nition at At =Lt we get by di¤erentiation w.r.t. t

Lt A_ t At L_ t A_ t
a_ t = = at n:
L2t Lt
Substitution of (10.4) gives the dynamic budget identity in per capita terms:

a_ t = (rt n)at + wt ct ; a0 given. (10.6)

By inserting At at Lt = at L0 ent ; the NPG condition (10.5) can be rewritten

Rt
lim at e 0 (rs n)ds
0; (10.7)
t!1

where the unimportant factor L0 has been eliminated.

We see that in both (10.6) and (10.7) a kind of corrected interest rate appears,
namely the interest rate, r; minus the family size growth rate, n. Although
deferring consumption gives a real interest rate of r, this return is diluted on
a per capita basis because it will have to be shared with more members of the
family when n > 0. In the form (10.7) the NPG condition requires that per capita
debt, if any, in the long run at most grows at a rate less than r n.

Solving the consumption/saving problem

The decision problem is now: choose (ct )0t=1 so as to a maximize U0 subject to
the constraints: ct 0; (10.6), and (10.7). To solve the problem we apply the
Maximum Principle. So we follow the same solution procedure as in the alike
problem (apart from n = 0) of Section 9.4 of the previous chapter:
3
In the previous chapter we saw that the NPG condition, in combination with (10.4), is
equivalent to an ordinary intertemporal budget constraint which says that the present value of
the planned consumption path cannot exceed initial total wealth, i.e., the sum of the initial
…nancial wealth and the present value of expected future labor income.
Violating the NPG condition means running a “Ponzi game”, that is, trying to make a fortune
through the chain-letter principle where old investors are payed o¤ with money from the new
investors.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
396 MODEL: RAMSEY

1) Set up the current-value Hamiltonian

H(a; c; ; t) = u(c) + [(r n) a + w c] ;

where is the adjoint variable associated with the di¤erential equation (10.6).
2) Di¤erentiate H partially w.r.t. the control variable, c; and put the result
equal to zero:
@H
= u0 (c) = 0: (10.8)
@c
3) Di¤erentiate H partially w.r.t. the state variable, a, and put the result
equal to minus the time derivative of plus the e¤ective discount rate (appearing
in the integrand of the criterion function) multiplied by :

@H _ +(
= (r n) = n) : (10.9)
@a
4) Apply the Maximum Principle: an interior optimal path (at ; ct )1t=0 will
satisfy that there exists a continuous function = t such that for all t 0;
(10.8) and (10.9) hold along the path and the transversality condition,
( n)t
lim at t e = 0; (10.10)
t!1

is satis…ed.
The interpretation of these optimality conditions is as follows. The condition
(10.8) can be considered a M C = M B condition (in utility terms). It illustrates
together with (10.9) that the adjoint variable, ; constitutes the shadow price,
measured in current utility, of per head …nancial wealth along the optimal path.
In the di¤erential equation (10.9), n cancels out and rearranging (10.9) gives

r +_
= :

This can be interpreted as a no-arbitrage condition. The left-hand side gives the
actual rate of return, measured in utility units, on the marginal unit of saving:
r can be seen as a dividend and _ as a capital gain. The right-hand side is the
required rate of return in utility units, : The household is willing to save the
marginal unit of income only up to the point where the actual return on saving
equals the required return.
The transversality condition (10.10) says that for t ! 1; the present shadow
value of per capita …nancial wealth should go to zero. Combined with (10.8), the
condition is that
lim at u0 (ct )e ( n)t = 0 (10.11)
t!1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.2. The agents 397

must hold along the optimal path. This requirement is not surprising if we
compare with the case where limt!1 at u0 (ct )e ( n)t > 0: In this case there
would be over-saving; U0 could be increased by reducing the “ultimate” at and
thereby, before eternity, consume more and save less. The opposite case, limt!1
at u0 (ct )e ( n)t < 0; will not even satisfy the NPG condition in view of Proposi-
tion 2 of the previous chapter. In fact, from that proposition we know that the
transversality condition (10.10) is equivalent to the NPG condition (10.7) being
satis…ed with strict equality, i.e.,
Rt
lim at e 0 (rs n)ds
= 0: (10.12)
t!1

Recall that the Maximum Principle gives only necessary conditions for an
optimal plan. But since the Hamiltonian is jointly concave in (a; c) for every t,
the necessary conditions are also su¢ cient, by Mangasarian’s su¢ ciency theorem.
The …rst-order conditions (10.8) and (10.9) give the Keynes-Ramsey rule:
c_t 1
= (rt ); (10.13)
ct (ct )
where (ct ) is the (absolute) elasticity of marginal utility,
ct
(ct ) 0
u00 (ct ) > 0: (10.14)
u (c t)

As we know from previous chapters, this elasticity indicates the consumer’s wish
to smooth consumption over time. The inverse of (ct ) is the elasticity of in-
tertemporal substitution in consumption. It indicates the willingness to vary
consumption over time in response to a change in the interest rate.
Note that the population growth rate, n; does not appear in the Keynes-
Ramsey rule. Going from n = 0 to n > 0 implies that rt is replaced by rt n in the
dynamic budget identity and is replaced by n in the criterion function. Hence
n cancels out in the Keynes-Ramsey rule. Yet n appears in the transversality
condition and thereby also in the level of consumption for given wealth, cf. (10.18)
below.

CRRA utility
In order that the model can accommodate Kaldor’s stylized facts, it should be
capable of generating a balanced growth path. When the population grows at
the same constant rate as the labor force, here n; by de…nition balanced growth
requires that per capita output, per capita capital, and per capita consumption
grow at constant rates. At the same time another of Kaldor’s stylized facts is
that the general rate of return in the economy tends to be constant. But (10.13)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
398 MODEL: RAMSEY

shows that having a constant per capita consumption growth rate at the same
time as r is constant, is only possible if the elasticity of marginal utility does not
vary with c: Hence, it makes sense to assume that the right-hand-side of (10.14)
is a positive constant, . We thus assume that the instantaneous utility function
is of CRRA form:
c1
u(c) = ; > 0; (10.15)
1
here, for = 1; the right-hand side should be interpreted as ln c as explained in
Section 3.3 of Chapter 3.
So our Keynes-Ramsey rule simpli…es to
c_t 1
= (rt ): (10.16)
ct

The consumption function The Keynes-Ramsey rule characterizes the opti-

mal rate of change of consumption. The optimal initial level of consumption, c0 ;
will be the highest feasible c0 which is compatible with both the Keynes-Ramsey
rule and the NPG condition. And for this reason the choice of c0 will exactly
comply with the transversality condition (10.12). Although at this stage an ex-
plicit determination of c0 is not necessary to pin down the equilibrium path of the
economy (see below), we note in passing that c0 can be found by the method de-
scribed at the end of Chapter 9. Indeed, given the book-keeping relation (10.6),
we know from Proposition 1 of that chapter that the transversality condition
(10.12) is equivalent to satisfying the intertemporal budget constraint with strict
equality: Z 1 Rt
ct e 0 (rs n)ds
dt = a0 + h0 : (10.17)
0
1
Rt
Solving the di¤erential equation (10.16), we get ct = c0 e 0 (rs )ds
which we
substitute for ct in (10.17). Isolating c0 now gives4

c0 = 0 (a0 + h0 ); where (10.18)

1
0 = R Rt (1 )rs
; and
1 ( +n)ds
e0 dt
Z0 1 Rt
h0 = wt e 0 (rs n)ds
dt:
0

Initial consumption is thus proportional to total wealth. The factor of propor-

tionality is 0 ; also called the marginal (and average) propensity to consume out
4
These formulas can also be derived directly from Example 1 of Chapter 9.5 by replacing
r( ) and by r( ) n and n; respectively. As to h0 ; see the hint in Exercise 9.1.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 399

of wealth. We see that the entire expected future evolution of wages and inter-
est rates a¤ects c0 through 0 : Moreover, 0 is less, the greater is the population
growth rate, n:5 The explanation is that the e¤ective utility discount rate, n; is
less, the greater is n. The propensity to save is greater the more mouths to feed in
the future. The initial saving level will be r0 a0 + w0 c0 = r0 a0 + w0 0 (a0 + h0 ):
gt
In case rt = r for all t and wt = w0 e ; where g < r n; we get 0 =
[( 1)r + n] = and a0 + h0 = a0 + w0 =(r n g):
In the Solow growth model the saving-income ratio is a parameter, a given
constant. The Ramsey model endogenizes the saving-income ratio. Solow’s para-
metric saving-income ratio is replaced by two “deeper” parameters, the rate of
impatience, ; and the desire for consumption smoothing, . As we shall see, the
resulting saving-income ratio will not generally be constant outside the steady
state of the dynamic system implied by the Ramsey model. But …rst we need a
description of production.

10.2.2 Firms
There is a large number of …rms. They have the same neoclassical production
function with CRS,
Yt = F (Ktd ; Tt Ldt ) (10.19)
where Yt is supply of output, Ktd is capital input, and Ldt is labor input, all
measured per time unit, at time t. The superscript d on the two inputs indicates
that these inputs are seen from the demand side. The factor Tt represents the
economy-wide level of technology as of time t and is exogenous. We assume there
is technological progress at a constant rate g ( 0) :
Tt = T0 egt ; T0 > 0: (10.20)
Thus the economy features Harrod-neutral technological progress, as is needed
for compliance with Kaldor’s stylized facts.
Necessary and su¢ cient conditions for the factor combination (Ktd ; Ldt ); where
Ktd > 0 and Ldt > 0; to maximize pro…ts under perfect competition are that
F1 (Ktd ; Tt Ldt ) = r^t ; (10.21)
F2 (Ktd ; Tt Ldt )Tt = wt : (10.22)

10.3 General equilibrium and dynamics

We now consider the economy as a whole and thereby the interaction between
households and …rms in the various markets. For simplicity, we assume that
5
This holds also if = 1; i. e., u(c) = ln c; since in that case 0 = n:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
400 MODEL: RAMSEY

the number of households is the same as the number of …rms. We normalize

this number to one so that F ( ; ) from now on is interpreted as the aggregate
production function and Ct as aggregate consumption.

Factor markets

In the short term, i.e., for …xed t, the available quantities of labor, Lt = L0 ent ;
and capital, Kt ; are predetermined. The factor markets clear at all points in time,
that is,
Ktd = Kt ; and Ldt = Lt; for all t 0: (10.23)

It is the rental rate, r^t ; and the wage rate, wt ; which adjust (immediately) so that
this is achieved for every t. Aggregate output can be written

Yt = F (Kt ; Tt Lt ) = Tt Lt F (k~t ; 1) Tt Lt f (k~t ); f 0 > 0; f 00 < 0; (10.24)

where k~t kt =Tt Kt =(Tt Lt ) is the technology-corrected capital labor ratio, also
sometimes just called the “capital intensity”. Substituting (10.23) into (10.21)
and (10.22), we …nd the equilibrium interest rate and wage rate:

@(Tt Lt f (k~t ))
rt = r^t = = f 0 (k~t ) ; (10.25)
@Kt
@(Tt Lt f (k~t )) h i
wt = Tt = f (k~t ) k~t f 0 (k~t ) Tt w(
~ k~t )Tt ; (10.26)
@(Tt Lt )

where k~t is at any point in time predetermined and where in (10.25) we have used
the no-arbitrage condition (10.1).

Capital accumulation

From now we leave out the explicit dating of the variables when not needed for
clarity. By national product accounting we have

K_ = Y C K: (10.27)

Let us check whether we get the same result from the wealth accumulation equa-
tion of the household. Because physical capital is the only asset in the economy,
aggregate …nancial wealth, A; at time t equals the total quantity of capital, K;

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 401

at time t:6 From (10.4) we thus have

A_ = K_ = rK + wL cL
= ~
(f 0 (k) ~
)K + (f (k) ~ 0 (k))T
kf ~ L cL (from (10.25) and (10.26))
= ~ L
f (k)T K cL (by rearranging and use of K kT ~ L)
= F (K; T L) K C=Y C K (by C cL):

Hence the book-keeping is in order (the national product account is consistent

with the national income account).
We now face an important di¤erence as compared with models where house-
holds have a …nite horizon, such as the Diamond OLG model. Current consump-
tion cannot be determined independently of the expected entire future evolution
of the economy. Consumption and saving, as we saw in Section 10.2, depend on
the expectations of the future path of wages and interest rates. And given the
presumption of perfect foresight, the households’expectations are identical to the
prediction that can be calculated from the model. In this way there is mutual de-
pendence between expectations and the level and evolution of consumption. We
can determine the level of consumption only in the context of the overall dynamic
analysis. In fact, the economic agents are in some sense in the same situation as
the outside analyst. They, too, have to think through the entire dynamics of the
economy in order to form their rational expectations.

The dynamic system

We get a concise picture of the dynamics by reducing the model to the minimum
number of coupled di¤erential equations. This minimum number is two. The key
endogenous variables are k~ K=(T L) and c~ C=(T L) c=T . Using the rule
for the growth rate of a fraction, we get

k~ K_ T_ L_ K_
= = (g + n) (from (10.2) and (10.20))
k~ K T L K
F (K; T L) C K
= (g + n) (from (10.27))
K
~
f (k) c~
= ( + g + n) (from (10.24)).
k~
6
Whatever …nancial claims on each other the households might have, they net out for the
household sector as a whole.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
402 MODEL: RAMSEY

The associated di¤erential equation for c~ is obtained in a similar way:

c~ c_ T_ 1
= = (rt ) g (from the Keynes-Ramsey rule)
c~ c T
1h 0 ~ i
= f (k) g (from (10.25)).

We thus end up with the dynamic system

k~ = f (k)
~ c~ ~
( + g + n)k; k~0 > 0 given, (10.28)
h
1 0 ~ i
c~ = f (k) g c~: (10.29)

There is no given initial value of c: Instead we have the transversality condition

(10.12). Using at = Kt =Lt k~t Tt = k~t T0 egt and rt = f 0 (k~t ) , we see that
(10.12) is equivalent to
Rt
lim k~t e 0 (f 0 (k~s ) g n)ds
= 0: (10.30)
t!1

Fig. 10.1 is an aid for the construction of the phase diagram in Fig. 10.2.
The curve OEB in Fig. 10.2 represents the points where k~ = 0 and is called the
nullcline for the di¤erential equation (10.28). We see from (10.28) that

k~ = 0 for c~ = f (k)
~ ( + g + n)k~ ~
c~(k): (10.31)

~ as the vertical distance between the curve

Fig. 10.1 displays the value of c~(k)
y~ = f (k) and the line y~ = ( + g + n)k~ (to save space the proportions are
~
somewhat distorted).7 The maximum value of c~(k); ~ if it exists, is reached at the
point where the tangent to the OEB curve in Fig. 10.2 is horizontal, i.e., where
~ = f 0 (k)
c~0 (k) ~ ( + g + n) = 0 or f 0 (k)
~ = g + n: The value of k~ which satis…es
this is the golden rule capital intensity, k~GR :

f 0 (k~GR ) = g + n: (10.32)

From (10.28) we see that for points above the k~ = 0 locus we have k~ < 0, whereas
for points below the k~ = 0 locus, k~ > 0. The horizontal arrows in the …gure
~
indicate these directions of movement for k.
7
As the graph is drawn, f (0) = 0; i.e., capital is assumed essential. But none of the conclu-
sions we are going to consider depends on this.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 403

We also need the nullcline for the di¤erential equation (10.29). We see from
(10.29) that
~ = + + g
c~ = 0 for f 0 (k) or c~ = 0: (10.33)
Let k~ > 0 satisfy the equation f 0 (k~ ) = + g: Then the vertical line k~ = k~
represents points where c~ = 0 (and so does of course the horizontal half-line
c~ = 0; k~ 0): For points to the left of the k~ = k~ line we have, according to
(10.29), c~ > 0: And for points to the right of the k~ = k~ line we have c~ < 0:
The vertical arrows in Fig. 10.2 indicate these directions of movement for c~. Four
illustrative examples of solution curves (I, II, III, and IV ) are drawn in the …gure.

Steady state
The point E has coordinates (k~ , c~ ) and represents the unique steady state.8
From (10.33) and (10.31) follows that

f 0 (k~ ) = + + g; and (10.34)

c~ = f (k~ ) ( + g + n)k~ : (10.35)

From (10.34) it can be seen that the real interest rate in steady state is

r = f 0 (k~ ) = + g: (10.36)

The e¤ective capital-labor ratio satisfying this equation is known as the modi…ed-
golden-rule capital intensity, k~M GR . The modi…ed golden rule is the rule saying
that for a representative agent economy to be in steady state, the capital intensity
must be such that the net marginal productivity of capital equals the required
rate of return, taking into account the pure rate of time preference, ; and the
desire for consumption smoothing, .9
We show below that the steady state is, in a speci…c sense, asymptotically sta-
ble. First we have to make sure, however, that the steady state is consistent with
8
As (10.33) shows, if c~t = 0; then c~ = 0: Therefore, mathematically, point B (if it exists) in
Fig. 10.2 is also a stationary point of the dynamic system. And if f (0) = 0; then according to
(10.29) and (10.31) also the point (0; 0) in the …gure is a stationary point. But these stationary
points have zero consumption forever and are therefore not steady states of any economic
system. From an economic point of view they are “trivial” steady states.
9
The of the Ramsey model corresponds to the intergenerational discount rate R of the
Barro dynasty model in Chapter 7. Indeed, in the discrete time Barro model we have 1 + r
= (1+R)(1+g) ; which, by taking logs on both sides and using …rst-order Taylor approximations
of ln(1 + x) around x = 0 gives r ln(1 + r ) = ln(1 + R) + ln(1 + g) R + g: Recall,
however, that in view of the considerable period length (about 25-30 years) of the Barro model,
this approximation may not be good.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
404 MODEL: RAMSEY

ỹ

ỹ = f (k̃)

δ+g+n

δ + ρ + θg
ỹ = (δ + g + n)k̃

0 k̃
k̃M GR k̃GR ¯
k̃

Figure 10.1: Building blocks for the phase diagram.

c̃
c̃˙ = 0

V IV
II
˙
k̃ = 0
c̃∗ E

c̃A A
VI

III
k̃
∗
k̃0 k̃ = k̃M GR k̃GR ¯
k̃

Figure 10.2: Phase diagram for the Ramsey model.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 405

general equilibrium. This consistency requires that the household’s transversality

condition (10.30) holds in the point E, where k~t = k~ and f 0 (k~t ) = + g for
all t: So the condition (10.30) becomes
lim k~ e ( + g g n)t
= 0: (10.37)
t!1

This is ful…lled if and only if + g > g + n; that is,

n > (1 )g: (A1)
This inequality also ensures that the improper integral U0 is bounded from above
(see Appendix B). If 1; (A1) is ful…lled as soon as the e¤ective utility discount
rate, n; is positive; (A1) may even hold for a negative n if not “too”
negative. If < 1, (A1) requires n to be “su¢ ciently positive”.
Since the parameter restriction (A1) can be written + g > g + n; it implies
that the steady-state interest rate, r , given in (10.36), is higher than the “nat-
ural”growth rate, g + n: If this did not hold, the transversality condition (10.12)
would fail at the steady-state point E. Indeed, along the steady-state path we
have
at e (r n)t = kt e (r n)t = k0 egt e (r n)t = k0 e(g+n r )t ;
which would take the constant positive value k0 for all t 0 if r = g + n and
would go to 1 for t ! 1 if r < g + n: The individual households would thus
be over-saving. Each household would in this situation alter its behavior and the
steady state could not be an equilibrium path.
Another way of seeing that r g + n can not be an equilibrium in a Ramsey
model is to recognize that this condition would make the in…nitely-lived house-
hold’s human wealth = 1 because wage income, wL; would grow at a rate, g + n;
at least as high as the real interest rate, r : This would motivate an immediate
increase in consumption and so the considered steady-state path would again not
be an equilibrium.
To have a model of interest, from now on we assume that the preference and
technology parameters satisfy the inequality (A1). As an implication, the capital
intensity in steady state, k~ ; is less than the golden-rule value k~GR . Indeed,
f 0 (k~ ) = + g > g + n = f 0 (k~GR ) , so that k~ < k~GR ; in view of f 00 < 0:
So far we have only ensured that if the steady state, E, exists, it is consistent
with general equilibrium. Existence of a steady state requires that the marginal
productivity of capital is su¢ ciently sensitive to variation in the capital intensity:
~
lim f 0 (k) > ~
+ g > lim f 0 (k) :
~
k!0 ~
k!1

We could proceed with this assumption. To allow comparison of the steady

state of the model with the golden rule allocation, we make the slightly stronger

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
406 MODEL: RAMSEY

assumption that f has the properties

~
lim f 0 (k) > + g and ~
lim f 0 (k) < g + n: (A2)
~
k!0 ~
k!1

Together with (A1) this implies limk!0 ~

~
f 0 (k) > + g > g+n > limk!1~
~
f 0 (k)
0 ~
: By continuity of f ; these inequalities ensure the existence of both k > 0 and
~
kGR > 0:10 Moreover, as illustrated by Fig. 10.1, the inequalities also ensure
existence of a k~ > 0 with the property that f (k) ~ ( + g + n)k~ = 0.11 Because
f (k) > 0 for all k~ > 0; it is implicit in the technology assumption (A2) that +
0 ~

g + n > 0: Even without deciding on the sign of n (a decreasing workforce should

not be ruled out in our days), this inequality seems a plausible presumption.

Trajectories in the phase diagram

A …rst condition for a path (k~t , c~t ); with k~t > 0 and c~t > 0 for all t 0; to
be a solution to the model is that it satis…es the system of di¤erential equations
(10.28)-(10.29). Indeed, to be technically feasible, it must satisfy (10.28) and to
comply with the Keynes-Ramsey rule, it must satisfy (10.29). Technical feasibility
of the path also requires that the initial value for k~ equals the historically given
value k~0 K0 =(T0 L0 ). In contrast, for c~ we have no given initial value. This is
because c~0 is a jump variable, also known as a forward-looking variable. By this
is meant an endogenous variable which can immediately shift to another value if
new information arrives so as to alter expectations about the future. We shall
see that the terminal condition (10.30), re‡ecting the transversality condition of
the households, makes up for this lack of an initial condition for c.
In Fig. 10.2 we have drawn some paths that could be solutions as t increases.
We are especially interested in the paths which are consistent with the historically
given k~0 , that is, paths starting at some point on the stippled vertical line in the
…gure. If the economy starts out with a high value of c~, it will follow a curve like
II in the …gure. The low level of saving implies that the capital stock goes to
zero in …nite time (see Appendix C). If the economy starts out with a low level
of c~, it will follow a curve like III in the …gure.
_
The high level of saving implies
that the capital intensity converges towards k~ in the …gure.
All in all this suggests the existence of an initial level of consumption some-
where in between, which results in a path like I. Indeed, since the curve II
emerged with a high c~0 , then by lowering this c~0 slightly, a path will emerge in
10
The often presumed Inada conditions, limk!0 ~ = 1 and lim~
f 0 (k) 0 ~
~ k!1 f (k) = 0; are stricter
than (A2) and not necessary.
11
We
_ _
claim that
_
k~ > k~GR must hold. Indeed, this inequality follows from f 0 (k~GR ) = + n + g
~ k~ > f 0 ( k);
f ( k)= ~ the latter inequality being due to f 00 < 0 and f (0) 0.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 407

which the maximal value of k~ on the k~ = 0 locus is greater than curve II’s maxi-
mal k~ value.12 We continue lowering c~0 until the path’s maximal k~ value is exactly
equal to k~ . The path which emerges from this, namely the path I, starting at
the point A, is special in that it converges towards the steady-state point E. No
other path starting at the stippled line, k~ = k~0 ; has this property. Paths starting
above A do not, as we just saw. Neither do paths starting below A, like path
III. Either this path never reaches the consumption level c~A in which case it can
not converge to E, of course. Or, after a while its consumption level reaches c~A ;
but at the same time it must have k~ > k~0 : From then on, as long as k~ k~ , for
every c~-value that path III has in common with path I, path III has a higher
k~ and a lower c~ than path I (use (10.28) and (10.29)). Hence, path III diverges
from point E.
Had we considered a value of k~0 > k~ , there would similarly be a unique value
of c~0 such that the path starting from (k~0 , c~0 ) would converge to E (see path IV
in Fig. 10.2).
The point E is a saddle point. By this is meant a steady-state point with
the following property: there exists exactly two paths, one from each side of k~ ,
that converge towards the steady-state point; all other paths (at least starting
in a neighborhood of the steady state) move away from the steady state and
asymptotically approach one of the two diverging paths, the stippled North-West
and South-East curves in Fig. 10.2. The two converging paths together make up
what is known as the stable branch (or stable arm); on their own they are referred
to as saddle paths (sometimes referred to in the singular as the saddle path).13
The stippled diverging paths in Fig. 10.2 together make up the unstable branch
(or unstable arm).

The equilibrium path

A solution to the model is a path which is technically feasible and in addition
satis…es a set of equilibrium conditions. In analogy with the de…nition in discrete
time (see Chapter 3) a path (k~t ; c~t )1
t=0 is called a technically feasible path if (i) the
~
path has kt 0 and c~t 0 for all t 0; (ii) it satis…es the accounting equation
(10.28); and (iii) it starts out, at t = 0, with the historically given initial capital
intensity. An equilibrium path with perfect foresight is then a technically feasible
path (k~t ; c~t )1
t=0 with the properties that the path (a) is consistent with …rms’

12
As an implication of the uniqueness theorem for di¤erential equations (see Math tools), two
solution paths in the phase plane cannot intersect.
13
An algebraic de…nition of a saddle point, in terms of eigenvalues, is given in Appendix A.
There it is also shown that if limk!0
~
~ = 0; then the saddle path on the left side of the steady
f (k)
state in Fig. 10.2 will start out in…nitely close to the origin.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
408 MODEL: RAMSEY

pro…t maximization and households’ optimization given their expectations; (b)

is consistent with market clearing for all t 0; and (c) has the property that
~ k~t )Tt and rt = f 0 (k~t )
the evolution of the pair (wt , rt ); where wt = w( ; is as
expected by the households. Among other other things these conditions require
the transformed Keynes-Ramsey rule (10.29) and the transversality condition
(10.30) to hold for all t 0:
Consider the case where 0 < k~0 < k~ ; as illustrated in Fig. 10.2. Then, the
path starting at point A and following the saddle path towards the steady state
is an equilibrium path because, by construction, it is technically feasible and in
addition has the required properties, (a), (b), and (c). More intuitively: if the
households expect an evolution of wt and rt corresponding to this path (that is,
expect a corresponding underlying movement of k~t ; which we know unambigu-
ously determines rt and wt ), then these expectations will induce a behavior the
aggregate result of which is an actual path for (k~t ; c~t ) that con…rms the expecta-
tions. And along this path the households …nd no reason to correct their behavior
because the path allows both the Keynes-Ramsey rule and the transversality con-
dition to be satis…ed.
No other path than the saddle path can be an equilibrium. This is because
no other technically feasible path is compatible with the households’individual
utility maximization under perfect foresight. An initial point above point A can
be excluded in that the implied path of type II does not satisfy the household’s
NPG condition (and, consequently, not at all the transversality condition).14 If
the individual household expected an evolution of rt and wt corresponding to path
II, then the household would immediately choose a lower level of consumption,
that is, the household would deviate in order not to su¤er the same fate as Charles
Ponzi. In fact all the households would react in this way. Thus path II would not
be realized and the expectation that it would, can not be a rational expectation.
Likewise, an initial point below point A can be ruled out because the implied
path of type III does not satisfy the household’s transversality condition but
implies over-saving. Indeed, at some point in the future, say at time t1 ; the
economy’s capital intensity would pass the golden rule value so that for all t > t1 ;
rt < g + n: But with a rate of interest permanently below the growth rate of
wage income of the household, the present value of human wealth is in…nite.
This motivates a higher consumption level than that along the path. Thus,
if the household expects an evolution of rt and wt corresponding to path III,
then the household will immediately deviate and choose a higher initial level of
consumption. But so will all the households react and the expectation that the
economy will follow path III can not be rational.
We have presumed 0 < k~0 < k~ : If instead k~0 > k~ ; the economy would move
14
This is shown in Appendix C.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.3. General equilibrium and dynamics 409

along the saddle path from above. Paths like V and V I in Fig. 10.2 can be ruled
out because they violate the NPG condition and the transversality condition,
respectively. With this we have shown:
PROPOSITION 1 Assume (A1) and (A2). Let there be a given k~0 > 0: Then
the Ramsey model exhibits a unique equilibrium path, characterized by (k~t , c~t )
converging, for t ! 1, toward a unique steady state with a capital intensity k~
satisfying f 0 (k~ ) = + g: In the steady state the real interest rate is given by
the modi…ed-golden-rule formula, r = + g, the per capita consumption path
is ct = c~ T0 egt ; where c~ = f (k~ ) ( + g + n)k~ ; and the real wage path is wt
= w(~ k~ )T0 egt :
A numerical example based on one year as the time unit: = 2; g = 0:02;
n = 0:01 and = 0:01: Then, r = 0:05 > 0:03 = g + n.
So output per capita, yt Yt =Lt y~t Tt ; tends to grow at the rate of techno-
logical progress, g :

y_ t y~t T_t f 0 (k~t )k~t

+ = +g !g for t ! 1;
yt y~t Tt f (k~t )

in view of k~t ! 0. This is also true for the growth rate of consumption per capita
and the real wage, since ct c~t Tt and wt = w( ~ k~t )Tt :
The intuition behind the convergence lies in the neoclassical principle of di-
minishing marginal productivity of capital. Starting from a low capital intensity
and therefore a high marginal and average productivity of capital, the resulting
high aggregate saving15 will be more than enough to maintain the capital inten-
sity which therefore increases. But when this happens, the marginal and average
productivity of capital decreases and the resulting saving, as a proportion of the
capital stock, declines until eventually it is only su¢ cient to replace worn-out ma-
chines and equip new “e¤ective”workers with enough machines to just maintain
the capital intensity. If instead we start from a high capital intensity, a similar
story can be told in reverse. In the long run the capital intensity settles down
at the steady-state level, k~ ; where the marginal saving and investment yields
a return as great as the representative household’s willingness to postpone the
marginal unit of consumption. Since the adjustment process is based on capital
accumulation, it is slow. The “speed of adjustment”, in the sense of the propor-
tionate rate of decline per year of the distance to the steady state, k~ k~ ; is
generally assessed to be in the interval (0.02, 0.10), assuming absence of distur-
bances to the system during the adjustment.
15
Saving will be high because the negative substitution and wealth e¤ects on current con-
sumption of the high interest rate dominate the income e¤ect.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
410 MODEL: RAMSEY

The equilibrium path generated by the Ramsey model is necessarily dynam-

ically e¢ cient and satis…es the modi…ed golden rule in the long run. Why this
contrast to Diamonds OLG model where equilibrium paths may be dynamically
ine¢ cient? The reason lies in the fact that only a “single in…nity”, not a “double
in…nity”, is involved in the Ramsey model. The time horizon of the economy is
in…nite but the number of decision makers is …nite. Births (into adult life) do
not re‡ect the emergence of new economic agents with separate interests. It is
otherwise in the Diamond OLG model where births imply entrance of new eco-
nomic decision makers whose preferences no-one cared about in advance. In that
model neither is there any …nal date, nor any …nal decision maker. Because of
this di¤erence, in some respects the two models give di¤erent results. A type of
equilibria, namely dynamically ine¢ cient ones, can be realized in the Diamond
model but not so in the Ramsey model. A rate of time preference low enough
to generate a tendency to a long-run interest rate below the income growth rate
is inconsistent with the conditions needed for general equilibrium in the Ramsey
model. And such a low rate of time preference is in fact ruled out in the Ramsey
model by the parameter restriction (A1).

The concept of saddle-point stability

The steady state of the model is globally asymptotically stable for arbitrary initial
values of the capital intensity (the phase diagram only veri…es local asymptotic
stability, but the extension to global asymptotic stability is veri…ed in Appendix
A). If k~ is hit by a shock at time 0 (say by a discrete jump in the technology level
T0 ), the economy will converge toward the same unique steady state as before. At
…rst glance this might seem peculiar considering that the steady state is a saddle
point. Such a steady state is unstable for arbitrary values of both coordinates in
the initial point (k~0 ; c~0 ). But the crux of the matter is that it is only the initial k~
that is arbitrary. The model assumes that the decision variable c0 ; and therefore
the value of c~0 c0 =T0 ; immediately adjusts to the given circumstances and
the available information about the future. That is, the model assumes that c~0
always takes the value needed for the household’s transversality condition under
perfect foresight to be satis…ed. This ensures that the economy is initially on
the saddle path, cf. the point A in Fig. 10.2. In the language of di¤erential
equations conditional asymptotic stability is present. The condition that ensures
the stability in our case is the transversality condition.
We shall follow the common terminology in macroeconomics and call a steady
state of a two-dimensional dynamic system (locally) saddle-point stable if:

1. the steady state is a saddle point;

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.4. Comparative analysis 411

2. one of the two endogenous variables is predetermined while the other is a

jump variable;

3. the saddle path is not parallel to the jump variable axis;

4. there is a boundary condition on the system such that the diverging paths
are ruled out as solutions.

Thus, to establish saddle-point stability all four properties must be veri…ed.

If for instance point 1 and 2 hold but, contrary to point 3, the saddle path is
parallel to the jump variable axis, then saddle-point stability does not obtain.
Indeed, given that the predetermined variable initially deviated from its steady-
state value, it would not be possible to …nd any initial value of the jump variable
such that the solution of the system would converge to the steady state for t ! 1:
In the present case, we have already veri…ed point 1 and 2. And as the phase
diagram indicates, the saddle path is not vertical. So also point 3 holds. The
transversality condition ensures that also point 4 holds. Thus, the Ramsey model
is saddle-point stable. In Appendix A it is shown that the positively-sloped saddle
path in Fig. 10.2 ranges over all k~ > 0 (there is nowhere a vertical asymptote
to the saddle path). Hence, the steady state is globally saddle-point stable. All
in all, these characteristics of the Ramsey model are analogue to those of Barro’s
dynasty model in discrete time when the bequest motive is operative.

10.4 Comparative analysis

10.4.1 The role of key parameters
The conclusion that in the long run the real interest rate is given by the modi…ed
golden rule formula, r = + g, tells us that only three parameters matter:
the rate of time preference, the elasticity of marginal utility, and the rate of
technological progress. A higher , i.e., more impatience and thereby less willing-
ness to defer consumption, implies less capital accumulation and thus in the long
run smaller capital intensity, higher interest rate, and lower consumption than
otherwise. The long-run growth rate is una¤ected.
A higher will have a similar e¤ect. As is a measure of the desire for
consumption smoothing, a higher implies that a larger part of the greater wage
income in the future (re‡ecting a positive g) will be consumed immediately. This
implies less saving and thereby less capital accumulation and so a lower k~ and
higher r . Similarly, the long-run interest rate will depend positively on the
technology growth rate g because the higher g is, the greater is the expected
future wage income. Thereby the consumption possibilities in the future are

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
412 MODEL: RAMSEY

greater even without any current saving. This discourages current saving and we
end up with lower capital accumulation and lower e¤ective capital intensity in
the long run, hence higher interest rate. It is also true that the higher is g; the
higher is the rate of return needed to induce the saving required for maintaining
a steady state and resist the desire for more consumption smoothing.
The long-run interest rate is independent of the particular form of the ag-
gregate production function, f . This function matters for what e¤ective capital
intensity and what consumption level per unit of e¤ective labor are compatible
with the long-run interest rate. This kind of results are speci…c to representative
agent models. This is because only in these models will the Keynes-Ramsey rule
hold not only for the individual household, but also at the aggregate level.
Unlike the Solow growth model, the Ramsey model provides a theory of the
evolution and long-run level of the saving rate. The endogenous gross saving rate
of the economy is

Yt Ct K_ t + Kt K_ t =Kt + k~t =k~t + g + n +

st = = =
Yt Yt Yt =Kt f (k~t )=k~t
g+n+
! s for t ! 1: (10.38)
f (k~ )=k~
By determining the path of k~t , the Ramsey model determines how st moves over
time and adjusts to its constant long-run level. Indeed, for any given k~ > 0,
the equilibrium value of c~t is uniquely determined by the requirement that the
economy must be on the saddle path. Since this de…nes c~t as a function, c~(k~t );
of k~t ; there is a corresponding function for the saving rate in that st = 1
c~(k~t )=f (k~t ) s(k~t ); so s(k~ ) = s :
We note that the long-run saving rate is a decreasing function of the rate of
impatience, ; and the desire of consumption smoothing, ; it is an increasing
function of the capital depreciation rate, ; and the rate of population growth, n:
For an example with an explicit formula for the long-run saving rate, consider:
EXAMPLE 1 Suppose the production function is Cobb-Douglas:
~ = Ak~ ;
y~ = f (k) A > 0; 0 < < 1: (10.39)
~ = A k~
Then f 0 (k) 1 ~ k:
= f (k)= ~ In steady state we get, by use of the steady-
state result (10.34),
f (k~ ) 1 + + g
= f 0 (k~ ) = :
k~
Substitution in (10.38) gives
+g+n
s = < ; (10.40)
+ + g

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.4. Comparative analysis 413

where the inequality follows from our parameter restriction (A1). Indeed, (A1)
implies + g > g + n: The long-run saving rate depends positively on the
following parameters: the elasticity of production w.r.t. to capital, ; the capital
depreciation rate, ; and the population growth rate, n: The long-run saving rate
depends negatively on the rate of impatience, ; and the desire for consumption
smoothing, : The role of the rate of technological progress is ambiguous.16
A numerical example (time unit = 1 year): If n = 0:005; g = 0:015; = 0:025;
= 3; and = 0:07; then s = 0:21: With the same parameter values except
= 0:05; we get s = 0:19:
It can be shown (see Appendix D) that if, by coincidence, = 1=s ; then
0 ~
s (k) = 0; that is, the saving rate st is also outside of steady state equal to s . In
view of (10.40), the condition = 1=s is equivalent to the “knife-edge”condition
= ( + )= [ ( + g + n) g] : More generally, assuming ( + g + n) > g
(which seems likely empirically), we have that if Q 1=s (i.e., Q ); then s0 (k) ~ Q
0; respectively (and if instead ( +g +n) g; then s0 (k) ~ < 0; unconditionally):17
Data presented in Barro and Sala-i-Martin (2004, p. 15) indicate no trend for
the US saving rate, but a positive trend for several other developed countries
since 1870. One interpretation is that whereas the US has for a long time been
close to its steady state, the other countries are still in the adjustment process
toward the steady state. As an example, consider the parameter values = 0:05;
= 0:02; g = 0:02 and n = 0:01: In this case we get = 10 if = 0:33; given
< 10; these other countries should then have s0 (k) ~ < 0 which, according to the
model, is compatible with a rising saving rate over time only if these countries
are approaching their steady state from above (i.e., they should have k~0 > k~ ):
It may be argued that should also re‡ect the role of education and R&D in
production and thus be higher; with = 0:75 we get = 1:75: Then, if > 1:75;
these countries would have s0 (k)~ > 0 and thus approach their steady state from
below (i.e., k~0 < k~ ):

10.4.2 Solow’s growth model as a special case

The above results give a hint that Solow’s growth model, with a given constant
saving rate s 2 (0; 1) and given ; g; and n (with +g + n > 0); can, under
certain circumstances, be interpreted as a special case of the Ramsey model. The
Solow model in continuous time is given by

k~t = sf (k~t ) ( + g + n)k~t :

16
Partial di¤erentiation w.r.t. g yields @s =@g = [ n ( 1) ] =( + + g)2 ; the sign
of which cannot be determined a priori.
17
See Appendix D.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
414 MODEL: RAMSEY

The constant saving rate implies proportionality between consumption and in-
come. In growth-corrected terms per capita consumption is

c~t = (1 s)f (k~t ):

For the Ramsey model to yield this, the production function must be like in
(10.39) (i.e., Cobb-Douglas) with > s: And the elasticity of marginal utility, ;
must satisfy = 1=s: Finally, the rate of time preference, ; must be such that
(10.40) holds with s replaced by s; which implies = ( + g + n)=s g:
It remains to show that this satis…es the inequality, n > (1 )g, which is
necessary for existence of an equilibrium in the Ramsey model. Since =s > 1;
the chosen satis…es > +g +n g = n+(1 )g; which was to be proved.
Thus, in this case the Ramsey model generates an equilibrium path which implies
an evolution identical to that generated by the Solow model with s = 1= .18
With this foundation of the Solow model, it will always hold that s = s <
sGR , where sGR is the golden rule saving rate. Indeed, from (10.38) and (10.32),
respectively,

( + g + n)k~GR f 0 (k~GR )k~GR

sGR = = = >s ;
f (k~GR ) f (k~GR )

from the Cobb-Douglas speci…cation and (10.40), respectively.

A point of the Ramsey model vis-a-vis the Solow model is to replace a me-
chanical saving rule by maximization of discounted utility and thereby, on the
one hand, open up for a wider range of possible evolutions, welfare analysis,
and analysis of incentive e¤ects of economic policy on households’ saving. On
the other hand, in some respects the Ramsey model narrows down the range of
possibilities, for example by ruling out over-accumulation (dynamic ine¢ ciency).

10.5 A social planner’s problem

Another implication of the Ramsey framework is that the decentralized market
equilibrium (within the idealized presumptions of the model) brings about the
same allocation of resources as would a social planner facing the same technology
and initial resources as described above and having the same criterion function
as the representative household.
18
A more elaborate account of the Solow model as a special case of the Ramsey model is
given in Appendix D.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.5. A social planner’s problem 415

10.5.1 The equivalence theorem

As in Chapter 8, by a social planner we mean a …ctional central authority who is
”all-knowing and all-powerful”and is constrained only by the limitations arising
from technology and initial resources. Within these con…nes the social planner
can fully decide on the resource allocation. Since we consider a closed economy,
the social planner has no access to an international loan market.
Let the economy be closed and let the social welfare function be time separable
with constant elasticity, ^; of marginal utility and a pure rate of time preference
^:19 Then the social planner’s optimization problem is
Z 1 ^
c1t (^ n)t
max W0 = e dt s.t. (10.41)
(ct )1
t=0 0 1 ^
ct 0; (10.42)
ct
k~t = f (k~t ) ( + g + n)k~t ; (10.43)
Tt
k~t 0 for all t 0: (10.44)

We assume ^ > 0 and ^ n > (1 ^)g in line with the assumption (A1) for
^
the market economy above. In case ^ = 1; the expression ct1 =(1 ^) should
be interpreted as ln ct : No market prices or other elements belonging to the spe-
ci…c market institutions of the economy enter the social planner’s problem. The
dynamic constraint (10.43) re‡ects the national product account. Because the
economy is closed, the social planner does not have the opportunity of borrowing
or lending from abroad. Hence there is no solvency requirement. Instead we just
impose the de…nitional constraint (10.44) of non-negativity of the state variable
~
k.
The problem is the continuous time analogue of the social planner’s problem in
discrete time in Chapter 8. Note, however, a minor conceptual di¤erence, namely
that in continuous time there is in the hshort run no upper bound i on the ‡ow
~ ~
variable ct ; that is, no bound like ct Tt f (kt ) ( + g + n)kt : A consumption
intensity ct which is higher than the right-hand side of this inequality will just be
re‡ected in a negative value of the ‡ow variable k~t :20
19
Possible reasons for allowing these two preference parameters to deviate from the corre-
sponding parameters in the private sector are given Section 8.1.1.
20
As usual we presume that capital can be “eaten”. That is, we consider the capital good to
be instantaneously convertible to a consumption good. Otherwise there would be at any time
an upper bound on c, namely c T f (k); ~ saying that the per capita consumption ‡ow cannot
exceed the per capita output ‡ow. The role of such constraints is discussed in Feichtinger and
Hartl (1986).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
416 MODEL: RAMSEY

To solve the problem we apply the Maximum Principle. The current-value

Hamiltonian is

c1
^ h c i
~
H(k; c; ; t) = ~
+ f (k) ~
( + g + n)k ;
1 ^ T

where is the adjoint variable associated with the dynamic constraint (10.43).
An interior optimal path (k~t ; ct )1
t=0 will satisfy that there exists a continuous
function = (t) such that, for all t 0;

@H ^ ^
= c = 0; i.e., c = ; and (10.45)
@c T T
@H ~ _
= (f 0 (k) g n) = (^ n) (10.46)
@ k~
hold along the path and the transversality condition,

lim k~t t e (^ n)t

= 0; (10.47)
t!1

is satis…ed.21
The condition (10.45) can be seen as a M C = M B condition and illustrates
that t is the social planner’s shadow price, measured in terms of current utility,
of k~t along the optimal path.22 The di¤erential equation (10.46) tells us how this
shadow price evolves over time. The transversality condition, (10.47), together
with (10.45), entails the condition
^
lim k~t ct egt e (^ n)t
= 0;
t!1

where the unimportant factor T0 has been eliminated. Imagine the opposite were
^
true, namely that limt!1 k~t ct e[g (^ n)]t > 0. Then, intuitively U0 could be
increased by reducing the long-run value of k~t , i.e., consume more and save less:
By taking logs in (10.45) and di¤erentiating w.r.t. t, we get ^c=c
_ = _= g:
Inserting (10.46) and rearranging gives the condition

c_ 1 _ 1
= (g ~
) = (f 0 (k) ^): (10.48)
c ^ ^
21
The in…nite-horizon Maximum Principle itself does not guarantee validity of such a straight-
forward extension of a necessary transversality condition from a …nite horizon to an in…nite hori-
zon. Yet, this extension is valid for the present problem when ^ n > (1 ^)g, cf. Appendix
E.
22
Decreasing ct by one unit, increases k~t by 1=Tt units, each of which are worth t utility
units to the social planner.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.5. A social planner’s problem 417

This is the social planner’s Keynes-Ramsey rule. If the rate of time preference, ^;
~
is lower than the net marginal productivity of capital, f 0 (k) , the social planner
will let per capita consumption be relatively low in the beginning in order to attain
greater per capita consumption later. The lower the impatience relative to the
return to capital, the more favorable it becomes to defer consumption.
Because c~ c=T; we get from (10.48) qualitatively the same di¤erential equa-
tion for c~ as we obtained in the decentralized market economy. And the dynamic
resource constraint (10.43) is of course identical to that of the decentralized mar-
ket economy. Thus, the dynamics are in principle unaltered and the phase dia-
gram in Fig. 10.2 is still valid. The solution of the social planner implies that
the economy will move along the saddle path towards the steady state. This
trajectory, path I in the diagram, satis…es both the …rst-order conditions and
the transversality condition. However, paths such as III in the …gure do not
satisfy the transversality condition of the social planner but imply permanent
over-saving. And paths such as II in the …gure will experience a sudden end
when all the capital has been used up. Intuitively, they cannot be optimal. A
rigorous argument is given in Appendix E, based on the fact that the Hamil-
~ c~): Thence, not only is the saddle path an optimal
tonian is strictly concave in (k;
solution, it is the unique optimal solution.
Comparing with the market solution of the previous section, we have estab-
lished:
PROPOSITION 2 (equivalence theorem) Consider an economy with neoclassical
CRS technology as described above and a representative in…nitely-lived household
with preferences as in (10.3) with u(c) = c1 =(1 ): Assume (A1) and (A2).
~
Let there be a given k0 > 0: Then perfectly competitive markets bring about the
same resource allocation as that brought about by a social planner with the same
criterion function as the representative household, i.e., with ^ = and ^ = .
This is a continuous time analogue to the discrete time equivalence theorem of
Chapter 8.
The capital intensity k~ in the social planner’s solution will not converge to-
wards the golden rule level, k~GR ; but towards a level whose distance to the golden
rule level depends on how much ^ + ^g exceeds the natural growth rate, g + n:
Even if society would be able to consume more in the long term if it aimed for
the golden rule level, this would not compensate for the reduction in current con-
sumption which would be necessary to achieve it. This consumption is relatively
more valuable, the greater is the social planner’s e¤ective rate of time preference,
^ n. In line with the market economy, the social planner’s solution ends up in
a modi…ed golden rule. In the long term, net marginal productivity of capital is
determined by preference parameters and productivity growth and equals ^ + ^g
> g + n: Hereafter, given the net marginal productivity of capital, the capital

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
418 MODEL: RAMSEY

intensity and the level of the consumption path is determined by the production
function.

Classical versus average utilitarianism*

In the above analysis the social planner maximizes the sum of discounted per
capita utilities weighted by generation size. We call this classical utilitarianism.
As an implication, the e¤ective utility discount rate, n; varies negatively (one
to one) with the population growth rate. Since this corresponds to how the per
capita rate of return on saving, r n; is “diluted” by population growth, the
net marginal productivity of capital in steady state becomes independent of n;
namely equal to ^ + ^g:
Some textbooks, Blanchard and Fischer (1989) for instance, assumes what
might be called average utilitarianism. Here the social planner maximizes the
sum of discounted per capita utilities without weighing by generation size. Then
the e¤ective utility discount rate is independent of the population growth rate,
n. With ^ still denoting the pure rate of time preference, the criterion function
becomes Z 1 1 ^
ct
W0 = e ^t dt:
0 1 ^
The social planner’s solution then converges towards a steady state with the net
marginal productivity of capital

f 0 (k~ ) = ^ + n + ^g: (10.49)

Here, an increase in n will imply higher long-run net marginal productivity of

capital and lower capital intensity, everything else equal. The representative
household in the Ramsey model may of courseR also have a criterion function in
1
line with average utilitarianism, that is, U0 = 0 u(ct )e t dt. Then, the interest
rate in the economy will in the long run be r = + n + g and so an increase in
n will increase r and decrease k~ :
The more common approach is classical utilitarianism, which may be based
on the argument: “if more people bene…t, so much the better”.

10.5.2 Ramsey’s original zero discount rate and the over-

taking criterion*
It was mostly the perspective of a social planner, rather than the market mecha-
nism, which was at the center of Ramsey’s original analysis. The case considered
by Ramsey has g = n = 0: Ramsey maintained that the social planner should
“not discount later enjoyments in comparison with earlier ones, a practice which

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.5. A social planner’s problem 419

is ethically indefensible and arises merely from the weakness of the imagination”
(Ramsey 1928). So Ramsey has n = = 0: Given the instantaneous utility
0 00
function, u; where u > 0; u < 0; and given = 0; Ramsey’s original problem
was: choose (ct )1t=0 so as to optimize (in some sense, see below)
Z 1
W0 = u(ct )dt s.t.
0
ct 0;
_kt = f (kt ) ct kt ;
kt 0 for all t 0:

A condition corresponding to our assumption (A1) above does not apply.

So the improper integral W0 will generally not be bounded23 and Ramsey can
not use maximization of W0 as an optimality criterion. Instead he considers a
criterion akin to the overtaking criterion we considered in a discrete time context
in Chapter 8. We only have to reformulate this criterion for a continuous time
setting.
Let (ct )1
t=0 be the consumption path associated with an arbitrary technically
feasible path and let (^ct ) be the consumption path associated with our candidate
as an optimal path, that is, the path we wish to test for optimality. De…ne
Z T Z T
DT u(^
ct )dt u(ct )dt: (10.50)
0 0

Then the feasible path (^ ct )1

t=0 is overtaking optimal, if for any feasible path,
(ct )t=0 ; there exists a number T 0 0 such that DT
1
0 for all T T 0 . That is,
if for every alternative feasible path, the candidate path has from some date on,
cumulative utility up to all later dates at least as great as that of the alternative
feasible path, then the candidate path is overtaking optimal.
We say that the candidate path is weakly preferred in case we just know that
DT 0 for all T T 0 . If DT 0 can be replaced by DT > 0; we say it is strictly
preferred.24
Optimal control theory is also applicable with this criterion. The current-value
HamiItonian is
H(k; c; ; t) = u(c) + [f (k) c k] :
R1
23
Suppose for instance that ct ! c for t ! 1: Then 0 u(ct )dt = 1 for u(c) ? 0;
respectively.
24
A slightly more generally applicable optimality criterion is the catching-up criterion. The
meaning of this criterion in continuous time is analogue to its meaning in discrete time, cf.
Chapter 8.3. The overtaking as well as the catching-up criterion entail generally only a par-
tial.ordering of alternative technically feasible paths.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
420 MODEL: RAMSEY

The Maximum Principle states that an interior overtaking-optimal path will sat-
isfy that there exists an adjoint variable such that for all t 0 it holds along
this path that

@H
= u0 (c) = 0; and (10.51)
@c
@H _:
= (f 0 (k) )= (10.52)
@k
Since = 0; the Keynes-Ramsey rule reduces to
c_t 1 c
= (f 0 (kt ) ); where (c) u00 (c):
ct (ct ) u0 (c)

One might conjecture that also the transversality condition,

lim kt t = 0; (10.53)
t!1

is necessary for optimality but, as we will see below, this turns out to be wrong
in this case with no discounting.
Our assumption (A2) here reduces to limk!0 f 0 (k) > > limk!1 f 0 (k) (which
requires > 0): Apart from this, the phase diagram is fully analogue to that in
Fig. 10.2, except that the steady state, E, is now at the top of the k_ = 0 curve.
This is because in steady state, f 0 (k ) = 0: This equation also de…nes kGR in
this case: It can be shown that the saddle path is again the unique solution to
the optimization problem (the method is essentially the same as in the discrete
time case of Chapter 8). The intuitive background is that failing to approach the
golden rule would imply a forgone “opportunity of in…nite gain”.
A noteworthy feature is that in this case the Ramsey model constitutes a
counterexample to the widespread presumption that an optimal plan with in…nite
horizon must satisfy a transversality condition like (10.53). Indeed, by (10.51),
0 0
t = u (ct ) ! u (c ) for t ! 1 along the overtaking-optimal path (the saddle
path). Thus, instead of (10.53), we get

lim kt t = k u0 (c ) > 0:
t!1

With CRRA utility it is straightforward to generalize these results to the case

g 0; n 0 and ^ n = (1 ^)g: The social planner’s overtaking-optimal
solution is still the saddle path approaching the golden rule steady state; and this
solution violates the seemingly “natural” transversality condition. The reason
is essentially that we have no condition like the parameter restriction (A1) in
Section 10.3, ensuring boundedness from above of the utility integral.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.6. Concluding remarks 421

Note also that with zero e¤ective utility discounting, there can not be equi-
librium in the market economy version of this story. The real interest rate would
in the long run be zero and thus the human wealth of the in…nitely-lived house-
hold would be in…nite. But then the demand for consumption goods would be
unbounded and equilibrium thus be impossible.

10.6 Concluding remarks

The Ramsey model has played an important role as a way of structuring econo-
mists’ thoughts about many macrodynamic phenomena. As illustrated in Fig.
10.3, the model can be seen as situated at one end of a line segment where the
Diamond OLG model is situated at the opposite end. Both models build on
idealized assumptions. The Diamond model ignores any bequest motive and em-
phasizes life-cycle behavior and heterogeneity in the population. The Ramsey
model implicitly assumes an altruistic bequest motive which is always operative
and which turns households into homogeneous, in…nitely-lived agents. In this way
the Ramsey model ends up as an easy-to-apply framework, suggesting inter alia a
clear-cut theory of the level of the real interest rate in the long run. The model’s
usefulness lies in allowing general equilibrium analysis of an array of problems in
a “vacuum”.
The next chapter discusses di¤erent applications of the Ramsey model. Be-
cause of the model’s simplicity, one should always be aware of the risk of non-
robust conclusions. The assumption of a representative household is a main lim-
itation. Indeed, it is not easy to endow the dynasty portrait of households with
plausibility. The lack of heterogeneity in the model’s population of households
implies a danger that important interdependencies between di¤erent classes of
agents are unduly neglected. For some problems these interdependencies may be
of only secondary importance, but they are crucial for others (for instance, issues
concerning public debt or interaction between private debtors and creditors or
issues where income and wealth distribution matter).
Another problematic feature of the model is that it endows the households
with an extreme amount of information about the future. There can be good
reasons for bearing in mind the following warning (by Solow, 1990, p. 221) against
overly reliance on saddle-point stability in the analysis of a market economy:

“The problem is not just that perfect foresight into the inde…nite future
is so implausible away from steady states. The deeper problem is that in
practice if there is any practice miscalculations about the equilibrium
path may not reveal themselves for a long time. The mistaken path gives
no signal that it will be ”ultimately“ infeasible. It is natural to comfort

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
422 MODEL: RAMSEY

oneself: whenever the error is perceived there will be a jump to a better

approximation to the converging arm. But a large jump may be required.
In a decentralized economy it will not be clear who knows what, or where
the true converging arm is, or, for that matter, exactly where we are now,
given that some agents (speculators) will already have perceived the need
for a mid-course correction while others have not. This thought makes it
hard even to imagine what a long-run path would look like. It strikes me
as more or less devastating for the interpretation of quarterly data as the
solution of an in…nite time optimization problem.”

As we saw in Section 10.5.2, Ramsey’s original analysis (Ramsey 1928) dealt

with a social planner’s in…nite horizon optimal control problem. In that opti-
mization problem there are well-de…ned shadow prices. In a decentralized market
economy, however, there are a multitude of both agents and prices and no god-
like auctioneer to ensure that the long-term price expectations coincide with the
long-term shadow prices in the social planner’s optimal control problem.
While the Ramsey and the Diamond model are polar cases along the line
segment in Fig. 10.3, less abstract macro models are scattered between these
poles, some being closer to one pole than to the other. Sometimes a given model
open up for alternative regimes, one close to Ramsey’s pole, another close to
Diamond’s. An example is Robert Barro’s model with parental altruism discussed
in Chapter 7. When the bequest motive in the Barro model is operative, the
model coincides with a Ramsey model (in discrete time) as was shown in Chapter
8. But when the bequest motive is not operative, the Barro model coincides
with a Diamond OLG model. Blanchard’s OLG model in continuous time (to
be analyzed in chapters 12, 13, and 15) also belongs to the interior of the line
segment, although closer to Diamond’s pole than to Ramsey’s.

Fig. 10.3 about here (not yet available)

10.7 Literature notes

1. Frank Ramsey (1903-1930) died at the age of 26 but he managed to publish
several path-breaking articles in economics. Ramsey discussed economic issues
with, among others, John Maynard Keynes. In an obituary published in the Eco-
nomic Journal (March 1932) after Ramsey’s death, Keynes described Ramsey’s
article about the optimal savings as “one of the most remarkable contributions to
mathematical economics ever made, both in respect of the intrinsic importance
and di¢ culty of its subject, the power and elegance of the technical methods

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.8. Appendix 423

employed, and the clear purity of illumination with which the writer’s mind is
felt by the reader to play about its subject”.
2. The version of the Ramsey model we have considered is in accordance with
the general tenet of neoclassical preference theory: saving is motivated only by
higher consumption in the future. Extended versions assume that accumulation of
wealth is to some extend an end in itself or perhaps motivated by a desire for social
prestige and economic and political power rather than consumption. In Kurz
(1968b) an extended Ramsey model is studied where wealth is an independent
argument in the instantaneous utility function.
Also Tournemaine and Tsoukis (2008) and Long and Shimomura (2004).
3. The equivalence in the Ramsey model between the decentralized market
equilibrium and the social planner’s solution can be seen as an extension of the
…rst welfare theorem as it is known from elementary textbooks, to the case where
the market structure stretches in…nitely far out in time, and the …nite number of
economic agents (family dynasties) face an in…nite time horizon: in the absence of
externalities etc., the allocation of resources under perfect competition will lead
to a Pareto optimal allocation. The Ramsey model is indeed a special case in that
all households are identical. But the result can be shown in a far more general
setup, cf. Debreu (1954). The result, however, does not hold in overlapping
generations models where an unbounded sequence of new generations enter and
the “interests”of the new households have not been accounted for in advance.
4. Cho and Graham (1996) consider the empirical question whether countries
tend to be above or below their steady state. Based on the Penn World Table
they …nd that on average, countries with a relatively low income per adult are
above their steady state and that countries with a higher income are below.

10.8 Appendix
A. Algebraic analysis of the dynamics around the steady state
To supplement the graphical approach of Section 10.3 with an exact analysis of
the adjustment dynamics of the model, we compute the Jacobian matrix for the
system of di¤erential equations (10.28) - (10.29):
2 3
~ k~ @ k=@~
~ c 0 ~
~ c~) = 4 @ k=@
J(k; 5 = f1 (k) ( + g + n) 1
:
00 ~ 1 0 ~
@ c~=@ k~ @ c~=@~
c f (k)~ c (f (k) + g)

Evaluated in the steady state this reduces to

n (1 )g 1
J(k~ ; c~ ) = 1
f (k~ )~
00
c 0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
424 MODEL: RAMSEY

This matrix has the determinant

1
f 00 (k~ )~
c < 0:

Since the product of the eigenvalues of the matrix equals the determinant, the
eigenvalues are real and opposite in sign.
In standard math terminology a steady-state point in a two dimensional
continuous-time dynamic system is called a saddle point if the associated eigen-
values are opposite in sign.25 For the present case we conclude that the steady
state is a saddle point. This mathematical de…nition of a saddle point is equiv-
alent to that given in the text of Section 10.3. Indeed, with two eigenvalues of
opposite sign, there exists, in a small neighborhood of the steady state, a stable
arm consisting of two saddle paths which point in opposite directions. From the
phase diagram in Fig. 10.2 we know that the stable arm has a positive slope.
At least for k~0 su¢ ciently close to k~ it is thus possible to start out on a saddle
path. Consequently, there is a (unique) value of c~0 such that (k~t ; c~t ) ! (k~ ; c~ ) for
t ! 1. Finally, the dynamic system has exactly one jump variable, c~; and one
predetermined variable, k. ~ It follows that the steady state is (locally) saddle-point
stable.
We claim that for the present model this can be strengthened to global saddle-
point stability. Indeed, for any k~0 > 0, it is possible to start out on the saddle
path. For 0 < k~0 k~ , this is obvious in that the extension of the saddle path
towards the left reaches the y-axis at a non-negative value of c~ . That is to say
that the extension of the saddle path cannot, according to the uniqueness theorem
~
for di¤erential equations, intersect the k-axis for k~ > 0 in that the positive part
~
of the k-axis is a solution of (10.28) - (10.29).26
For k~0 > k~ , our claim can be veri…ed in the following way: suppose, contrary
to our claim, that there exists a k~1 > k~ such that the saddle path does not
intersect that region of the positive quadrant where k~ k~1 . Let k~1 be chosen as
the smallest possible value with this property. The slope, d~ ~ of the saddle
c=dk;
~ ~
path will then have no upper bound when k approaches k1 from the left. Instead
c~ will approach 1 along the saddle path. But then ln c~ will also approach 1
along the saddle path for k~ ! k~1 (k~ < k~1 ): It follows that d ln c~=dk~ = (d~ ~ c,
c=dk)=~
computed along the saddle path, will have no upper bound. Nevertheless, we
25
Note the di¤erence compared to a discrete time system, cf. Appendix D of Chapter 8. In
the discrete time system we have next period’s k~ and c~ on the left-hand side of the dynamic
equations, not the increase in k~ and c~; respectively. Therefore, the criterion for a saddle point
looks di¤erent in discrete time.
26
Because the extension of the saddle path towards the left in Fig. 10.1 can not intersect the
c~-axis at a value of c~ > f (0); it follows that if f (0) = 0; the extension of the saddle path ends
up in the origin.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.8. Appendix 425

have
d ln c~ d ln c~=dt c~=~c 1 ~
(f 0 (k) g)
= = = :
dk~ ~
dk=dt ~
f (k) c~ ( + g + n)k~
k~
When k~ ! k~1 and c~ ! 1 , the numerator in this expression is bounded, while
the denominator will approach 1. Consequently, d ln c~=dk~ will approach zero
from above, as k~ ! k~1 . But this contradicts that d ln c~=dk~ has no upper bound,
when k~ ! k~1 . Thus, the assumption that such a k~1 exists is false and our original
hypothesis holds true.

B. Boundedness of the utility integral

We claimed in Section 10.3 that if the parameter restriction

n > (1 )g (A1)
R1 1
holds, then the utility integral, U0 = 0 c1 e ( n)t dt; is bounded, from above
as well as from below, along the steady-state path, ct = c~ Tt . The proof is as
follows. Recall that > 0 and g 0: For 6= 1;
Z 1 Z 1
1 ( n)t
(1 )U0 = ct e dt = (c0 egt )1 e ( n)t dt
0 0
Z 1
c0
= c0 e[(1 )g ( n)]t dt = , (10.54)
0 n (1 )g
which by (A1) is …nite and positive since c0 > 0. If = 1; so that u(c) = ln c; we
get Z 1
U0 = (ln c0 + gt)e ( n)t dt; (10.55)
0

which is also …nite, in view of (A1) implying n > 0 in this case (the exponential
( n)t
term, e ; declines faster than the linear term gt increases). It follows that
also any path converging to the steady state will entail bounded utility, when
(A1) holds.
On the other hand, suppose that (A1) does not hold, i.e., n (1 )g:
Then by the third equality in (10.54) and c0 > 0 follows that (1 )U0 = 1 if
6= 0: If instead = 1; (10.55) implies U0 = 1:

C. The diverging paths

In Section 10.3 we stated that paths of types II and III in the phase diagram
in Fig. 10.2 can not be equilibria with perfect foresight. Given the expectation
corresponding to any of these paths, every single household will choose to deviate

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
426 MODEL: RAMSEY

from the expected path (i.e., deviate from the expected “average behavior”in the
economy). We will now show this formally.
We …rst consider a path of type III. A path of this type will not be able _to reach
the horizontal axis in Fig. 10.2. It will only converge towards the point (k; ~ 0) for
t ! 1. This claim follows from the uniqueness theorem for di¤erential equations
with continuously di¤erentiable right-hand sides. The uniqueness implies that
two solution curves cannot intersect. And we see from (10.29) that the positive
part of the_x-axis is from a mathematical point of view a solution curve (and
~ 0) is a trivial steady state). This rules out another solution curve
the point (k;
hitting the x-axis. _ _
~ ~
The convergence of k towards k implies limt!1 rt = f (k) 0 ~
< g + n; where
_
the inequality follows from k~ > k~GR . So,
Rt Rt Rt _
0 ~
lim at e 0 (rs n)ds = lim k~t e 0 (rs g n)ds = lim k~t e 0 (f (ks ) g n)ds ~ 1 > 0:
= ke
t!1 t!1 t!1
(10.56)
Hence the transversality condition of the households is violated. Consequently,
the household will choose higher consumption than along this path and can do
so without violating the NPG condition.
Consider now instead a path of type II. We shall …rst show that if the economy
follows such a path, then depletion of all capital occurs in …nite time. Indeed, in
the text it was shown that any path of type II will pass the k~ = 0 locus in Fig.
10.2. Let t0 be the point in time where this occurs. If path II lies above the k~
= 0 locus for all t 0, then we set t0 = 0. For t > t0 , we have

k~t = f (k~t ) c~t ( + g + n)k~t < 0:

By di¤erentiation w.r.t. t we get

k~t = f 0 (k~t )k~t c_t ( + g + n)k~t = [f 0 (k~t ) g n]k~t c_t < 0;

where the inequality comes from k~t < 0 combined with the fact that k~t < k~GR
implies f 0 (k~t ) > f 0 (k~GR ) = g + n: Therefore, there exists a t1 > t0 0
such that Z t1
k~t = k~t +1 k~t dt = 0;
0
t0

as was to be shown. At time t1 ; k~ cannot fall any further and c~t immediately
drops to f (0) and stay there hereafter.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.8. Appendix 427

Yet, this result does not in itself explain why the individual household will
deviate from such a path. The individual household has a negligible impact on
the movement of k~t in society and correctly perceives rt and wt as essentially
independent of its own consumption behavior. Indeed, the economy-wide k~ is
not the household’s concern. What the household cares about is its own …nancial
wealth and budget constraint. In the perspective of the household nothing pre-
vents it from planning a negative …nancial wealth, a; and possibly a continuously
declining …nancial wealth, if only the NPG condition,
Rt
lim at e 0 (rs n)ds
0;
t!1
is satis…ed.
But we can show that paths of type II will violate the NPG condition. The
reasoning is as follows. The household plans to follow the Keynes-Ramsey rule.
Given an expected evolution of rt and wt corresponding to path II, this will
imply a planned gradual transition from positive …nancial wealth to debt. The
transition to positive net debt, d~t a
~t at =Tt > 0, takes place at time t1
de…ned above.
The continued growth in the debt will meanwhile be so fast that the NPG
condition is violated. To see this, note that the NPG condition implies the re-
quirement Rt
lim d~t e 0 (rs g n)ds 0; (NPG)
t!1

that is, the productivity-corrected debt, d~t , is allowed to grow in the long run
only at a rate less than the growth-corrected real interest rate. For t > t1 we get
from the accounting equation a_ t = (rt n)at + wt ct that

d~t = (rt g n)d~t + c~t w~t > 0;

where d~t > 0; rt > + g > g + n; and where c~t grows exponentially according
to the Keynes-Ramsey rule, while w~t is non-increasing in that k~t does not grow.
This implies
d~t
lim lim (rt g n);
t!1 d~t t!1

which is in con‡ict with (NPG).

Consequently, the household will choose a lower consumption path and thus
deviate from the reference path considered. Every household will do this and the
evolution of rt and wt corresponding to path II is thus not an equilibrium with
perfect foresight.
The conclusion is that all individual households understand that the only
evolution which can be expected rationally is the one corresponding to the saddle
path.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
428 MODEL: RAMSEY

D. A constant saving rate as a special case

As we noted in Section 10.4, Solow’s growth model can be seen as a special case of
the Ramsey model. Indeed, a constant saving rate may, under certain conditions,
emerge as an endogenous result in the Ramsey model.
Let the rate of saving, (Yt Ct )=Yt , be st . We have generally

c~t = (1 st )f (k~t ); and so (10.57)

k~t = f (k~t ) c~t ( + g + n)k~t = st f (k~t ) ( + g + n)k~t : (10.58)

In the Solow model the rate of saving is a constant, s, and we then get, by
di¤erentiating with respect to t in (10.57) and using (10.58),

c~t ( + g + n)k~t
= f 0 (k~t )[s ]: (10.59)
c~t f (k~t )

By maximization of discounted utility in the Ramsey model, given a rate of

time preference and an elasticity of marginal utility , we get in equilibrium

c~t 1
= (f 0 (k~t ) g): (10.60)
c~t
There will not generally exist a constant, s; such that the right-hand sides of
(10.59) and (10.60), respectively, are the same for varying k~ (that is, outside
steady state). But Kurz (1968a) showed the following:
CLAIM Let ; g; n; ; and be given. If the elasticity of marginal utility is
greater than 1 and the production function is y~ = Ak~ with 2 (1= ; 1), then a
Ramsey model with = ( + g + n) g will generate a constant saving
rate s = 1= : Thereby the same resource allocation and transitional dynamics
arise as in the corresponding Solow model with s = 1= .
Proof. Let 1= < < 1 and f (k) ~ = Ak~ : Then f 0 (k)
~ = A k~ 1
: The right-hand-
side of the Solow equation, (10.59), becomes

( + g + n)k~t
A k~ 1
[s ] = sA k~ 1
( + g + n): (10.61)
Ak~
The right-hand-side of the Ramsey equation, (10.60), becomes

1 + + g
A k~ 1
:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

10.8. Appendix 429

By inserting = ( + g + n) g; this becomes

1 + ( + g + n) g+ g
A k~ 1

1
= A k~ 1
( + g + n): (10.62)

For the chosen we have = ( + g + n) g > n + (1 )g; because

> 1 and + g + n > 0: Thus, n > (1 )g and existence of equilibrium in
the Ramsey model with this is ensured. We can now make (10.61) and (10.62)
the same by inserting s = 1= : This also ensures that the two models require the
same k~ to obtain a constant c~ > 0: With this k~ , the requirement k~t = 0 gives
the same steady-state value of c~ in both models, in view of (10.58). It follows
that (k~t ; c~t ) is the same in the two models for all t 0:
On the other hand, maintaining y~ = Ak~ , but allowing 6= ( + g + n)
g; so that 6= 1=s ; then s0 (k) ~ 6= 0; i.e., the Ramsey model does not
generate a constant saving rate except in steady state. De…ning s as in (10.40)
and ( + )= [ ( + g + n) g], we have: When ( + g + n) > g (which
seems likely empirically), it holds that if Q 1=s (i.e., if Q ); then s0 (k) ~
Q 0; respectively; if instead ( + g + n) g; then < 1=s and s0 (k)~ < 0;
unconditionally: These results follow by considering the slope of the saddle path
in a phase diagram in the (k; ~ c~=f (k))
~ plane and using that s(k)~ = 1 c~=f (k);
~
~
cf. Exercise 10.?? The intuition is that when k is rising over time (i.e., society is
becoming wealthier), then, when the desire for consumption smoothing is “high”
( “high”), the prospect of high consumption in the future is partly taken out as
high consumption already today, implying that saving is initially low, but rising
over time until it eventually settles down in the steady state. But if the desire
for consumption smoothing is “low” ( “low”), saving will initially be high and
then gradually fall in the process towards the steady state. The case where k~ is
falling over time gives symmetric results.

E. The social planner’s solution

In the text of Section 10.5 we postponed some of the more technical details.
First, by (A2), the existence of the steady state, E, and the saddle path in
Fig. 10.2
Rt 0
is ensured. Solving the linear di¤erential equation (10.46) gives t
~
= 0 e 0 (ks ) ^ g)ds : Substituting this into the transversality condition (10.47)
(f

gives Rt 0
~
lim k~t e 0 (f (ks ) g n)ds = 0; (10.63)
t!1
^
where we have eliminated the unimportant positive factor 0 = c0 T0 :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 10. THE BASIC REPRESENTATIVE AGENT
430 MODEL: RAMSEY

This condition is essentially the same as the transversality condition (10.30)

for the market economy and holds in the steady state, given the parameter re-
striction ^ n > (1 ^)g; which is analogue to (A1). Thus, (10.63) also holds
along the saddle path. Since we must have k~ 0 for all t 0; (10.63) has the
form required by Mangasarian’s su¢ ciency theorem. If we can show that the
Hamiltonian is jointly concave in (k;~ c) for all t 0; then the saddle path is a
solution to the social planner’s problem. And if we can show strict concavity, the
saddle path is the unique solution. We have:
@H ~ @H ^
= (f 0 (k) ( + g + n)); =c ;
@ k~ @c T
@ 2H 2
~ < 0 (by = c ^ T > 0); @ H = ^c ^ 1
= f 00 (k) < 0;
@ k~2 @c2
2
@ H
= 0:
~
@ k@c
So the leading principal minors of the Hessian matrix of H are
2
@ 2H @ 2H @ 2H @ 2H
D1 = > 0; D2 = > 0:
@ k~2 @ k~2 @c2 ~
@ k@c

Hence, H is strictly concave in (k;~ c) and the saddle path is the unique optimal
solution.
It also follows that the transversality condition (10.47) is a necessary optimal-
ity condition when the parameter restriction ^ n > (1 ^)g holds. Note that
we have had to derive this conclusion in a di¤erent way than when solving the
household’s consumption/saving problem in Section 10.2. There we could appeal
to a link between the No-Ponzi-Game condition (with strict equality) and the
transversality condition to verify necessity of the transversality condition. But
that proposition does not cover the social planner’s problem where there is no
NPG condition.
As to the diverging paths in Fig. 10.2, note that paths of type II (those paths
which, as shown in Appendix C, in …nite time deplete all capital) can not be
optimal, in spite of the temporarily high consumption level. This follows from
the fact that the saddle path is the unique solution. Finally, paths of type III
in Fig. 10.2 behave as in (10.56) and thus violate the transversality condition
(10.47), as claimed in the text.

10.9 Exercises

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Advanced Macroeconomics Short Note 1

01.10.2015. Christian Groth

A glimpse of theory of
the “level of interest rates”

This short note provides a brief sketch of what macroeconomics says about the general
level around which rates of return ‡uctuate. We also give a “broad”summary of di¤erent
circumstances that give rise to di¤erences in rates of return on di¤erent assets.

In non-monetary models without uncertainty there is in equilibrium only one rate of

return, r: If in addition there is a) perfect competition in all markets, b) the consumption
good is physically indistinguishable from the capital good, and c) there are no capital
adjustment costs, as in simple neoclassical models (like the Diamond OLG model and
the Ramsey model), then the equilibrium real interest rate is at any time equal to the
current net marginal productivity of capital evaluated at full employment (r = @Y =@K
in standard notation). Moreover, under conditions ensuring “well-behavedness” of these
models, they predict that in the absence of disturbances, the technology-corrected capital-
labor ratio, and thereby the marginal productivity of capital, adjusts over time to some
long-run level (on which more below).

Di¤erent rates of return In simple neoclassical models with perfect competition and
no uncertainty, the equilibrium short-term real interest rate is at any time equal to the
net marginal productivity of capital (r = @Y =@K ): In turn the marginal productivity
of capital adjusts over time, via changes in the capital intensity, to some long-run level
(on this more below). As we saw in Chapter 14, existence of convex capital installation
costs loosens the link between r and @Y =@K. The convex adjustment costs create a
wedge between the price of investment goods and the market value of the marginal unit
of installed capital. Besides the marginal productivity of capital, the possible capital gain
in the market value of installed capital as well as the e¤ect of the marginal unit of installed
capital on future installation costs enter as co-determinants of the current rate of return
on capital.

1
 Arithmetic Standard Geometric
average deviation average
----------------- Percent -----------------
Nominal values
Small Company Stocks 17,3 33,2 12,5
Large Company Stocks 12,7 20,2 10,7
Long-Term Corporate Bonds 6,1 8,6 5,8
Long-Term Government Bonds 5,7 9,4 5,3
Intermediate-Term Government Bonds 5,5 5,7 5,3
U.S. Treasury Bills 3,9 3,2 3,8
Cash 0,0 0,0 0,0
Inflation rate 3,1 4,4 3,1
Real values
Small Company Stocks 13,8 32,6 9,2
Large Company Stocks 9,4 20,4 7,4
Long-Term Corporate Bonds 3,1 9,9 2,6
Long-Term Government Bonds 2,7 10,6 2,2
Intermediate-Term Government Bonds 2,5 7,0 2,2
U.S. Treasury Bills 0,8 4,1 0,7
Cash -2,9 4,2 -3,0

Table 1: Average annual rates of return on a range of U.S. asset portfolios, 1926-2001.
Source: Stocks, Bonds, Bills, and In‡ation: Yearbook 2002, Valuation Edition. Ibbotson
Associates, Inc.

When imperfect competition in the output markets rules, prices are typically set as a
mark-up on marginal cost. This implies a wedge between the net marginal productivity
of capital and capital costs. And when uncertainty and limited opportunities for risk
diversi…cation are added to the model, a wide spectrum of expected rates of return on
di¤erent …nancial assets and expected marginal productivities of capital in di¤erent pro-
duction sectors arise, depending on the risk pro…les of the di¤erent assets and production
sectors. On top of this comes the presence of taxation which may complicate the picture
because of di¤erent tax rates on di¤erent asset returns.

Nominal and real average annual rates of return on a range of U.S. asset portfolios for
the period 1926–2001 are reported in Table 1. By a portfolio of n assets, i = 1; 2; : : : ; n
is meant a “basket”, (v1 ; v2 ; : : : ; vn ); of the n assets in value terms, that is, vi = pi xi is
the value of the investment in asset i; the price of which is denoted pi and the quantity
P
of which is denoted xi . The total investment in the basket is V = ni=1 vi : If Ri denotes
the gross rate of return on asset i; the overall gross rate of return on the portfolio is
Pn
vi Ri X
n
R= i = wi Ri ;
V i=1

2
where wi vi =V is the weight or fraction of asset i in the portfolio. De…ning Ri 1 + ri ;
where ri is the net rate of return on asset i; the net rate of return on the portfolio can be
written
X
n X
n X
n X
n
r=R 1= wi (1 + ri ) 1= wi + wi ri 1= wi ri :
i=1 i=1 i=1 i=1

The net rate of return is often just called “the rate of return”.

In Table 1 we see that the portfolio consisting of small company stocks throughout the
period 1926-2001 had an average annual real rate of return of 13.8 per cent (the arithmetic
average) or 9.2 per cent (the geometric average). This is more than the annual rate of
return of any of the other considered portfolios. Small company stocks are also seen to
be the most volatile. The standard deviation of the annual real rate of return of the
portfolio of small company stocks is almost eight times higher than that of the portfolio
of U.S. Treasury bills (government zero coupon bonds with 30 days to maturity), with
an average annual real return of only 0.8 per cent (arithmetic average) or 0.7 per cent
(geometric average) throughout the period. The displayed positive relation between high
returns and high volatility is not without exceptions, however. The portfolio of long-term
corporate bonds has performed better than the portfolio of long-term government bonds,
although they have been slightly less volatile as here measured. The data is historical and
expectations are not always met. Moreover, risk depends signi…cantly on the covariance
of asset returns within the total set of assets and speci…cally on the correlation of asset
returns with the business cycle, a feature that can not be read o¤ from Table 1. Share
prices, for instance, are very sensitive to business cycle ‡uctuations.

The need for means of payment money is a further complicating factor. That is,
besides dissimilarities in risk and expected return across di¤erent assets, also dissimilar-
ities in their degree of liquidity are important, not least in times of …nancial crisis. The
expected real rate of return on cash holding is minus the expected rate of in‡ation and
is therefore negative in an economy with in‡ation, cf. the last row in Table 1. When
agents nevertheless hold cash in their portfolios, it is because the low rate of return is
compensated by the liquidity services of money. In the Sidrauski model of Chapter 17 this
is modeled in a simple way, albeit ad hoc, by including real money holdings directly as an
argument in the utility function. Another dimension along which the presence of money
interferes with returns is through in‡ation. Real assets, like physical capital, land, houses,
etc. are better protected against ‡uctuating in‡ation than are nominally denominated
bonds (and money of course).

3
 Without claiming too much we can say that investors facing such a spectrum of rates
of return choose a composition of assets so as to balance the need for liquidity, the wish
for a high expected return, and the wish for low risk. Finance theory teaches us that
adjusted for di¤erences in risk and liquidity, asset returns tend to be the same. This
raises the question: at what level? This is where macroeconomics as an empirically
oriented theory about the economy as a whole comes in.

Macroeconomic theory of the “average rate of return” The point of departure

is that market forces by and large may be thought of as anchoring the rate of return of
an average portfolio of interest-bearing assets to the net marginal productivity of capital
in an aggregate production function, assuming a closed economy. Some popular phrases
are:

the net marginal productivity of capital acts as a centre of gravitation for the spec-
trum of asset returns; and

movements of the rates of return are in the long run held in check by the net marginal
productivity of capital.

Though such phrases seem to convey the right ‡avour, in themselves they are not
very informative. The net marginal productivity of capital is not a given, but an endoge-
nous variable which, via changes in the capital intensity, adjusts through time to more
fundamental factors in the economy.

The di¤erent macroeconomic models we have encountered in previous chapters bring

to mind di¤erent presumptions about what these fundamental factors are.

1. Solow’s growth model The Solow growth model leads to the fundamental di¤er-
ential equation (standard notation)

k~t = sf (k~t ) ( + g + n)k~t ;

where s is an exogenous and constant aggregate saving-income ratio, 0 < s < 1: In steady
state

r = f 0 (k~ ) ; (1)

4
where k~ is the unique steady state value of the (e¤ective) capital intensity, k;
~ satisfying

sf (k~ ) = ( + g + n)k~ : (2)

In society there is a debate and a concern that changed demography and less growth
in the source of new technical ideas, i.e., the stock of educated human beings, will in the
future result in lower n and lower g; respectively, making …nancing social security more
di¢ cult: On the basis ~
h of the Solow model i 1we …nd by implicit di¤erentiation in (2) @ k =@n
= @ k~ =@g = k~ + g + n sf 0 (k~ ) ; which is negative since sf 0 (k~ ) < sf (k~ )=k~
= + g + n: Hence, by (1),
@r @r @r @ k~ k~
= = = f 00 (k~ ) > 0;
@n @g @ k~ @n + g + n sf 0 (k~ )
since f 00 (k~ ) < 0: It follows that

n # or g #) r # : (3)

A limitation of this theory is of course the exogeneity of the saving-income ratio, which
is a key co-determinant of k~ ; hence of r : The next models are examples of di¤erent ways
of integrating a theory of saving into the story about the long-run rate of return.

2. The Diamond OLG model In the Diamond OLG model, based on a life-cycle
theory of saving, we again arrive at the formula r = f 0 (k~ ) . Like in the Solow model,
the long-run rate of return thus depends on the aggregate production function and on k~ :
But now there is a logically complete theory about how k~ is determined. In the Diamond
model k~ depends in a complicated way on the lifetime utility function and the aggregate
production function. The steady state of a well-behaved Diamond model will nevertheless
have the same qualitative property as indicated in (3).

3. The Ramsey model Like the Solow and Diamond models, the Ramsey model
implies that rt = f 0 (k~t ) for all t: But unlike in the Solow and Diamond models, the net
marginal productivity of capital now converges in the long run to a speci…c value given
by the modi…ed golden rule formula. In a continuous time framework this formula says:

r = + g; (4)

where the new parameter, ; is the (absolute) elasticity of marginal utility of consumption.
Because the Ramsey model is a representative agent model, the Keynes-Ramsey rule holds

5
not only at the individual level, but also at the aggregate level. This is what gives rise to
this simple formula for r .

Here there is no role for n; only for g: On the other hand, there is an alternative
speci…cation of the Ramsey model, namely the “average utilitarianism” speci…cation. In
this version of the model, we get r = f 0 (k~ ) = + n + g; so that not only a lower
g; but also a lower n implies lower r :

Also the Sidrauski model, i.e., the monetary Ramsey model of Chapter 17, results in
the modi…ed golden rule formula.1

4. Blanchard’s OLG model A continuous time OLG model with emphasis on life-
cycle aspects is Blanchard’s model, Blanchard (1985). In that model the net marginal
productivity of capital adjusts to a value where, in addition to the production function,
technology growth, and preference parameters, also demographic parameters, like birth
rate, death rate, and retirement rate, play a role. One of the results is that when = 1;

+g <r < + g + b;

where is the retirement rate (re‡ecting how early in life the “average” person retire
from the labor market) and b is the (crude) birth rate. The population growth rate is the
di¤erence between the birth rate, b; and the (crude) mortality rate, m; so that n = b m:
The qualitative property indicated in (3) becomes conditional. It still holds if the fall in
n re‡ects a lower b; but not necessarily if it re‡ects a higher m.

5. What if technological change is embodied? The models in the list above assume
a neoclassical aggregate production function with CRS and disembodied Harrod-neutral
technological progress, that is,

Yt = F (Kt ; Tt Lt ) Tt Lt f (k~t ); f 0 > 0; f 00 < 0: (5)

This amounts to assuming that new technical knowledge advances the combined pro-
ductivity of capital and labor independently of whether the workers operate old or new
machines.

In contrast, we say that technological change is embodied if taking advantage of new

technical knowledge requires construction of new investment goods. The newest technol-
ogy is incorporated in the design of newly produced equipment; and this equipment will
1
See Chapter 10, Section 10.5.

6
not participate in subsequent technological progress. Both intuition and empirics suggest
that most technological progress is of this form. Indeed, Greenwood et al. (1997) estimate
for the U.S. 1950-1990 that embodied technological change explains 60% of the growth in
output per man hour.

So a theory of the rate of return should take this into account. Fortunately, this can
be done with only minor modi…cations. We assume that the link between investment and
capital accumulation takes the form

K_ t = Qt It Kt ; (6)

where It is gross investment (I = Y C) and Qt measures the “quality” (e¢ ciency) of

newly produced investment goods. Suppose for instance that

Qt = Q0 e t ; > 0:

Then, even if no technological change directly appears in the production function, that
is, even if (5) is replaced by

Yt = F (Kt ; Lt ) = Kt L1t ; 0< < 1;

the economy will still experience a rising standard of living.2 A given level of gross
investment will give rise to greater and greater additions to the capital stock K; measured
in e¢ ciency units. Since at time t; Qt capital goods can be produced at the same cost as
one consumption good, the price, pt ; of capital goods in terms of the consumption good
must in competitive equilibrium equal the inverse of Qt ; that is, pt = 1=Qt : In this way
embodied technological progress results in a steady decline in the relative price of capital
equipment.

This prediction is con…rmed by the data. Greenwood et al. (1997) …nd for the U.S.
that the relative price of capital equipment has been declining at an average rate of 0:03
per year in the period 1950-1990, a trend that has seemingly been forti…ed in the wake of
the computer revolution.

Along a balanced growth path the constant growth rate of K will now exceed that
of Y; and Y =K thus be falling. The output-capital ratio in value terms, Y =(pK); will be
constant, however. Embedding these features in a Ramsey-style framework, we …nd the
2
We specify F to be Cobb-Douglas, because otherwise a model with embodied technical progress in
the form (6) will not be able to generate balanced growth and comply with Kaldor’s stylized facts.

7
long-run rate of return to be3
r = + :
1
This is of the same form as (4) since growth in output per unit of labor in steady state is
exactly g = =(1 ):

Adding uncertainty and risk of bankruptcy Although absent from many simple
macroeconomic models, uncertainty and risk of bankruptcy are signi…cant features of
reality. Bankruptcy risk may lead to a con‡ict of interest between share owners and
managers. Managers may want less debt and more equity than the share owners because
bankruptcy can be very costly to managers who loose a well-paid job and a promising
carrier. So managers are unwilling to …nance all new capital investment by new debt in
spite of the associated lower capital cost (there is generally a lower rate of return on debt
than on equity). In this way the excess of the rate of return on equity over that on debt,
the equity premium, is sustained.

A rough behavioral theory of the equity premium goes as follows.4 Firm managers
prefer a payout structure with a fraction, sf ; going to equity and the remaining fraction,
1 sf ; to debt (corporate bonds). That is, out of each unit of expected operating pro…t,
managers are unwilling to commit more than 1 sf to bond owners. This is to reduce the
risk of a failing payment ability in case of a bad market outcome. And those who …nance
…rms by loans de…nitely also want debtor …rms to have some equity at stake.

We let households’ preferred portfolio consist of a fraction sh in equities and the

remainder, 1 sh , in bonds. In view of households’risk aversion and memory of historical
stock market crashes, it is plausible to assume that sh < sf .

As a crude adaptation of for instance the Blanchard OLG model to these features, we
interpret the model’s r as an average rate of return across …rms. Let time be discrete
and let aggregate …nancial wealth be A = pK; where p is the price of capital equipment
in terms of consumption goods. In the frameworks 1 to 4 above we have p 1; but in
framework 5 the relative price p equals 1=Q and is falling over time. Anyway, given A
at time t; the aggregate gross return or payout is (1 + r )A. Out of this, (1 + r )Asf
constitutes the gross return to the equity owners and (1 + r )A(1 sf ) the gross return
3
See Exercise 18.??
4
The following is inspired by Baker, DeLong, and Krugman (2005). These authors discuss the implied
predictions for U.S. rates of return in the future and draw implications of relevance for the debate on
social security reform.

8
to the bond owners. Let re denote the rate of return on equity and rb the rate of return
on bonds.

To …nd re and rb we have

(1 + re )Ash = (1 + r )Asf ;
(1 + rb )A(1 sh ) = (1 + r )A(1 sf ):

Thus,
sf
1 + re = (1 + r ) >1+r ;
sh
1 sf
1 + rb = (1 + r ) <1+r :
1 sh
We may de…ne the equity premium, ; by 1 + (1 + re )=(1 + rb ): Then

sf (1 sh )
= 1 > 0:
sh (1 sf )

Of course these formulas have their limitations. The key variables sf and sh will
depend on a lot of economic circumstances and should be endogenous in an elaborate
model. Yet, the formulas may be helpful as a way of organizing one’s thoughts about
rates of return in a world with asymmetric information and risk of bankruptcy.

There is evidence that in the last decades of the twentieth century the equity premium
had become lower than in the long aftermath of the Great Depression in the 1930s.5 A
likely explanation is that sh had gone up, along with rising con…dence. The computer
and the World Wide Web have made it much easier for individuals to invests in stocks of
shares. On the other hand, the recent …nancial and economic crisis, known as the Great
Recession 2007- , and the associated rise in mistrust seems to have halted and possibly
reversed this tendency for some time (source ??).

5
Blanchard (2003, p. 333).

9
Chapter 11

Applications of the Ramsey

model

The Ramsey representative agent framework has, rightly or wrongly, been a work-
horse for the study of many macroeconomic issues. Among these are public …-
nance themes and themes relating to endogenous productivity growth. In this
chapter we consider issues within these two themes. Section 11.1 deals with a
market economy with a public sector. The focus is on general equilibrium e¤ects
of government spending and taxation, including e¤ects of shifts in …scal policy,
both anticipated and unanticipated shifts. In Section 11.2 we set up and analyze
a model of technology growth based on learning by investing. The analysis leads
to a characterization of a “…rst-best policy”.

11.1 Market economy with a public sector

In this section we extend the Ramsey model of a competitive market economy by
adding a government sector that spends on goods and services, makes transfers
to the private sector, and levies taxes.
Subsection 11.1.1 considers the e¤ect of government spending on goods and
services, assuming a balanced budget where all taxes are lump sum. The issue
what is really meant by one-o¤ shocks in a perfect foresight model is addressed,
including how to model the e¤ects of such shocks. In subsections 11.1.2 and
11.1.3 we consider income taxation and how the economy responds to the arrival
of new information about future …scal policy. Finally, subsection 11.1.4 introduces
…nancing by temporary budget de…cits. In view of the Ramsey model being a
representative agent model, it is not surprising that Ricardian equivalence will
hold in the model.

431
432 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

11.1.1 Public consumption …nanced by lump-sum taxes

The representative household (or family dynasty) has Lt = L0 ent members each
of which supplies one unit of labor inelastically per time unit, n 0. The
household’s preferences can be represented by a time separable utility function
Z 1
u~(ct ; Gt )Lt e t dt;
0

where ct Ct =Lt is consumption per family member and Gt is public consumption

in the form of a service delivered by the government, while is the rate of time
preference. We assume, for simplicity, that the instantaneous utility function is
additive: u~(c; G) = u(c) + v(G); where u0 > 0; u00 < 0; i.e., there is positive but
diminishing marginal utility of private consumption; the properties of the utility
function v are immaterial for the questions to be studied (but hopefully v 0 > 0).
The public service might consist in making a non-rival good, say “law and order”
or TV-transmitted theatre, available for the households free of charge.
Throughout this section the government budget is always balanced. In the
present subsection the government spending, Gt , is …nanced by a per capita lump-
sum tax, t , so that
t Lt = Gt : (11.1)
To allow for balanced growth under technological progress we assume that u
is a CRRA function. Thus, the criterion function of the representative household
can be written Z 1
c1t
U0 = + v(Gt ) e ( n)t dt; (11.2)
0 1
where > 0 is the constant (absolute) elasticity of marginal utility of private
consumption.
As usual, let the real interest rate and the real wage be denoted rt and wt ;
respectively. The household’s dynamic book-keeping equation reads
a_ t = (rt n)at + wt t ct ; a0 given, (11.3)
where at is per capita …nancial wealth. The …nancial wealth is held in claims of
a form similar to a variable-rate deposit in a bank. Hence, at any point in time
at is historically determined and independent of the current and future interest
rates. The No-Ponzi-Game condition (solvency condition) is
Rt
lim at e 0 (rs n)ds
0: (NPG)
t!1

We see from (11.2) that leisure does not enter the instantaneous utility func-
tion. So per capita labor supply is exogenous. We …x its value to be one unit of
labor per time unit, as is indicated by (11.3).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 433

In view of the additive instantaneous utility function in (11.2), marginal utility

of private consumption is not a¤ected by Gt . The Keynes-Ramsey rule resulting
from the household’s optimization will therefore be as if there were no government
sector:
c_t 1
= (rt ):
ct
The transversality condition of the household is that (NPG) holds with strict
equality, i.e.,
Rt
lim at e 0 (rs n)ds = 0:
t!1

GDP is produced through an aggregate neoclassical production function with

CRS:
Yt = F (Ktd ; Tt Ldt );
where Ktd and Ldt are inputs of capital and labor, respectively, and Tt is the
technology level, assumed to grow at an exogenous and constant rate g 0:
For simplicity we assume that F satis…es the Inada conditions. It is further
assumed that in the production of Gt the same technology (production function)
is applied as in the production of the other components of GDP; thereby the
same unit production costs are involved. A possible role of Gt for productivity is
ignored (so we should not interpret Gt as related to such things as infrastructure,
health, education, or research).
All capital in the economy is assumed to belong to the private sector. The
economy is closed. In accordance with the standard Ramsey model, there is
perfect competition in all markets. Hence there is market clearing so that Ktd =
Kt and Ldt = Lt for all t:

General equilibrium and dynamics

The increase in the capital stock, K; per time unit equals aggregate gross saving:

K_ t = Yt Ct Gt Kt = F (Kt ; Tt Lt ) ct Lt Gt Kt ; K0 > 0 given: (11.4)

We assume Gt is proportional to the work force measured in e¢ ciency units, that

is
Gt = ~ Tt Lt ; (11.5)
where the size of ~ 0 is decided by the government. The balanced budget
(11.1) now implies that the per capita lump-sum tax grows at the same rate as
technology:
gt gt
t = Gt =Lt = ~ Tt = ~ T0 e = 0 e : (11.6)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

434 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

De…ning k~t Kt =(Tt Lt ) kt =Tt and c~t Ct =(Tt Lt ) ct =Tt ; the dynamic
aggregate resource constraint (11.4) can be written

k~t = f (k~t ) c~t ~ ( + g + n)k~t ; k~0 > 0 given, (11.7)

where f is the production function in intensive form, f 0 > 0; f 00 < 0: As F satis…es

the Inada conditions, we have

f (0) = 0; ~ = 1;
lim f 0 (k) ~ = 0:
lim f 0 (k)
~
k!0 ~
k!1

As usual, by the golden-rule capital intensity, k~GR ; we mean that capital

intensity which maximizes sustainable consumption per unit of e¤ective labor,
c~ + ~ : By setting the left-hand side of (11.7) to zero, eliminating the time indices
on the right-hand side, and rearranging, we get c~ + ~ = f (k) ~ ( + g + n)k~
~ In view of the Inada conditions, the problem max~ c(k)
c(k): ~ has a unique
k
solution, k > 0; characterized by the condition f (k) = + g + n: This k~ is, by
~ 0 ~

de…nition, k~GR :
In general equilibrium the real interest rate, rt ; equals f 0 (k~t ) : Expressed
in terms of c~; the Keynes-Ramsey rule thus becomes
1h 0 ~ i
c~t = f (kt ) g c~t : (11.8)

Moreover, we have at = kt k~t Tt = k~t T0 egt ; and so the transversality condition

of the representative household can be written
Rt
lim k~t e 0 (f 0 (k~s ) n g )ds
= 0: (11.9)
t!1

The phase diagram of the dynamic system (11.7) - (11.8) is shown in Fig.
11.1 where, to begin with, the k~ = 0 locus is represented by the stippled inverse
U curve. Apart from a vertical downward shift of the k~ = 0 locus, when we
have ~ > 0 instead of ~ = 0; the phase diagram is similar to that of the Ramsey
model without government. Although the per capita lump-sum tax is not visible
in the reduced form of the model consisting of (11.7), (11.8), and (11.9), it is
indirectly present because it ensures that for all t 0; the c~t and k~t appearing
in (11.7) represent exactly the consumption demand and net saving coming from
the households’intertemporal budget constraint (which depends on the lump-sum
tax, cf. (11.11). Otherwise, equilibrium would not be maintained.
We assume ~ is of “moderate size” compared to the productive capacity of
the economy so as to not rule out the existence of a steady state. Moreover, to

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 435

Figure 11.1: Phase portrait of an unanticipated permanent increase in government

spending from ~ to ~ 0 > ~ .

guarantee bounded discounted utility and existence of general equilibrium, we

impose the parameter restriction
n > (1 )g: (A1)

How to model e¤ects of unanticipated policy shifts

In a perfect foresight model, as the present one, agents’expectations and actions
never incorporate that unanticipated events, “shocks”, may arrive. That is, if
a shock occurs in historical time, it must be treated as a complete surprise, a
one-o¤ shock not expected to be replicated in any sense.
Suppose that up until time t0 > 0 government spending maintains the given
ratio Gt =(Tt Lt ) = ~ : Suppose further that before time t0 ; the households expected
this state of a¤airs to continue forever. But, unexpectedly, at time t0 there is a
shift to a higher constant spending ratio, ~ 0 ; which is maintained for a long time.
We assume that the upward shift in public spending goes hand in hand with
higher lump-sum taxes so as to maintain a balanced budget. Thereby the after-
tax human wealth of the household is at time t0 immediately reduced. As the
households are now less wealthy, private consumption immediately drops.
Mathematically, the time path of ct will therefore have a discontinuity at
t = t0 : To …x ideas, we will generally consider control variables, e.g., consumption,

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

436 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

to be right-continuous functions of time in such cases. This means that ct0 =

limt!t+0 ct : Likewise, at such points of discontinuity of the control variable the
“time derivative” of the state variable a in (11.3) is generally not well-de…ned
without an amendment. In line with the right-continuity of the control variable,
we de…ne the time derivative of a state variable at a point of discontinuity of the
control variable as the right-hand time derivative, i.e., a_ t0 = limt!t+0 (at at0 )=(t
t0 ):1 We say that the control variable has a jump at time t0 ; we call the point
where this jump occurs a switch point, and we say that the state variable, which
remains a continuous function of t, has a kink at time t0 :
In line with this, control variables are called jump variables or forward-looking
variables. The latter name comes from the notion that a decision variable can
immediately shift to another value if new information arrives. In contrast, a state
variable is said to be pre-determined because its value is an outcome of the past
and it cannot jump.

An unanticipated permanent shift in government spending Returning

to our speci…c example, suppose that the economy has been in steady state for
t < t0 : Then, unexpectedly, the new spending policy ~ 0 > ~ is introduced, followed
by an increase in taxation so as to maintain a balanced budget. Let the households
rightly expect this new policy to be maintained forever. As a consequence, the k~
= 0 locus in Fig. 11.1 is shifted downwards while the c~ = 0 locus remains where
it is. It follows that k~ stays unchanged at its old steady-state level, k~ ; while c~
jumps down to the new steady-state value, c~ 0 : There is immediate crowding out
of private consumption to the exact extent of the rise in public consumption.2
To understand the mechanism, note that Per capita consumption of the house-
hold is
ct = t (at + ht ); (11.10)
where ht is the after-tax human wealth per family member and is given by
Z 1 Rs
ht = (ws t (rz n)dz ds; (11.11)
s )e
t

and t is the propensity to consume out of wealth,

1
t =R Rs (1 )rz
; (11.12)
1 ( +n)dz
t
e t ds
1
While these conventions help to …x ideas, they are mathematically inconsequential. Indeed,
the value of the consumption intensity at each isolated point of discontinuity will a¤ect neither
the utility integral of the household nor the value of the state variable, a:
2
The conclusion is modi…ed, of course, if Gt encompasses public investments and if these
have an impact on the productivity of the private sector.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 437

as derived in the previous chapter. The upward shift in public spending is accom-
panied by higher lump-sum taxes, 0t = ~ 0 Lt ; forever, implying that ht is reduced,
which in turn reduces consumption.
Had the unanticipated shift in public spending been downward, say from ~ 0 to
~ ; the e¤ect would be an upward jump in consumption but no change in k; ~ that
is, a jump E’to E in Fig. 11.1.
Many kinds of disturbances of a steady state will result in a gradual adjust-
ment process, either to a new steady state or back to the original steady state. It
is otherwise in this example where there is an immediate jump to a new steady
state.

11.1.2 Income taxation

We now replace the assumed lump-sum taxation by income taxation of di¤erent
kinds. In addition, we introduce lump-sum income transfers to the households.

Taxation of labor income

Consider a tax on wage income at the constant rate w ; 0 < w < 1. Since labor
supply is exogenous, it is una¤ected by the wage income tax. While (11.7) is
still the dynamic resource constraint of the economy, the household’s dynamic
book-keeping equation now reads
a_ t = (rt n)at + (1 w )wt + xt ct ; a0 given,
where xt is the per capita lump-sum transfers at time t: Maintaining the assump-
tion of a balanced budget, the tax revenue at every t exactly covers government
spending on goods and services and the lump-sum transfers to the private sector.
This means that
w wt Lt = Gt + xt Lt for all t 0:
As Gt and w are given, the interpretation is that for all t 0; transfers adjust
so as to balance the budget. This requires that xt = w wt Gt =Lt = w wt ~ Tt ;
for all t 0; if xt need be negative to satisfy this equation, so be it. Then xt
would act as a positive lump-sum tax.
Disposable income at time t is
(1 w )wt + xt = w t ~ Tt ;
and human wealth at time t per member of the representative household is thus
Z 1 Rs
Z 1 Rs
(r n)dz t (rz n)dz ds:
ht = [(1 w )w s + x s ] e t z
ds = (w s ~ T s )e
t t
(11.13)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

438 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Owing to the given ~ , a shift in the value of w is immediately compensated

by an adjustment of the path of transfers in the same direction so as to maintain a
balanced budget. Neither disposable income nor ht is a¤ected. So the shift in w
leaves the determinants of per capita consumption una¤ected. As also disposable
income is una¤ected, it follows that private saving is una¤ected. This is why
w nowhere enters the model in its reduced form, consisting of (11.7), (11.8),
and (11.9). The phase diagram for the economy with labor income taxation is
completely identical to that in Fig. 11.1 where there is no tax on labor income.
The evolution of the economy is independent of the size of w (if the model were
extended with endogenous labor supply, the result would generally be di¤erent).
The intuitive explanation is that the three conditions: (a) inelastic labor supply,
(b) a balanced budget,3 and (c) a given path for Gt , imply that a labor income tax
a¤ects neither the marginal trade-o¤s (consumption versus saving and working
versus enjoying leisure) nor the intertemporal budget constraint of the household.

Taxation of capital income

It is di¤erent when it comes to a tax on capital income because saving in the
Ramsey model responds to incentives. Consider a constant capital income tax at
the rate r , 0 < r < 1: The household’s dynamic budget identity becomes

a_ t = [(1 r )rt n] at + wt + xt ct ; a0 given,

where, if at < 0; the tax acts as a rebate. As above, xt is a per capita lump-sum
transfer. In view of a balanced budget, we have at the aggregate level

Gt + xt Lt = r rt Kt :

As Gt and r are given, the interpretation is that for all t 0; transfers adjust so
as to balance the budget. This requires that xt = r rt kt Gt =Lt = r rt kt ~ Tt :
The No-Ponzi-Game condition is now
Rt
lim at e 0 [(1 r )rs n]ds
0;
t!1

and the Keynes-Ramsey rule becomes

c_t 1
= [(1 r )rt ]:
ct
3
In fact, as we shall see in Section 11.1.4, the key point is not that, to …x ideas, we have
assumed the budget is balanced for every t: It is enough that the government satis…es its
intertemporal budget constraint.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 439

In general equilibrium we have

1h 0 ~
i
c~t = (1 r )(f (kt ) ) g c~t : (11.14)

The di¤erential equation for k~ is still (11.7).

In steady state we get (f 0 (k~ ) )(1 r) = + g, that is,

+ g
f 0 (k~ ) = > + g > g + n;
1 r

where the last inequality comes from the parameter condition (A1). Because
f 00 < 0, k~ is lower than if r = 0. Consequently, in the long run consumption
is lower as well.4 The resulting resource allocation is not Pareto optimal. There
exist an alternative technically feasible resource allocation that makes everyone
in society better o¤. This is because the capital income tax implies a wedge
between the marginal rate of transformation over time in production, f 0 (k~t ) ,
and the marginal rate of transformation over time to which consumers adapt,
0 ~
(1 r )(f (kt ) ).

11.1.3 E¤ects of shifts in the capital income tax rate

We shall study e¤ects of a rise in the tax on capital income. The e¤ects depend
on whether the change is anticipated in advance or not and whether the change
is permanent or only temporary. So there are four cases to consider.

(i) Unanticipated permanent shift in r

Until time t0 the economy has been in steady state with a tax-transfer scheme
based on some given constant tax rate, r ; on capital income. At time t0 , un-
expectedly, the government introduces a new tax-transfer scheme, involving a
higher constant tax rate, 0r , on capital income, i.e., 0 < r < 0r < 1: The path of
spending on goods and services remains unchanged, i.e., Gt = ~ Tt Lt for all t 0:
The lump-sum transfers, xt ; are raised so as to maintain a balanced budget. We
assume it is credibly announced that the new tax-transfer scheme will be adhered
to forever. So households expect the real after-tax interest rate (rate of return
0
on saving) to be (1 r )rt for all t t0 :
For t < t0 the dynamics are governed by (11.7) and (11.14) with 0 < r < 1:
The corresponding steady state, E, has k~ = k~ and c~ = c~ as indicated in the
4
In the Diamond OLG model a capital income tax, which …nances lump-sum transfers to the
old generation, has an ambiguous e¤ect on capital accumulation, depending on whether < 1
or > 1, cf. Exercise 5.?? in Chapter 5.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

440 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Figure 11.2: Phase portrait of an unanticipated permanent rise in r.

phase diagram in Fig. 11.2. The new tax-transfer scheme ruling after time t0
shifts the steady state point to E’with k~ = k~ 0 and c~ = c~ . The new c~ = 0 line
0

and the new saddle path are to the left of the old, i.e., k~ 0 < k~ : Until time t0
the economy is at the point E. Immediately after the shift in the tax on capital
income, equilibrium requires that the economy is on the new saddle path. So
there will be a jump from point E to point A in Fig. 11.2.
This upward jump in consumption is intuitively explained the following way.
We know that individual consumption immediately after the policy shock satis…es

ct0 = (at + ht0 ); where (11.15)

Z t01 0 Rt 0 )r
t0 ((1 n)dz
ht0 = (wt + 0r rt kt ~ Tt )e r z
dt; and
t0
1
= Rt :
t0 R1 (
(1 )(1 0 )r
r z +n)dz
t0
e t0
dt

Two e¤ects are present. First, both the higher transfers and the lower after-
tax rate of return after time t0 contribute to a higher ht0 ; there is thereby a
positive wealth e¤ect on current consumption through a higher ht0 . Second, the
propensity to consume, t0 ; will generally be a¤ected. If < 1; the reduction in
the after-tax rate of return will have a positive e¤ect on t0 : The positive e¤ect
on t0 when < 1 re‡ects that the positive substitution e¤ect on ct0 of a lower
after-tax rate of return dominates the negative income e¤ect. If instead > 1;
the positive substitution e¤ect on ct0 is dominated by the negative income e¤ect.
Whatever happens to t0 ; however, the phase diagram shows that in general

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 441

equilibrium there will necessarily be an upward jump in ct0 . We get this result
even if is much higher than 1. The explanation lies in the assumption that all
the extra tax revenue obtained by the rise in r is immediately transferred back
to the households lump sum, thereby strengthening the positive wealth e¤ect
0
on current consumption through the lower discount rate implied by (1 r )rz
< (1 r )rz :
In response to the rise in r ; we thus have c~t0 > f (k~t0 ) ( +g +n)k~t0 ; implying
~ which thus begins to fall. This results in lower
that saving is too low to sustain k,
real wages and higher before-tax interest rates, that is two negative feedbacks
on human wealth. Could these feedbacks not fully o¤set the initial tendency for
(after-tax) human wealth to rise? The answer is no, see Box 11.1.
As indicated by the arrows in Fig. 11.2, the economy moves along the new
saddle path towards the new steady state E’. Because k~ is lower in the new
steady state than in the old, so is c~: The evolution of the technology level, T; is
by assumption exogenous; thus, also actual per capita consumption, c c~ T; is
lower in the new steady state.

Box 11.1. A mitigating feedback can not instantaneously fully o¤set the
force that activates it.

Can the story told by Fig. 11.2 be true? Can it be true that the net e¤ect of
the higher tax on capital income is an upward jump in consumption at time
t0 as indicated in Fig. 11.2? Such a jump means that c~t0 > f (k~t0 )
( + g + n)k~t0 and the resulting reduced saving will make the future k lower
than otherwise and thereby make expected future real wages lower and
expected future before-tax interest rates higher. Both feedbacks partly
counteract the initial upward shift in human wealth due to higher transfers
and a lower e¤ective discount rate that were the direct result of the rise in
w : Could the two mentioned counteracting feedbacks fully o¤set the initial
tendency for (after-tax) human wealth, and therefore current consumption, to
rise?
The phase diagram says no. But what is the intuition? That the two feed-
backs can not fully o¤set (or even reverse) the tendency for (after-tax) human
wealth to rise at time t0 is explained by the fact that if they could, then the two
feedbacks would not be there in the …rst place. We cannot at the same
time have both a rise in the human wealth that triggers higher consumption
(and thereby lower saving and investment in the economy) and a neutrali-
zation, or a complete reversal, of this rise in the human wealth caused by
the higher consumption. The two feedbacks can only partly o¤set the initial
tendency for human wealth to rise.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

442 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Instead of all the extra tax revenue obtained being transferred back lump sum
to the households, we may alternatively assume that a major part of it is used to
…nance a rise in government consumption to the level G0t = ~ 0 Tt Lt ; where ~ 0 > ~ :5
In addition to the leftward shift of the c~ = 0 locus this will result in a downward
shift of the k~ = 0 locus. The phase diagram would look like a convex combination
of Fig. 11.1 and Fig. 11.2. Then it is possible that the jump in consumption at
time t0 becomes downward instead of upward.
Returning to the case where the extra tax revenue is fully transferred, the
next subsection splits the change in taxation policy into two events.

(ii) Anticipated permanent shift in r

Until time t0 the economy has been in steady state with a tax-transfer scheme
based on some given constant tax rate, r ; on capital income. At time t0 , unex-
pectedly, the government announces that a new tax-transfer policy with 0r > r
is to be implemented at time t1 > t0 . We assume people believe in this announce-
ment and that the new policy is implemented at time t1 as announced. The shock
to the economy is now not the event of a higher tax being implemented at time
t1 ; this event is expected after time t0 : The shock occurs at time t0 in the form
of the unexpected announcement. The path of spending on goods and services
remains unchanged throughout, i.e., Gt = ~ Tt Lt for all t 0:
The phase diagram in Fig. 11.3 illustrates the evolution of the economy for
t t0 : There are two time intervals to consider. For t 2 [t1 ; 1) ; the dynamics
are governed by (11.7) and (11.14) with r replaced by 0r ; starting from whatever
value obtained by k~ at time t1 :
In the time interval [t0 ; t1 ) ; however, the “old dynamics”, with the lower tax
rate, r ; in a sense still hold. Yet the path the economy follows immediately after
time t0 is di¤erent from what it would be without the information that capital
income will be taxed heavily from time t1 , where also transfers will become higher.
Indeed, the expectation of a lower after-tax interest rate until time t1 ; combined
with higher transfers from time t1 implies higher present value of future labor and
transfer income. Already at time t0 this induces an upward jump in consumption
to the point C in Fig. 11.3 because people feel more wealthy.
Since the low r rules until time t1 ; the point C is below the point A, which
is the same as that in Fig. 11.2. How far below? The answer follows from the
fact that there cannot be an expected discontinuity of marginal utility at time t1 ;
since that would contradict the preference for consumption smoothing over time
It is understood that also ~ 0 is not larger than what allows a steady state to exist. Moreover,
5

the government budget is still balanced for all t so that any temporary surplus or shortage of
tax revenue, 0r rt Kt G0t ; is immediately transferred or collected lump-sum.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 443

Figure 11.3: Phase portrait of an anticipated permanent rise in r.

implied by u00 (c) < 0 (strict concavity of the instantaneous utility function) and
re‡ected in the Keynes-Ramsey rule. To put it di¤erently: the shift to 0r does
not occur immediately, as in (11.15), but in the future, and as long as the shift
is known to occur at a given time in the future. The shift, when it takes place,
namely at the announced time t1 , will not trigger a jump in human wealth; ht1 :6
Hence, at time t1 ; there will be no jump in consumption, ct1 .
The intuitive background for this is that a consumer will never plan a jump
in consumption. To see this, consider a consumption path in the time inter-
val (t0 ; t2 ); where t2 > t1 : Suppose there is a discontinuity in ct at time t1 . In
view of the strict concavity of the utility function, there would then be gains to
be obtained by smoothing out consumption. Recalling the optimality condition
u0 (ct1 ) = t1 ; we could also say that along an optimal path there can be no ex-
pected discontinuity in the shadow price of …nancial wealth, t1 . This is analogue
to the fact that in an asset market, arbitrage rules out the existence of a generally
expected jump in the price of the asset to occur at some future time t1 . If we
imagine the expected jump is upward, an in…nite positive rate of return could
be obtained by buying the asset immediately before the jump. This generates
excess demand of the asset before time t1 and drives its price up in advance thus
preventing an expected upward jump at time t1 . And if we on the other hand
imagine the expected jump is downward, an in…nite negative rate of return could
be avoided by selling the asset immediately before the jump. This generates ex-

6
Replace t0 in the formula for human wealth in (11.15) by some t 2 (t0 ; t1 ); and consider ht
as the sum of the integrals from t to t1 and from t1 to 1; respectively, and let then t approach
t1 from below.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

444 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

cess supply of the asset before time t1 and drives its price down in advance thus
preventing an expected downward jump at t1 .
To avoid existence of an expected discontinuity in consumption, the point C
on the vertical line k~ = k~ in Fig. 11.3 must be such that, following the “old
dynamics”, it takes exactly t1 t0 time units to reach the new saddle path. This
dictates a unique position of the point C between E and A. If C were at a lower
position, the journey to the saddle path would take longer than t1 t0 : And if C
were at a higher position, the journey would not take as long as t1 t0 :
Immediately after time t0 , k~ will be decreasing (because saving is smaller than
what is required to sustain a constant k); ~ and c~ will be increasing in view of the
Keynes-Ramsey rule, since the rate of return on saving is above + g as long
as k~ < k~ and r low. Precisely at time t1 the economy reaches the new saddle
path, the high taxation of capital income begins, and the after-tax rate of return
becomes lower than + g: Hence, per-capita consumption begins to fall and the
economy gradually approaches the new steady state E’.
This analysis illustrates that when economic agents’ behavior depend on
forward-looking expectations, a credible announcement of a future change in pol-
icy has an e¤ect already before the new policy is implemented. Such e¤ects are
known as announcement e¤ects or anticipation e¤ects.

(iii) Unanticipated temporary shift in r

Once again we change the scenario. The economy with low capital taxation
has been in steady state up until time t0 . Then a new tax-transfer scheme is
unexpectedly introduced. At the same time it is credibly announced that the high
taxes on capital income and the corresponding transfers will cease at time t1 > t0 .
The path of spending on goods and services remains unchanged throughout, i.e.,
Gt = ~ Tt Lt for all t 0:
The phase diagram in Fig. 11.4 illustrates the evolution of the economy for
t t0 : For t t1 ; the dynamics are governed by (11.7) and (11.14), again with
the old r ; starting from whatever value obtained by k~ at time t1 :
In the time interval [t0 ; t1 ) the “new, temporary dynamics” with the high 0r
and high transfers hold sway. Yet the path that the economy takes immediately
after time t0 is di¤erent from what it would have been without the information
that the new tax-transfers scheme is only temporary. Indeed, the expectation of
a shift to a higher after-tax rate of return and cease of high transfers as of time
t1 implies lower present value of expected future labor and transfer earnings than
without this information. Hence, the upward jump in consumption at time t0 is
smaller than in Fig. 11.2. How much smaller? Again, the answer follows from
the fact that there can not be an expected discontinuity of marginal utility at time
t1 ; since that would violate the principle of smoothing of planned consumption.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 445

Figure 11.4: Phase portrait of an unanticipated temporary rise in r.

Thus the point F on the vertical line k~ = k~ in Fig. 11.4 must be such that,
following the “new, temporary dynamics”, it takes exactly t1 time units to reach
the solid saddle path in Fig. 11.4 (which is in fact the same as the saddle path
before time t0 ). The implied position of the economy at time t1 is indicated by
the point G in the …gure.
Immediately after time t0 , k~ will be decreasing (because saving is smaller than
what is required to sustain a constant k) ~ and c~ will be decreasing in view of the
Keynes-Ramsey rule in a situation with an after-tax rate of return lower than
+ g. Precisely at time t1 , when the temporary tax-transfers scheme based
on 0r is abolished (as announced and expected), the economy reaches the solid
saddle path. From that time the return on saving is high both because of the
abolition of the high capital income tax and because k~ is relatively low. The
general equilibrium e¤ect of this is higher saving, and so the economy moves
along the solid saddle path back to the original steady-state point E.
There is a last case to consider, namely an anticipated temporary shift in r :
We leave that for an exercise, see Exercise 11.??

11.1.4 Ricardian equivalence

We now drop the balanced budget assumption and allow public spending to be
…nanced partly by issuing government bonds and partly by lump-sum taxation.
Transfers and gross tax revenue as of time t are called Xt and T~t respectively,
while the real value of government net debt is called Bt : For simplicity, we assume
all public debt is short-term. Ignoring any money-…nancing of the spending, the
increase per time unit in government debt is identical to the government budget

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

446 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

de…cit:
B_ t = rt Bt + Gt + Xt T~t : (11.16)
As we ignore uncertainty, on its debt the government has to pay the same interest
rate, rt ; as other borrowers.
Along an equilibrium path in the Ramsey model the long-run interest rate
necessarily exceeds the long-run GDP growth rate. As we saw in Chapter 6, to
remain solvent, the government must then, as a debtor, ful…l a solvency require-
ment analogous to that of the households in the Ramsey model:
Rt
rs ds
lim Bt e 0 0: (11.17)
t!1

This NPG condition says that the debt is in the long run allowed to grow at most
at a rate less than the interest rate. As in discrete time, given the accounting
relationship (11.16), the NPG condition is equivalent to the intertemporal budget
constraint Z 1 Z 1
Rt Rt
(Gt + Xt )e 0 rs ds
dt T~t e 0 rs ds dt B0 : (11.18)
0 0
This says that the present value of the credibly planned public expenditure cannot
exceed government net wealth consisting of the present value of the expected
future tax revenues minus initial government debt, i.e., assets minus liabilities.
Assuming that the government does not want to be a net creditor to the
private sector in the long run, it will not collect more taxes than is necessary to
satisfy (11.18). Hence, we replace “ ”by “=”and rearrange to obtain
Z 1 Rt
Z 1 Rt
~
Tt e 0 rs ds
dt = (Gt + Xt )e 0 rs ds dt + B0 : (11.19)
0 0

Thus, for a given path of Gt and Xt ; the stream of the expected tax revenue
must be such that its present value equals the present value of total liabilities on
the right-hand-side of (11.19). A temporary budget de…cit leads to more debt
and therefore also higher taxes in the future. A budget de…cit merely implies a
deferment of tax payments. The condition (11.19) can be reformulated as
Z 1 Rt
(T~t Gt Xt )e 0 rs ds dt = B0 ;
0

showing that if net debt is positive today, then the government has to run a
positive primary budget surplus (that is, T~t Gt Xt > 0) in a su¢ ciently long
time in the future.
We will now show that when taxes are lump sum, then Ricardian equivalence
holds in the Ramsey model with a public sector.7 That is, a temporary tax
7
It is enough that just those taxes that are varied in the thought experiment are lump-sum.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.1. Market economy with a public sector 447

cut will have no consequences for aggregate consumption. The time pro…le of
lump-sum taxes does not matter.
Consider the intertemporal budget constraint of the representative household,
Z 1 Rt
ct Lt e 0 rs ds dt A0 + H0 = K0 + B0 + H0 ; (11.20)
0
where H0 is human wealth of the household. This says, that the present value of
the planned consumption stream can not exceed the total wealth of the household.
In the optimal plan of the household, we have strict equality in (11.20).
Let t denote the lump-sum per capita net tax: Then, T~t Xt = t Lt and
Z 1 Rt
Z 1 Rt
H0 = h0 L0 = (wt t )Lt e 0 rs ds
dt = (wt Lt + Xt T~t )e 0 rs ds dt
0 0
Z 1 Rt
= (wt Lt Gt )e 0 rs ds dt B0 ; (11.21)
0

where the last equality comes from rearranging (11.19). It follows that
Z 1 Rt
B0 + H0 = (wt Lt Gt )e 0 rs ds dt:
0
We see that the time pro…les of transfers and taxes have fallen out. What matters
for total wealth of the forward-looking household is just the spending on goods
and services, not the time pro…le of transfers and taxes. A higher initial debt
has no e¤ect on the sum, B0 + H0 ; because H0 ; which incorporates transfers
and taxes, becomes equally much lower. Total private wealth is thus una¤ected
by government debt. So is therefore also private consumption when net taxes
are lump sum. A temporary tax cut will not make people feel wealthier and
induce them to consume more. Instead they will increase their saving by the
same amount as taxes have been reduced, thereby preparing for the higher taxes
in the future.
This is the Ricardian equivalence result, which we encountered also in Barro’s
discrete time dynasty model in Chapter 7:
In a representative agent model with full employment, rational
expectations, and no credit market imperfections, if taxes are lump
sum, then, for a given evolution of public expenditure, aggregate pri-
vate consumption is independent of whether current public expen-
diture is …nanced by taxes or by issuing bonds. The latter method
merely implies a deferment of tax payments. Given the government’s
intertemporal budget constraint, (11.19), a cut in current taxes has
to be o¤set by a rise in future taxes of the same present value. Since,
with lump-sum taxation, it is only the present value of the stream of
taxes that matters, the “timing”is irrelevant.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

448 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

The assumptions of a representative agent and a long-run interest rate in

excess of the long-run GDP growth rate are of key importance. As pointed
out in Chapter 6, Ricardian equivalence breaks down in OLG models without
an operative Barro-style bequest motive. Such a bequest motive is implicit in
the in…nite horizon of the Ramsey household. In OLG models, where …nite life
time is emphasized, there is a turnover in the population of tax payers so that
taxes levied at di¤erent times are levied on partly di¤erent sets of agents. In
the future there are newcomers and they will bear part of the higher future tax
burden. Therefore, a current tax cut makes current generations feel wealthier and
this leads to an increase in current consumption, implying a decrease in national
saving, as a result of the temporary de…cit …nance. The present generations
bene…t, but future generations bear the cost in the form of smaller national
wealth than otherwise. We return to further reasons for absence of Ricardian
equivalence in chapters 13 and 19.

11.2 Learning by investing and investment-enhancing

policy
In endogenous growth theory the Ramsey framework has been applied extensively
as a simplifying description of the household sector. In most endogenous growth
theory the focus is on mechanisms that generate and shape technological change.
Di¤erent hypotheses about the generation of new technologies are then often
combined with a simpli…ed picture of the household sector as in the Ramsey
model. Since this results in a simple determination of the long-run interest rate
(the modi…ed golden rule), the analyst can in a …rst approach concentrate on the
main issue, technological change, without being disturbed by aspects that are
often secondary to this issue.
As an example, let us consider one of the basic endogenous growth models,
the learning-by-investing model, sometimes called the learning-by-doing model.
Learning from investment experience and di¤usion across …rms of the resulting
new technical knowledge (positive externalities) play an important role.
There are two popular alternative versions of the model. The distinguishing
feature is whether the learning parameter (see below) is less than one or equal
to one. The …rst case corresponds to a model by Nobel laureate Kenneth Arrow
(1962). The second case has been drawn attention to by Paul Romer (1986) who
assumes that the learning parameter equals one. We …rst consider the common
framework shared by these two models. Next we describe and analyze Arrow’s
model (in a simpli…ed version) and …nally we compare it to Romer’s.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 449

11.2.1 The common framework

We consider a closed economy with …rms and households interacting under con-
ditions of perfect competition. Later, a government attempting to internalize the
positive investment externality is introduced.
Let there be N …rms in the economy (N “large”). Suppose they all have
the same neoclassical production function, F; with CRS. Firm no. i faces the
technology
Yit = F (Kit ; Tt Lit ); i = 1; 2; :::; N; (11.22)
where the economy-wide technology level Tt is an increasing function of society’s
previous experience, proxied by cumulative aggregate net investment:
Z t
Tt = Isn ds = Kt ; 0< 1; (11.23)
1
P
where Isn is aggregate net investment and Kt = i Kit :8
The idea is that investment the production of capital goods as an unin-
tended by-product results in experience. The …rm and its employees learn from
this experience. Producers recognize opportunities for process and quality im-
provements. In this way knowledge is achieved about how to produce the capital
goods in a cost-e¢ cient way and how to design them so that in combination
with labor they are more productive and better satisfy the needs of the users.
Moreover, as emphasized by Arrow,

“each new machine produced and put into use is capable of changing
the environment in which production takes place, so that learning is
taking place with continually new stimuli”(Arrow, 1962).9

The learning is assumed to bene…t essentially all …rms in the economy. There
are knowledge spillovers across …rms and these spillovers are reasonably fast rel-
ative to the time horizon relevant for growth theory. In our macroeconomic ap-
proach both F and T are in fact assumed to be exactly the same for all …rms in the
economy. That is, in this speci…cation the …rms producing consumption-goods
bene…t from the learning just as much as the …rms producing capital-goods.
The parameter indicates the elasticity of the general technology level, T ;
with respect to cumulative aggregate net investment and is named the “learning
8
For arbitrary units of measurement for labor and output the hypothesis is Tt = BKt ;
B > 0: In (11.23) measurement units are chosen such that B = 1.
9
Concerning empirical evidence of learning-by-doing and learning-by-investing, see Liter-
ature Notes. The citation of Arrow indicates that it was experience from cumulative gross
investment he had in mind as the basis for learning. Yet, to simplify, we stick to the hypothesis
in (11.23), where it is cumulative net investment that matters.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

450 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

parameter”. Whereas Arrow assumes < 1; Romer focuses on the case = 1:

The case of > 1 is ruled out since it would lead to explosive growth (in…nite
output in …nite time) and is therefore not plausible.

The individual …rm

In the simple Ramsey model we assumed that households directly own the capital
goods in the economy and rent them out to the …rms. When discussing learning-
by-investment, it somehow …ts the intuition better if we (realistically) assume
that the …rms generally own the capital goods they use. They then …nance their
capital investment by issuing shares and bonds. Households’ …nancial wealth
then consists of these shares and bonds.
Consider …rm i: There is perfect competition in all markets. So the …rm is
a price taker. Its problem is to choose a production and investment plan which
maximizes the present value, Vi ; of expected future cash-‡ows. Thus the …rm
chooses (Lit ; Iit )1
t=0 to maximize
Z 1 Rt
Vi0 = [F (Kit ; Tt Lit ) wt Lit Iit ] e 0 rs ds dt
0

subject to K_ it = Iit Kit : Here wt and It are the real wage and gross investment,
respectively, at time t, rs is the real interest rate at time s; and 0 is the capital
depreciation rate. Rising marginal capital installation costs and other kinds of
adjustment costs are assumed minor and can be ignored. It can be shown, cf.
Chapter 14, that in this case the …rm’s problem is equivalent to maximization of
current pure pro…ts in every short time interval. So, as hitherto, we can describe
the …rm as just solving a series of static pro…t maximization problems.
We suppress the time index when not needed for clarity. At any date …rm i
maximizes current pure pro…ts, i = F (Ki ; T Li ) (r + )Ki wLi : This leads
to the …rst-order conditions for an interior solution:

@ i =@Ki = F1 (Ki ; T Li ) (r + ) = 0; (11.24)

@ i =@Li = F2 (Ki ; T Li )T w = 0:

Behind (11.24) is the presumption that each …rm is small relative to the economy
as a whole, so that each …rm’s investment has a negligible e¤ect on the economy-
wide technology level Tt . Since F is homogeneous of degree one, by Euler’s
theorem,10 the …rst-order partial derivatives, F1 and F2 ; are homogeneous of
degree 0. Thus, we can write (11.24) as

F1 (ki ; T ) = r + ; (11.25)
10
See Math tools.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 451

where ki Ki =Li . Since F is neoclassical, F11 < 0: Therefore (11.25) determines

ki uniquely. From (11.25) follows that the chosen capital-labor ratio, ki ; will be
the same for all …rms, say k:

The individual household

The household sector is described by our standard Ramsey framework with in-
elastic labor supply and a constant population growth rate n 0. The households
have CRRA instantaneous utility with parameter > 0: The pure rate of time
preference is a constant, . The ‡ow budget identity in per capita terms is

a_ t = (rt n)at + wt ct ; a0 given,

where a is per capita …nancial wealth. The NPG condition is

Rt
lim at e 0 (rs n)ds
0:
t!1

The resulting consumption-saving plan implies that per capita consumption fol-
lows the Keynes-Ramsey rule,
c_t 1
= (rt );
ct
and the transversality condition that the NPG condition is satis…ed with strict
equality. In general equilibrium of our closed economy with no role for natural
resources and no government debt, at will equal Kt =Lt :

Equilibrium in factor markets

P P
For every t we have in equilibrium that i Ki = K and i Li = L; where K
and L are the available amounts
P ofPcapital and
P labor, respectively (both pre-
determined). Since K = i Ki = i ki Li = i kLi = kL; the chosen capital
intensity, ki ; satis…es
K
ki = k = k; i = 1; 2; :::; N: (11.26)
L
As a consequence we can use (11.25) to determine the equilibrium interest rate:

rt = F1 (kt ; Tt ) : (11.27)

That is, whereas in the …rm’s …rst-order condition (11.25) causality goes from rt
to kit ; in (11.27) causality goes from kt to rt : Note also that in our closed economy
with no natural resources and no government debt, at will equal kt :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

452 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

The implied aggregate production function is

X X X X
Y = Yi yi Li = F (ki ; T )Li = F (k; T )Li
i i i i
X
= F (k; T ) Li = F (k; T )L = F (K; T L) = F (K; K L), (11.28)
i

where we have used (11.22), (11.26), and (11.23) and the assumption that F is
homogeneous of degree one.

11.2.2 The arrow case: <1

The Arrow case is the robust case where the learning parameter satis…es 0 <
< 1: The method for analyzing the Arrow case is analogue to that used in
the study of the Ramsey model with exogenous technical progress. In particular,
aggregate capital per unit of e¤ective labor, k~ K=(T L); is a key variable. Let
y~ Y =(T L): Then
F (K; T L) ~ 1) ~
y~ = = F (k; f (k); f 0 > 0; f 00 < 0: (11.29)
TL
We can now write (11.27) as

rt = f 0 (k~t ) ; (11.30)

where k~t is pre-determined.

Dynamics
From the de…nition k~ K=(T L) follows

k~ K_ T_ L_ K_ K_
= = n (by (11.23))
k~ K T L K K
Y C K y~ c~ k~ C c
= (1 ) n = (1 ) n; where c~ :
K k~ TL T
Multiplying through by k~ we have

k~ = (1 ~
)(f (k) c~) [(1 ~
) + n] k: (11.31)

In view of (11.30), the Keynes-Ramsey rule implies

c_ 1 1 ~
gc = (r )= f 0 (k) : (11.32)
c
c Groth, Lecture notes in macroeconomics, (mimeo) 2015.
11.2. Learning by investing and investment-enhancing policy 453

De…ning c~ c=A; now follows

:
c~ c_ T_ c_ K_ c_ Y cL K c_ ~
= = = = (~
y c~ k)
c~ c T c K c K c k~
1 0 ~ ~
= (f (k) ) (~
y c~ k):
k~
Multiplying through by c~ we have

1 ~ ~ ~ c~:
c~ = (f 0 (k) ) (f (k) c~ k) (11.33)
k~

The two coupled di¤erential equations, (11.31) and (11.33), determine the
evolution over time of the economy.

Phase diagram Fig. 11.5 depicts the phase diagram. The k~ = 0 locus comes
from (11.31), which gives

n
k~ = 0 for c~ = f (k)
~ ( + ~
)k; (11.34)
1

where we realistically may assume that + n=(1 ) > 0: As to the c~ = 0 locus,

we have

~ k~
c~ = 0 for c~ = f (k) k~ ~
(f 0 (k) )

~ k~
= f (k) k~ gc ~
c(k) (from (11.32)). (11.35)

Before determining the slope of the c~ = 0 locus, it is convenient to consider

the steady state, (k~ ; c~ ).

Steady state In a steady state c~ and k~ are constant so that the growth rate of
_ + n; i.e.,
C as well as K equals A=A

C_ K_ T_ K_
= = +n= + n:
C K T K
Solving gives
C_ K_ n
= = :
C K 1

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

454 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Figure 11.5: Phase diagram for the Arrow model.

Thence, in a steady state

C_ n n
gc = n= n= gc ; and (11.36)
C 1 1
T_ K_ n
= = = gc : (11.37)
T K 1
~ respectively, will therefore satisfy, by (11.32),
The steady-state values of r and k;

n
r = f 0 (k~ ) = + gc = + : (11.38)
1
To ensure existence of a steady state we assume that the private marginal pro-
ductivity of capital is su¢ ciently sensitive to capital per unit of e¤ective labor,
from now called the “capital intensity”:

~ > + + n ~
lim f 0 (k) > lim f 0 (k): (A1)
~
k!0 1 ~
k!1

The
R t transversality condition of the representative household is that limt!1
at e 0 (rs n)ds = 0; where a is per capita …nancial wealth. In general equilibrium
t

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 455

at = kt k~t Tt ; where Tt in steady state grows according to (11.37). Thus, in

steady state the transversality condition can be written

lim k~ e(gc r +n)t

= 0: (TVC)
t!1

For this to hold, we need

n
r > gc + n = ; (11.39)
1
by (11.36). In view of (11.38), this is equivalent to

n
n > (1 ) ; (A2)
1
which we assume satis…ed.
As to the slope of the c~ = 0 locus we have from (11.35),

~
1 ~ f 00 (k) 1
0 ~ = f 0 (k)
c (k) ~ (k ~
+ gc ) > f 0 (k) gc ; (11.40)

since f 00 < 0: At least in a small neighborhood of the steady state we can sign
the right-hand side of this expression. Indeed,

1 1 n n n
f 0 (k~ ) gc = + g c gc = + = n (1 ) > 0;
1 1 1
(11.41)
~
by (11.36) and (A2). So, combining with (11.40), we conclude that c (k ) > 0: 0
~
By continuity, in a small neighborhood of the steady state, c0 (k) c0 (k~ ) > 0:
Therefore, close to the steady state, the c~ = 0 locus is positively sloped, as
indicated in Fig. 11.5.
Still, we have to check the following question: In a neighborhood of the steady
state, which is steeper, the c~ = 0 locus or the k~ = 0 locus? The slope of the latter
~
is f 0 (k) n=(1 ); from (11.34): At the steady state this slope is

1
f 0 (k~ ) gc 2 (0; c0 (k~ ));

in view of (11.41) and (11.40). The c~ = 0 locus is thus steeper. So, the c~ = 0
locus crosses the k~ = 0 locus from below and can only cross once.
The assumption (A1) ensures existence of a k~ > 0 satisfying (11.38). As
Fig. 11.5 is drawn, a little more is implicitly assumed namely that there exists a

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

456 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

k^ > 0 such that the private net marginal productivity of capital equals the the
steady-state growth rate of output, i.e.,

^ Y_ T_ L_ n n
f 0 (k) =( ) =( ) + = +n= ; (11.42)
Y T L 1 1

where we have used (11.37). Thus, the tangent to the k~ = 0 locus at k~ = k^ is

horizontal and k^ > k~ as indicated in the …gure.
Note, however, that k^ is not the golden-rule capital intensity. The latter is the
capital intensity, k~GR ; at which the social net marginal productivity of capital
equals the steady-state growth rate of output (see Appendix). If k~GR exists, it will
be larger than k^ as indicated in Fig. 11.5. To see this, we now derive a convenient
expression for the social marginal productivity of capital. From (11.28) we have

@Y 1 ~ + F2 ( )K L( K
= F1 ( ) + F2 ( ) K L = f 0 (k) 1
) (by (11.29))
@K
~ + (F ( ) F1 ( )K) K 1
= f 0 (k) (by Euler’s theorem)
~ + (f (k)K
= f 0 (k) ~ ~
L f 0 (k)K) K 1
(by (11.29) and (11.23))
~ ~ 0 ~
~ + (f (k)K
= f 0 (k) ~ 1
L ~
f 0 (k)) ~ + f (k) kf (k) > f 0 (k):
= f 0 (k) ~
k~

in view of k~ = K=(K L) = K 1 L 1 and f (k)= ~ k~ f 0 (k)

~ > 0: As expected, the
positive externality makes the social marginal productivity of capital larger than
the private one. Since we can also write @Y =@K = (1 ~ + f (k)=
)f 0 (k) ~ k;
~ we see
~ 0 ~ ~ ~
that @Y =@K is a decreasing function of k (both f (k) and f (k)=k are decreasing
~
in k:
Now, the golden-rule capital intensity, k~GR ; will be that capital intensity which
satis…es
f (k~GR ) k~GR f 0 (k~GR ) Y_ n
f 0 (k~GR ) + =( ) = :
k~GR Y 1
To ensure there exists such a k~GR ; we strengthen the right-hand side inequality
in (A1) by the assumption
!
f ( ~
k) ~ 0 (k)
kf ~ n
~ +
lim f 0 (k) < + : (A3)
~
k!1 k~ 1

This, together with (A1) and f < 0, implies existence of a unique k~GR , and in
00

view of our additional assumption (A2), we have 0 < k~ < k^ < k~GR ; as displayed
in Fig. 11.5.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 457

Stability The arrows in Fig. 11.5 indicate the direction of movement as de-
termined by (11.31) and (11.33). We see that the steady state is a saddle point.
The dynamic system has one pre-determined variable, k; ~ and one jump variable,
c~: The saddle path is not parallel to the jump variable axis. We claim that for
a given k~0 > 0; (i) the initial value of c~0 will be the ordinate to the point where
the vertical line k~ = k~0 crosses the saddle path; (ii) over time the economy will
move along the saddle path towards the steady state. Indeed, this time path is
consistent with all conditions of general equilibrium, including the transversality
condition (TVC). And the path is the only technically feasible path with this
property. Indeed, all the divergent paths in Fig. 11.5 can be ruled out as equi-
librium paths because they can be shown to violate the transversality condition
of the household.
In the long run c and y Y =L y~T = f (k~ )T grow at the rate n=(1 );
which is positive if and only if n > 0: This is an example of endogenous growth in
the sense that the positive long-run per capita growth rate is generated through an
internal mechanism (learning) in the model (in contrast to exogenous technology
growth as in the Ramsey model with exogenous technical progress).

Two types of endogenous growth

One may distinguish between two types of endogenous growth. One is called fully
endogenous growth which occurs when the long-run growth rate of c is positive
without the support by growth in any exogenous factor (for example exogenous
growth in the labor force); the Romer case, to be considered in the next section,
provides an example. The other type is called semi-endogenous growth and is
present if growth is endogenous but a positive per capita growth rate can not be
maintained in the long run without the support by growth in some exogenous
factor (for example growth in the labor force). Clearly, in the Arrow model of
learning by investing, growth is “only” semi-endogenous. The technical reason
for this is the assumption that the learning parameter is below 1; which implies
diminishing returns to capital at the aggregate level. If and only if n > 0; do we
_ > 0 in the long run.11 In line with this, @gy =@n > 0:
have c=c
The key role of population growth derives from the fact that although there
are diminishing marginal returns to capital at the aggregate level, there are in-
creasing returns to scale w.r.t. capital and labor. For the increasing returns to
be exploited, growth in the labor force is needed. To put it di¤erently: when
there are increasing returns to K and L together, growth in the labor force not
only counterbalances the falling marginal productivity of aggregate capital (this
11
Note, however, that the model, and therefore (11.36), presupposes n 0: If n < 0; then
K would tend to be decreasing and so, by (11.23), the level of technical knowledge would be
decreasing, which is implausible, at least for a modern industrialized economy.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

458 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

counter-balancing role re‡ects the complementarity between K and L), but also
upholds sustained productivity growth.
Note that in the semi-endogenous growth case @gy =@ = n=(1 )2 > 0 for
n > 0: That is, a higher value of the learning parameter implies higher per capita
growth in the long run, when n > 0. Note also that @gy =@ = 0 = @gy =@ ; that
is, in the semi-endogenous growth case preference parameters do not matter for
long-run growth. As indicated by (11.36), the long-run growth rate is tied down
by the learning parameter, ; and the rate of population growth, n: But, like in
the simple Ramsey model, it can be shown that preference parameters matter for
the level of the growth path. This suggests that taxes and subsidies do not have
long-run growth e¤ects, but “only”level e¤ects (see Exercise 11.??).

11.2.3 Romer’s limiting case: = 1; n = 0

We now consider the limiting case = 1: We should think of it as a thought
experiment because, by most observers, the value 1 is considered an unrealistically
high value for the learning parameter. To avoid a forever rising growth rate we
have to add the restriction n = 0:
The resulting model turns out to be extremely simple and at the same time
it gives striking results (both circumstances have probably contributed to its
popularity).
First, with = 1 we get T = K and so the equilibrium interest rate is, by
(11.27),
r = F1 (k; K) = F1 (1; L) r;

where we have divided the two arguments of F1 (k; K) by k K=L and again
used Euler’s theorem. Note that the interest rate is constant “from the beginning”
and independent of the historically given initial value of K; K0 . The aggregate
production function is now

Y = F (K; KL) = F (1; L)K; L constant, (11.43)

and is thus linear in the aggregate capital stock. In this way the general neo-
classical presumption of diminishing returns to capital has been suspended and
replaced by exactly constant returns to capital. So the Romer model belongs to a
class of models known as AK models, that is, models where in general equilibrium
the interest rate and the output-capital ratio are necessarily constant over time
whatever the initial conditions.
The method for analyzing an AK model is di¤erent from the one used for a
diminishing returns model as above.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 459

Dynamics
The Keynes-Ramsey rule now takes the form
c_ 1 1
= (r ) = (F1 (1; L) ) ; (11.44)
c
which is also constant “from the beginning”. To ensure positive growth, we
assume
F1 (1; L) > : (A1’)
And to ensure bounded intertemporal utility (and existence of equilibrium), it is
assumed that
> (1 ) and therefore < + = r: (A2’)
Solving the linear di¤erential equation (11.44) gives

ct = c0 e t ; (11.45)

where c0 is unknown so far (because c is not a predetermined variable). We shall

…nd c0 by applying the households’transversality condition
rt rt
lim at e = lim kt e = 0: (TVC)
t!1 t!1

First, note that the dynamic resource constraint for the economy is

K_ = Y cL K = F (1; L)K cL K;

or, in per-capita terms,

k_ = [F (1; L) ]k c0 e t : (11.46)

In this equation it is important that F (1; L) > 0: To understand this

inequality, note that, by (A2’), F (1; L) > F (1; L) r = F (1; L) F1 (1; L)
= F2 (1; L)L > 0; where the …rst equality is due to r = F1 (1; L) and the second
is due to the fact that since F is homogeneous of degree 1, we have, by Euler’s
theorem, F (1; L) = F1 (1; L) 1 + F2 (1; L)L > F1 (1; L) > ; in view of (A1’). The
key property F (1; L) F1 (1; L) > 0 is illustrated in Fig. 11.6.
_ + ax(t) = ceht ;
The solution of a linear di¤erential equation of the form x(t)
with h 6= a; is
c c ht
x(t) = (x(0) )e at + e : (11.47)
a+h a+h
Thus the solution to (11.46) is
c0 c0
kt = (k0 )e(F (1;L) )t
+ e t: (11.48)
F (1; L) F (1; L)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

460 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Figure 11.6: Illustration of the fact that for L given, F (1; L) > F1 (1; L).

To check whether (TVC) is satis…ed we consider

rt c0 c0
kt e = (k0 )e(F (1;L) r)t
+ e( r)t
F (1; L) F (1; L)
c0
! (k0 )e(F (1;L) r)t
for t ! 1;
F (1; L)
since r > ; by (A2’). But r = F1 (1; L) < F (1; L) ; and so (TVC) is only
satis…ed if
c0 = (F (1; L) )k0 : (11.49)
If c0 is less than this, there will be over-saving and (TVC) is violated (at e rt ! 1
for t ! 1; since at = kt ). If c0 is higher than this, both the NPG and (TVC)
are violated (at e rt ! 1 for t ! 1).
Inserting the solution for c0 into (11.48), we get
c0
kt = e t = k0 e t ;
F (1; L)
that is, k grows at the same constant rate as c “from the beginning”: Since y
Y =L = F (1; L)k; the same is true for y: Hence, from start the system is in
balanced growth (there is no transitional dynamics).
This is a case of fully endogenous growth in the sense that the long-run growth
rate of c is positive without the support by growth in any exogenous factor. This
outcome is due to the absence of diminishing returns to aggregate capital, which
is implied by the assumed high value of the learning parameter. The empirical
foundation for being in a neighborhood of this high value is weak, however, cf.
Literature notes. A further problem with this special version of the learning
model is that the results are non-robust. With slightly less than 1, we are back

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 461

in the Arrow case and growth peters out, since n = 0: With slightly above 1, it
can be shown that growth becomes explosive (in…nite output in …nite time).12
The Romer case, = 1; is thus a knife-edge case in a double sense. First,
it imposes a particular value for a parameter which apriori can take any value
within an interval. Second, the imposed value leads to theoretically non-robust
results; values in a hair’s breadth distance result in qualitatively di¤erent behavior
of the dynamic system. Still, whether the Romer case - or, more generally, a
fully-endogenous growth case - can be used as an empirical approximation to
its semi-endogenous “counterpart” for a su¢ ciently long time horizon to be of
interest, is a debated question within growth analysis.
It is noteworthy that the causal structure in the long run in the diminishing
returns case is di¤erent than in the AK-case of Romer. In the diminishing returns
case the steady-state growth rate is determined …rst, as gc in (11.36), and then r
is determined through the Keynes-Ramsey rule; …nally, Y =K is determined by the
technology, given r : In contrast, the Romer case has Y =K and r directly given
as F (1; L) and r; respectively. In turn, r determines the (constant) equilibrium
growth rate through the Keynes-Ramsey rule.

Economic policy in the Romer case

In the AK case, that is, the fully endogenous growth case, we have @ =@ < 0 and
@ =@ < 0: Thus, preference parameters matter for the long-run growth rate and
not “only”for the level of the growth path. This suggests that taxes and subsidies
can have long-run growth e¤ects. In any case, in this model there is a motivation
for government intervention due to the positive externality of private investment.
This motivation is present whether < 1 or = 1: Here we concentrate on the
latter case, which is the simpler one. We …rst …nd the social planner’s solution.

The social planner The social planner faces the aggregate production function
Yt = F (1; L)Kt or, in per capita terms, yt = F (1; L)kt : The social planner’s
problem is to choose (ct )1
=0 to maximize

Z 1
c1t t
U0 = e dt s.t.
0 1
ct 0;
_kt = F (1; L)kt ct kt ; k0 > 0 given, (11.50)
kt 0 for all t > 0: (11.51)
12
See Solow (1997).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

462 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

The current-value Hamiltonian is

c1
H(k; c; ; t) = + (F (1; L)k c k) ;
1
where = t is the adjoint variable associated with the state variable, which is
capital per unit of labor. Necessary …rst-order conditions for an interior optimal
solution are
@H
= c = 0, i.e., c = ; (11.52)
@c
@H
= (F (1; L) )= _+ : (11.53)
@k
We guess that also the transversality condition,
t
lim kt t e = 0; (11.54)
t!1

must be satis…ed by an optimal solution. This guess will be of help in …nding a

candidate solution. Having found a candidate solution, we shall invoke a theorem
on su¢ cient conditions to ensure that our candidate solution is really a solution.
Log-di¤erentiating w.r.t. t in (11.52) and combining with (11.53) gives the
social planner’s Keynes-Ramsey rule,
c_t 1
= (F (1; L) ) SP : (11.55)
ct
We see that SP > : This is because the social planner internalizes the economy-
wide learning e¤ect associated with capital investment, that is, the social planner
takes into account that the “social” marginal productivity of capital is @yt =@kt
= F (1; L) > F1 (1; L): To ensure bounded intertemporal utility we sharpen (A2’)
to
> (1 ) SP : (A2”)
To …nd the time path of kt , note that the dynamic resource constraint (11.50)
can be written
k_ t = (F (1; L) )kt c0 e SP t ;
in view of (11.55). By the general solution formula (11.47) this has the solution
c0 c0
kt = (k0 )e(F (1;L) )t
+ e SP t : (11.56)
F (1; L) SP F (1; L) SP

In view of (11.53), in an interior optimal solution the time path of the adjoint
variable is
[(F (1;L) ]t
t = 0e ;

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.2. Learning by investing and investment-enhancing policy 463

where 0 = c0 > 0; by (11.52): Thus, the conjectured transversality condition

(11.54) implies
lim kt e (F (1;L) )t = 0; (11.57)
t!1
where we have eliminated 0 : To ensure that this is satis…ed, we multiply kt from
(11.56) by e (F (1;L) )t to get
c0 c0
kt e (F (1;L) )t = k0 + e[ SP (F (1;L) )]t
F (1; L) SP F (1; L) SP
c0
! k0 for t ! 1;
F (1; L) SP

since, by (A2”), SP < + SP = F (1; L) in view of (11.55). Thus, (11.57)

is only satis…ed if
c0 = (F (1; L) SP )k0 : (11.58)
Inserting this solution for c0 into (11.56), we get
c0
kt = e SP t = k0 e SP t ;
F (1; L) SP

that is, k grows at the same constant rate as c “from the beginning”: Since y
Y =L = F (1; L)k; the same is true for y: Hence, our candidate for the so-
cial planner’s solution is from start in balanced growth (there is no transitional
dynamics).
The next step is to check whether our candidate solution satis…es a set of
su¢ cient conditions for an optimal solution. Here we can use Mangasarian’s
theorem. Applied to a continuous-time optimization problem like this, with one
control variable and one state variable, the theorem says that the following con-
ditions are su¢ cient:
(a) Concavity: For all t 0 the Hamiltonian is jointly concave in the control
and state variables, here c and k.
(b) Non-negativity: There is for all t 0 a non-negativity constraint on the
state variable; in addition, the co-state variable, ; is non-negative for all
t 0 along the optimal path.
(c) TVC: The candidate solution satis…es the transversality condition
limt!1 kt t e t = 0; where t e t is the discounted co-state variable.
In the present case we see that the Hamiltonian is a sum of concave func-
tions and therefore is itself concave in (k; c): Further, from (11.51) we see that
condition (b) is satis…ed. Finally, our candidate solution is constructed so as to
satisfy condition (c). The conclusion is that our candidate solution is an optimal
solution. We call it an SP allocation.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

464 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Implementing the SP allocation in the market economy Returning to

the competitive market economy, we assume there is a policy maker, the govern-
ment, with only two activities. These are (i) paying an investment subsidy, s; to
the …rms so that their capital costs are reduced to

(1 s)(r + )

per unit of capital per time unit; (ii) …nancing this subsidy by a constant con-
sumption tax rate :
Let us …rst …nd the size of s needed to establish the SP allocation. Firm i
now chooses Ki such that

@Yi
jK …xed = F1 (Ki ; KLi ) = (1 s)(r + ):
@Ki

By Euler’s theorem this implies

F1 (ki ; K) = (1 s)(r + ) for all i;

so that in equilibrium we must have

F1 (k; K) = (1 s)(r + );

where k K=L; which is pre-determined from the supply side. Thus, the equi-
librium interest rate must satisfy

F1 (k; K) F1 (1; L)
r= = ; (11.59)
1 s 1 s
again using Euler’s theorem.
It follows that s should be chosen such that the “right” r arises. What is
the “right” r? It is that net rate of return which is implied by the production
technology at the aggregate level, namely @Y =@K = F (1; L) : If we can
obtain r = F (1; L) ; then there is no wedge between the intertemporal rate of
transformation faced by the consumer and that implied by the technology. The
required s thus satis…es

F1 (1; L)
r= = F (1; L) ;
1 s
so that
F1 (1; L) F (1; L) F1 (1; L) F2 (1; L)L
s=1 = = :
F (1; L) F (1; L) F (1; L)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.3. Concluding remarks 465

It remains to …nd the required consumption tax rate : The tax revenue will
be cL; and the required tax revenue is

T = s(r + )K = (F (1; L) F1 (1; L)) K = cL:

Thus, with a balanced budget the required tax rate is

T F (1; L) F1 (1; L) F (1; L) F1 (1; L)

= = = > 0; (11.60)
cL c=k F (1; L) SP

where we have used that the proportionality in (11.58) between c and k holds for
all t 0: Substituting (11.55) into (11.60), the solution for can be written

[F (1; L) F1 (1; L)] F2 (1; L)L

= = :
( 1)(F (1; L) )+ ( 1)(F (1; L) )+

The required tax rate on consumption is thus a constant. It therefore does not
distort the consumption/saving decision on the margin, cf. Appendix B.
It follows that the allocation obtained by this subsidy-tax policy is the SP
allocation. A policy, here the policy (s; ); which in a decentralized system in-
duces the SP allocation, is called a …rst-best policy. In a situation where for some
reason it is impossible to obtain an SP allocation in a decentralized way (because
of adverse selection and moral hazard problems, say), a government’s optimiza-
tion problem would involve additional constraints to those given by technology
and initial resources. A decentralized implementation of the solution to such a
problem is called a second-best policy.

11.3 Concluding remarks

(not yet available)

11.4 Literature notes

(incomplete)
As to empirical evidence of learning-by-doing and learning-by-investing, see
...
As noted in Section 11.2.1, the citation of Arrow indicates that it was expe-
rience from cumulative gross investment, rather than net investment, he had in
mind as the basis for learning. Yet the hypothesis in (11.23) is the more popu-
lar one - seemingly for no better reason than that it leads to simpler dynamics.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

466 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

Another way in which (11.23) deviates from Arrow’s original ideas is by assum-
ing that technical progress is disembodied rather than embodied, a distinction we
touched upon in Chapter 2. Moreover, we have assumed a neoclassical technology
whereas Arrow assumed …xed technical coe¢ cients.

11.5 Appendix
A. The golden-rule capital intensity in Arrow’s growth model
In our discussion of Arrow’s learning-by-investing model in Section 11.2.2 (where
0 < < 1); we claimed that the golden-rule capital intensity, k~GR ; will be that ef-
fective capital-labor ratio at which the social net marginal productivity of capital
equals the steady-state growth rate of output. In this respect the Arrow model
with endogenous technical progress is similar to the standard neoclassical growth
model with exogenous technical progress. This claim corresponds to a very gen-
eral theorem, valid also for models with many capital goods and non-existence of
an aggregate production function. This theorem says that the highest sustainable
path for consumption per unit of labor in the economy will be that path which
results from those techniques which pro…t maximizing …rms choose under perfect
competition when the real interest rate equals the steady-state growth rate of
GNP (see Gale and Rockwell, 1975).
To prove our claim, note that in steady state, (11.35) holds whereby consump-
tion per unit of labor (here the same as per capita consumption as L = labor
force = population) can be written

~ n
ct c~t Tt = f (k) ( + )k~ Kt
1
~ n n n
= f (k) ( + )k~ K0 e 1 t
(by gK = )
1 1
~ n 1 n Kt K1
= f (k) ( + )k~ ~ 0) 1 e 1
(kL t
(from k~ = = 0 )
1 Kt Lt L0
~ n n n
= f (k) ( + )k~ k~ 1 L0 1 e 1 t ~ 0 1 e1
'(k)L t
;
1

~ in the obvious way.

de…ning '(k)
We look for that value of k~ at which this steady-state path for ct is at the
n
highest technically feasible level. The positive coe¢ cient, L0 1 e 1 t , is the only
time dependent factor and can be ignored since it is exogenous. The problem is
thereby reduced to the static problem of maximizing '(k) ~ with respect to k~ > 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.5. Appendix 467

We …nd

~ = ~ n n
'0 (k) f 0 (k) ( + ) k~ 1 ~
+ f (k) ( + )k~ k~ 1 1
1 1 1
" ! #
n ~
f (k) n
~
= f 0 (k) ( + )+ ( + ) k~ 1
1 k~ 1 1
" #
~
f (k) n k~ 1
= (1 ~
)f 0 (k) (1 ) n+ ( + )
k~ 1 1
" #
~
f (k) n k~ 1 ~1
= (1 ~
)f 0 (k) + ~ k
(k) ; (11.61)
k~ 1 1 1

de…ning (k) ~ in the obvious way. The …rst-order condition for the problem,
~ = 0; is equivalent to (k)
'0 (k) ~ = 0: After ordering this gives

~ ~ 0 (k)
~
~ + f (k)
f 0 (k)
kf
=
n
: (11.62)
k~ 1
We see that
~ R 0 for
'0 (k) ~ R 0;
(k)
respectively. Moreover,
~
f (k) ~ 0 (k)
kf ~
0 ~ = (1
(k) ~
)f 00 (k) < 0;
k~2
~ k~ > f 0 (k):
in view of f 00 < 0 and f (k)= ~ So a k~ > 0 satisfying (k) ~ = 0 is the
unique maximizer of '(k):~ By (A1) and (A3) in Section 11.2.2 such a k~ exists
and is thereby the same as the k~GR we were looking for.
The left-hand side of (11.62) equals the social marginal productivity of capital
and the right-hand side equals the steady-state growth rate of output. At k~ = k~GR
it therefore holds that !
@Y Y_
= :
@K Y

This con…rms our claim in Section 11.2.2 about k~GR .

Remark about the absence of a golden rule in the Romer case. In the Romer case
the golden rule is not a well-de…ned concept for the following reason. Along any
balanced growth path we have from (11.50),

k_ t ct c0
gk = F (1; L) = F (1; L) ;
kt kt k0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

468 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

because gk (= gK ) is by de…nition constant along a balanced growth path, whereby

also ct =kt must be constant. We see that gk is decreasing linearly from F (1; L)
to when c0 =k0 rises from nil to F (1; L): So choosing among alternative techni-
cally feasible balanced growth paths is inevitably a choice between starting with
low consumption to get high growth forever or starting with high consumption
to get low growth forever. Given any k0 > 0; the alternative possible balanced
growth paths will therefore sooner or later cross each other in the (t; ln c) plane.
Hence, for the given k0 ; there exists no balanced growth path which for all t 0
has ct higher than along any other technically feasible balanced growth path.
B. Consumption taxation
Is a consumption tax distortionary - always? never? sometimes?
The answer is the following.
1. Suppose labor supply is elastic (due to leisure entering the utility func-
tion). Then a consumption tax (whether constant or time-dependent) is generally
distortionary (not neutral). This is because it reduces the e¤ective opportunity
cost of leisure by reducing the amount of consumption forgone by working one
hour less. Indeed, the tax makes consumption goods more expensive and so the
amount of consumption that the agent can buy for the hourly wage becomes
smaller. The substitution e¤ect on leisure of a consumption tax is thus positive,
while the income and wealth e¤ects will be negative. Generally, the net e¤ect
will not be zero, but can be of any sign; it may be small in absolute terms.
2. Suppose labor supply is inelastic (no trade-o¤ between consumption and
leisure). Then, at least in the type of growth models we consider in this course,
a constant (time-independent) consumption tax acts as a lump-sum tax and is
thus non-distortionary. If the consumption tax is time-dependent, however, a
distortion of the intertemporal aspect of household decisions tends to arise.
To understand answer 2, consider a Ramsey household with inelastic labor
supply. Suppose the household faces a time-varying consumption tax rate t > 0:
To obtain a consumption level per time unit equal to ct per capita, the household
has to spend
ct = (1 + t )ct
units of account (in real terms) per capita. Thus, spending ct per capita per time
unit results in the per capita consumption level
1
ct = (1 + t) ct : (11.63)
In order to concentrate on the consumption tax as such, we assume the tax
revenue is simply given back as lump-sum transfers and that there are no other
government activities. Then, with a balanced government budget, we have
xt Lt = t ct Lt ;

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

11.5. Appendix 469

where xt is the per capita lump-sum transfer, exogenous to the household, and
Lt is the size of the representative household.
Assuming CRRA utility with parameter > 0, the instantaneous per capita
utility can be written

c1t (1 + t)
1 1
ct
u(ct ) = = :
1 1
In our standard notation the household’s intertemporal optimization problem is
then to choose (ct )1
t=0 so as to maximize
Z 1 1 1
(1 + t) ct ( n)t
U0 = e dt s.t.
0 1
ct 0;
a_ t = (rt n)at + wt + xt ct ; a0 given,
R1
(rs n)ds
lim at e 0 0:
t!1

From now, we let the timing of the variables be implicit unless needed for
clarity. The current-value Hamiltonian is
1 1
(1 + ) c
H= + [(r n)a + w + x c] ;
1
where is the co-state variable associated with …nancial per capita wealth, a: An
interior optimal solution will satisfy the …rst-order conditions
@H 1 1
= (1 + ) c = 0; so that (1 + ) c = ; (FOC1)
@c
@H _ +(
= (r n) = n) ; (FOC2)
@a
and a transversality condition which amounts to
R1
(rs n)ds
lim at e 0 = 0: (TVC)
t!1

We take logs in (FOC1) to get

( 1) log(1 + ) log c = log :

Di¤erentiating w.r.t. time, taking into account that = t; gives

_ c _
( 1) = = r:
1+ c

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

470 CHAPTER 11. APPLICATIONS OF THE RAMSEY MODEL

By ordering, we …nd the growth rate of consumption spending,

c 1 _
= r+( 1) :
c 1+
Using (11.63), this gives the growth rate of consumption,
c_ c _ 1 _ _ 1 _
= = r+( 1) = (r ):
c c 1+ 1+ 1+ 1+
Assuming …rms maximize pro…t under perfect competition, in equilibrium the
real interest rate will satisfy
@Y
r= : (11.64)
@K
But the e¤ective real interest rate, r^; faced by the consuming household, is
_
r^ = r Q r for _ R 0;
1+
respectively. If for example the consumption tax is increasing, then the e¤ective
real interest rate faced by the consumer is smaller than the market real interest
rate, given in (11.64), because saving implies postponing consumption and future
consumption is more expensive due to the higher consumption tax rate.
The conclusion is that a time-varying consumption tax rate is distortionary.
It implies a wedge between the intertemporal rate of transformation faced by the
consumer, re‡ected by r^; and the intertemporal rate of transformation o¤ered by
the technology of society, indicated by r in (11.64). On the other hand, if the
consumption tax rate is constant, the consumption tax is non-distortionary when
there is no utility from leisure.
A remark on tax smoothing
Outside steady state it is often so that maintaining constant tax rates is incon-
sistent with maintaining a balanced government budget. Is the implication of
this that we should recommend the government to let tax rates be continually
adjusted so as to maintain a forever balanced budget? No! As the above exam-
ple as well as business cycle theory suggest, maintaining tax rates constant (“tax
smoothing”), and thereby allowing government de…cits and surpluses to arise, will
generally make more sense. In itself, a budget de…cit is not worrisome. It only
becomes worrisome if it is not accompanied later by su¢ cient budget surpluses
to avoid an exploding government debt/GDP ratio to arise. This requires that
the tax rates taken together have a level which in the long run matches the level
of government expenses.

11.6 Exercises

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Chapter 14

Fixed capital investment and

Tobin’s q

The models considered so far (the OLG models as well as the representative agent
models) have ignored capital adjustment costs. In the closed-economy version
of the models aggregate investment is merely a re‡ection of aggregate saving
and appears in a “passive” way as just the residual of national income after
households have chosen their consumption. We can describe what is going on by
telling a story in which …rms just rent capital goods owned by the households
and households save by purchasing additional capital goods. In these models
only households solve intertemporal decision problems. Firms merely demand
labor and capital services with a view to maximizing current pro…ts. This may
be a legitimate abstraction in some contexts within long-run analysis. In short-
and medium-run analysis, however, the dynamics of …xed capital investment is
important. So a more realistic approach is desirable.
In the real world the capital goods used by a production …rm are usually
owned by the …rm itself rather than rented for single periods on rental markets.
This is because inside the speci…c plant in which these capital goods are an
integrated part, they are generally worth much more than outside. So in practice
…rms acquire and install …xed capital equipment to maximize discounted expected
earnings in the future.
Tobin’s q-theory of investment (after the American Nobel laureate James To-
bin, 1918-2002) is an attempt to model these features. In this theory,

(a) …rms make the investment decisions and install the purchased capital goods
in their own businesses;

(b) there are certain adjustment costs associated with this investment: in ad-
dition to the direct cost of buying new capital goods there are costs of

573
 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
574 TOBIN’S Q

installation, costs of reorganizing the plant, costs of retraining workers to

operate the new machines etc.;
(c) the adjustment costs are strictly convex so that marginal adjustment costs
are increasing in the level of investment think of constructing a plant in
a month rather than a year.
The strict convexity of adjustment costs is the crucial constituent of the the-
ory. It is that element which assigns investment decisions an active role in the
model. There will be both a well-de…ned saving decision and a well-de…ned in-
vestment decision, separate from each other. Households decide the saving, …rms
the physical capital investment; households accumulate …nancial assets, …rms ac-
cumulate physical capital. As a result, in a closed economy interest rates have to
adjust for aggregate demand for goods (consumption plus investment) to match
aggregate supply of goods. The role of interest rate changes is no longer to clear
a rental market for capital goods.
To …x the terminology, from now the adjustment costs of setting up new
capital equipment in the …rm and the associated costs of reorganizing work
processes will be subsumed under the term capital installation costs. When faced
with strictly convex installation costs, the optimizing …rm has to take the fu-
ture into account, that is, …rms’forward-looking expectations become important.
To smooth out the adjustment costs, the …rm will adjust its capital stock only
gradually when new information arises. We thereby avoid the counterfactual im-
plication from earlier chapters that the capital stock in a small open economy with
perfect mobility of goods and …nancial capital is instantaneously adjusted when
the interest rate in the world …nancial market changes. Moreover, sluggishness in
investment is exactly what the data show. Some empirical studies conclude that
only a third of the di¤erence between the current and the “desired”capital stock
tends to be covered within a year (Clark 1979).
The q-theory of investment constitutes one approach to the explanation of this
sluggishness in investment. Under certain conditions, to be described below, the
theory gives a remarkably simple operational macroeconomic investment function,
in which the key variable explaining aggregate investment is the valuation of the
…rms by the stock market relative to the replacement value of the …rms’physical
capital. This link between asset markets and …rms’aggregate investment is an
appealing feature of Tobin’s q-theory.

14.1 Convex capital installation costs

Let the technology of a single …rm be given by
Y~ = F (K; L);

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 575

where Y~ ; K; and L are “potential output” (to be explained), capital input, and
labor input per time unit, respectively, while F is a concave neoclassical produc-
tion function. So we allow decreasing as well as constant returns to scale (or a
combination of locally CRS and locally DRS), whereas increasing returns to scale
is ruled out. Until further notice technological change is ignored for simplicity.
Time is continuous. The dating of the variables will not be explicit unless needed
for clarity. The increase per time unit in the …rm’s capital stock is given by

K_ = I K; > 0; (14.1)

where I is gross …xed capital investment per time unit and is the rate of wearing
down of capital (physical capital depreciation). To …x ideas, we presume the
realistic case with positive capital depreciation, but most of the results go through
even for = 0:
Let J denote the …rm’s capital installation costs (measured in units of output)
per time unit. The installation costs imply that a part of the potential output, Y~ ;
is “used up”in transforming investment goods into installed capital; only Y~ J
is “true output”available for sale.
Assuming the price of investment goods is one (the same as that of output
goods), then total investment costs per time unit are I +J; i.e., the direct purchase
costs, 1 I; plus the indirect cost associated with installation etc., J: The q-theory
of investment assumes that the capital installation cost, J; is a strictly convex
function of gross investment and is either independent of or a decreasing function
of the current capital stock. Thus,

J = G(I; K);

where the installation cost function G satis…es

G(0; K) = 0; GI (0; K) = 0; GII (I; K) > 0; and GK (I; K) 0 (14.2)

for all K and all (I; K); respectively. For …xed K = K the graph is as shown
in Fig. 14.1. Also negative gross investment, i.e., sell o¤ of capital equipment,
involves costs (for dismantling, reorganization etc.). Therefore GI < 0 for I < 0.
The important assumption is that GII > 0 (strict convexity in I); implying that
the marginal installation cost is increasing in the level of gross investment. If the
…rm wants to accomplish a given installation project in only half the time, then
the installation costs are more than doubled (the risk of mistakes is larger, the
problems with reorganizing work routines are larger etc.).
The strictly convex graph in Fig. 14.1 illustrates the essence of the matter.
Assume the current capital stock in the …rm is K and that the …rm wants to
increase it by a given amount K: If the …rm chooses the investment level I >

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
576 TOBIN’S Q

Figure 14.1: Installation costs as a function of gross investment when K = K.

0 per time unit in the time interval [t; t + t), then, in view of (14.1), K
(I K) t: So it takes t K=(I K) units of time to accomplish
the desired increase K. If, however, the …rm slows down the adjustment and
invests only half of I per time unit, then it takes approximately twice as long
time to accomplish K. Total costs of the two alternative courses of action are
approximately G(I; K) t and G( 12 I; K)2 t; respectively (ignoring discounting
and assuming the initial increase in capital is small in relation to K). By drawing
a few straight line segments in Fig. 14.1 the reader will be convinced that the
last-mentioned cost is smaller than the …rst-mentioned due to strict convexity of
installation costs (see Exercise 14.1). Haste is waste.

On the other hand, there are of course limits to how slow the adjustment
to the desired capital stock should be. Slower adjustment means postponement
of the potential bene…ts of a higher capital stock. So the …rm faces a trade-o¤
between fast adjustment to the desired capital stock and low adjustment costs.

In addition to the strict convexity of G with respect to I, (14.2) imposes

the condition GK (I; K) 0: Indeed, it often seems realistic to assume that
GK (I; K) < 0 for I 6= 0. A given amount of investment may require more
reorganization in a small …rm than in a large …rm (size here being measured
by K): When installing a new machine, a small …rm has to stop production
altogether, whereas a large …rm can to some extent continue its production by
shifting some workers to another production line. A further argument is that
the more a …rm has invested historically, the more experienced it is now. So,
for a given I today; the associated installation costs are lower, given a larger
accumulated K.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 577

14.1.1 The decision problem of the …rm

In the absence of tax distortions, asymmetric information, and problems with
enforceability of …nancial contracts, the Modigliani-Miller theorem (Modigliani
and Miller, 1958) says that the …nancial structure of the …rm is both indeter-
minate and irrelevant for production decisions (see Appendix A). Although the
conditions required for this theorem are very idealized, the q-theory of investment
accepts them because they allow the analyst to concentrate on the production
aspects in a …rst approach.
With the output good as unit of account, let the operating cash ‡ow (the net
payment stream to the …rm before interest payments on debt, if any) at time t
be denoted Rt (for “receipts”). Then

Rt F (Kt ; Lt ) G(It ; Kt ) wt Lt It ; (14.3)

where wt is the wage per unit of labor at time t. As mentioned, the installation
cost G(It ; Kt ) implies that a part of production, F (Kt ; Lt ); is used up in trans-
forming investment goods into installed capital; only the di¤erence F (Kt ; Lt )
G(It ; Kt ) is available for sale.
We ignore uncertainty and assume the …rm is a price taker. The interest rate
is rt , which we assume to be positive, at least in the long run. The decision
problem, as seen from time 0, is to choose a plan (Lt ; It )1 t=0 so as to maximize the
…rm’s market value, i.e., the present value of the future stream of expected cash
‡ows: Z 1 Rt
max1 V0 = Rt e 0 rs ds dt s.t. (14.3) and (14.4)
(Lt ;It )t=0 0

Lt 0; It free (i.e., no restriction on It ); (14.5)

K_ t = It Kt ; K0 > 0 given, (14.6)
Kt 0 for all t: (14.7)
There is no speci…c terminal condition but we have posited the feasibility condi-
tion (14.7) saying that the …rm can never have a negative capital stock.1
In the previous chapters the …rm was described as solving a series of static
pro…t maximization problems. Such a description is no longer valid, however,
when there is dependence across time, as is the case here. When installation
1
It is assumed that wt is a piecewise continuous function. At points of discontinuity (if
any) in investment, we will consider investment to be a right-continuous function of time.
That is, It0 = limt!t+ It : Likewise, at such points of discontinuity, by the “time derivative”
0

of the corresponding state variable, K; we mean the right-hand time derivative, i.e., K_ t0 =
limt!t+ (Kt Kt0 )=(t t0 ): Mathematically, these conventions are inconsequential, but they
0
help the intuition.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
578 TOBIN’S Q

costs are present, current decisions depend on the expected future circumstances.
The …rm makes a plan for the whole future so as to maximize the value of the …rm,
which is what matters for the owners. This is the general neoclassical hypothesis
about …rms’behavior. As shown in Appendix A, when strictly convex installation
costs or similar dependencies across time are absent, then value maximization is
equivalent to solving a sequence of static pro…t maximization problems, and we
are back in the previous chapters’description.
To solve the problem (14.4) (14.7), where Rt is given by (14.3), we apply
the Maximum Principle. The problem has two control variables, L and I, and
one state variable, K. We set up the current-value Hamiltonian:

H(K; L; I; q; t) F (K; L) wL I G(I; K) + q(I K); (14.8)

where q (to be interpreted economically below) is the adjoint variable associated

with the dynamic constraint (14.6). For each t 0 we maximize H w.r.t. the
control variables. Thus, @H=@L = FL (K; L) w = 0; i.e.,

FL (K; L) = w; (14.9)

and @H=@I = 1 GI (I; K) + q = 0; i.e.,

1 + GI (I; K) = q: (14.10)

Next, we partially di¤erentiate H w.r.t. the state variable and set the result equal
to rq q,_ where r is the discount rate in (14.4):
@H
= FK (K; L) GK (I; K) q = rq q:
_ (14.11)
@K
Then, the Maximum Principle says that for an interior optimal path (Kt ; Lt ; It )
there exists an adjoint variable q; which is a continuous function of t; written qt ;
such that for all t 0 the conditions (14.9), (14.10), and (14.11) hold and the
transversality condition Rt
lim Kt qt e 0 rs ds = 0 (14.12)
t!1

is satis…ed.
The optimality condition (14.9) is the usual employment condition equalizing
the marginal product of labor to the real wage. In the present context with
strictly convex capital installation costs, this condition attains a distinct role as
labor will in the short run be the only variable input. This is because the strictly
convex capital installation costs imply that the …rm’s installed capital in the
short run is a quasi-…xed production factor. So, e¤ectively there are diminishing
returns (equivalent with rising marginal costs) in the short run even though the
production function might have CRS.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 579

The left-hand side of (14.10) gives the cost of acquiring one extra unit of
installed capital at time t (the sum of the cost of buying the marginal investment
good and the cost of its installation). That is, the left-hand side is the marginal
cost, MC, of increasing the capital stock in the …rm. Since (14.10) is a necessary
condition for optimality, the right-hand side of (14.10) must be the marginal
bene…t, MB, of increasing the capital stock. Hence, qt represents the value to
the optimizing …rm of having one more unit of (installed) capital at time t: To
put it di¤erently: the adjoint variable qt can be interpreted as the shadow price
(measured in current output units) of capital along the optimal path.2
As to the interpretation of the di¤erential equation (14.11), a condition for
optimality must be that the …rm acquires capital up to the point where the
“marginal productivity of capital”, FK GK ; equals “capital costs”, rt qt + ( qt
q_t ); the …rst term in this expression represents interest costs and the second
economic depreciation. In (14.11) the “marginal productivity of capital”appears
as FK GK ; because we should take into account the potential reduction, GK ; of
installation costs in the next instant brought about by the marginal unit of already
installed capital. The shadow price qt appears as the “overall”price at which the
…rm can buy and sell the marginal unit of installed capital. In fact, in view of qt =
1 + GI (Kt ; Lt ) along the optimal path (from (14.10)), qt measures, approximately,
both the “overall” cost increase associated with increasing investment by one
unit and the “overall” cost saving associated with decreasing investment by one
unit. In the …rst case the …rm not only has to pay one extra unit of account
in the investment goods market but must also bear an installation cost equal to
GI (Kt ; Lt ); thereby in total investing qt units of account: And in the second case
the …rm recovers qt by saving both on installation costs and purchases in the
investment goods market. Continuing along this line of thought, by reordering in
(14.11) we get the “no-arbitrage”condition

FK GK q + q_
= r; (14.13)
q

saying that along the optimal path the rate of return on the marginal unit of
installed capital must equal the interest rate.
The transversality condition (14.12) says that the present value of the capital
stock “left over” at in…nity must be zero. That is, the capital stock should not
in the long run grow too fast, given the evolution of its discounted shadow price.
In addition to necessity of (14.12) it can be shown3 that the discounted shadow
2
Recall that a shadow price, measured in some unit of account, of a good, from the point of
view of the buyer, is the maximum number of units of account that he or she is willing to o¤er
for one extra unit of the good.
3
See Appendix B.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
580 TOBIN’S Q

price itself in the far future must along an optimal path be asymptotically nil,
i.e.,
Rt
rs ds
lim qt e 0 = 0: (14.14)
t!1

If along the optimal path, Kt grows without bound, then not only must (14.14)
hold but, in view of (14.12), the discounted shadow price must in the long run
approach zero faster than Kt grows. Intuitively, otherwise the …rm would be
“over-accumulating”. The …rm would gain by reducing the capital stock “left
over” for eternity (which is like“money left on the table”), since reducing the
ultimate investment and installation costs would raise the present value of the
…rm’s expected cash ‡ow.
In connection with (14.10) we claimed that qt can be interpreted as the shadow
price (measured in current output units) of capital along the optimal path. A
con…rmation of this interpretation isR obtained by solving the di¤erential equation
t
(14.11). Indeed, multiplying by e 0 (rs + )ds on both sides of (14.11), we get by
integration and application of (14.14),4
Z 1 R
(rs + )ds
qt = [FK (K ; L ) GK (I ; K )] e t d : (14.15)
t

The right-hand side of (14.15) is the present value, as seen from time t; of expected
future increases of the …rm’s cash-‡ow that would result if one extra unit of
capital were installed at time t; indeed, FK (K ; L ) is the direct contribution
to output of one extra unit of capital, while GK (I ; K ) 0 represents the
potential reduction of installation costs in the next instant brought about by the
marginal unit of installed capital. However, future increases of cash-‡ow should
be discounted at a rate equal to the interest rate plus the capital depreciation
rate; from one extra unit of capital at time t there are only e ( t) units left at
time :
To concretize our interpretation of qt as representing the value to the opti-
mizing …rm at time t of having one extra unit of installed capital, let us make
a thought experiment. Assume that a extra units of installed capital at time t
drops down from the sky. At time > t there are a e ( t) units of these still
in operation so that the stock of installed capital is

K0 = K + a e ( t)
; (14.16)

where K denotes the stock of installed capital as it would have been without
this “injection”. Now, in (14.3) replace t by and consider the optimizing …rm’s
4
For details, see Appendix A.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 581

cash-‡ow R as a function of (K ; L ; I ; ; t; a). Taking the partial derivative of

R w.r.t. a at the point (K ; L ; I ; ; t; 0); we get

@R ( t)
= [FK (K ; L ) GK (I ; K )] e : (14.17)
@a ja=0

Considering the value of the optimizing …rm at time t as a function of installed

capital, Kt ; and t itself, we denote this function V (Kt ; t). Then at any point
where V is di¤erentiable, we have
Z 1 R
@V (Kt ; t) @R rs ds
= e t d
@Kt t @a ja=0
Z 1 R
(rs + )ds
= [FK (K ; L ) GK (I ; K )]e t d = qt (14.18)
t

when the …rm moves along the optimal path. The second equality sign comes
from (14.17) and the third is implied by (14.15). So the value of the adjoint
variable, q, at time t equals the contribution to the …rm’s maximized value of a
…ctional marginal “injection” of installed capital at time t: This is just another
way of saying that qt represents the bene…t to the …rm of the marginal unit of
installed capital along the optimal path.
This story facilitates the understanding that the control variables at any point
in time should be chosen so that the Hamiltonian function is maximized. Thereby
one maximizes the properly weighted sum of the current direct contribution to the
criterion function and the indirect contribution, which is the bene…t (as measured
approximately by qt Kt ) of having a higher capital stock in the future.
As we know, the Maximum Principle gives only necessary conditions for an
optimal path, not su¢ cient conditions. We use the principle as a tool for …nding
candidates for a solution. Having found in this way a candidate, one way to pro-
ceed is to check whether Mangasarian’s su¢ cient conditions are satis…ed. Given
the transversality condition (14.12) and the non-negativity of the state variable,
K; the only additional condition to check is whether the Hamiltonian function
is jointly concave in the endogenous variables (here K, L; and I): If it is jointly
concave in these variables, then the candidate is an optimal solution. Owing
to concavity of F (K; L), inspection of (14.8) reveals that the Hamiltonian func-
tion is jointly concave in (K; L; I) if G(I, K) is jointly concave in (I; K): This
condition is equivalent to G(I; K) being jointly convex in (I; K); an assumption
allowed within the con…nes of (14.2); for example, G(I; K) = ( 12 ) I 2 =K as well as
the simpler G(I; K) = ( 12 ) I 2 (where in both cases > 0) will do. Thus, assum-
ing joint convexity of G(I; K); the …rst-order conditions and the transversality
condition are not only necessary, but also su¢ cient for an optimal solution.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
582 TOBIN’S Q

14.1.2 The implied investment function

From condition (14.10) we can derive an investment function. Rewriting (14.10),
we have that an optimal path satis…es

GI (It ; Kt ) = qt 1: (14.19)

Combining this with the assumption (14.2) on the installation cost function, we
see that
It T 0 for qt T 1; respectively, (14.20)
cf. Fig. 14.2.5 In view of GII 6= 0; (14.19) implicitly de…nes optimal investment,
It , as a function of the shadow price, qt ; and the state variable, Kt :

It = M(qt ; Kt ); (14.21)

where, in view of (14.20), M(1; Kt ) = 0. By implicit di¤erentiation w.r.t. qt and

Kt ; respectively, in (14.19), we …nd

@It 1 @It GIK (It ; Kt )

= > 0; and = ,
@qt GII (It ; Kt ) @Kt GII (It ; Kt )

where the latter cannot be signed without further speci…cation.

It follows that optimal investment is an increasing function of the shadow
price of installed capital. In view of (14.20), M(1; K) = 0. Not surprisingly, the
investment rule is: invest now, if and only if the value to the …rm of the marginal
unit of installed capital is larger than the price of the capital good (which is
1, excluding installation costs). At the same time, the rule says that, because
of the convex installation costs, invest only up to the point where the marginal
installation cost, GI (It ; Kt ), equals qt 1; cf. (14.19):
Condition (14.21) shows the remarkable information content that the shadow
price qt has. As soon as qt is known (along with the current capital stock Kt ),
the …rm can decide the optimal level of investment through knowledge of the
installation cost function G alone (since, when G is known, so is in principle the
inverse of GI w.r.t. I, the investment function M). All the information about the
production function, input prices, and interest rates now and in the future that
is relevant to the investment decision is summarized in one number, qt : The form
of the investment function, M; depends only on the installation cost function G:
These are very useful properties in theoretical and empirical analysis.
5
From the assumptions made in (14.2), we only know that the graph of GI (I; K) is an
upward-sloping curve going through the origin. Fig. 14.2 shows the special case where this
curve happens to be linear.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 583

Figure 14.2: Marginal installation costs as a function of the gross investment level, I,
for a given amount, K, of installed capital. The optimal gross investment, It , when
q = qt is indicated.

14.1.3 A not implausible special case

We now introduce the convenient case where the installation function G is homo-
geneous of degree one w.r.t. I and K so that we can, for K > 0; write
I I
J = G(I; K) = G( ; 1)K g( )K; or (14.22)
K K
J I
= g( );
K K
where g( ) represents the installation cost-capital ratio and g(0) G(0; 1) = 0;
by (14.2):
LEMMA 1 The function g( ) has the following properties:
(i) g 0 (I=K) = GI (I; K);
(ii) g 00 (I=K) = GII (I; K)K > 0 for K > 0; and
(iii) g(I=K) g 0 (I=K)I=K = GK (I; K) < 0 for I 6= 0:
Proof. (i) GI = Kg 0 =K = g 0 ; (ii) GII = g 00 =K; (iii) GK = @(g(I=K)K)=@K
= g(I=K) g 0 (I=K)I=K < 0 for I 6= 0 since, in view of g 00 > 0 and g(0) = 0; we
have g(x)=x < g 0 (x) for all x 6= 0:
The graph of g(I=K) is qualitatively the same as that in Fig. 14.1 (imagine we
have K = 1 in that graph). The installation cost relative to the existing capital
stock is now a strictly convex function of the investment-capital ratio, I=K:
EXAMPLE 1 Let J = G(I; K) = 12 I 2 =K; where > 0: Then G is homogeneous
of degree one w.r.t. I and K and gives J=K = 12 (I=K)2 g(I=K):

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
584 TOBIN’S Q

A further important property of (14.22) is that the cash-‡ow function in (14.3)

becomes homogeneous of degree one w.r.t. K; L; and I in the “normal”case where
the production function has CRS: This has two implications. First, Hayashi’s
theorem applies (see below). Second, the q-theory can easily be incorporated
into a model of economic growth.6
Does the hypothesis of linear homogeneity of the cash ‡ow in K; L; and I
make economic sense? According to the replication argument it does. Suppose a
given …rm has K units of installed capital and produces Y units of output with
L units of labor. When at the same time the …rm invests I units of account
in new capital, it obtains the cash ‡ow R after deducting the installation costs,
G(I; K): Then it makes sense to assume that the …rm could do the same thing at
another place, hereby doubling its cash-‡ow. (Of course, owing to the possibility
of indivisibilities, this reasoning does not take us all the way to linear homogeneity.
Moreover, the argument ignores that also land is a necessary input. As discussed
in Chapter 2, the empirical evidence on linear homogeneity is mixed.)
In view of (i) of Lemma 1, the linear homogeneity assumption for G allows us
to write (14.19) as
g 0 (I=K) = q 1: (14.23)
This equation de…nes the investment-capital ratio, I=K , as an implicit function,
m; of q :
It 1
= m(qt ); where m(1) = 0 and m0 = 00 > 0; (14.24)
Kt g
by implicit di¤erentiation in (14.23). In this case q encompasses all information
that is of relevance to the decision about the investment-capital ratio.
In Example 1 above we have g(I=K) = 12 (I=K)2 ; in which case (14.23) gives
I=K = (q 1)= . So in this case we have m(q) = q= 1= ; a linear investment
function, as illustrated in Fig. 14.3: The parameter can be interpreted as
the degree of sluggishness in the capital adjustment. The degree of sluggishness
re‡ects the degree of convexity of installation costs.7 The stippled lines in Fig.
14.3 are explained below. Generally the graph of the investment function is
positively sloped, but not necessarily linear.
To see how the shadow price q changes over time along the optimal path, we
rearrange (14.11):
q_t = (rt + )qt FK (Kt ; Lt ) + GK (It ; Kt ): (14.25)
6
The relationship between the function g and other ways of formulating the theory is com-
mented on in Appendix C.
7
For a twice di¤erentiable function, f (x); with f 0 (x) 6= 0; we de…ne the degree of convexity
in the point x by f 00 (x)=f 0 (x): So the degree of convexity of g(I=K) is g 00 =g 0 = (I=K) 1
= (q 1) 1 and thereby we have = (q 1)g 00 =g 0 : So, for given q; the degree of sluggishness
is proportional to the degree of convexity of adjustment costs.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.1. Convex capital installation costs 585

Figure 14.3: Optimal investment-capital ratio as a function of the shadow price of

installed capital when g(I=K) = 21 (I=K)2 .

Recall that GK (It ; Kt ) indicates how much lower the installation costs are as
a result of the marginal unit of installed capital. In the special case (14.22) we
have from Lemma 1

I I I
GK (I; K) = g( ) g0( ) = g(m(q)) (q 1)m(q);
K K K

using (14.24) and (14.23).

Inserting this into (14.25) gives

q_t = (rt + )qt FK (Kt ; Lt ) + g(m(qt )) (qt 1)m(qt ): (14.26)

This di¤erential equation is very useful in macroeconomic analysis, as we will

soon see, cf. Fig. 14.4 below.
In a macroeconomic context, for steady state to achievable, gross investment
must be large enough to match not only capital depreciation, but also growth in
the labor input. Otherwise a constant capital-labor ratio can not be sustained.
That is, the investment-capital ratio, I=K; must be equal to the sum of the
depreciation rate and the growth rate of the labor force, i.e., + n: The level of q
which is required to motivate such an investment-capital ratio is called q in Fig.
14.3.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
586 TOBIN’S Q

14.2 Marginal q and average q

Our q above, determining investment, should be distinguished from what is usu-
ally called Tobin’s q or average q. In a more general context, let pIt denote
the current purchase price (in terms of output units) per unit of the invest-
ment good (before installment). Then Tobin’s q or average q; qta ; is de…ned as
qta Vt =(pIt Kt ); that is, Tobin’s q is the ratio of the market value of the …rm to
the replacement value of the …rm in the sense of the “reacquisition value of the
capital goods before installment costs”(the top index “a”stands for “average”).
In our simpli…ed context we have pIt 1 (the price of the investment good is the
same as that of the output good). Therefore Tobin’s q can be written
Vt V (Kt ; t)
qta = ; (14.27)
Kt Kt
where the equality holds for an optimizing …rm. Conceptually this is di¤erent
from the …rm’s internal shadow price on capital, i.e., what we have denoted qt
in the previous sections. In the language of the q-theory of investment this qt is
the marginal q, representing the value to the …rm of one extra unit of installed
capital relative to the price of un-installed capital equipment. The term marginal
q is natural since along the optimal path, as a slight generalization of (14.18), we
must have qt = (@V =@Kt )=pIt : Letting qtm (“m”for “marginal”) be an alternative
symbol for this qt ; we have in our model above, where we consider the special
case pIt 1;
@V
qtm qt = : (14.28)
@Kt
The two concepts, average q and marginal q, have not always been clearly dis-
tinguished in the literature. What is directly relevant to the investment decision
is marginal q. Indeed, the analysis above showed that optimal investment is an
increasing function of q m . Further, the analysis showed that a “critical”value of
q m is 1 and that only if q m > 1; is positive gross investment warranted.
The importance of q a is that it can be measured empirically as the ratio of the
sum of the share market value of the …rm and its debt to the current acquisition
value of its total capital before installment. Since q m is much harder to measure
than q a , it is important to know the relationship between q m and q a . Fortunately,
we have a simple theorem giving conditions under which q m = q a .
THEOREM (Hayashi, 1982) Assume the …rm is a price taker, that the production
function F is jointly concave in (K; L); and that the installation cost function G
is jointly convex in (I; K):8 Then, along an optimal path we have:
8
That is, in addition to (14.2), we assume GKK 0 and GII GKK G2IK 0: The speci…-
cation in Example 1 above satis…es this.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.3. Applications 587

(i) qtm = qta for all t 0; if F and G are homogeneous of degree 1.

(ii) qtm < qta for all t; if F is strictly concave in (K; L) and/or G is strictly
convex in (I; K):
Proof. See Appendix D.
The assumption that the …rm is a price taker may, of course, seem critical.
The Hayashi theorem has been generalized, however. Also a monopolistic …rm,
facing a downward-sloping demand curve and setting its own price, may have a
cash ‡ow which is homogeneous of degree one in the three variables K; L; and I:
If so, then the condition qtm = qta for all t 0 still holds (Abel 1990). Abel and
Eberly (1994) present further generalizations.
In any case, when q m is approximately equal to (or just proportional to)
q a , the theory gives a remarkably simple operational investment function, I =
m(q a )K; cf. (14.24). At the macro level we interpret q a as the market valuation
of the …rms relative to the replacement value of their total capital stock. This
market valuation is an indicator of the expected future earnings potential of the
…rms. Under the conditions in (i) of the Hayashi theorem the market valuation
also indicates the marginal earnings potential of the …rms, hence, it becomes a
determinant of their investment. This establishment of a relationship between the
stock market and …rms’aggregate investment is the basic point in Tobin (1969).

14.3 Applications
Capital installation costs in a closed economy
Allowing for convex capital installation costs in the economy has far-reaching
implications for the causal structure of a model of a closed economy. Investment
decisions attain an active role in the economy and forward-looking expectations
become important for these decisions. Expected future market conditions and an-
nounced future changes in corporate taxes and depreciation allowance will a¤ect
…rms’investment already today.
The essence of the matter is that current and expected future interest rates
have to adjust for aggregate saving to equal aggregate investment, that is, for the
output and asset markets to clear. Given full employment (Lt = Lt ); the output
market clears when

F (Kt ; Lt ) G(It ; Kt ) = value added GDPt = Ct + It ;

where Ct is determined by the intertemporal utility maximization of the forward-

looking households, and It is determined by the intertemporal value maximization
of the forward-looking …rms facing strictly convex installation costs. Like in the
determination of Ct ; current and expected future interest rates now also matter

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
588 TOBIN’S Q

for the determination of It : This is the …rst time in this book where clearing in the
output market is assigned an active role. In the earlier models investment was just
a passive re‡ection of household saving. Desired investment was automatically
equal to the residual of national income left over after consumption decisions had
taken place. Nothing had to adjust to clear the output market, neither interest
rates nor output. In contrast, in the present framework adjustments in interest
rates and/or the output level are needed for the continuous clearing in the output
market and these adjustments are decisive for the macroeconomic dynamics.
In actual economies there may of course exist “secondary markets” for used
capital goods and markets for renting capital goods owned by others. In view of
installation costs and similar, however, shifting capital goods from one plant to
another is generally costly. Therefore the turnover in that kind of markets tends
to be limited and there is little underpinning for the earlier models’supposition
that the current interest rate should be tied down by a requirement that such
markets clear.
In for instance Abel and Blanchard (1983) a Ramsey-style model integrating
the q-theory of investment is presented. The authors study the two-dimensional
general equilibrium dynamics resulting from the adjustment of current and ex-
pected future (short-term) interest rates needed for the output market to clear.
Adjustments of the whole structure of interest rates (the yield curve) take place
and constitute the equilibrating mechanism in the output and asset markets.
By having output market equilibrium playing this role in the model, a …rst
step is taken towards medium- and short-run macroeconomic theory. We take
further steps in later chapters, by allowing imperfect competition and nominal
price rigidities to enter the picture. Then the demand side gets an active role
both in the determination of q (and thereby investment) and in the determination
of aggregate output and employment. This is what Keynesian theory (old and
new) deals with.
In the remainder of this chapter we will still assume perfect competition in all
markets including the labor market. In this sense we will stay within the neoclas-
sical framework (supply-dominated models) where, by instantaneous adjustment
of the real wage, labor demand continuously matches labor supply. The next
two subsections present examples of how Tobin’s q-theory of investment can be
integrated into the neoclassical framework. To avoid the more complex dynamics
arising in a closed economy, we shift the focus to a small open economy. This
allows concentrating on a dynamic system with an exogenous interest rate.

A small open economy with capital installation costs

By introducing convex capital installation costs in a model of a small open econ-
omy (SOE), we avoid the counterfactual outcome that the capital stock adjusts

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.3. Applications 589

instantaneously when the interest rate in the world …nancial market changes.
In the standard neoclassical growth model for a small open economy, without
convex capital installation costs, a rise in the interest rate leads immediately to
a complete adjustment of the capital stock so as to equalize the net marginal
productivity of capital to the new higher interest rate. Moreover, in that model
expected future changes in the interest rate or in corporate taxes and deprecia-
tion allowances do not trigger an investment response until these changes actually
happen. In contrast, when convex installation costs are present, expected future
changes tend to in‡uence …rms’investment already today.
We assume:

1. Perfect mobility across borders of goods and …nancial capital.

2. Domestic and foreign …nancial claims are perfect substitutes.

3. No mobility across borders of labor.

4. Labor supply is inelastic and constant and there is no technological progress.

5. The capital installation cost function G(I; K) is homogeneous of degree 1.

In this setting the SOE faces an exogenous interest rate, r; given from the
world …nancial market. We assume r is a positive constant. The aggregate pro-
duction function, F (K; L); is neoclassical and concave as in the previous sections.
With L > 0 denoting the constant labor supply, continuous clearing in the labor
market under perfect competition gives Lt = L for all t 0 and

wt = FL (Kt ; L) w(Kt ): (14.29)

At any time t; Kt is predetermined in the sense that due to the convex installation
costs, changes in K take time. Thus (14.29) determines the market real wage wt :
To pin down the evolution of the economy, we now derive two coupled di¤er-
ential equations in K and q. Inserting (14.24) into (14.6) gives

K_ t = (m(qt ) )Kt ; K0 > 0 given: (14.30)

As to the dynamics of q; we have (14.26). Since the capital installation cost

function G(I; K) is assumed to be homogeneous of degree 1, point (iii) of Lemma
1 applies and we can write (14.26) as

q_t = (r + )qt FK (Kt ; L) + g(m(qt )) (qt 1)m(qt ): (14.31)

As r and L are exogenous, the capital stock, K; and its shadow price, q; are
the only endogenous variables in the di¤erential equations (14.30) and (14.31).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
590 TOBIN’S Q

Figure 14.4: Phase diagram for investment dynamics in a small open economy (a case
where > 0).

In addition, we have an initial condition for K and a necessary transversality

condition involving q, namely
rt
lim Kt qt e = 0: (14.32)
t!1

Fig. 14.4 shows the phase diagram for these two coupled di¤erential equations.
Let q be de…ned as the value of q satisfying the equation m(q) = : Since m0 > 0,
q is unique. Suppressing for convenience the explicit time subscripts, we then
have
K_ = 0 for m(q) = ; i.e., for q = q :
As > 0; we have q > 1: This is so because also mere reinvestment to o¤set
capital depreciation requires an incentive, namely that the marginal value to
the …rm of replacing worn-out capital is larger than the purchase price of the
investment good (since the installation cost must also be compensated). From
(14.30) is seen that

K_ ? 0 for m(q) ? ; respectively, i.e., for q ? q ; respectively,

cf. the horizontal arrows in Fig. 14.4.

From (14.31) we have

q_ = 0 for 0 = (r + )q FK (K; L) + g(m(q)) (q 1)m(q): (14.33)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.3. Applications 591

If, in addition K_ = 0 (hence, q = q and m(q) = m(q ) = ), this gives

0 = (r + )q FK (K; L) + g( ) (q 1) ; (14.34)

where the right-hand-side is increasing in K; in view of FKK < 0. Hence, there

exists at most one value of K such that the steady state condition (14.34) is
satis…ed;9 this value is denoted K ; corresponding to the steady state point E
in Fig. 14.4. The question is now: what is the slope of the q_ = 0 locus? In
Appendix E it is shown that at least in a neighborhood of the steady state point
E, this slope is negative in view of the assumption r > 0 and FKK < 0. From
(14.31) we see that

q_ 7 0 for points to the left and to the right, respectively, of the q_ = 0 locus,

since FKK (Kt ; L) < 0: The vertical arrows in Fig. 14.4 show these directions of
movement.
Altogether the phase diagram shows that the steady state E is a saddle point,
and since there is one predetermined variable, K; and one jump variable, q; and
the saddle path is not parallel to the jump variable axis, the steady state is
saddle-point stable. At time 0 the economy will be at the point B in Fig. 14.4
where the vertical line K = K0 crosses the saddle path. Then the economy
will move along the saddle path towards the steady state. This solution satis…es
the transversality condition (14.32) and is the unique solution to the model (for
details, see Appendix F).

The e¤ect of an unanticipated rise in the interest rate Suppose that

until time 0 the economy has been in the steady state E in Fig. 14.4. Then,
an unexpected shift in the interest rate occurs so that the new interest rate is
a constant r0 > r: We assume that the new interest rate is rightly expected to
remain at this level forever. From (14.30) we see that q is not a¤ected by this
shift, hence, the K_ = 0 locus is not a¤ected. However, (14.33) implies that the
q_ = 0 locus and K shift to the left, in view of FKK (K; L) < 0:
Fig. 14.5 illustrates the situation for t > 0: At time t = 0 the shadow price q
jumps down to a level corresponding to the point B in Fig. 14.5. There is now
a heavier discounting of the future bene…ts that the marginal unit of capital can
provide. As a result the incentive to invest is diminished and gross investment
will not even compensate for the depreciation of capital. Hence, the capital
stock decreases gradually. This is where we see a crucial role of convex capital
installation costs in an open economy. For now, the installation costs are the costs
9
And assuming that F satis…es the Inada conditions, we are sure that such a value exists
since (14.34) gives FK (K; L) = rq + g( ) + > 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
592 TOBIN’S Q

Figure 14.5: Phase portrait of an unanticipated rise in r (the case > 0).

associated with disinvestment (dismantling and selling out of machines). If these

convex costs were not present, we would get the same counterfactual prediction
as from the previous open-economy models in this book, namely that the new
steady state is attained immediately after the shift in the interest rate.
As the capital stock is diminished, the marginal productivity of capital rises
and so does q: The economy moves along the new saddle path and approaches
the new steady state E’ as time goes by.
Suppose that for some reason such a decrease in the capital stock is not
desirable from a social point of view; this could be because of positive external
e¤ects of capital and investment, e.g., a kind of “learning by doing”. Then the
government could decide to implement an investment subsidy ; 0 < < 1; so
that to attain an investment level I; purchasing the investment goods involves a
cost of (1 )I: Assuming the subsidy is …nanced by some tax not a¤ecting …rms’
behavior (for example a constant tax on households’consumption), investment is
increased again and the economy may in the long run end up at the old steady-
state level of K (but the new q will be lower than the old).

A growing small open economy with capital installation costs*

The basic assumptions are the same as in the previous section except that now
labor supply, Lt , grows at the constant rate n 0; while the technology level, T;
grows at the constant rate 0 (both rates exogenous and constant) and the
production function is neoclassical with CRS. We assume that the world market
real interest rate, r; is a constant and satis…es r > + n: Still assuming full
employment, we have Lt = Lt = L0 ent :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.3. Applications 593

In this setting the production function on intensive form is useful:

K ~ L;
Y = F (K; T L) = F ( ; 1)T L f (k)T
TL
where k~ K=(T L) and f satis…es f 0 > 0 and f 00 < 0: Still assuming perfect
competition, the market-clearing real wage at time t is determined as
h i
wt = F2 (Kt ; Tt Lt )Tt = f (k~t ) k~t f 0 (k~t ) Tt w(
~ k~t )Tt ;

where both k~t and Tt are predetermined. By log-di¤erentiation of k~ K=(T L)

w.r.t. time we get k~t =k~t = K_ t =Kt ( + n): Substituting (14.30), we get

k~t = [m(qt ) ( + + n)] k~t : (14.35)

The change in the shadow price of capital is now described by

q_t = (r + )qt f 0 (k~t ) + g(m(qt )) (qt 1)m(qt ); (14.36)

from (14.26). In addition, the transversality condition,

lim k~t qt e (r n)t

= 0; (14.37)
t!1

must hold.
The di¤erential equations (14.35) and (14.36) constitute our new dynamic
system. Fig. 14.6 shows the phase diagram, which is qualitatively similar to that
in Fig. 14.4. We have

k~ = 0 for m(q) = + + n; i.e., for q = q ;

where q now is de…ned by the requirement m(q ) = + + n: Notice, that when

+ n > 0; we get a larger steady state value q than in the previous section. This
is so because now a higher investment-capital ratio is required for a steady state
to be possible. Moreover, the transversality condition (14.12) is satis…ed in the
steady state.
From (14.36) we see that q_ = 0 now requires

0 = (r + )q ~ + g(m(q))
f 0 (k) (q 1)m(q):

If, in addition k~ = 0 (hence, q = q and m(q) = m(q ) = + + n), this gives

0 = (r + )q ~ + g( +
f 0 (k) + n) (q 1)( + + n):

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
594 TOBIN’S Q

Figure 14.6: Phase portrait of an unanticipated fall in r (a growing economy with

+ +n + n > 0).

Here, the right-hand-side is increasing in k~ (in view of f 00 (k)

~ < 0). Hence, the
steady state value k~ of the e¤ective capital-labor ratio is unique, cf. the steady
state point E in Fig. 14.6.
By the assumption r > + n we have, at least in a neighborhood of E in
Fig. 14.6, that the q_ = 0 locus is negatively sloped (see Appendix E).10 Again
the steady state is a saddle point, and the economy moves along the saddle path
towards the steady state.
In Fig. 14.6 it is assumed that until time 0, the economy has been in the
steady state E. Then, an unexpected shift in the interest rate to a lower constant
level, r0 , takes place. The q_ = 0 locus is shifted to the right, in view of f 00 < 0:
The shadow price, q; immediately jumps up to a level corresponding to the point
B in Fig. 14.6. The economy moves along the new saddle path and approaches
the new steady state E’ with a higher e¤ective capital-labor ratio as time goes
by. In Exercise 14.2 the reader is asked to examine the analogue situation where
an unanticipated downward shift in the rate of technological progress takes place.

10
In our perfect foresight model we in fact have to assume r > +n for the …rm’s maximization
problem to be well-de…ned. If instead r + n; the market value of the representative …rm
would be in…nite, and maximization would loose its meaning.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.4. Concluding remarks 595

14.4 Concluding remarks

Tobin’s q-theory of investment gives a remarkably simple operational macroeco-
nomic investment function, in which the key variable explaining aggregate invest-
ment is the valuation of the …rms by the stock market relative to the replacement
value of the …rms’physical capital. This link between asset markets and …rms’
aggregate investment is an appealing feature of Tobin’s q-theory.
When faced with strictly convex installation costs, the …rm has to take the
future into account to invest optimally. Therefore, the …rm’s expectations be-
come important. Owing to the strictly convex installation costs, the …rm adjusts
its capital stock only gradually when new information arises. This investment
smoothing is analogue to consumption smoothing.
By incorporating these features, Tobin’s q-theory helps explaining the slug-
gishness in investment we see in the empirical data. And the theory avoids the
counterfactual outcome from earlier chapters that the capital stock in a small
open economy with perfect mobility of goods and …nancial capital is instanta-
neously adjusted when the interest rate in the world market changes. So the
theory takes into account the time lags in capital adjustment in real life, a fea-
ture which may, perhaps, be abstracted from in long-run analysis and models of
economic growth, but not in short- and medium-run analysis.
Many econometric tests of the q theory of investment have been made, often
with quite critical implications. Movements in q a , even taking account of changes
in taxation, seemed capable of explaining only a minor fraction of the movements
in investment. And the estimated equations relating …xed capital investment
to q a typically give strong auto-correlation in the residuals. Other variables, in
particular availability of current corporate pro…ts for internal …nancing, seem
to have explanatory power independently of q a (see Abel 1990, Chirinko 1993,
Gilchrist and Himmelberg, 1995). So there is reason to be somewhat sceptical
towards the notion that all information of relevance for the investment decision
is re‡ected by the market valuation of …rms. This throws doubt on the basic
assumption in Hayashi’s theorem or its generalization, the assumption that …rms’
cash ‡ow tends to be homogeneous of degree one w.r.t. K; L; and I:
Going outside the model, there are further circumstances relaxing the link
between q a and investment. In the real world with many production sectors,
physical capital is heterogeneous. If for example a sharp unexpected rise in the
price of energy takes place, a …rm with energy-intensive technology will loose in
market value. At the same time it has an incentive to invest in energy-saving
capital equipment. Hence, we might observe a fall in q a at the same time as
investment increases.
Imperfections in credit markets are ignored by the model. Their presence

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
596 TOBIN’S Q

further loosens the relationship between q a and investment and may help explain
the observed positive correlation between investment and corporate pro…ts.
We might also question that capital installation costs really have the hy-
pothesized strictly convex form. It is one thing that there are costs associated
with installation, reorganizing and retraining etc., when new capital equipment
is procured. But should we expect these costs to be strictly convex in the vol-
ume of investment? To think about this, let us for a moment ignore the role
of the existing capital stock. Hence, we write total installation costs J = G(I)
with G(0) = 0. It does not seem problematic to assume G0 (I) > 0 for I > 0.
The question concerns the assumption G00 (I) > 0. According to this assumption
the average installation cost G(I)=I must be increasing in I:11 But against this
speaks the fact that capital installation may involve indivisibilities, …xed costs,
acquisition of new information etc. All these features tend to imply decreasing
average costs. In any case, at least at the microeconomic level one should ex-
pect unevenness in the capital adjustment process rather than the above smooth
adjustment.
Because of the mixed empirical success of the convex installation cost hypoth-
esis other theoretical approaches that can account for sluggish and sometimes
non-smooth and lumpy capital adjustment have been considered: uncertainty,
investment irreversibility, indivisibility, or …nancial problems due to bankruptcy
costs (Nickell 1978, Zeira 1987, Dixit and Pindyck 1994, Caballero 1999, Adda and
Cooper 2003). These approaches notwithstanding, it turns out that the q-theory
of investment has recently been somewhat rehabilitated from both a theoretical
and an empirical point of view. At the theoretical level Wang and Wen (2010)
show that …nancial frictions in the form of collateralized borrowing at the …rm
level can give rise to strictly convex adjustment costs at the aggregate level yet
at the same time generate lumpiness in plant-level investment. For large …rms,
unlikely to be much a¤ected by …nancial frictions, Eberly et al. (2008) …nd that
the theory does a good job in explaining investment behavior.
In any case, the q-theory of investment is in di¤erent versions widely used
in short- and medium-run macroeconomics because of its simplicity and the ap-
pealing link it establishes between asset markets and …rms’investment. And the
q-theory has also had an important role in studies of the housing market and the
role of housing prices for household wealth and consumption, a theme to which
we return in the next chapter.

11
Indeed, for I 6= 0 we have d[G(I)=I]=dI = [IG0 (I) G(I)]=I 2 > 0; when G is strictly convex
(G00 > 0) and G(0) = 0:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.5. Literature notes 597

14.5 Literature notes

A …rst sketch of the q-theory of investment is contained in Tobin (1969). Later
advances of the theory took place through the contributions of Hayashi (1982)
and Abel (1990).
Both the Ramsey model and the Blanchard OLG model for a closed market
economy may be extended by adding strictly convex capital installation costs, see
Abel and Blanchard (1983) and Lim and Weil (2003). Adding a public sector,
such a framework is useful for the study of how di¤erent subsidies, taxes, and
depreciation allowance schemes a¤ect investment in physical capital as well as
housing, see, e.g., Summers (1981), Abel and Blanchard (1983), and Dixit (1990).
Groth and Madsen (2013) study medium-term ‡uctuations arising in a Ramsey-
Tobin’s q framework when extended by sluggishness in real wage adjustments.

14.6 Appendix
A. When value maximization is - and is not - equivalent with continuous
static pro…t maximization

For the idealized case where tax distortions, asymmetric information, and prob-
lems with enforceability of …nancial contracts are absent, the Modigliani-Miller
theorem (Modigliani and Miller, 1958) says that the …nancial structure of the …rm
is both indeterminate and irrelevant for production outcomes. Considering the
…rm described in Section 14.1, the implied separation of the …nancing decision
from the production and investment decision can be exposed in the following way.

Simple version of the Modigliani-Miller theorem Although the theorem

allows for risk, we here ignore risk. Let the real debt of the …rm be denoted Bt
and the real dividends, Xt : We then have the accounting relationship

B_ t = Xt (F (Kt ; Lt ) G(It ; Kt ) wt Lt It rt Bt ) :

A positive Xt represents dividends in the usual meaning (payout to the owners

of the …rm), whereas a negative Xt can be interpreted as emission of new shares
of stock. Since we assume perfect competition, the time path of wt and rt is
exogenous to the …rm.
We …rst consider the …rm’s combined …nancing and production-investment
problem, which we call Problem I. We assume that those who own the …rm at
time 0 want it to maximize its net worth, i.e., the present value of expected future

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
598 TOBIN’S Q

dividends:
Z 1 Rt
max 1 V~0 = Xt e 0 rs ds
dt s.t.
(Lt ;It ;Xt )t=0 0
Lt 0; It free,
K_ t = It Kt ; K0 > 0 given, Kt 0 for all t;
B_ t = Xt (F (Kt ; Lt ) G(It ; Kt ) wt Lt It rt Bt ) ;
where B0 is given, (14.38)
Rt
rs ds
lim Bt e 0 0: (NPG)
t!1

The last constraint is a No-Ponzi-Game condition, saying that a positive debt

should in the long run at most grow at a rate which is less than the interest rate.
In Section 14.1 we considered another problem, namely a separate investment-
production problem:
Z 1 Rt
max1 V0 = Rt e 0 rs ds dt s.t.,
(Lt ;It )t=0 0
Rt F (Kt ; Lt ) G(It ; Kt ) wt Lt It ;
Lt 0; It free,
_
Kt = It Kt ; K0 > 0 given, Kt 0 for all t:

Let this problem, where the …nancing aspects are ignored, be called Problem
II. When considering the relationship between Problem I and Problem II, the
following mathematical fact is useful.
LEMMA A1 Consider a continuous function a(t) and a di¤erentiable function
f (t). Then
Z t1 Rt R t1
a(s)ds
(f 0 (t) a(t)f (t))e t0 dt = f (t1 )e t0 a(s)ds f (t0 ):
t0

Proof. Integration by parts from time t0 to time t1 yields

Z t1 Rt Rt Z t1 Rt
0 a(s)ds a(s)ds t a(s)ds
f (t)e t0 dt = f (t)e t0 t0 +
1
f (t)a(t)e t0
dt:
t0 t0

Hence,
Z t1 Rt
a(s)ds
(f 0 (t) a(t)f (t))e t0
dt
t0
R t1
a(s)ds
= f (t1 )e t0
f (t0 ):

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.6. Appendix 599

CLAIM 1 If (Kt ; Bt ; Lt ; It ; Xt )1 1
t=0 is a solution to Problem I, then (Kt ; Lt ; It )t=0
is a solution to Problem II.
Proof. By (14.38) and the de…nition of Rt , Xt = Rt + B_ t rt Bt so that
Z 1 Rt
Z 1 Rt
V~0 = Xt e 0 rs ds
dt = V0 + (B_ t rt Bt )e 0 rs ds
dt: (14.39)
0 0

In Lemma A1, let f (t) = Bt ; a(t) = rt ; t0 = 0; t1 = T and consider T ! 1:

Then
Z T Rt RT
lim (B_ t rt Bt )e 0 rs ds dt = lim BT e 0 rs ds B0 B0 ;
T !1 0 T !1

where the weak inequality is due to (NPG). Substituting this into (14.39), we
see that maximum
RT
of net worth V~0 is obtained by maximizing V0 and ensuring
r ds
limT !1 BT e 0 s
= 0; in which case net worth equals ((maximized V0 ) B0 );
where B0 is given. So a plan that maximizes net worth of the …rm must also
maximize V0 in Problem II.
Consequently it does not matter for the …rm’s production and investment
behavior whether the …rm’s investment is …nanced by issuing new debt or by
issuing shares of stock. Moreover, if we assume investors do not care about
whether they receive the …rm’s earnings in the form of dividends or valuation
gains on the shares, the …rm’s dividend policy is also irrelevant. Hence, from now
on we can concentrate on the investment-production problem, Problem II above.

The case with no capital installation costs Suppose the …rm has no capital
installation costs. Then the cash ‡ow reduces to Rt = F (Kt ; Lt ) wt Lt It :
CLAIM 2 When there are no capital installation costs, Problem II can be reduced
to a series of static pro…t maximization problems.
Proof. Current (pure) pro…t is de…ned as

t = F (Kt ; Lt ) wt Lt (rt + )Kt (Kt ; Lt ):

It follows that Rt can be written

Rt = F (Kt ; Lt ) wt Lt (K_ t + Kt ) = t + (rt + )Kt (K_ t + Kt ): (14.40)

Hence, Z Z
1 Rt 1 Rt
V0 = te 0 rs ds
dt + (rt Kt K_ t )e 0 rs ds
dt: (14.41)
0 0

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
600 TOBIN’S Q

The …rst integral on the right-hand side of this expression is independent of the
second. Indeed, the …rm can maximize the …rst integral by renting capital and
labor, Kt and Lt ; at the going factor prices, rt + and wt ; respectively, such that
t = (Kt ; Lt ) is maximized at each t. The factor costs are accounted for in the
de…nition of t :
The second integral on the right-hand side of (14.41) is the present value of
net revenue from renting capital out to others. In Lemma A1, let f (t) = Kt ;
a(t) = rt ; t0 = 0; t1 = T and consider T ! 1: Then
Z T Rt RT
lim (rt Kt K_ t )e 0 rs ds
dt = K0 lim KT e 0 rs ds
= K0 ; (14.42)
T !1 0 T !1

where the last equality comes from the fact that maximization of V0 requires
maximization of the left-hand side of (14.42)
RT
which in turn, since K0 is given,
rs ds
requires minimization of limT !1 KT e 0 . The latter expression is always
non-negative and can be made zero by choosing any time path for Kt such that
limT !1 KT = 0: (We may alternatively put it this way: it never pays Rthe …rm to
T
accumulate costly capital so fast in the long run that limT !1 KT e 0 rs ds > 0;
that is, to maintain accumulation of capital at a rate equal Rto or higher
Rt than the
1
interest rate:) Substituting (14.42) into (14.41), we get V0 = 0 t e 0 rs ds dt+K0 :
The conclusion is that, given K0 ,12 V0 is maximized if and only if Kt and Lt
are at each t chosen such that t = (Kt ; Lt ) is maximized.

The case with strictly convex capital installation costs Now we rein-
troduce the capital installation cost function G(It ; Kt ); satisfying in particular
the condition GII (I; K) > 0 for all (I; K): Then, as shown in the text, the …rm
adjusts to a change in its environment, say a downward shift in r; by a gradual
adjustment of K; in this case upward, rather than attempting an instantaneous
maximization of (Kt ; Lt ): The latter would entail an instantaneous upward jump
in Kt of size Kt = a > 0, requiring It t = a for t = 0: This would require
It = 1; which implies G(It ; Kt ) = 1; which may interpreted either as such a
jump being impossible or at least so costly that no …rm will pursue it.
12
Note that in the absence of capital installation costs, the historically given K0 is no more
“given” than the …rm may instantly let it jump to a lower or higher level. In the …rst case the
…rm would immediately sell a bunch of its machines and in the latter case it would immediately
buy a bunch of machines. Indeed, without convex capital installation costs nothing rules out
jumps in the capital stock. But such jumps just re‡ect an immediate jump, in the opposite
direction, in another asset item in the balance sheet and leave the maximized net worth of the
…rm unchanged.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.6. Appendix 601

Proof that qt satis…es (14.15) along an interior optimal pathR Rearrang-

t
ing (14.11) and multiplying through by the integrating factor e 0 (rs + )ds ; we
get
Rt Rt
[(rt + )qt q_t ] e 0 (rs + )ds = (FKt GKt ) e 0 (rs + )ds ; (14.43)
where FKt FK (Kt ; Lt ) and GKt GK (It ; Kt ): In Lemma A1, let f (t) = qt ;
a(t) = rt + ; t0 = 0; t1 = T: Then
Z T Rt RT
[(rt + )qt q_t ] e 0 (rs + )ds
dt = q0 qT e 0 (rs + )ds
0
Z T Rt
= (FKt GKt ) e 0 (rs + )ds
dt;
0

where the last equality comes from (14.43). Letting T ! 1; we get

RT
Z 1 Rt
(r + )ds
q0 lim qT e 0 s
= q0 = (FKt GKt ) e 0 (rs + )ds dt; (14.44)
T !1 0

where the …rst equality follows from the transversality condition (14.14), which
we repeat here:
Rt
lim qt e 0 rs ds = 0: (*)
t!1
RT
Indeed, since 0; limT !1 (e 0 rs ds e T ) = 0; when (*) holds. Initial time
is arbitrary, and so we may replace 0 and t in (14.44) by t and ; respectively.
The conclusion is that (14.15) holds along an interior optimal path, given the
transversality condition (*). A proof of necessity of the transversality condition
(*) is given in Appendix B.13

B. Transversality conditions
Rt
In view of (14.44), a quali…ed conjecture is that the condition limt!1 qt e 0 (rs + )ds
= 0 is necessary for optimality. This is indeed true, since this condition follows
from the stronger transversality condition (*) in Appendix A, the necessity of
which along an optimal path we will now prove.

Proof of necessity of (14.14) As the transversality condition (14.14) is the

same as (*) in Appendix A, from now we refer to (*).
13
An equivalent approach to derivation of (14.15) can be based on applying the transversality
condition (*) to the general solution formula for linear inhomogeneous …rst-order di¤erential
equations. Indeed, the …rst-order condition (14.11) provides such a di¤erential equation in qt :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
602 TOBIN’S Q

Rt
rs ds
Rearranging (14.11) and multiplying through by the integrating factor e 0 ;
we have Rt Rt
(rt qt q_t )e 0 rs ds = (FKt GKt qt ) e 0 rs ds :
In Lemma A1, let f (t) = qt ; a(t) = rt ; t0 = 0; t1 = T . Then
Z T Rt RT
Z T Rt
rs ds rs ds rs ds
(rt qt q_t )e 0 dt = q0 qT e 0 = (FKt GKt qt ) e 0 dt:
0 0

Rearranging and letting T ! 1; we see that

Z 1 Rt RT
q0 = (FKt GKt qt ) e 0 rs ds dt + lim qT e 0 rs ds
: (14.45)
0 T !1

RT
If, contrary to (*), limT !1 qT e 0 rs ds > 0 along the optimal path, then (14.45)
shows that the …rm is over-investing. By reducing initial investment by one unit,
the …rm would save approximately 1 + GI (I0 ; K0 ) = q0 ; by (14.10), which would
be more than the present value of the stream of potential net gains coming from
this marginal unit of installed capital (the …rst term on the right-hand side of
(14.45)). RT
Suppose instead that limT !1 qT e 0 rs ds < 0: Then, by a symmetric argu-
ment, the …rm has under-invested initially.

Necessity of (14.12) In cases where along an optimal path, Kt remains bounded

from above for t ! 1; the transversality condition (14.12) is implied by (*). In
cases where along an optimal path, Kt is not bounded from above for t ! 1; the
transversality condition (14.12) is stronger than (*). A proof of the necessity of
(14.12) in this case can be based on Weitzman (2003) and Long and Shimomura
(2003).

C. On di¤erent speci…cations of the q-theory

The simple relationship we have found between I and q can easily be generalized
to the case where the purchase price on the investment good, pIt ; is allowed to
di¤er from 1 (its value above) and the capital installation cost is pIt G(It ; Kt ).
In this case it is convenient to replace q in the Hamiltonian function by, say, .
Then the …rst-order condition (14.10) becomes pIt + pIt GI (It ; Kt ) = t ; implying

t
GI (It ; Kt ) = 1;
pIt

and we can proceed, de…ning as before qt by qt t =pIt :

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.6. Appendix 603

Sometimes in the literature installation costs, J, appear in a slightly di¤erent

form compared to the above exposition. But applied to a model with economic
growth this will result in installation costs that rise faster than output and ulti-
mately swallow the total produce.
Abel and Blanchard (1983), followed by Barro and Sala-i-Martin (2004, p.
152-160), introduce a function, ; representing capital installation costs per unit
of investment as a function of the investment-capital ratio. That is, total in-
stallation cost is J = (I=K)I; where (0) = 0; 0 > 0: This implies that
J=K = (I=K)(I=K): The right-hand side of this equation may be called g(I=K);
and then we are back at the formulation in Section 14.1. Indeed, de…ning
x I=K; we have installation costs per unit of capital equal to g(x) = (x)x;
and assuming (0) = 0; 0 > 0; it holds that

g(x) = 0 for x = 0; g(x) > 0 for x 6= 0;

g 0 (x) = (x) + x 0 (x) R 0 for x R 0; respectively, and
g 00 (x) = 2 0 (x) + x 00 (x):

Now, g 00 (x) must be positive for the theory to work. But the assumptions (0) =
0; 0 > 0; and 00 0; imposed in p. 153 and again in p. 154 in Barro and
Sala-i-Martin (2004), are not su¢ cient for this (since x < 0 is possible). Since
in macroeconomics x < 0 is seldom, this is only a minor point, of course. Yet,
from a formal point of view the g( ) formulation may seem preferable to the ( )
formulation.
It is sometimes convenient to let the capital installation cost G(I, K) appear,
not as a reduction in output, but as a reduction in capital formation so that

K_ = I K G(I; K): (14.46)

This approach is used in Hayashi (1982) and Heijdra and Ploeg (2002, p. 573 ¤.).
For example, Heijdra and Ploeg write the rate of capital accumulation as K=K _
= '(I=K) ; where the “capital installation function”'(I=K) can be interpreted
as '(I=K) [I G(I; K)] =K = I=K g(I=K); the latter equality comes from
assuming G is homogeneous of degree 1. In one-sector models, as we usually
consider in this text, this changes nothing of importance. In more general models
this installation function approach may have some analytical advantages; what
gives the best …t empirically is an open question. In our housing market model
in the next chapter we apply a speci…cation analogue to (14.46), interpreting K_
as the number of new houses per time unit.
Finally, some analysts assume that installation costs are a strictly convex
function of net investment, I K; not gross investment, I: This agrees well with
intuition if mere replacement investment occurs in a smooth way not involving

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
604 TOBIN’S Q

new technology, work interruption, and reorganization. To the extent capital

investment involves indivisibilities and embodies new technology, it may seem
more plausible to specify the installation costs as a convex function of gross
investment.

D. Proof of Hayashi’s theorem

For convenience we repeat:
THEOREM (Hayashi) Assume the …rm is a price taker, that the production
function F is jointly concave in (K; L), and that the installation cost function G
is jointly convex in (I, K). Then, along the optimal path we have:
(i) qtm = qta for all t 0; if F and G are homogeneous of degree 1.
(ii) qtm < qta for all t; if F is strictly concave in (K; L) and/or G is strictly
convex in (I; K):
Proof. The value of the …rm as seen from time t is
Z 1 R
rs ds
Vt = (F (K ; L ) G(I ; K ) w L I )e t d : (14.47)
t

We introduce the functions

A = A(K; L) F (K; L) FK (K; L)K FL (K; L)L; (14.48)

B = B(I; K) GI (I; K)I + GK (I; K)K G(I; K): (14.49)

Then the cash-‡ow of the …rm at time can be written

R = F (K ; L ) FL L G(I ; K ) I
= A(K ; L ) + FK K + B(I ; K ) GI I GK K I ;

where we have used …rst FL = w and then the de…nitions of A and B above.
Consequently, when moving along the optimal path,
Z 1 R
Vt = V (Kt ; t) = (A(K ; L ) + B(I ; K )) e t rs ds d (14.50)
t
Z 1 R
+ [(FK GK )K (1 + GI )I ]e t rs ds d
Z 1t R
= (A(K ; L ) + B(I ; K ))e t rs ds d + qt Kt ;
t

cf. Lemma D1 below. Isolating qt ; it follows that

Z 1 R
m Vt 1 rs ds
qt qt = [A(K ; L ) + B(I ; K )]e t d ; (14.51)
Kt K t t

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.6. Appendix 605

when moving along the optimal path.

Since F is concave and F (0; 0) = 0, we have for all K and L, A(K; L) 0
with equality sign, if and only if F is homogeneous of degree one. Similarly, since
G is convex and G(0; 0) = 0, we have for all I and K, B(I; K) 0 with equality
sign, if and only if G is homogeneous of degree one. Now the conclusions (i) and
(ii) follow from (14.51) and the de…nition of q a in (14.27):
LEMMA D1 The last integral on the right-hand side of (14.50) equals qt Kt , when
investment follows the optimal path.
Proof. We want to characterize a given optimal path (K ; I ; L )1=t . Keeping t
…xed and using z as our varying time variable, we have

(FKz GKz )Kz (1 + GIz )Iz = [(rz + )qz q_z ]Kz (1 + GIz )Iz

= [(rz + )qz q_z ]Kz qz (K_ z + Kz ) = rz qz Kz (q_z Kz + qz K_ z ) = rz uz u_ z ;

where we have used (14.11), (14.10), (14.6), and the de…nition uz qz Kz . We
look at this as a di¤erential equation: u_ z rz uz = 'z , where 'z [(FKz
GKz )Kz (1 + GIz )Iz ] is considered as some given function of z. The solution of
this linear di¤erential equation is
Rz
Z z Rz
r ds
uz = ut e t s
+ ' e rs ds d ;
t
Rz
rs ds
implying, by multiplying through by e t ; reordering, and inserting the de…-
nitions of u and ';
Z z R
rs ds
[(FK GK )K (1 + GI )I ]e t d
t
Rz
rs ds
= qt Kt qz Kz e t ! qt Kt for z ! 1;

from the transversality condition (14.12) with t replaced by z and 0 replaced by

t.
A di¤erent and perhaps more illuminating way of understanding (i) in
Hayashi’s theorem is the following.
Suppose F and G are homogeneous of degree one. Then A = B = 0; GI I +
GK K = G = g(I=K)K; and FK = f 0 (k); where f is the production function in
intensive form: Consider an optimal path (K ; I ; L )1=t and let k K =L and
x I =K along this path which we now want to characterize. As the path is
assumed optimal, from (14.47) follows
Z 1 R
Vt = V (Kt ; t) = [f 0 (k ) g(x ) x ]K e t rs ds d : (14.52)
t

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
606 TOBIN’S Q

R
From K_ t = (xt )Kt follows K = Kt e t (xs )ds
. Substituting this into (14.52)
yields Z 1 R
V (Kt ; t) = Kt [f 0 (k ) g(x ) x ]e t (rs xs + )ds
d :
t
In view of (14.24), with t replaced by ; the optimal investment ratio x depends,
for all ; only on q ; not on K ; hence not on Kt : Therefore,
Z 1 R
@V =@Kt = [f 0 (k ) g(x ) x ]e t (rs xs + )ds d = Vt =Kt :
t

Hence, from (14.28) and (14.27), we conclude qtm = qta .

Remark. We have assumed throughout that G is strictly convex in I. This does
not imply that G is jointly strictly convex in (I; K). For example, the function
G(I; K) = I 2 =K is strictly convex in I (since GII = 2=K > 0): But at the same
time this function has B(I; K) = 0 and is therefore homogeneous of degree one.
Hence, it is not jointly strictly convex in (I; K).

E. The slope of the q_ = 0 locus in the SOE case

First, we shall determine the sign of the slope of the q_ = 0 locus in the case
g + n = 0; considered in Fig. 14.4. Taking the total di¤erential in (14.33) w.r.t.
K and q gives
0 = FKK (K; L)dK + fr + + g 0 (m(q))m0 (q) [m(q) + (q 1)m0 (q)]g dq
= FKK (K; L)dK + [r + m(q)] dq;
since g 0 (m(q)) = q 1, by (14.23) and (14.24). Therefore
dq FKK (K; L)
= for r + 6= m(q):
dK jq=0
_ r+ m(q)
From this it is not possible to sign dq=dK at all points along the q_ = 0 locus. But
in a neighborhood of the steady state we have m(q) ; hence r + m(q)
r > 0: And since FKK < 0; this implies that at least in a neighborhood of E in
Fig. 14.4 the q_ = 0 locus is negatively sloped.
Second, consider the case g + n > 0; illustrated in Fig. 14.6. Here we get in
a similar way
dq f 00 (k~ )
= for r + 6= m(q):
dk~ jq=0
_ r+ m(q)
From this it is not possible to sign dq=dk~ at all points along the q_ = 0 locus. But
in a small neighborhood of the steady state we have m(q) + + n; hence
00
r+ m(q) r n: Since f < 0; then, at least in a small neighborhood of
E in Fig. 14.6, the q_ = 0 locus is negatively sloped, when r > + n.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

14.7. Exercises 607

F. The divergent paths

Text not yet available.

14.7 Exercises
14.1 (induced sluggish capital adjustment). Consider a …rm with capital instal-
lation costs J = G(I; K); satisfying

G(0; K) = 0; GI (0; K) = 0; GII (I; K) > 0; and GK (I; K) 0:

a) Can we from this conclude anything as to strict concavity or strict convexity

of the function G? If yes, with respect to what argument or arguments?

b) For two values of K; K and K; illustrate graphically the capital installation

costs J in the (I; J) plane. Comment.

c) By drawing a few straight line segments in the diagram, illustrate that

G( 21 I; K)2 < G(I; K) for any given I > 0:

14.2 (see end of Section 14.3)

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

 CHAPTER 14. FIXED CAPITAL INVESTMENT AND
608 TOBIN’S Q

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Makroøkonomi. Note 2.

18.10.2015 Christian Groth

Uncertainty, expectations,
and asset price bubbles
This lecture note provides a framework for addressing themes where expectations
in uncertain situations are important elements. Our previous models have not taken
seriously the problem of uncertainty. Where agent’s expectations about future variables
were involved and these expectations were assumed to be model-consistent (“rational”),
we only considered a special case: perfect foresight. Shocks were treated in a peculiar
(almost self-contradictory) way: they might occur, but only as a complete surprise, a
once-for-all event. Agents’ expectations and actions never incorporated that new shocks
could arrive.

We will now allow recurrent shocks to take place. The environment in which the
economic agents act will be considered inherently uncertain. How can this be modeled
and how can we solve the resultant models? Since it is easier to model uncertainty
in discrete rather than continuous time, we examine uncertainty and expectations in a
discrete time framework.

Our emphasis will be on the hypothesis that when facing uncertainty a dominating
fraction of the economic agents form “rational expectations” in the sense of making prob-
abilistic forecasts which coincide with the forecast calculated on the basis of the “relevant
economic model”. But we begin with simple mechanistic expectation formation hypothe-
ses that have been used to describe day-to-day expectations of people who do not at all
think about the probabilistic properties of the economic environment.

1 Simple expectation formation hypotheses

One simple supposition is that expectations change gradually to correct past expectation
errors. Let  denote the general price level in period  and   ≡ ( − −1 )−1
the corresponding inflation rate. Further, let  −1 denote the “subjective expectation”,
formed in period  − 1 of   i.e., the inflation rate from period  − 1 to period  We may

1
think of the “subjective expectation” as the expected value in a vaguely defined subjective
conditional probability distribution.

The hypothesis of adaptive expectations (the AE hypothesis) says that the expectation
is revised in proportion to the past expectation error,

 −1 =  −2−1 + ( −1 −  −2−1 ) 0   ≤ 1 (1)

where the parameter  is called the adjustment speed. If  = 1 the formula reduces to

 −1 =  −1  (2)

This limiting case is known as static expectations or myopic expectations; the subjective
expectation is that the inflation rate will remain the same or at least that it is not more
likely to go up than down.

We may write (1) on the alternative form

 −1 =  −1 + (1 − )−2−1  (3)

This says that the expected value concerning this period (period ) is a weighted average
of the actual value for the last period and the expected value for the last period. By
backward substitution we find

 −1 =  −1 + (1 − )[−2 + (1 − ) −3−2 ]

=  −1 + (1 − ) −2 + (1 − )2 [−3 + (1 − ) −4−3 ]
X
=  (1 − )−1 − + (1 − )  −−1− 
=1

Since (1 − ) → 0 for  → ∞, we have (for −−1− bounded as  → ∞)

X
∞
 −1 = (1 − )−1  −  (4)
=1

Thus, according to the AE hypothesis with 0    1 the expected inflation rate is a

weighted average of the historical inflation rates back in time. The weights are geomet-
rically declining with increasing time distance from the current period. The weights sum
P
to one (in that ∞ =1 (1 − )
−1
= (1 − (1 − ))−1 = 1)

The formula (4) can be generalized to the general backward-looking expectations for-
mula,
X
∞ X
∞
 −1 =   −1−  where  = 1 (5)
=1 =1

2
If the weights  in (5) satisfy  = (1 − )−1   = 1 2. . .  we get the AE formula (4).
If the weights are

1 = 1 +  2 = −  = 0 for  = 3 4 . . . ,

we get
 −1 = (1 + ) −1 −  −2  =  −1 + ( −1 −  −2 ) (6)
This is called the hypothesis of extrapolative expectations and says:

if   0 then the recent direction of change in  is expected to continue;

if   0 then the recent direction of change in  is expected to be reversed;
if  = 0 then expectations are static as in (2).

There are cases where for instance myopic expectations are “rational” (in a sense to
be defined below). Exercise 1 provides an example. But in many cases purely backward-
looking formulas are too rigid, too mechanistic. They will often lead to systematic expec-
tation errors to one side or the other. It seems implausible that people should not then
respond to their experience and revise their expectations formula. And when expectations
are about things that really matter for people, they are likely to listen to professional fore-
casters who build their forecasting on statistical or econometric models. Such models are
based on a formal probabilistic framework, take the interaction between different variables
into account, and incorporate new information about future possible events.

2 The rational expectations hypothesis

2.1 Preliminaries

We first recapitulate a few concepts from statistics. A sequence { } of random variables
indexed by time is called a stochastic process. A stochastic process { } is called white
noise if for all   has zero expected value, constant variance, and zero covariance across
time.1 A stochastic process { } is called a first-order autoregressive process, abbreviated
AR(1), if  =  0 +  1 −1 +   where  0 and  1 are constants, and { } is white noise;
if | 1 |  1 then { } is called a stationary RA(1) process. A stochastic process { } is
called a random walk if  = −1 +   where { } is white noise.
1
The expression white noise derives from electrotechnics. In electrotechnical systems signals will often
be subject to noise. If this noise is arbitrary and has no dominating frequence, it looks like white light.
The various colours correspond to a certain wave length, but white light is light which has all frequences
(no dominating frequence).

3
 Before defining the term rational expectation, it is useful to clarify a distinction be-
tween two ways in which expectations, whatever their nature, may enter a macroeconomic
model.

2.1.1 Two model types

Type A: models with past expectations of current endogenous variables Sup-

pose a given macroeconomic model can be reduced to two equations, the first being

 =  −1 +     = 0 1 2  (7)

where  is some endogenous variable (not necessarily  )  and  are given constant
coefficients, and  is an exogenous random variable which follows some specified stochas-

tic process. In line with the notation from Section 1, −1 is the subjective expectation
formed in period −1 of the value of the variable  in period  The economic agents are in
simple models assumed to have the same expectations. Or, at least there is a dominating

expectation, −1  in the society. What the equation (7) claims is that the endogenous
variable,  , depends, in the specified linear way, on the “generally held” expectation of
 , formed in the previous period. It is natural to think of the outcome  as being the
aggregate result of agents’ decisions and market mechanisms, the decisions being made at
discrete points in time     −2 −1      immediately after the uncertainty concerning
the period in question is resolved.

The second equation specifies how the subjective expectation is formed. To fix ideas,
let us assume myopic expectations,

−1 = −1  (8)

as in (2) above. A solution to the model is a stochastic process for  such that (7) holds,
given the expectation formation (8) and the stochastic process which  follows.

EXAMPLE 1 (imported raw materials and the domestic price level) Let the endogenous
variable in (7) represent the domestic price level (the consumer price index)   and let
 be the price level of imported raw materials. Suppose the price level is determined
through a markup on unit costs,
1
 = ( +  )(1 + ) 0  (*)
1+
where  is the nominal wage level in period  = 0 1 2    , and  and  are positive tech-
nical coefficients representing the assumed constant labor and raw meterials requirements,

4
respectively, per unit of output;  is a constant markup. Assume further that workers in
period  − 1 negotiate next period’s wage level,   so as to achieve, in expected value, a
certain target real wage which we normalize to 1, i.e.,


= 1
−1

Inserting into (*), we have


 =  −1 +    0   = (1 + )  1 0   = (1 + ) (9)

Suppose  = ̄ +   where ̄ is a positive constant and { } is white noise. Assuming

myopic expectations,

−1 = −1  (10)

the solution for the evolution of the price level is

 =  −1 + (̄ +  )  = 0 1 2    

Without shocks, and starting from an arbitrary −1  0 the time path of the price
level would be  = (−1 −  ∗ )+1 +  ∗  where  ∗ = ̄(1 − ) Shocks to the price of
imported raw materials result in transitory deviations from  ∗  But as the shocks are only
temporary and ||  1 the domestic price level gradually returns towards the constant
level  ∗  The intervening changes in wage demands in response to the changes in the price
level changes prolong the time it takes to return to  ∗ in the absence of new shocks. ¤


Equation (7) can also be interpreted as a vector equation (such that  and −1 are
-vectors,  is an  ×  matrix,  an  ×  matrix, and  an -vector). The crucial
feature is that the endogenous variables dated  only depend on previous expectations of
date- values of these variables and on the exogenous variables.

Models with past expectations of current endogenous variables will serve as our point
of reference when introducing the concept of rational expectations below.

Type B: models with forward-looking expectations Another way in which agents’

expectations may enter is exemplified by


 =  +1 +     = 0 1 2  (11)


Here +1 is the subjective expectation, formed in period  of the value of  in period
 + 1. Example: the equity price today depends on what the equity price is expected to be

5
tomorrow. Or more generally: the current expectation of a future value of an endogenous
variable influences the current value of this variable. We name this the case of forward-

looking expectations. (In “everyday language” also −1 in model type 1 can be said to
be a forward-looking variable as seen from period  − 1. But the dividing line between
the two model types, (7) and (11), is whether current expectations of future values of the
endogenous variables do or do not influence the current values of these.)

The complete model with forward-looking expectations will include an additional equa-

tion, specifying how the subjective expectation, +1  is formed. We might again impose

myopic expectations, +1 =   A solution to the model is a stochastic process for 
satisfying (11), given the stochastic process followed by  and given the specified ex-
pectation formation and perhaps some additional restrictions in the form of boundary
conditions or similar. The case of forward-looking expectations is important in connec-
tion with many topics in macroeconomics, including the evolution of asset prices, and
issues of asset price bubbles. This case will be dealt with in sections 3 and 4 below.

In passing we note that in both model type 1 and model type 2, it is the mean (in the
subjective probability distribution) of the random variable(s) that enters. This is typical
of simple macroeconomic models which often ignore other measures such as the median,
mode, or higher-order moments. The latter, say the variance of  , may be included in
more advanced models where for instance behavior towards risk is important.

2.1.2 The concept of a model-consistent expectation

The concepts of a rational expectation and model-consistent expectation are closely related,
but not the same. We start with the latter.

Let there be given a stochastic model represented by (7) combined with some given
expectation formation (8), say. We put ourselves in the position of the investigator or
model builder and ask what the model-consistent expectation of the endogenous variable
 is as seen from period  − 1. It is the mathematical conditional expectation that can
be calculated on the basis of the model and available relevant data revealed up to and
including period  − 1. Let us denote this expectation

( |−1 ) (12)

where  is the expectation operator and −1 denotes the information available at time
 − 1. We think of period  − 1 as the half-open time interval [ − 1 ) and imagine that
the uncertainty concerning the exogenous random variable −1 is resolved at time  − 1

6
So −1 includes knowledge of −1 and thereby, via the model, also of −1 2

The information −1 may comprise knowledge of the realized values of  and  up
until and including period  − 1 Instead of (12) we could, for instance, write

( |−1 = −1      − = − ; −1 = −1      − = − )

Here information (some of which may be redundant) goes back to a given initial period,
say period 0, in which case  equals  Alternatively, perhaps information goes back to
“ancient times”, possibly represented by  = ∞ Anyway, as time proceeds, in general
more and more realizations of the exogenous and endogenous variables become known
and in this sense the information −1 expands with rising . The information −1 may
also be interpreted as “partial lack of uncertainty”, so that an “increasing amount of
information” and “reduced uncertainty” are seen as two sides of the same thing. The
“reduced uncertainty” lies in the fact that the space of possible time paths {(   )}+
−
as of time  shrinks as time proceeds ( denotes the time horizon as seen from time ).3
Indeed, this space shrinks precisely because more and more realizations of the variables
take place (more information appears) and thereby rule out an increasing subset of paths
that were earlier possible.

In Example 1, as long as the subjective expectation is the myopic expectation (10),

the model-consistent expectation is

( |−1 ) =  −1 + ̄

Inserting the investigator’s estimated values of the coefficients  and  the investigator’s
forecast of  is obtained.

2.2 The rational expectations hypothesis

Unsatisfied with mechanistic formulas like those of Section 1, the American economist
John F. Muth (1961) introduced a radically different approach, the hypothesis of rational
expectations. Muth stated the hypothesis the following way:

I should like to suggest that expectations, since they are informed predictions
of future events, are essentially the same as the predictions of the relevant
2
We refer to −1 as the “available information” rather than the “information set” which is an alterna-
tive term used in the literature. The latter term is tricky, however, and has different meanings in different
branches of economics, hence we are hesitant to use it. The subtleties are accounted for in Appendix B,
dealing with mathematical conditional expectations in general.
3
By “possible” is meant “ex ante feasible according to a given model”.

7
 economic theory. At the risk of confusing this purely descriptive hypothesis
with a pronouncement as to what firms ought to do, we call such expectations
’rational’ (Muth 1961).

Muth applied this hypothesis to simple microeconomic problems. The hypothesis was
subsequently extended and applied to general equilibrium theory and macroeconomics by
what since the early 1970s became known as the New Classical Macroeconomics school.
Nobel laureate Robert E. Lucas from the University of Chicago lead the way by a series of
papers starting with Lucas (1972) and Lucas (1973). Assuming rational expectations in a
model instead of, for instance, adaptive expectations may radically change the dynamics
and impact of economic policy.

2.2.1 The concept

Assuming the economic agents have rational expectations (RE) is to assume that their
subjective expectation equals the model-consistent expectation, that is, the mathematical
conditional expectation that can be calculated on the basis of the model and available
relevant information about the exogenous stochastic variables. In connection with the
model ingredient (7), assuming the agents have rational expectations thus means that

−1 = ( |−1 ) (13)

i.e., agents’ subjective conditional expectation coincides with the “objective” or “true”
conditional expectation, given the model (7).

Together, the equations (7) and (13) constitute a simple rational expectations model
(henceforth an RE model). We may write the model in compact form as

 = ( |−1 ) +     = 0 1 2  (14)

The assumption of rational expectations thus relies on idealized conditions.

2.2.2 Solving a simple RE model

To solve the model means to find the stochastic process followed by   given the sto-
chastic process followed by the exogenous variable   For a linear RE model with past
expectations of current endogenous variables, the solution procedure is the following.

1. By substitution, reduce the RE model (or the relevant part of the model) into a
form like (14) expressing the endogenous variable in period  in terms of its past

8
 expectation and the exogenous variable(s). (The case with multiple endogenous
variables is treated similarly.)

2. Take the conditional expectation on both sides of the equation and solve for the
conditional expectation of the endogenous variable.

3. Insert into the reduced form and rearrange.

In practice there is often a fourth step, namely to express other endogenous variables
in the model in terms of those found in step 3. Let us see how the procedure works by
way of the following example.

EXAMPLE 2 We modify Example 1 by replacing myopic expectations by rational expec-


tations, i.e., (10) is replaced by −1 = ( |−1 ) Now “available information” includes
that the subjective expectations are rational expectations. Step 1:

 = ( |−1 ) +    0    1   0 (15)

Step 2: ( |−1 ) = ( |−1 ) + ̄ implying

̄
( |−1 ) =  
1−
Step 3: Insert into (15) to get
̄
 =  + (̄ +  )
1−
This is the solution of the model in the sense of a specification of the stochastic process
followed by  .

To compare with myopic expectations, suppose the event  6= 0 is relatively seldom

and that at  = 0 1  0 − 1 it so happens that  = 0 hence  = ̄(1 − ) ≡  ∗ 
Then, at  = 0  0  0 so that 0 =  ∗ + 0   ∗  But for  = 0 + 1 0 + 2  0 + 
there is again a sequence of periods with  = 0 Then, under RE, domestic price level
returns to  ∗ already in period 0 + 1.

With myopic expectations, combined with −1 =  ∗  say, the positive shock to import
prices at  = 0 will imply 0 =  ∗ + (̄ + 0 ) =  ∗ + 0  0 +1 = ( ∗ +  ) + ̄
=  ∗ +   0 + =  ∗ +   for  = 1 2   After 0 there is a systematic positive
forecast error. This is because the mechanical expectation does not consider how the
economy really functions. ¤

9
 Returning to the general form (14), without specifying the process { }  the second
step gives
( |−1 )
( |−1 ) =   (16)
1−
when  6= 14 Then, in the third step we get

( |−1 ) + (1 − )  − ( − ( |−1 ))

 =  =  (17)
1− 1−

EXAMPLE 3 Let  follow the process  = ̄ + −1 +   where 0    1 and 

has zero expected value, given all observed past values of  and  Then (17) yields the
solution
 −  ̄ + −1 + (1 − )
 =  =   = 0 1 2 .
1− 1−
In Exercise 2 you are asked to solve a simple Keynesian model of this form and compare
the solution under rational expectations with the solution under static expectations. ¤

Rational expectations should be viewed as a simplifying assumption that at best offers

an approximation. First, the assumption entails essentially that the economic agents
share one and the same understanding about how the economic system functions (and in
this chapter they also share one and the same information, −1 ). This is already a big
mouthful. Second, this perception is assumed to comply with the model of the informed
economic specialist. Third, this model is supposed to be the true model of the economic
process, including the true parameter values as well as the true stochastic process which

 follows. By equalizing −1 with the true conditional expectation, ( |−1 ) and not
at most some econometric estimate of this, it is presumed that agents know the true values
of the parameters  and  in the data-generating process which the model is supposed
to mimic. In practice it is not possible to attain such a model, at least not unless the
considered economic system has reached some kind of steady state and no structural
changes occur.

Nevertheless, a model based on the rational expectations hypothesis can in many

contexts be seen as a useful cultivation of a theoretical research question. The results
that emerge cannot be due to systematic expectation errors from the economic agents’
side. In this sense the assumption of rational expectations makes up a theoretically
interesting benchmark case.
4
If  = 1, the model (14) is inconsistent unless ( |−1 )) = 0 in which case there are multiple
solutions. Indeed, for any number  ∈ (−∞, +∞), the process  =  +  solves the model when
( |−1 ) = 0

10
 We shall stick to the term “rational expectation” because it is standard. The term
can easily be misunderstood, however. Usually, in economists’ terminology “rational”
refers to behavior based on optimization subject to the constraints faced by the agent.
So one might think that the RE hypothesis stipulates that economic agents try to get the
most out of a situation with limited information, contemplating the benefits and costs
of gathering more information and using adequate statistical estimation methods. But
this is a misunderstanding. The RE hypothesis presumes that the true model is already
known to the agents. The “rationality” refers to taking this assumed knowledge fully into
account.

2.2.3 The forecast error*


Let the forecast of some variable  one period ahead be denoted −1 . Suppose the
forecast is determined by some given function,  , of realizations of  and  up to and

including period  − 1 that is, −1 = (−1  −2   −1  −2  ) Such a function is
known as a forecast function. It might for instance be one of the mechanistic forecasting
principles in Section 1. At the other extreme the forecast function might, at least theo-
retically, coincide with the a model-consistent conditional expectation. In the latter case
it is a model-consistent forecast function and we can write

(−1  −2   −1  −2  ) = ( |−1 ) (18)

= ( |−1 = −1  −2 = −2   −1 = −1  −2 = −2  ) 

The forecast error is the difference between the actually occurring future value,   of

a variable and the forecasted value. So, for a given forecast, −1  the forecast error is

 ≡  − −1 and is itself a stochastic variable.

If the forecast function in (18) complies with the true data-generating process (a big
“if”), then the implied forecasts would have several ideal properties:

(a) the forecast error would have zero mean;

(b) the forecast error would be uncorrelated with any of the variable in the information
−1 and therefore also with its own past values; and

(c) the expected squared forecast error would be minimized.

To see these properties, note that the model-consistent forecast error is  =  −

( |−1 )  From this follows that ( |−1 ) = 0 cf. (a). Also the unconditional expec-

11
tation is nil, i.e., ( ) = 0; this is because (( |−1 )) = (0) = 0 at the same time as
(( |−1 )) = ( ) by the law of iterated expectations from statistics saying that the
unconditional expectation of the conditional expectation of a stochastic variable  is given
by the unconditional expectation of , cf. Appendix B. Considering the specific model
(7), the model-consistent-forecast error is  =  − ( |−1 ) = ( − ( |−1 )) by
(16) and (17). An ex post error ( 6= 0) thus emerges if and only if the realization of the
exogenous variable deviates from its conditional expectation as seen from the previous
period.

As to property (b), for  = 1 2  let − be some variable value belonging to the
information − . Then, property (b) is the claim that the (unconditional) covariance
between  and − is zero, i.e., Cov( − ) = 0 for  = 1 2 . This follows from the
orthogonality property of model-consistent expectations (see Appendix C). In particular,
with − = −  we get Cov( − ) = 0 i.e., the forecast errors exhibit lack of serial
correlation. If the covariance were not zero, it would be possible to improve the forecast
by incorporating the correlation into the forecast. In other words, under the assumption of
rational expectations economic agents have no more to learn from past forecast errors. As
remarked above, the RE hypothesis precisely refers to a fictional situation where learning
has been completed and underlying mechanisms do not change.

Finally, a desirable property of a forecast function (·) is that it maximizes “accuracy”,

i.e., minimizes an appropriate loss function. A popular loss function,  in this context is
the expected squared forecast error conditional on the information −1 ,

 = (( − (−1  −2   −1  −2  ))2 |−1 ) 

Assuming   −1   −1  −2   are jointly normally distributed, then the solution to
the problem of minimizing  is to set (·) equal to the conditional expectation ( |−1 )
based on the data-generating model as in (18).5 This is what property (c) refers to.

EXAMPLE 4 Let  = ( |−1 ) +   with  = ̄ +   where ̄ is a constant and

 is white noise with variance  2 . Then (17) applies, so that
̄
 = +    = 0 1 
1−
with variance 2  2  The model-consistent forecast error is  =  − ( |−1 ) =  with
conditional expectation equal to ( |−1 ) = 0 This forecast error itself is white noise
and is therefore uncorrelated with the information on which the forecast is based. ¤
5
For proof, see Pesaran (1987). Under the restriction of only linear forecast functions, property (c)
holds even without the joint normality assumption, see Sargent (1979).

12
 It is worth emphasizing that the “true” conditional expectation can not usually be
known − neither to the economic agents nor to the investigator. At best there can be a
reasonable estimate, probably somewhat different across the agents because of differences
in information and conceptions of how the economic system functions. A deeper model of
expectations would give an account of the mechanisms through which agents learn about
the economic environment. An important ingredient here would be how agents contem-
plate the costs and potential gains associated with further information search needed
to reduce systematic expectation errors where possible. This contemplation is intricate
because information search often means entering unknown territory. Moreover, for a sig-
nificant subset of the agents the costs may be prohibitive. A further complicating factor
involved in learning is that when the agents have obtained some knowledge about the
statistical properties of the economic variables, the resulting behavior of the agents may
change these statistical properties. The rational expectations hypothesis sets these prob-
lems aside. It is simply assumed that the structure of the economy remains unchanged
and that the learning process has been completed.

2.3 Perfect foresight as a special case

The notion of perfect foresight corresponds to the limiting case where the variance of
the exogenous variable(s) is zero so that with probability one,  = ( |−1 ) for all
. Then we have a non-stochastic model where rational expectations imply that agents’
ex post forecast error with respect to  is zero.6 To put it differently: rational expec-
tations in a non-stochastic model is equivalent to perfect foresight. Note, however, that
perfect foresight necessitates the exogenous variable  to be known in advance. Real-
world situations are usually not like that. If we want our model to take this into account,
the model ought to be formulated in an explicit stochastic framework. And assumptions
should be stated about how the economic agents respond to the uncertainty. The ra-
tional expectations assumption is a one approach to the problem and has been much
applied in macroeconomics in recent decades, perhaps due to lack of compelling tractable
alternatives.
6
Here we disregard zero probability events.

13
3 Models with rational forward-looking expectations

We here turn to models where current expectations of a future value of an endogenous

variable have an influence on the current value of this variable, that is, the case exemplified
by equation (11). At the same time we introduce two simplifications in the notation. First,
instead of using capital letters to denote the stochastic variables (as we did above and
is common in mathematical statistics), we follow the tradition in macroeconomics and
use lower case letters. So a lower case letter may from now on represent a stochastic
variable or a specific value of this variable, depending on the context. So an equation

like (11) will now read  =  +1 +    Under rational expectations it takes the form
 = (+1 | ) +     = 0 1 2    . Second, from now on we write this equation as

 =  +1 +        = 0 1 2      6= 0 (19)

That is, the expected value of a stochastic variable, +  conditional on the information
 , will be denoted  + 

A stochastic difference equation of the form (19) is called a linear expectation difference
equation of first order with constant coefficient .7 A solution is a specified stochastic
process { } which satisfies (19), given the stochastic process followed by  . In the
economic applications usually no initial value, 0 , is given. On the contrary, the interpre-
tation is that  depends, for all  on expectations about the future.8 So  is considered
a jump variable that can immediately shift its value in response to the emergence of new
information about the future ’s. For example, a share price may immediately jump to a
new value when the accounts of the firm become publicly known (often even before, due
to sudden rumors).

Due to the lack of an initial condition for   there can easily be infinitely many
processes for  satisfying our expectation difference equation. We have an infinite forward-
looking “regress”, where a variable’s value today depends on its expected value tomorrow,
this value depending on the expected value the day after tomorrow and so on. Then usu-
ally there are infinitely many expected sequences which can be self-fulfilling in the sense
that if only the agents expect a particular sequence, then the aggregate outcome of their
behavior will be that the sequence is realized. It “bites its own tail” so to speak. Yet, when
7
To keep things simple, we let the coefficients  and  be constants, but a generalization to time-
dependent coefficients is straightforward.
8
The reason we say “depends on” is that it would be inaccurate to say that  is determined (in a
one-way-sense) by expectations about the future. Rather there is mutual dependence. In view of  being
an element in the information   the expectation of +1 in (19) may depend on  just as much as 
depends on the expectation of +1 .

14
an equation like (19) is part of a larger model, there will often (but not always) be con-
ditions that allow us to select one of the many solutions to (19) as the only economically
relevant one. For example, an economy-wide transversality condition or another general
equilibrium condition may rule out divergent solutions and leave a unique convergent
solution as the final solution.

We assume  6= 0 since otherwise (19) itself is already the unique solution. It turns
out that the set of solutions to (19) takes a different form depending on whether ||  1
or ||  1:

The case ||  1 In general, there is a unique fundamental solution and infinitely many
explosive bubble solutions.

The case ||  1 In general, there is no fundamental solution but infinitely many non-
explosive solutions. (The case || = 1 resembles this.)

In the case ||  1 the expected future has modest influence on the present. Here we
will concentrate on this case, since it is the case most frequently appearing in macroeco-
nomic models with rational expectations.

4 Solutions when ||  1

Various solution methods are available. Repeated forward substitution is the most easily
understood method.

4.1 Repeated forward substitution

Repeated forward substitution consists of the following steps. We first shift (19) one
period ahead:
+1 =  +1 +2 +  +1 

Then we take the conditional expectation on both sides to get

 +1 =   (+1 +2 ) +   +1 =   +2 +   +1  (20)

where the second equality sign is due to the law of iterated expectations, which says that

 (+1 +2 ) =  +2  (21)

15
see Box 1. Inserting (20) into (19) then gives

 = 2  +2 +   +1 +    (22)

The procedure is repeated by forwarding (19) two periods ahead; then taking the condi-
tional expectation and inserting into (22), we get

 = 3  +3 + 2   +2 +   +1 +   

We continue in this way and the general form (for  = 0 1 2 ) becomes

+ =  + (++1 ) +  + 

 + =   ++1 +   + 

X

+1
 =   ++1 +  +    +  (23)
=1

Box 1. The law of iterated expectations

The method of repeated forward substitution is based on the law of iterated expecta-
tions which says that  ( +1 +2 ) =   +2  as in (21). The logic is the fol-
lowing. Events in period  + 1 are stochastic as seen from period  and so +1 +2
(the expectation conditional on these events) is a stochastic variable. Then the law
of iterated expectations says that the conditional expectation of this stochastic variable
as seen from period  is the same as the conditional expectation of +2 itself as seen
from period  So, given that expectations are rational, then an earlier expectation of
a later expectation of  is just the earlier expectation of . Put differently: my best
forecast today of how I am going to forecast tomorrow a share price the day after
tomorrow, will be the same as my best forecast today of the share price the day after
tomorrow. If beforehand we have good reasons to expect that we will revise our
expectations upward, say, when next period’s additional information arrives, the
original expectation would be biased, hence not rational.9

4.2 The fundamental solution

PROPOSITION 1 Consider the expectation difference equation (19), where  6= 0 If

X

lim   + exists, (24)
→∞
=1

9
A formal account of conditional expectations and the law of iterated expectations is given in Appendix
B.

16
then
X
∞ X
∞

 =    + =  +    + ≡ ∗   = 0 1 2  (25)
=0 =1

is a solution to the equation.

Proof Assume (24). Then the formula (25) is meaningful. In view of (23), it satisfies
(19) if and only if lim→∞ +1  ++1 = 0 Hence, it is enough to show that the process
(25) satisfies this latter condition.
P∞
In (25), replace  by  +  + 1 to get ++1 =  =0  ++1 ++1+  Using the law
of iterated expectations, this yields
X
∞
 ++1 =    ++1+ so that
=0
X
∞ X
∞
+1 +1 
  ++1 =     ++1+ =    + 
=0 =+1

P∞
It remains to show that lim→∞ =+1   + = 0 From the identity

X
∞ X
 X
∞
 
  + =   + +   +
=1 =1 =+1

follows
X
∞ X
∞ X

 
  + =   + −   + 
=+1 =1 =1

Letting  → ∞ this gives

X
∞ X
∞ X
∞
 
lim   + =   + −   + = 0
→∞
=+1 =1 =1

which was to be proved. ¤

The solution (25) is called the fundamental solution of (19), often marked by an
asterisk ∗ . The fundamental solution is (for  6= 0) defined only when the condition (24)
holds. In general this condition requires that ||  1 In addition, (24) requires that the
absolute value of the expectation of the exogenous variable does not increase “too fast”.
More precisely, the requirement is that | + |, when  → ∞, has a growth factor less
than ||−1  As an example, let 0    1 and   0, and suppose that  +  0 for 
= 0 1 2  and that 1 +  is an upper bound for the growth factor of  +  Then

 + ≤ (1 + ) +−1 ≤ (1 + )   = (1 + )  

17
Multiplying by  , we get   + ≤  (1 + )   By summing from  = 1 to 
X
 X


  + ≤  [(1 + )] 
=1 =1

Letting  → ∞ we get
X X
(1 + )
lim 
  + ≤  lim [(1 + )] =   ∞
→∞
=1
→∞
=1
1 − (1 + )
if 1 +   −1  using the sum rule for an infinite geometric series.

As noted in the proof of Proposition 1, the fundamental solution, (25), has the property
that
lim   + = 0 (26)
→∞
That is, the expected value of  is not “explosive”: its absolute value has a growth factor
less than ||−1 . Given ||  1 the fundamental solution is the only solution of (19) with
this property. Indeed, it is seen from (23) that whenever (26) holds, (25) must also hold.
In Example 1 below,  is interpreted as the market price of a share and  as dividends.
Then the fundamental solution gives the share price as the present value of the expected
future flow of dividends.

EXAMPLE 1 (the fundamental value of an equity share) Consider arbitrage between

shares of stock and a riskless asset paying the constant rate of return   0. Let period
 be the current period. Let + be the market price of the share at the beginning of
period  +  and + the dividend paid out at the end of that period,  +   = 0 1 2 .
As seen from period  there is uncertainty about + and + for  = 1 2 . An investor
who buys  shares at time  (the beginning of period ) thus invests  ≡   units
of account at time  At the end of the period the gross return comes out as the known
dividend   and the potential sales value of the shares at the beginning of next period.
This is unlike standard accounting and finance notation in discrete time, where  would
be the end-of-period- market value of the stock of shares that begins to yield dividends
in period  + 1.10
10
Our use of  for the price of a share bought at the beginning of period  is not inconsistent with
our use, in earlier chapters, of  to denote the price, possibly in the same unit of account, per unit
of consumption in period  but paid for at the end of the period. At the beginning of period  after
the uncertainty pertaining to period  has been resolved (thus updating the available information), a
consumer-investor will decide both the investment and the consumption flow for the period. But only
the investment expence,   is disbursed immediately.
It is convenient to think of the course of actions such that receipt of the previous period’s dividend,
−1  and payment for that period’s consumption, at the price −1  occur right before period  begins
and the new information arrives. Indeed, the resolution of uncertainty at discrete points in time motivates
a distinction between “end of” period  − 1 and “beginning of” period , where the new information has
just arrived.

18
 Suppose investors have rational expectations and care only about expected return.
Then the no-arbitrage condition reads
 +  +1 − 
=   0 (27)

This can be written
1 1
 = +1 +   (28)
1+ 1+
which is of the same form as (19) with  =  = 1(1 + ) ∈ (0 1). Assuming dividends do
not grow “too fast”, we find the fundamental solution, denoted ∗  as

1 X X
∞ ∞
1 1 1
∗ =  + 
 
 + =  
+1  +
(29)
1+ 1 +  =1 (1 + ) =0
(1 + )

The fundamental solution is simply the present value of expected future dividends.

If the dividend process is +1 =  + +1  where +1 is white noise, then the dividend
process is known as a random walk and  + =  for  = 1 2   Thus ∗ =  , by
the sum rule for an infinite geometric series. In this case the fundamental value is thus
itself a random walk. More generally, the dividend process could be a martingale, that is,
a sequence of stochastic variables with the property that the expected value next period
exists and equals the current actual value, i.e.,  +1 =  ; but in a martingale, +1
≡ +1 −  need not be white noise; it is enough that  +1 = 011 Given the constant
required return  we still have ∗ =   So the fundamental value itself is in this case a
martingale. ¤

In finance theory the present value of the expected future flow of dividends on an
equity share is referred to as the fundamental value of the share. It is by analogy with
this that the general designation fundamental solution has been introduced for solutions
of form (25). We could also think of  as the market price of a house rented out and
 as the rent. Or  could be the market price of an oil well and  the revenue (net of
extraction costs) from the extracted oil in period 

4.3 Bubble solutions

Other than the fundamental solution, the expectation difference equation (19) has infi-
nitely many bubble solutions. In view of ||  1, these are characterized by violating the
condition (26). That is, they are solutions whose expected value explodes over time.
11
A random walk is thus a special case of a martingale.

19
 It is convenient to first consider the homogenous expectation equation associated with
(19). This is defined as the equation emerging when setting  = 0 in (19):

 =  +1  (30)

Every stochastic process { } of the form

+1 = −1  + +1 , where  +1 = 0 (31)

has the property that

 =  +1  (32)

and is thus a solution to (30). The “disturbance” +1 represents “new information” which
may be related to movements in “fundamentals”, +1  But it does not have to. In fact,
+1 may be related to conditions that per se have no economic relevance whatsoever.

For ease of notation, from now on we just write  even if we think of the whole process
{ } rather than the value taken by  in the specific period  The meaning should be clear
from the context. A solution to (30) is referred to as a homogenous solution associated
with (19). Let  be a given homogenous solution and let  be an arbitrary constant.
Then  =  is also a homogenous solution (try it out for yourself). Conversely, any
homogenous solution  associated with (19) can be written in the form (31). To see this,
let  be a given homogenous solution, that is,  =  +1 . Let +1 = +1 −  +1 .
Then
+1 =  +1 + +1 = −1  + +1 

where  +1 =  +1 −  +1 = 0. Thus,  is of the form (31).

For convenience we here repeat our original expectation difference equation (19):

 =  +1 +        = 0 1 2      6= 0 (*)

PROPOSITION 2 Consider the expectation difference equation (*). Let ̃ be a particular
solution to the expectation difference equation (19), where  6= 0 Then:

(i) every stochastic process of the form

 = ̃ +   (33)

where  satisfies (31), is a solution to (*);

(ii) every solution to (*) can be written in the form (33) with  being an appropriately
chosen homogenous solution associated with (*).

20
Proof. Let some particular solution ̃ be given. (i) Consider  = ̃ +  where  satisfies
(31). Since ̃ satisfies (*), we have  =   ̃+1 +   +  . Consequently, by (30),

 =   ̃+1 +   +   +1 =   (̃+1 + +1 ) +   =   +1 +   

saying that (33) satisfies (*). (ii) Let  be an arbitrary solution to (*). Define  =  − ̃ .
Then we have

 =  − ̃ =  +1 +  − ( ̃+1 +  )

=  (+1 − ̃+1 ) =  +1 

where the second equality follows from the fact that both  and ̃ are solutions to (*).
This shows that  is a solution to the homogenous equation (30) associated with (*).
Since  = ̃ +  , the proposition is hereby proved. ¤

Proposition 2 holds for any  6= 0 In case the fundamental solution (25) exists and
||  1, it is convenient to choose this solution as the particular solution in (33). Thus,
referring to the right-hand side of (25) as ∗ , we can use the particular form,

 = ∗ +   (34)

When the component  is different from zero, the solution (34) is called a bubble
solution and  is called the bubble component. In the typical economic interpretation
the bubble component shows up only because it is expected to show up next period, cf.
(32). The name bubble springs from the fact that the expected value conditional on the
information available in period  explodes over time when ||  1. To see this, as an
example, let 0    1 Then, from (30), by repeated forward substitution we get

 =   (+1 +2 ) = 2  +2 =  =   +   = 1 2 

It follows that  + = −  , and from this follows that the bubble, for  going to infinity,
is unbounded in expected value:
½
∞, if   0
lim  + =  (35)
→∞ −∞ if   0
Indeed, the absolute value of  + will for rising  grow geometrically towards infinity
with a growth factor equal to 1  1

Let us consider a special case of (*19) that allows a simple graphical illustration of
both the fundamental solution and some bubble solutions.

21
 y
II

cx I
1 a

III

Figure 1: Deterministic bubbles (the case 0    1   0 and  = ̄)

4.3.1 When  has constant mean

Suppose the stochastic process  (the “fundamentals”) takes the form  = ̄ +   where
̄ is a constant and  is white noise. Then

 =   +1 + (̄ +  ) 0  ||  1 (36)

The fundamental solution is

X
∞
̄ ̄
∗ =   +   ̄ = ̄ +  +  = +  
=1
1− 1−

Referring to (i) of Proposition 2,

̄
 = +  +  (37)
1−
is thus also a solution of (36) if  is of the form (31).

It may be instructive to consider the case where all stochastic features are eliminated.
So we assume  ≡  ≡ 0. Then we have a model with perfect foresight; the solution (37)
simplifies to
̄
+ 0 − 
 = (38)
1−
where we have used repeated backward substitution in (31). By setting  = 0 we see that
̄
0 = 1−
+ 0  Inserting this into (38) gives

̄ ̄
 = + (0 − )−  (39)
1− 1−

In Fig. 1 we have drawn three trajectories for the case 0    1,   0. Trajectory

I has 0 = ̄(1 − ) and represents the fundamental solution. Trajectory II, with 0

22
 ̄(1 − ) and trajectory III, with 0  ̄(1 −) are bubble solutions. Since we have
imposed no boundary condition apriori, one 0 is as good as any other. The interpretation
is that there are infinitely many trajectories with the property that if only the economic
agents expect the economy will follow that particular trajectory, the aggregate outcome of
their behavior will be that this trajectory is realized. This is the potential indeterminacy
arising when  is not a predetermined variable. However, as alluded to above, in a
complete economic model there will often be restrictions on the endogenous variable(s)
not visible in the basic expectation difference equation(s), here (36). It may be that
the economic meaning of  precludes negative values (a share certificate would be an
example). In that case no-one can rationally expect a path such as III in Fig. 1. Or
perhaps, for some reason, there is an upper bound on  (think of the full-employment
ceiling for output in a situation where the “natural” growth factor for output is smaller
than −1 ). Then no one can rationally expect a trajectory like II in the figure.

To sum up: in order for a solution of a first-order linear expectation difference equation
with constant coefficient , where ||  1 to differ from the fundamental solution, the
solution must have the form (34) where  has the form described in (31). This provides
a clue as to what asset price bubbles might look like.

4.3.2 Asset price bubbles

A stylized fact of stock markets is that stock price indices are quite volatile on a month-to-
month, year-to-year, and especially decade-to-decade scale, cf. Fig. 2. There are different
views about how these swings should be understood. According to the Efficient Market
Hypothesis the swings just reflect unpredictable changes in the “fundamentals”, that is,
changes in the present value of rationally expected future dividends. This is for instance
the view of Nobel laureate Eugene Fama (1970, 2003) from University of Chicago.

In contrast, Nobel laureate Robert Shiller (1981, 2003, 2005) from Yale University,
and others, have pointed to the phenomenon of “excess volatility”. The view is that asset
prices tend to fluctuate more than can be rationalized by shifts in information about
fundamentals (present values of dividends). Although in no way a verification, graphs
like those in Fig. 2 and Fig. 3 are suggestive. Fig. 2 shows the monthly real Standard
and Poors (S&P) composite stock prices and real S&P composite earnings for the period
1871-2008. The unusually large increase in real stock prices since the mid-90’s, which
ended with the collapse in 2000, is known as the “dot-com bubble”. Fig. 3 shows, on a
monthly basis, the ratio of real S&P stock prices to an average of the previous ten years’

23
 2000 500

1800 450

Real S&P composite stock price index

1600 400

Real S&P composite earnings

1400 350

1200 300

1000 250

800 200

600 150

400 100

200 50

0 0
1860 1880 1900 1920 1940 1960 1980 2000 2020

Real price Real earnings

Figure 2: Monthly real S&P composite stock prices from January 1871 to January 2008 (left)
and monthly real S&P composite earnings from January 1871 to September 2007 (right). Source:
http://www.econ.yale.edu/~shiller/data.htm.

real S&P earnings along with the long-term real interest rate. It is seen that this ratio
reached an all-time high in 2000, by many observers considered as “the year the dot-com
bubble burst”.

Shiller’s interpretation of the large stock market swings is that they are due to fads,
herding, and shifts in fashions and “animal spirits” (the latter being a notion from
Keynes).

A third possible source of large stock market swings was pointed out by Blanchard
(1979) and Blanchard and Watson (1982). They argued that bubble phenomena need
not be due to irrational behavior and absence of rational expectations. This lead to the
theory of rational bubbles − the idea that excess volatility can be explained as speculative
bubbles arising from self-fulfilling rational expectations.

Consider an asset which yields either dividends or services in production or consump-

tion in every period in the future. The fundamental value of the asset is, at the theoretical
level, defined as the present value of the expected future flow of dividends or services.12
An asset price bubble (or a speculative bubble) is then defined as a positive deviation of
12
In practice there are many ambiguities involved in this definition of the fundamental value because
it relates to an unknown future.

24
 50 20

45 18

40 16

Long-term real interest rate

35 14
Price-earnings ratio

30 12

25 10

20 8

15 6

10 4

5 2

0 0
1860 1880 1900 1920 1940 1960 1980 2000 2020
Price earnings ratio Year Long-term real interes t rate

Figure 3: S&P price-earnings ratio and long-term real interest rates from January 1881
to January 2008. The earnings are calculated as a moving average over the preceding
ten years. The long-term real interest rate is the 10-year Treasury rate from 1953 and
government bond yields from Sidney Homer, “A History of Interest Rates” from before
1953. Source: http://www.econ.yale.edu/~shiller/data.htm.

25
the market price,   of the asset from its fundamental value, ∗ :

 =  − ∗  (40)

An asset price bubble that emerges in a setting where the no-arbitrage condition (27)
holds under rational expectations, is called a rational bubble. It emerges only because
there is an economy-wide self-fulfilling expectation that it will appreciate at a rate high
enough to warrant the overcharge involved. In the definition in (40) and in the discussion
below we ignore that at a less abstract level it is a systematic deviation, rather than just
a temporary noise deviation, of  from ∗ which qualifies for an asset price bubble.

EXAMPLE 2 (an ever-expanding rational bubble) Consider again an equity share for
which the no-arbitrage condition is
 +  +1 − 
=   0 (41)

As in Example 1, the implied expectation difference equation is  =  +1 +   with 
=  = 1(1 + ) ∈ (0 1) Let the price of the share at time  be  = ∗ +   where ∗ is the
fundamental value and   0 a bubble component following the deterministic process,
+1 = (1+)  0  0 so that  = 0 (1+)  This is called a deterministic rational bubble.
Agents may be ready to pay a price over and above the fundamental value (whether or
not they know the “true” fundamental value) if they expect they can sell at a sufficiently
higher price later; trading with such motivation is called speculative behavior. If generally
held and lasting for some time, this expectation may be self-fulfilling. Note that (41)
implies that the asset price ultimately grows at the rate . Indeed, let  = 0 (1 + ) 
   (if  ≤  the asset price would be infinite). By the rule of the sum of an infinite
geometrice series, we then have ∗ =  ( −) showing that the fundamental value grows
at the rate  Consequently,   = (∗ +  ) = ∗  + 1 → 1 as    It follows that
the asset price in the long run grows at the same rate as the bubble, the rate 

We are not acquainted with ever-expanding incidents of that caliber in real world
situations, however. A deterministic rational bubble is implausible. ¤

In some contexts it may not matter whether or not we think of the “rational” market
participants as knowing the probability distribution of the “fundamentals”, hence knowing
∗ (by “fundamentals” is meant any information relating to the future dividend or service
capacity of an asset: a firm’s technology, resources, market conditions etc.). All the same,
it seems common to imply such a high level of information in the term “rational bubbles”.
Unless otherwise indicated, we shall let this implication be understood.

26
 While a deterministic rational bubble was found implausible, let us now consider an
example of a stochastic rational bubble which sooner or later bursts.

EXAMPLE 3 (a bursting bubble) Once again we consider the no-arbitrage condition is

(41) where for simplicity we still assume the required rate of return is constant, though
possibly including a risk premium. Following Blanchard (1979), we assume that the
market price,   of the share contains a stochastic bubble of the following form:
½ 1+
 with probability  
 
+1 = (42)
0 with probability 1 −  
where  = 0 1 2  and 0  0. In addition we may assume that  = (∗   ) ∗ ≥ 0
 ≤ 0 If ∗  0 the probability that the bubble persists at least one period ahead is
higher the greater the fundamental value has become. If   0 the probability that
the bubble persists at least one period ahead is less, the greater the bubble has already
become. In this way the probability of a crash becomes greater and greater as the share
price comes further and further away from fundamentals. As a compensation, the longer
time the bubble has lasted, the higher is the expected growth rate of the bubble in the
absence of a collapse.

This bubble satisfies the criterion for a rational bubble. Indeed, (42) implies
1+
 +1 = (  )+1 + 0 · (1 − +1 ) = (1 + ) 
+1
This is of the form (31) with −1 = 1 +  and the bubble is therefore a stochastic
rational bubble. The stochastic component is +1 = +1 −  +1 = +1 − (1 + )
and has conditional expectation equal to zero. Although +1 must have zero conditional
expectation, it need not be white noise (it can for instance have varying variance). ¤

As this example illustrates, a stochastic rational bubble does not have the implausible
ever-expanding form of a deterministic rational bubble. Yet, under certain conditions
even stochastic rational bubbles can be ruled out or at least be judged implausible. The
next section reviews some arguments.

4.4 When rational bubbles in asset prices can or can not be

ruled out

We concentrate on assets whose services are valued independently of the price.13 Let 
be the market price and ∗ the fundamental value of the asset as of time . Even if the
13
This is in contrast to assets that serve as means of payment.

27
asset yields services rather than dividends, we think of ∗ as in principle the same for all
agents. This is because a user who, in a given period, values the service flow of the asset
relatively low can hire it out to the one who values it highest (the one with the highest
willingness to pay). Until further notice we assume ∗ known to the market participants.

4.4.1 Partial equilibrium arguments

The principle of reasoning to be used is called backward induction: If we know something

about an asset price in the future, we can conclude something about the asset price today.

(a) Assets which can be freely disposed of (“free disposal”) Can a rational asset
price bubble be negative? The answer is no. The logic can be illustrated on the basis
of Example 2 above. For simplicity, let the dividend be the same constant   0 for all
 = 0 1 2 . Then, from the formula (39) we have

 − ∗ = (0 − ∗ )(1 + ) 

where   0 and ∗ =  Suppose there is a negative bubble in period 0, i.e., 0 −∗  0
In period 1, since 1 +   1 the bubble is greater in absolute value. The downward
movement of  continues and sooner or later  is negative. The intuition is that the
low 0 in period 0 implies a high dividend-price ratio. Hence a negative capital gain
(+1 −   0) is needed for the no-arbitrage condition (41) to hold. Thereby 1  0 
and so on.

But in a market with self-interested rational agents, an object which can be freely
disposed of can never have a negative price. A negative price means that the “seller”
has to pay to dispose of the object. Nobody will do that if the object can just be
thrown away. An asset which can be freely disposed of (share certificates for instance)
can therefore never have a negative price. We conclude that a negative rational bubble
can not be consistent with rational expectations. Similarly, with a stochastic dividend,
a negative rational bubble would imply that in expected value the share price becomes
negative at some point in time, cf. (35). Again, rational expectations rule this out.

Hence, if we imagine that for a short moment   ∗ , then everyone will want to buy
the asset and hold it forever, which by own use or by hiring out will imply a discounted
value equal to ∗  There is thus excess demand until  has risen to ∗ 

When a negative rational bubble can be ruled out, then, if at the first date of trading
of the asset there were no positive bubble, neither can a positive bubble arise later. Let

28
us make this precise:

PROPOSITION 3 Assume free disposal of a given asset. Then, if a rational bubble in the
asset price is present today, it must be positive and must have been present also yesterday
and so on back to the first date of trading the asset. And if a rational bubble bursts, it
will not restart later.

Proof As argued above, in view of free disposal, a negative rational bubble in the asset
price can be ruled out. It follows that  =  − ∗ ≥ 0 for  = 0 1 2  where  = 0 is
the first date of trading the asset. That is, any rational bubble in the asset price must be
a positive bubble. We now show by contradiction that if, for an arbitrary  = 1 2  it
holds that   0 then −1  0. Let   0 Then, if −1 = 0 we have −1  = −1 
= 0 (from (31) with  replaced by  − 1), implying, since   0 is not possible, that  = 0
with probability one as seen from period  − 1 Ignoring zero probability events, this rules
out   0 and we have arrived at a contradiction. Thus −1  0 Replacing  by  − 1
and so on backward in time, we end up with 0  0. This reasoning also implies that if
a bubble bursts in period , it can not restart in period  + 1 nor, by extension, in any
subsequent period. ¤

This proposition (due to Diba and Grossman, 1988) claims that a rational bubble in
an asset price must have been there since trading of the asset began. Yet such a conclusion
is not without ambiguities. If new information about radically new technology comes up
at some point in time, is a share in the firm then the same asset as before? In a legal
sense the firm is the same, but is the asset also the same? Even if an earlier bubble has
crashed, cannot a new rational bubble arise later in case of an utterly new situation?

These ambiguities reflect the difficulty involved in the concepts of rational expectations
and rational bubbles when we are dealing with uncertainties about future developments of
the economy. The market’s evaluation of many assets of macroeconomic importance, not
the least shares in firms, depends on vague beliefs about future preferences, technologies,
and societal circumstances. The fundamental value can not be determined in any objective
way. There is no well-defined probability distribution over the potential future outcomes.
Fundamental uncertainty, also called Knightian uncertainty,14 is present.

(b) Bonds with finite maturity The finite maturity ensures that the value of the bond
is given at some finite future date. Therefore, if there were a positive bubble in the market
14
After the Chicago of University economist Frank Knight who in his book, Risk, Uncertainty, and
Profit (1921), coined the important distinction between measurable risk and unmeasurable uncertainty.

29
price of the bond, no rational investor would buy just before that date. Anticipating this,
no one would buy the date before, and so on. Consequently, nobody will buy in the first
place. By this backward-induction argument follows that a positive bubble cannot get
started. And since there also is “free disposal”, all rational bubbles can be precluded.

From now on we take as given that negative rational bubbles are ruled out. So, the
discussion is about whether positive rational asset price bubbles may exist or not.

(c) Assets whose supply is elastic Real capital goods (including buildings) can be
reproduced and have clearly defined costs of reproduction. This precludes rational bubbles
on this kind of assets, since a potential buyer can avoid the overcharge by producing
instead. Notice, however, that building sites with a specific amenity value and apartments
in attractive quarters of a city are not easily reproducible. Therefore, rational bubbles on
such assets are more difficult to rule out.

Here are a few intuitive remarks about bubbles on shares of stock in an established
firm. An argument against a rational bubble might be that if there were a bubble, the
firm would tend to exploit it by issuing more shares. But thereby market participants
mistrust is raised and may pull market evaluation back to the fundamental value. On
the other hand, the firm might anticipate this adverse response from the market. So the
firm chooses instead to “fool” the market by steady financing behavior, calmly enjoying
its solid equity and continuing as if no bubble were present. It is therefore not obvious
that this kind of argument can rule out rational bubbles on shares of stock.

(d) Assets for which there exists a “backstop-technology” For some articles of
trade there exists substitutes in elastic supply which will be demanded if the price of
the article becomes sufficiently high. Such a substitute is called a “backstop-technology”.
For example oil and other fossil fuels will, when their prices become sufficiently high,
be subject to intense competition from substitutes (renewable energy sources). This
precludes an unbounded bubble process in the price of oil.

On account of the arguments (c) and (d), it seems more difficult to rule out rational
bubbles when it comes to assets which are not reproducible or substitutable, let alone
assets whose fundamentals are difficult to ascertain. For some assets the fundamentals
are not easily ascertained. Examples are paintings of past great artists, rare stamps,
diamonds, gold etc. Also new firms that introduce completely novel products and tech-
nologies are potential candidates. Think of the proliferation of radio broadcasting in the

30
1920s before the wall Street crash in 1929 and the internet in the 1990s before the dotcom
bubble burst in 2000.

What these situations allow for may not be termed rational bubbles, if by definition
this concept requires a well-defined fundamental. Then we may think of a broader class
of real-world bubbly phenomena driven by self-reinforcing expectations.

4.4.2 Adding general equilibrium arguments

The above considerations are of a partial equilibrium nature. On top of this, general
equilibrium arguments can be put forward to limit the possibility of rational bubbles. We
may briefly give a flavour of two such general equilibrium arguments. We still consider
assets whose services are valued independently of the price and which, as in (a) above,
can be freely disposed of. A house, a machine, or a share in a firm yields a service in
consumption or production or in the form of a dividend stream. Since such an asset has
an intrinsic value, ∗  equal to the present value of the flow of services, one might believe
that positive rational bubbles on such assets can be ruled out in general equilibrium.
As we shall see, this is indeed true for an economy with a finite number of “neoclassical”
households (to be defined below), but not necessarily in an overlapping generations model.
Yet even there, rational bubbles can under certain conditions be ruled out.

(e) An economy with a finite number of infinitely-lived households Assume

that the economy consists of a finite number of infinitely-lived agents − here called house-
holds − indexed  = 1 2  . The households are “neoclassical” in the sense that they
save only with a view to future consumption.

Under free disposal in point (a) we saw that   ∗ can not be an equilibrium. We
now consider the case of a positive bubble, i.e.,   ∗  All owners of the bubble asset
who are users will in this case prefer to sell and then rent; this would imply excess supply
and could thus not be an equilibrium. Hence, we turn to households that are not users,
but speculators. Assuming “short selling” is legal, speculators may pursue “short selling”,
that is, they first rent the asset (for a contracted interval of time) and immediately sell
it at  . This results in excess supply and so the asset price falls towards ∗ . Within the
contracted interval of time the speculators buy the asset back and return it to the original
owners in accordance with the loan accord. So   ∗ can not be an equilibrium.

Even ruling out “short selling” (which is sometimes outright forbidden), we can ex-
clude positive bubbles in the present setup with a finite number of households. To assume

31
that owners who are not users would want to hold the bubble asset forever as a permanent
investment will contradict that these owners are “neoclassical”. Indeed, their transver-
sality condition would be violated because the value of their wealth would grow at a rate
asymptotically equal to the rate of interest. This would allow them to increase their
consumption now without decreasing it later and without violating their No-Ponzi-Game
condition.

We have to instead imagine that the “neoclassical” households who own the bubble
asset, hold it against future sale. This could on the face of it seem rational enough
if there were some probability that not only would the bubble continue to exist, but
it would also grow so that the return would be at least as high as that yielded on an
alternative investment. Owners holding the asset in the expectation of a capital gain, will
thus plan to sell at some later point in time. Let  be the point in time where household
 wishes to sell and let
 = max{1  2    }

Then nobody will plan to hold the asset after  The household speculator,  having
 =  will thus not have anyone to sell to (other than people who will only pay ∗ )
Anticipating this, no-one would buy or hold the asset the period before, and so on. So
no-one will want to buy or hold the asset in the first place.

The conclusion is that   ∗ cannot be a rational expectations equilibrium in a setup

with a finite number of “neoclassical” households.

The same line of reasoning does not, however, go through in an overlapping generations
model where new households − that is, new traders − enter the economy every period.

(f) An economy with interest rate above the output growth rate In an overlap-
ping generations (OLG) model with an infinite sequence of new decision makers, rational
bubbles are under certain conditions theoretically possible. The argument is that with
 → ∞  as defined above is not bounded. Although this unboundedness is a necessary
condition for rational bubbles, it is not sufficient, however.

To see why, let us return to the arbitrage examples 1, 2, and 3 where we have −1 =
1 +  so that a hypothetical rational bubble has the form +1 = (1 + ) ++1  where
 +1 = 0 So in expected value the hypothetical bubble is growing at a rate equal to
the interest rate,  If at the same time  is higher than the long-run output growth rate,
the value of the expanding bubble asset would sooner or later be larger than GDP and
aggregate saving would not suffice to back its continued growth. Agents with rational

32
expectations anticipate this and so the bubble never gets started.

This point is valid when the interest rate in the OLG economy is higher than the
growth rate of the economy − which is normally considered the realistic case. Yet, the
opposite case is possible and in that situation it is less easy to rule out rational asset
price bubbles. This is also the case in situations with imperfect credit markets. It turns
out that the presence of segmented financial markets or externalities that create a wedge
between private and social returns on productive investment may increase the scope for
rational bubbles (Blanchard, 2008).

4.5 Conclusion

The empirical evidence concerning asset price bubbles in general and rational asset price
bubbles in particular seems inconclusive. It is very difficult to statistically distinguish
between bubbles and mis-specified fundamentals. Rational bubbles can also have more
complicated forms than the bursting bubble in Example 3 above. For example Evans
(1991) and Hall et al. (1999) study “regime-switching” rational bubbles.

Whatever the possible limits to the plausibility of rational bubbles in asset prices, it is
useful to be aware of their logical structure and the variety of forms they can take as logical
possibilities. Rational bubbles may serve as a benchmark for a variety of “behavioral asset
price bubbles”, i.e., bubbles arising through particular psychological mechanisms. This
would take us to behavioral finance theory. The reader is referred to, e.g., Shiller (2003).

For surveys on the theory of rational bubbles and econometric bubble tests, see Salge
(1997) and Gürkaynak (2008). For discussions of famous historical bubble episodes, see
the symposium in Journal of Economic Perspectives 4, No. 2, 1990, and Shiller (2005).

5 Appendix

A. The log-linear specification

In many macroeconomic models with rational expectations the equations are specified as
log-linear, that is, as being linear in the logarithms of the variables. If   and  are
the original positive stochastic variables, defining  = ln  ,  = ln  and  = ln , a
log-linear relationship between   and  is a relation of the form

 =  +  +  (43)

33
where   and  are constants. The motivation for assuming log-linearity can be:

(a) Linearity is convenient because of the simple rule for the expected value of a sum:
( +  + ) =  + () + (), where  is the expectation operator. Indeed,
for a non-linear function, ( ) we generally have (( )) 6= (() ()).

(b) Linearity in logs may often seem a more realistic assumption than linearity in any-
thing else.

(c) In time series models a logarithmic transformation of the variables followed by

formation of first differences can be the road to eliminating a trend in the mean
and variance.

As to point (b) we state the following:

CLAIM To assume linearity in logs is equivalent to assuming constant elasticities.

Proof Let the positive variables  ,  and  be related by  =  (, ), where  is a
continuous function with continuous partial derivatives. Taking the differential on both
sides of ln  = ln  ( ) we get
1  1 
 ln  =  +  (44)
 ( )   ( ) 
       
= + =   +   =     ln  +    ln 
       
where    and    are the partial elasticities of  w.r.t.  and , respectively. Thus,
defining  = ln  ,  = ln  and  = ln , gives

 =     +    (45)

Assuming constant elasticities amounts to putting   =  and   = , where  and

 are constants. Then we can write (45) as  =  + . By integration, we get (43)
where  is now an arbitrary integration constant. Hereby we have shown that constant
elasticities imply a log-linear relationship between the variables.

Now, let us instead start by assuming the log-linear relationship (43). Then,
 
=  =  (46)
 
But (43), together with the definitions of ,  and  implies that

 = ++ = + ln + ln  

34
from which follows that
 1  
=   so that   ≡ = 
   
and
 1  
=   so that    ≡ = 
   
That is, the partial elasticities are constant. ¤

So, when the variables are in logs, then the coefficients in the linear expressions are
the elasticities. Note, however, that the interest rate is normally an exception. It is often
regarded as more realistic to let the interest rate itself and not its logarithm enter linearly.
Then the associated coefficient indicates the semi-elasticity with respect to the interest
rate.

B. Conditional expectations and the law of iterated expectations

The mathematical conditional expectation is a weighted sum of the possible values of the
stochastic variable with weights equal to the corresponding conditional probabilities.

Let  and  be two discrete stochastic variables with joint probability function ( )
and marginal probability functions () and () respectively. If the conditional probabil-
ity function for  given  = 0 is denoted ( |0 )  we have ( |0 ) = ( 0 )(0 ) as-
suming (0 )  0 The conditional expectation of  given  = 0  denoted ( | = 0 )
is then
X ( 0 )
( | = 0 ) =   (47)

(0 )
where the summation is over all the possible values of 

This conditional expectation is a function of 0  Since 0 is just one possible value of

the stochastic variable  we interpret the conditional expectation itself as a stochastic
variable and write it as ( |) Generally, for a function of the discrete stochastic variable
 say () the expected value is
X
(()) = ()()


When we here let the conditional expectation ( |) play the role of () and sum over
all  for which ()  0 we get
à !
X X X ( )
(( |)) = ( |)() =  () (by (47))
  
()
à !
X X X
=  ( ) = () = ( )
  

35
 This result is a manifestation of the law of iterated expectations: the unconditional
expectation of the conditional expectation of  is given by the unconditional expectation
of 

Now consider the case where  and  are continuous stochastic variables with joint
probability density function ( ) and marginal density functions () and () respec-
tively. If the conditional density function for  given  = 0 is denoted ( |0 )  we have
( |0 ) = ( 0 )(0 ) assuming (0 )  0 The conditional expectation of  given
 = 0 is Z ∞
( 0 )
( | = 0 ) =   (48)
−∞ (0 )
where we have assumed that the range of  is (−∞ ∞) Again, we may view the condi-
tional expectation itself as a stochastic variable and write it as ( |) Generally, for a
function of the continuous stochastic variable  say () the expected value is
Z
(()) = ()()


where  stands for the range of  When we let the conditional expectation ( |) play
the role of () we get
Z Z µZ ∞ ¶
( )
(( |)) = ( |)() =   () (by (48))
  −∞ ()
Z ∞ µZ ¶ Z ∞
=  ( )  =  () = ( ) (49)
−∞  −∞

This shows us the law of iterated expectations in action for continuous stochastic
variables: the unconditional expectation of the conditional expectation of  is given by
the unconditional expectation of 

EXAMPLE Let the two stochastic variables,  and  follow a two-dimensional normal
distribution. Then, from mathematical statistics we know that the conditional expectation
of  given  satisfies
Cov( )
( |) = ( ) + ( − ())
Var()
Taking expectations on both sides gives
Cov( )
(( |)) = ( ) + (() − ()) = ( ) ¤
Var()

We may also express the law of iterated expectations in terms of subsets of the original
outcome space for a stochastic variable. Let the event A be a subset of the outcome space

36
for  and let B be a subset of A. Then the law of iterated expectations takes the form

(( |B)|A) = ( |A) (50)

That is, when B ⊆ A the expectation, conditional on A of the expectation of  , condi-

tional on B, is the same as the expectation, conditional on A, of 

In the text of this and the subsequent chapters we consider a dynamic context where
expectations are conditional on dated information − ( = 1 2 ). By a, so far, “informal
analogy” with (49) we then write the law of iterated expectations this way:

(( |− )) = ( ) for  = 1 2  (51)

In words: the unconditional expectation of the conditional expectation of   given the

information up to time  −  equals the unconditional expectation of   Similarly, by a,
so far, “informal analogy” with (50) we may write

((+2 |+1 )| ) = (+2 | ) (52)

That is, the expectation today of the expectation tomorrow, when more may be known,
of a variable the day after tomorrow is the same as the expectation today of the variable
the day after tomorrow. Intuitively: you ask a stockbroker in which direction she expects
to revise her expectations upon the arrival of more information. If the broker answers
“upward”, say, then another broker is recommended.

The notation used in the transition from (50) to (52) might seem problematic, though.
That is why we talk of “informal analogy”. The sets A and B are subsets of the outcome
space and B ⊆ A In contrast, the “information” or “information content” represented by
our symbol  will, for the uninitiated, inevitably be understood in a meaning not fitting
the inclusion +1 ⊆  . Intuitively “information” dictates the opposite inclusion, namely
as a set which expands over time − more and more “information” (like “knowledge” or
“available data”) is revealed as time proceeds.

It is possible, however, to interpret the information  from another angle so as to

make the notation in (52) fully comply with that in (50). Let the outcome space Ω denote
the set of ex ante possible15 sequences {(   )}=0  where  and  are vectors of
date- endogenous and exogenous stochastic variables, respectively, and where  is the
time horizon, possibly  = ∞. For  ∈ {0  0 + 1 . . .  0 +  }  let the subset Ω ⊆ Ω
be defined as the of time  still possible sequences {(   )}=
0 +
0
 Now, as time proceeds,
15
By “possible” is meant “ex ante feasible according to a given model”.

37
more and more realizations occur, that is, more and more of the ex ante random states
(   ) become historical data, (   ) Hence, as time proceeds, the subset Ω shrinks
in the sense that Ω+1 ⊆ Ω . The increasing amount of information and the “reduced
uncertainty” can thus be seen as two sides of the same thing. Interpreting  this way,
i.e., as “partial lack of uncertainty”, the expression (52) means the same thing as

((+2 |Ω+1 )|Ω ) = (+2 |Ω )

This is in complete harmony with (50).

C. Properties of the model-consistent forecast

As in the text of Section 24.2.2, let  denote the model-consistent forecast error  −
( |−1 ) Then, if −1 represents information contained in −1 ,

( |−1 ) = ( − ( |−1 ) |−1 ) = ( |−1 ) − (( |−1 ) |−1 )
= ( |−1 ) − ( |−1 ) = 0 (53)

where we have used that (( |−1 ) |−1 ) = ( |−1 )  by the law of iterated expec-
tations. With −1 = −1 we have, as a special case,

( |−1 ) = 0 as well as (54)

( ) = ( − ( |−1 )) = ( ) − (( |−1 )) = 0

in view of (51) with  = 1. This proves property (a) in Section 24.2.3.

As to property (b) in Section 24.2.2, for  = 1 2  let − be an arbitrary variable
value belonging to the information − . Then, ( − |− ) = − ( |− ) = 0 by
(53) with −1 = − (since − is contained in −1 ). Thus, by the principle (51),

( − ) =  (( − |− )) = (0) = 0 for  = 1 2  (55)

This result is known as the orthogonality property of model-consistent expectations (two

stochastic variables  and  are said to be orthogonal if ( ) = 0) From the general
formula for the (unconditional) covariance follows

Cov( − ) = ( − ) − ( )(− ) = 0 − 0 = 0 for  = 1 2 

by (54) and (55). In particular, with − = −  we get Cov( − ) = 0 This proves that
model-consistent forecast errors exhibit lack of serial correlation.

38
6 Exercises

1. Let { } be a stochastic process in discrete time. Suppose  =  +  , where

 = −1 +  and  and  are white noise.

a) Is { } a random walk? Why or why not?

b) Is { } a random walk? Why or why not?

c) Calculate the rational expectation of  conditional on all relevant information up

to and including period  − 1.

d) What is the rational expectation of  conditional on all relevant information up to

and including period  − 1?

e) Compare with the subjective expectation of  based om the adaptive expectations

formula with adjustment speed equal to one.

2. Consider a simple Keynesian model of a closed economy with constant wages and
prices (behind the scene), abundant capacity, and output determined by demand:

 =  =  + ¯ +   (1)

 =  + −1    0 0    1 (2)
 = (1 − )̄ + −1 +   ̄  0 0    1 (3)

where the endogenous variables are  = output (= income),  = aggregate demand,


 = consumption, and −1 = expected output (income) in period  as seen from period
−1 while  , which stands for government spending on goods and services, is considered
¯ and the parameters  
exogenous as is  , which is white noise. Finally, investment, ,
 and ̄ are given positive constants.

Suppose expectations are “static” in the sense that expected income in period  equals
actual income in the previous period.

a) Solve for  .

b) Find the income multiplier (partial derivative of  ) with respect to a change in

−1 and   respectively

39
 Suppose instead that expectations are rational.

c) Explain what this means.

d) Solve for  

e) Find the income multiplier with respect to a change in −1 and   respectively.

f) Compare the result under e) with that under b). Comment.

3. Consider arbitrage between equity shares and a riskless asset paying the constant
rate of return   0. Let  denote the price at the beginning of period  of a share that
at the end of period  yields the dividend  . As seen from period  there is uncertainty
about + and + for  = 1 2. . . . Suppose agents have rational expectations and care
only about expected return (risk neutrality).

a) Write down the no-arbitrage condition.

Suppose dividends follow the process  = ¯ +   where ¯ is a positive constant and

 is white noise, observable in period  but not known in advance.

b) Find the fundamental solution for  and let it be denoted ∗ . Hint: given 
P
=  +1 +    the fundamental solution is  =  +  ∞ 
=1   + 

Suppose someone claims that the share price follows the process

 = ∗ +  

with a given 0  0 and, for  = 0 1 2. . . ,

½ 1+
 with probability  
 
+1 =
0 with probability 1 −  
where  = ( )  0  0

c) What is an asset price bubble and what is a rational asset price bubble?

d) Can the described  process be a rational asset price bubble? Hint: a bubble
component associated with the inhomogenous equation  =  +1 +   is a
solution, different from zero, to the homogeneous equation,  =  +1 .

40
Chapter 16

Money in macroeconomics

Money buys goods and goods buy money; but goods do not buy goods.
Robert W. Clower (1967).

Up to now we have put monetary issues aside. The implicit assumption has
been that the exchange of goods and services in the market economy can be
carried out without friction as mere intra- or intertemporal barter. This is, of
course, not realistic. At best it can provide an acceptable approximation to reality
only for a limited set of macroeconomic issues. We now turn to models in which
there is a demand for money. We thus turn to monetary theory, that is, the study
of causes and consequences of the fact that a large part of the exchange of goods
and services in the real world is mediated through the use of money.

16.1 What is money?

16.1.1 The concept of money
In economics money is de…ned as an asset (a store of value) which functions as a
generally accepted medium of exchange, i.e., it can be used directly to buy any
good o¤ered for sale in the economy. A note of IOU (a bill of exchange) may
also be a medium of exchange, but it is not generally accepted and is therefore
not money.1 Moreover, the extent to which an IOU is acceptable in exchange
depends on the general state in the economy. In contrast, money is characterized
by being a fully liquid asset. An asset is fully liquid if it can be used directly,
instantly, and without any extra costs or restrictions to make payments.
1
Generally accepted mediums of exchange are also called means of payment.

645
646 CHAPTER 16. MONEY IN MACROECONOMICS

Figure 16.1: No direct exchange possible. A medium of exchange, here good 2, solves
the problem (details in text).

Generally, liquidity should be conceived as a matter of degree so that an asset

has a higher or lower degree of liquidity depending on the extent to which it can
easily be exchanged for money. By “easily”we mean “immediately, conveniently,
and cheaply”. So an asset’s liquidity is the ease with which the asset can be
converted into money or be used directly for making payments. Where to draw the
line between “money”and “non-money assets”depends on what is appropriate for
the problem at hand. In the list below of di¤erent monetary aggregates (Section
16.2), M1 corresponds most closely to the traditional de…nition of money. De…ned
as currency in circulation plus demand deposits held by the non-bank public in
commercial banks, M1 embraces all under “normal circumstances” fully liquid
assets in the hands of the non-bank public.
The reason that a market economy uses money is that money facilitates trade
enormously, thereby reducing transaction costs. Money helps an economy to avoid
the need for a “double coincidence of wants”. The classical way of illustrating
this is by the exchange triangle in Fig. 16.1. The individuals A, B, and C are
endowed with one unit of the goods 1, 3, and 2, respectively. But A, B, and C
want to consume 3, 2, and 1, respectively. Thus, no direct exchange is possible
between two individuals each wanting to consume the other’s good. There is
a lack of double coincidence of wants. The problem can be solved by indirect
exchange where A exchanges good 1 for good 2 with C and then, in the next
step, uses good 2 in an exchange for good 3 with B. Here good 2 serves as a
medium of exchange. If good 2 becomes widely used and accepted as a medium
of exchange, it is money. Extending the example to a situation with n goods,
we have that exchange without money (i.e., barter) requires n(n 1)=2 markets
(“trading spots”). Exchange with money, in the form of modern “paper money”,
requires only n markets.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.1. What is money? 647

16.1.2 Historical remarks

In the past, ordinary commodities, such as seashells, rice, cocoa, precious metals
etc., served as money. That is, commodities that were easily divisible, handy
to carry, immutable, and involved low costs of storage and transportation could
end up being used as money. This form of money is called commodity money.
Applying ordinary goods as a medium of exchange is costly, however, because
these goods have alternative uses. A more e¢ cient way to trade is by using
currency, i.e., coins and notes in circulation with little or no intrinsic value, or
pieces of paper, checks, representing claims on such currency. Regulation by a
central authority (the state or the central bank) has been of key importance in
bringing about this transition into the modern payment system.
Coins, notes, pieces of paper like checks, and electronic signals from smart
phones to accounts in a bank have no intrinsic value. Yet they may be generally
accepted media of exchange, in which case we refer to them as paper money. By
having these pieces of paper circulating and the real goods moving only once,
from initial producer to …nal consumer, the trading costs in terms of time and
e¤ort are minimized.
In the industrialized countries these paper monies were in the last third of
the nineteenth century and until the outbreak of the First World War backed
through the gold standard. And under the Bretton-Woods agreement, 1947-71,
the currencies of the developed Western countries outside the United States were
convertible into US dollars at a …xed exchange rate (or rather an exchange rate
which is adjustable only under speci…c circumstances); and US dollar reserves
of these countries were (in principle) convertible into gold by the United States
at a …xed price (though in practice with some discouragement from the United
States).
This indirect gold-exchange standard broke down in 1971-73, and nowadays
money in most countries is unbacked paper money (including electronic entries
in banks’ accounts). This feature of modern money makes its valuation very
di¤erent from that of other assets. A piece of paper money in a modern payments
system has no worth at all to an individual unless she expects other economic
agents to value it in the next instant. There is an inherent circularity in the
acceptance of money. Hence the viability of such a paper money system is very
much dependent on adequate juridical institutions as well as con…dence in the
ability and willingness of the government and central bank to conduct policies
that sustain the purchasing power of the currency. One elementary juridical
institution is that of “legal tender”, a status which is conferred to certain kinds
of money. An example is the law that a money debt can always be settled by
currency and a tax always be paid by currency. A medium of exchange whose
market value derives entirely from its legal tender status is called …at money

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

648 CHAPTER 16. MONEY IN MACROECONOMICS

(because the value exists through “…at”, a ruler’s declaration). In view of the
absence of intrinsic value, maintaining the exchange value of …at money over
time, that is, avoiding high or ‡uctuating in‡ation, is one of the central tasks of
monetary policy.

16.1.3 The functions of money

The following three functions are sometimes considered to be the de…nitional
characteristics of money:

1. It is a generally accepted medium of exchange.

2. It is a store of value.

3. It serves as a unit of account in which prices are quoted and books kept
(the numeraire).

On can argue, however, that the last function is on a di¤erent footing com-
pared to the two others. Thus, we should make a distinction between the func-
tions that money necessarily performs, according to our de…nition above, and the
functions that money usually performs. Property 1 and 2 certainly belong to the
essential characteristics of money. By its role as a device for making transactions
money helps an economy to avoid the need for a double coincidence of wants.
In order to perform this role, money must be a store of value, i.e., a device that
transfers and maintains value over time. The reason that people are willing to
exchange their goods for pieces of paper is exactly that these can later be used
to purchase other goods. As a store of value, however, money is dominated by
other stores of value such as bonds and shares that pay a higher rate of return.
When nevertheless there is a demand for money, it is due to the liquidity of this
store of value, that is, its service as a generally accepted medium of exchange.
Property 3, however, is not an indispensable function of money as we have
de…ned it. Though the money unit is usually used as the unit of account in which
prices are quoted, this function of money is conceptually distinct from the other
two functions and has sometimes been distinct in practice. During times of high
in‡ation, foreign currency has been used as a unit of account, whereas the local
money continued to be used as the medium of exchange. During the German
hyperin‡ation of 1922-23 US dollars were the unit of account used in parts of the
economy, whereas the mark was the medium of exchange; and during the Russian
hyperin‡ation in the middle of the 1990s again US dollars were often the unit of
account, but the rouble was still the medium of exchange.
This is not to say that it is of little importance that money usually serves
as numeraire. Indeed, this function of money plays an important role for the

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.2. The money supply 649

short-run macroeconomic e¤ects of changes in the money supply. These e¤ects

are due to nominal rigidities, that is, the fact that prices, usually denominated
in money, of most goods and services generally adjust only sluggishly (they are
not traded in auction markets).

16.2 The money supply

The money supply is the total amount of money available in an economy at a
particular point in time (a stock). As noted above, where to draw the line between
assets that should be counted as money and those that should not, depends on
the context.

16.2.1 Di¤erent measures of the money stock

Usually the money stock in an economy is measured as one of the following
alternative monetary aggregates:

M0 ; i.e., the monetary base, alternatively called base money, central bank
money, or high-powered money. The monetary base is de…ned as fully liquid
claims on the central bank held by the private sector, that is, currency (coins
and notes) in circulation plus demand deposits held by the commercial
banks in the central bank.2 This monetary aggregate is under the direct
control of the central bank and is changed by open-market operations, that
is, by the central bank trading bonds, usually short-term government bonds,
with the private sector. But clearly the monetary base is an imperfect
measure of the liquidity in the private sector.

M1 ; de…ned as currency in circulation plus demand deposits held by the

non-bank general public in commercial banks. These deposits are also called
checking accounts because they are deposits on which checks can be written
and payment cards (debit cards) be used. M1 does not include currency
held by commercial banks and demand deposits held by commercial banks
in the central bank. Yet M1 includes the major part of M0 and is generally
considerably larger than M0 : The measure M1 is intended to re‡ect the
quantity of assets serving as media of exchange in the hands of the non-
bank general public, i.e., the non-bank part of the private sector.
Broader categories of money include:
2
The commercial banks are usually part of the private sector and by law it is generally only
the commercial banks that are allowed to have demand deposits in the central bank the
“banks’bank”.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

650 CHAPTER 16. MONEY IN MACROECONOMICS

M2 = M1 plus savings deposits with unrestricted access and small-denomination

time deposits (say below e 100,000). Although these claims may not be
instantly liquid, they are close to.

M3 = M2 plus large-denomination (say above e 100,000) time-deposits.3

As we move down the list, the liquidity of the added assets decreases, while
their interest yield increases.4 Currency earns zero interest. When in macroeco-
nomic texts the term “money supply” is used, traditionally M1 or M2 has been
meant; there is, however, a rising tendency to focus on M3 . Along with currency,
the demand deposits in the commercial banks are normally fully liquid, at least
as long as they are guaranteed by a governmental deposit insurance (although
normally only up to a certain maximum per account). The interest earned on
these demand deposits is usually low (at least for “small”depositors) and in fact
often ignored in simple theoretical models.
A related and theoretically important simple classi…cation of money types is
the following:

1. Outside money = money that on net is an asset of the private sector.

2. Inside money = money that is not net wealth of the private sector.

Clearly M0 is outside money. Most money in modern economies is inside

money, however. Deposits at the commercial banks is an example of inside money.
These deposits are an asset to their holders, but a liability of the banks. Even
broader aggregates of money (or “near-money”) than M3 are sometimes consid-
ered. For instance, it has been argued that the amounts that people are allowed
to charge by using their credit cards should be included in the concept of “broad
money”. But this would involve double counting. Actually you do not pay when
you use a credit card at the store. It is the company issuing the credit card that
pays to the store (shortly after you made your purchases). You postpone your
payment until you receive your monthly bill from the credit card company. That
is, the credit card company does the payment for you and gives credit to you. It
is otherwise with a payment card where the amount for which you buy is instantly
charged your account in the bank.
3
In casual notation, M1 M2 M3 ; but M0 * M1 since only a part of M0 belongs to M1 :
4
This could be an argument for weighing the di¤erent components of a monetary aggregate
by their degree of liquidity (see Barnett, 1980, and Spindt, 1985).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.2. The money supply 651

16.2.2 The money multiplier

Bank lending is the channel through which the monetary base expands to an
e¤ective money supply, the “money stock”, considerably larger than the monetary
base. The excess of the deposits of the general public over bank reserves (“vault
cash” and demand deposits in the central bank) is lent out in the form of bank
loans, or government or corporate bonds etc. The non-bank public then deposits
a fraction of these loans on checking accounts. Next, the banks lend out a fraction
of these and so on. This process is named the money multiplier process. And the
ratio of the “money stock”, measured as M1 ; say, to the monetary base is called
the money multiplier.
Let

CU R = currency held by the non-bank general public,

DEP = demand deposits held by the non-bank general public,
CU R
= cd; the desired currency-deposit ratio,
DEP
RES = bank reserves = currency held by the commercial banks
(“vault cash”) plus their demand deposits in the central bank,
RES
= rd; the desired reserve-deposit ratio.
DEP
Notice that the currency-deposit ratio, cd; is chosen by the non-bank public,
whereas the reserve-deposit ratio, rd; refers to the behavior of commercial banks.
In many countries there is a minimum reserve-deposit ratio required by law to
ensure a minimum liquidity bu¤er to forestall “bank runs” (situations where
many depositors, fearing that their bank will be unable to repay their deposits in
full and on time, simultaneously try to withdraw their deposits). On top of the
minimum reserve-deposit ratio the banks may hold “excess reserves” depending
on their assessment of their lending risks and need for liquidity.
To …nd the money multiplier, note that

M1 = CU R + DEP = (cd + 1)DEP; (16.1)

where DEP is related to the monetary base, M0; through

M0 = CU R + RES = cdDEP + rdDEP = (cd + rd)DEP:

Substituting into (16.1) gives

cd + 1
M1 = M0 = mmM0 ; (16.2)
cd + rd
where mm = (cd + 1)=(cd + rd) is the money multiplier:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

652 CHAPTER 16. MONEY IN MACROECONOMICS

As a not unrealistic example consider cd 0:7 and rd 0:07: Then we get

mm 2:2: When broader measures of money supply are considered, then, of
course, a larger money multiplier arises. It should be kept in mind that both cd
and rd; and therefore also mm; are neither constant nor exogenous from the point
of view of monetary models. They are highly endogenous and depend on several
things, including degree of liquidity, expected returns, and risk on alternative
assets from the banks’perspective as well as the customers’. In the longer run
cd and rd are a¤ected by the evolution of payment technologies.
To some extent it is therefore a simple matter of identities and not particularly
informative, when we say that, given M0 and the currency-deposit ratio, the
money supply is smaller, the larger is the reserve-deposit ratio. Similarly, since
the latter ratio is usually considerably smaller than one, the money supply is
also smaller the larger is the currency-deposit ratio. Nevertheless, the money
multiplier turns out to be fairly stable under “normal circumstances”. But not
always. During 1929-33, in the early part of the Great Depression, the money
multiplier in the US fell sharply. Although M0 increased by 15% during the four-
year period, liquidity ( M1 ) declined by 27%.5 Depositors became nervous about
their bank’s health and began to withdraw their deposits (thereby increasing cd)
and this forced the banks to hold more reserves (thereby increasing rd). There is
general agreement that this banking panic contributed to the depression and the
ensuing de‡ation.
There is another way of interpreting the money multiplier. By de…nition
of cd; we have CU R = cdDEP: Let cm denote the non-bank public’s desired
currency-money ratio, i.e., cm = CU R=M1 : Suppose cm is a constant. Then

CU R = cmM1 = cm(cd + 1)DEP: (by (16.1))

It follows that cm = cd=(cd + 1) and 1 cm = 1=(cd + 1): Combining this with

(16.2) yields

1 1 1
M1 = cd 1
= = M0 = mmM0 :
cd+1
+ rd cd+1 cm + rd(1 cm) 1 (1 rd)(1 cm)
(16.3)
The way the central bank controls the monetary base is through open-market
operations, that is, by buying or selling bonds (typically short-term government
bonds) in the amount needed to sustain a desired level of the monetary base. In
the next stage the aim could be to obtain a desired level of M1 or a desired level
of the short-term interest rate or, in an open economy, a desired exchange rate
vis-a-vis other currencies.
5
Blanchard (2003).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.3. Money demand 653

An intuitive understanding of the money multiplier and the way commercial

banks “create”money can be attained by taking a dynamic perspective. Suppose
the central bank increases M0 by the amount M0 through an open-market
operation, thus purchasing bonds: This is the …rst round. The seller of the bonds
deposits the fraction 1 cm on a checking account in her bank and keeps the
rest as cash. The bank keeps the fraction rd of (1 cm) M0 as reserves and
provides bank loans or buys bonds with the rest. This is the second round.
Thus, in the …rst round money supply is increased by M0 ; in the second round
it is further increased by (1 rd)(1 cm) M0 ; in the third round further by
(1 rd)2 (1 cm)2 M0 ; etc.6 In the end, the total increase in money supply is

M1 = M0 + (1 rd)(1 cm) M0 + (1 rd)2 (1 cm)2 M0 + :::

1
= M0 = mm M0 :
1 (1 rd)(1 cm)
The second last equality comes from the rule for the sum of an in…nite geometric
series with quotient in absolute value less than one. The conclusion is that the
money supply is increased mm times the increase in the monetary base.

16.3 Money demand

Explaining in a precise way how paper money gets purchasing power and how
holding money - the “demand for money”in economists’traditional language - is
determined, is a di¢ cult task and not our endeavour here. Su¢ ce it to say that:

In the presence of sequential trades and the absence of complete information

and complete markets, there is a need for a generally accepted medium of
exchange money.

The demand for money, by which we mean the quantity of money held by the
non-bank public, should be seen as part of a broader portfolio decision by
which economic agents allocate their …nancial wealth to di¤erent existing
assets, including money, and liabilities. The portfolio decision involves a
balanced consideration of after-tax expected return, risk, and liquidity.

Money is demanded primarily because of its liquidity service in transactions.

Money holding therefore depends on the amount of transactions expected to be
carried out with money in the near future. Money holding also depends on the
need for ‡exibility in spending when there is uncertainty: it is convenient to have
ready liquidity in case favorable opportunities should turn up. Generally money
6
For simplicity, we assume here that cm and rd are constant.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

654 CHAPTER 16. MONEY IN MACROECONOMICS

earns no interest at all or at least less interest than other assets. Therefore money
holding involves a trade-o¤ between the need for liquidity and the wish for interest
yield.
The incorporation of a somewhat micro-founded money demand in macro-
models is often based on one or another kind of short-cut:

The cash-in-advance constraint (also called the Clower constraint).7 Gen-

erally, households’purchases of nondurable consumption goods are in every
short period paid for by money held at the beginning of the period. With
the cash-in-advance constraint it is simply postulated that to be able to
carry out most transactions, you must hold money in advance. In continu-
ous time models the household holds a stock of money which is an increasing
function of the desired level of consumption per time unit and a decreasing
function of the opportunity cost of holding money.

The shopping-costs approach. Here the liquidity services of money are mod-
elled as reducing shopping time or other kinds of non-pecuniary or pecu-
niary shopping costs. The shopping time needed to purchase a given level
of consumption, ct ; is decreasing in real money holdings and increasing in
ct :

The money-in-the-utility function approach. Here, the indirect utility that

money provides through reducing non-pecuniary as well as pecuniary trans-
action costs is modelled as if the economic agents obtain utility directly from
holding money. This will be our approach in the next chapter.

The money-in-the-production-function approach. Here money facilitates

the …rms’transactions, making the provision of the necessary inputs easier.
After all, typically around a third of the aggregate money stock is held by
…rms.

16.4 What is then the “money market”?

In macroeconomic theory, by the “money market” is usually meant an abstract
market place (not a physical location) where at any particular moment the ag-
gregate demand for money “meets”the aggregate supply of money. Suppose the
aggregate demand for real money balances can be approximated by the function
L(Y; i); where LY > 0 and Li < 0 (“L” for liquidity demand). The level of
aggregate economic activity, Y; enters as an argument because it is an (approxi-
mate) indicator of the volume of transactions in the near future for which money
7
After the American monetary theorist Robert Clower (1967).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.4. What is then the “money market”? 655

is needed. The short-term nominal interest rate, i, enters because it is the op-
portunity cost of holding cash instead of interest-bearing short-term securities,
for instance government bonds that mature in one year or less. 8 The latter
constitute a close substitute to money because they have a high degree of liquid-
ity. They are standardized and extensively traded in centralized auction markets
and under “normal circumstances” relatively safe. Because of the short term to
maturity, their market value is less volatile than longer-term securities.
Let the money supply in focus be M1 and let P be the general price level in
the economy (say the GDP de‡ator). Then money market equilibrium is present
if
M1 = P L(Y; i); (16.4)
that is, the available amount of money equals nominal money demand. Note that
supply and demand are in terms of stocks (amounts at a given point in time),
not ‡ows. One of the issues in monetary theory is to account for how this stock
equilibrium is brought about at any instant. Which of the variables M1 ; P; Y;
and i is the equilibrating variable? Presuming that the central bank controls M1 ;
classical (pre-Keynesian) monetary theory has P as the equilibrating variable
while in Keynes’monetary theory it is primarily i which has this role.9 Popular
speci…cations of the function L include L(Y; i) = Y i and L(Y; i) = Y e i ;
where and are positive constants.
One may alternatively think of the “money market”in a more narrow sense,
however. We may translate (16.4) into a description of demand and supply for
base money:
P
M0 = L(Y; i); (16.5)
mm
where mm is the money multiplier. The right-hand side of this equation re‡ects
that the demand for M1 via the actions of commercial banks is translated into a
demand for base money.10 If the public needs more cash, the demand for bank
loans rises and when granted, banks’ reserves are reduced. When in the next
round the deposits in the banks increase, then generally also the banks’reserves
8
To simplify, we assume that none of the components in the monetary aggregate considered
earns interest. In practice demand deposits in the central bank and commercial banks may
earn a small nominal interest.
9
If the economy has ended up in a “liquidity trap”with i at its lower bound, 0, an increase in
M1 will not generate further reductions in i. Agents would prefer holding cash at zero interest
rather than short-term bonds at negative interest. That is, the “=”in the equilibrium condition
(16.4) should be replaced by “ ” or, equivalently, L(Y; i) should at i = 0 be interpreted as a
“set-valued function”. The implications of this are taken up later in this book.
10
Although the money multiplier tends to depend positively on i as well as other interest
rates, this aspect is unimportant for the discussion below and is ignored in the notation in
(16.5).

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

656 CHAPTER 16. MONEY IN MACROECONOMICS

have to increase. To maintain the required reserve-deposit ratio, banks which for
a few days have too little liquidity, borrow from other banks or other institutions
which have too much.
This narrowly de…ned money market is closely related to what is by the practi-
tioners and in the …nancial market statistics called the “money market”, namely
the trade in short-term debt-instruments that are close substitutes to holding
central bank money (think of commercial paper and government bonds with ma-
turity of less than one year). The agents trading in this market not only include
the central bank and the commercial banks but also the mortgage credit institu-
tions, life insurance companies, and other …nancial institutions. What is in the
theoretical models called the “short-term nominal interest rate”can normally be
identi…ed with what is in the …nancial market statistics called the money market
rate or the interbank rate. This is the interest rate (usually measured as a per
year rate) at which the commercial banks provide unsecured loans (“signature
loans”) to each other, often on a day-to-day basis.

Open market operations The commercial banks may under certain condi-
tions borrow (on a secured basis) from the central bank at a rate usually called
the discount rate. This central bank lending rate will be somewhat above the
central bank deposit rate, that is, the interest rate, possibly nil, earned by the
commercial banks on their deposits in the central bank. The interval between the
discount rate and the deposit rate constitutes the interest rate corridor, within
which, under “normal circumstances”, the money market rate, i, ‡uctuates. The
central bank deposit rate acts as a ‡oor for the money market rate and the cen-
tral bank lending rate as a ceiling. Sometimes, however, the money market rate
exceeds the central bank lending rate. This may happen in a …nancial crisis
where the potential lenders are hesitant because of the risk that the borrowing
bank goes bankrupt and because there are constraints on how much and when, a
commercial bank in need of cash can borrow from the central bank.
If the money market rate, i; tends to deviate from what the central bank
aims at (the “target rate”, also called the “policy rate”), the central bank will
typically through open-market operations provide liquidity to the money market
or withhold liquidity from it. The mechanism is as follows. Consider a one-period
government bond with a secured payo¤ equal to 1 euro at the end of the period
and no payo¤s during the period (known as a zero-coupon bond or discount
bond). To …x ideas, let the period length be one month. In the …nancial market
language the maturity date is then one month after the issue date. Let v be the
market price (in euros) of the bond at the beginning of the month. The implicit
monthly interest rate, x; is then the solution to the equation v = (1 + x) 1 ; i.e.,
1
x=v 1:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.4. What is then the “money market”? 657

Translated into an annual interest rate, with monthly compounding, this amounts
to i = (1 + x)12 1 = v 12 1 per year. With v = 0:9975; we get i = 0:03049
per year.11
Suppose the central bank …nds that i is too high and buys a bunch of these
bonds. Then less of them are available for the private sector, which on the other
hand now has a larger money stock at its disposal. According to the Keynesian
monetary theory (which is by now quite commonly accepted), under normal cir-
cumstances the general price level for goods and services is sticky in the short
run. It will be the bond price, v; which responds. In the present case it moves
up, thus lowering i; until the available stocks of bonds and money are willingly
held. In practice this adjustment of v; and hence i; to a new equilibrium level
takes place rapidly.
In recent decades the short-term interest rate has been the main monetary
policy tool when trying to stimulate or dampen the general level of economic
activity and control in‡ation. Under normal circumstances the open market
operations give the central bank a narrow control over the short-term interest rate.
Central banks typically announce their target level for the short-term interest rate
and then adjust the monetary base such that the actual money market rate ends
up close to the announced interest rate. This is what the European Central
Bank (the ECB) does when it announces its target for EONIA (euro overnight
index average) and what the U.S. central bank, the Federal Reserve, does when
it announces its target for the federal funds rate. In spite of its name, the latter
is not an interest rate charged by the U.S. central bank but a weighted average
of the interest rates commercial banks in the U.S. charge each other, usually
overnight.
In the narrowly de…ned “money market”close substitutes to money are traded.
From a logical point of view a more appropriate name for this market would be
the “short-term bond market” or the “near-money market”. This would entail
using the term “market” in its general meaning as a “place” where a certain
type of goods or assets are traded for money. Moreover, speaking of a “short-
term bond market” would be in line with the standard name for market(s) for
…nancial assets with maturity of more than one year, namely market(s) for longer-
term bonds and equity; by practitioners these markets are also called the capital
markets. Anyway, in this book we shall use the term “money market”in its broad
theoretical meaning as an abstract market place where the aggregate demand
for money “meets” the aggregate supply of money. As to what kind of money,
“narrow”or “broad”, further speci…cation is always to be added.
The open-market operations by the central bank a¤ect directly or indirectly
11 i=12 1
With continuous compounding we have v = e so that i = 12 ln v = 0:03004 when v
= 0:9975:

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

658 CHAPTER 16. MONEY IN MACROECONOMICS

all the equilibrating prices in the …nancial markets as well as expectations about
the future path of these prices. This in‡uence derives from the direct control
over the monetary base, M0 : The central bank has no direct control, however,
over the money supply in the broader sense of M1 ; M2 ; or M3 : These broader
monetary aggregates are also a¤ected by the behavior of the commercial banks
and the non-bank public. The money supply in this broad sense can at most be
an intermediate target for monetary policy, that is, a target that can be reached
in some average-sense in the medium run.

16.5 Key questions in monetary theory and pol-

icy
Some of the central questions in monetary theory and policy are:

1. How is the level and the growth rate of the money supply (in the M0 sense,
say) linked to:

(a) the real variables in the economy (resource allocation),

(b) the price level and the rate of in‡ation?

2. How can monetary policy be designed to stabilize the purchasing power of

money and optimize the liquidity services to the inhabitants?

3. How can monetary policy be designed to stabilize the economy and “smooth”
business cycle ‡uctuations?

4. Do rational expectations rule out persistent real e¤ects of changes in the

money supply?

5. What kind of regulation of commercial banks is conducive to a smooth

functioning of the credit system and reduced risk of a …nancial crisis?

6. Is hyperin‡ation always the result of an immense growth in the money

supply or can hyperin‡ation be generated by self-ful…lling expectations?

As an approach to answering long-run monetary issues, we will in the next

chapter consider a kind of neoclassical monetary model by Sidrauski (1967). In
this model money enters as a separate argument in the utility function. The
model has been applied to the study of long-run aspects like the issues 1, 2, and 6
above. The model is less appropriate, however, for short- and medium-run issues
such as 3, 4, and 5 in the list. These issues are dealt with in later chapters.

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

16.6. Literature notes 659

16.6 Literature notes

In the Arrow-Debreu model, the basic microeconomic general equilibrium model,
there is assumed to exist a complete set of markets. That is, there is a market for
each “contingent commodity”, by which is meant that there are as many markets
as there are possible combinations of physical characteristics of goods, dates of
delivery, and “states of nature”that may prevail. In such an …ctional world any
agent knows for sure the consequences of the choices made. All trades can be
made once for all and there will thus be no need for any money holding (Arrow
and Hahn, 1971).
For a detailed account of the di¤erent ways of modelling money demand in
macroeconomics, the reader is referred to, e.g., Walsh (2003). Concerning “money
in the production function”, see Mankiw and Summers (1986).

16.7 Exercises

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

660 CHAPTER 16. MONEY IN MACROECONOMICS

c Groth, Lecture notes in macroeconomics, (mimeo) 2015.

Makroøkonomi 2 Note 3

22.10.2015 Christian Groth

Back to short-run macroeconomics

In this lecture note we shift the focus from long-run macroeconomics to short-run
macroeconomics. The long-run models concentrated on mechanisms that are important
for the economic evolution over a time horizon of at least 10-15 years. With such a hori-
zon it is the development on the supply side (think of capital accumulation, population
growth, and technical progress) that is the primary determinant of cumulative changes
in output and consumption − the trend. The demand side and monetary factors are im-
portant for the fluctuations about the trend. In a long-run perspective these fluctuations
have limited quantitative importance. But within a short horizon, say up to four years,
the demand-side, monetary factors, nominal rigidities, and expectation errors are quan-
titatively important. The present note re-introduces these short-run factors and aims at
suggesting how short-run and long-run theory are linked. This also implies a few remarks
about theory dealing with the medium run, say 4 to 15 years.1 The purpose of medium-
run theory is to explain the regularities in the fluctuations (business cycles) about the
trend and to study what can be accomplished by monetary and fiscal stabilization pol-
icy. In that context the dynamic interaction between demand and supply factors and
the time-consuming adjustment in relative prices play an important role. In this way
medium-run theory bridges the gap between the long run and the short run.

1 Stylized facts about the short run

The idea that prices of most goods and services are sticky in the short run rests on
the empirical observation that in the short run firms in the manufacturing and service
industries typically let output do the adjustment to changes in demand while keeping
prices unchanged. In industrialized societies firms are able to do that because under
“normal circumstances” there is “abundant production capacity” available in the economy.
Three of the most salient short-run features that arise from macroeconomic time series
1
These number-of-years declarations should not be understood as more than a rough indication. Their
appropriateness will depend on the specific historical circumstances and on the problem at hand.

1
analysis of industrialized market economies are the following (cf. Blanchard and Fischer,
1989, Christiano et al., 1999):

1) Shifts in aggregate demand (induced by sudden changes in the state of confidence,

exports, fiscal or monetary policy, or other events) are largely accommodated by
changes in quantities rather than changes in nominal prices − nominal price insen-
sitivity.

2) Even large movements in quantities are often associated with little or no movement
in relative prices − real price insensitivity. The real wage, for instance, exhibits
such insensitivity in the short run.

3) Nominal prices are sensitive to general changes in input costs.

These stylized facts pertain to final goods and services. It is not the case that all
nominal prices in the economy are in the short run insensitive vis-a-vis demand changes.
One must distinguish between production of most final goods and services on the one
hand and production of primary foodstuff and raw materials on the other. This leads to
the associated distinction between “cost-determined” and “demand- determined” prices.

Final goods and services are typically differentiated goods (imperfect substitutes).
Their production takes place under conditions of imperfect competition. As a result of
existing reserves of production capacity, generally speaking, the production is elastic w.r.t.
demand. An upward shift in demand tends to be met by a rise in production rather than
price. The price changes which do occur are mostly a response to general changes in costs
of production. Hence the name “cost-determined” prices.

For primary foodstuff and many raw materials the situation is different. To increase
the supply of most agricultural products requires considerable time. This is also true
(though not to the same extent) with respect to mining of raw materials as well as
extraction and transport of crude oil. When production is inelastic w.r.t. demand in
the short run, an increase in demand results in a diminution of stocks and a rise in
price. Hence the name “demand-determined prices”. The price rise may be enhanced by
a speculative element: temporary hoarding in the expectation of further price increases.
The price of oil and coffee − two of the most traded commodities in the world market
− fluctuate a lot. Through the channel of costs the changes in these demand-determined
prices spill over to the prices of final goods. Housing is also an area where, apart from
regulation, demand-determined prices is the rule in the short run.

2
 In industrialized economies manufacturing and services are the main sectors, and the
general price level is typically regarded as cost-determined rather than demand deter-
mined. Two further aspects are important. First, many wages and prices are set in
nominal terms by price setting agents like craft unions and firms operating in imperfectly
competitive output markets. Second, these wages and prices are in general deliberately
kept unchanged for some time even if changes in the environment of the agent occurs; this
aspect, possibly due to pecuniary or non-pecuniary costs of changing prices, is known as
nominal price stickiness. Both aspects have vast consequences for the functioning of the
economy as a whole compared with a regime of perfect competition and flexible prices.

2 A simple short-run model

The simple model presented below is close to what Paul Krugman named the World’s
Smallest Macroeconomic Model.2 The model is crude but nevertheless useful in at least
three ways:

• the model demonstrates the fundamental difference in the functioning of an economy

with flexible prices and one with sticky prices;

• by addressing spillovers across markets, the model is a suitable point of departure

for a definition of the Keynesian concept of effective demand;

• the model displays the logic behind the Keynesian refutation of Say’s law.

2.1 Elements of the model

We consider a monetary closed economy which produces a consumption good. There

are three sectors in the economy, a production sector, a household sector, and a public
sector with a consolidated government/central bank. Time is discrete. There is a current
period, of length a quarter of a year, say, and “the future”, compressing the next period
and onward. Labor is the only input in production. To simplify notation, the model
presents its story as if there is just one representative household and one representative
firm owned by the household, but the reader should imagine that there are numerous
agents of each kind.
2
Krugman (1999). Krugman tells he learned the model back in 1975 from Robert Hall. As presented
here there is also an inspiration from Barro and Grossman (1971).

3
 The production function has CRS,

 =    0 (1)

where  is aggregate output of a consumption good which is perishable and therefore

cannot be stored,  is a technology parameter and  is aggregate employment in the
current period. In short- and medium-run macroeconomics the tradition is to use 
to denote labor input (“number of hours”), while  is typically used for either money
demand (“liquidity demand”) or supply of bank loans. We will follow this tradition.

The price of the consumption good in terms of money, i.e., the nominal price, is  The
wage rate in terms of money, the nominal wage, is  We assume that the representative
firm maximizes profit, taking these current prices as given. The nominal profit, possibly
nil, is
Π =   −   (2)

There is free exit from the production sector in the sense that the representative firm can
decide to produce nothing. Hence, an equilibrium with positive production requires that
profits are non-negative.

The representative household lives only one period, but leaves a bequest for the next
generation. The household supplies labor inelastically in the amount ̄ and receives the
profit obtained by the firm, if any. The household demands the consumption good in the
amount   in the current period (since we want to allow cases of non-market clearing,
we distinguish between consumption demand,    and realized consumption, . Current
income not consumed is saved for the future. As the output good cannot be stored, the
only non-human asset available in the economy is fiat money, which is thus the only asset
on hand for saving. There is no private banking sector in the economy. So “money”
means the “currency in circulation” (the monetary base) and is on net an asset in the
private sector as a whole. Until further notice the money stock is constant.

The preferences of the household are given by the utility function,

̂
 = ln   +  ln  0    1 (3)


where ̂ is the amount of money transferred to “the future”, and   is the expected
future price level. The utility discount factor  (equal to (1 + )−1 if  is the utility
discount rate) reflects “patience”.

Consider the household’s choice problem. Facing  and  and expecting that the

4
future price level will be    the household chooses   and ̂ to maximize  s.t.

   + ̂ =  +   + Π ≡   ≤ ̄ (4)

Here,   0 is the stock of money held at the beginning of the current period and is
predetermined. The actual employment is denoted  and equals the minimum of the
amount of employment offered by the firm and the labor supply ̄ (the principle of
voluntary trade). The sum of initial financial wealth,  and nominal income,   + Π
constitutes the budget, 3 Nominal financial wealth at the beginning of the next period
is ̂ =  +   + Π −     i.e., the sum of initial financial wealth and planned saving
where the latter equals   + Π −     The benefit obtained by transferring ̂ depends
on the expected purchasing power of ̂ hence it is ̂  that enters the utility function.
Presumably the household expects some labor and profit income also in the future and
seemingly ownership rights to the firms’ profit are non-negotiable. How the decision
making is related to such matters is not specified in this minimalist way of representing
that there is a future.

Substituting ̂ =  −    into (3), we get the first-order condition

 1  

= 
+  
(−  ) = 0
   −  
which gives
1
  = (5)
1+
We see that the marginal (= average) propensity to consume is (1 + )−1  hence inversely
related to the patience parameter  The planned stock of money to be held at the end
of the period is
1 
̂ = (1 − ) = 
1+ 1+
So, the expected price level,    in the future does not affect the demands,   and ̂
This is a special feature caused by the additive-logarithmic specification of the utility
function in (3). Indeed, with this specification the substitution and income effects of a
rise in the expected real gross rate of return, (1  )(1 ) on savings exactly offset each
other, and there is no wealth effect in this model.

Inserting (4) and (2) into (5) gives


  +  + Π +
 = = =   (6)
 (1 + )  (1 + ) 1+
3
As time is discrete, expressions like  +   + Π are legitimate. Although it is meaningless to add a
stock and a flow (since they have different denominations), the sum  +   + Π should be interpreted
as  + (  + Π)∆ where ∆ is the period length. With the latter being the time unit, we have ∆
= 1

5
In our simple model output demand is the same as the consumption demand    So
clearing in the output market, in the sense of equality between demand and actual output,
requires   =  So, if this clearing condition holds, substituting into (6) gives the
relationship

 =  (7)

This is only a relationship between  and  not a solution for any of them since both
are endogenous variables so far. Moreover, the relationship is conditional on clearing in
the output market.

We have assumed that agents take prices as given when making their demand and
supply decisions. But we have said nothing about whether nominal prices are flexible or
rigid as seen from the perspective of the system as a whole.

2.2 The case of fully flexible  and 

What Keynes called “classical economics” is nowadays also often called “Walrasian macro-
economics” (sometime just “pre-Keynesian macroeconomics”). In this theoretical tradi-
tion both wages and prices are assumed fully flexible and all markets perfectly competitive.

Firms’ ex ante output supply conditional on a hypothetical wage-price pair (  ) and
the corresponding labor demand will be denoted   and   , respectively. As we know
from microeconomics, the pair (     ) need not be unique, it can easily be a “set-valued
function” of (  ) Moreover, with constant returns to scale in the production function,
the range of this function may for certain pairs (  ) include (∞ ∞).

The distinguishing feature of the Walrasian approach is that wages and prices are
assumed fully flexible. Both  and  are thought to adjust immediately so as to clear
the labor market and the output market like in a centralized auction market. Clearing
in the labor market requires that  and  are adjusted so that actual employment, 
equals labor supply,    which is here inelastic at the given level ̄ So

 =   = ̄ =    (8)

where the last equality indicates that this employment level is willingly demanded by the
firms.

We have assumed a constant-returns-to-scale production function (1). Hence, the

clearing condition (8) requires that firms have zero profit. In turn, by (1) and (2), zero

6
profit requires that the real wage equals labor productivity:

=  (9)

With clearing in the labor market, output must equal full-employment output,

 = ̄ ≡   =    (10)

where the superscript “ ” stands for “full employment”, and where the last equality
indicates that this level of output is willingly supplied by the firms. For this level of
output to match the demand,    coming from the households, the price level must be

 = ≡   (11)
 
in view of (7) with  =    This price level is the classical equilibrium price, hence the
superscript “”. Substituting into (9) gives the classical equilibrium wage

 =   ≡    (12)

For general equilibrium we also need that the desired money holding at the end of the
period equals the available money stock. By Walras’ law this equality follows automat-
ically from the household’s Walrasian budget constraint and clearing in the output and
labor markets. To see this, note that the Walrasian budget constraint is a special case of
the budget constraint (4), namely the case

   + ̂ =  +    + Π  (13)

where Π is the notional profit associated with the hypothetical production plan (     )
i.e.,
Π ≡    −     (14)
The Walrasian budget constraint thus imposes replacement of the term for actual employ-
ment,  with the households’ desired labor supply,   (= ̄) It also imposes replacement
of the term for actual profit, Π with the hypothetical profit Π (“” for “classical”) cal-
culated on the basis of the firms’ aggregate production plan (     ).

Now, let the Walrasian auctioneer announce an arbitrary price vector (  1) with
  0   0 and 1 being the price of the numeraire, money. Then the values of excess
demands add up to

 (  −   ) +  (  −   ) + ̂ − 
=    −    +    + ̂ −  −    (by rearranging)
=    −    + Π (by (13))
=    −    + Π ≡ 0 (from definition of Π in (14))

7
This exemplifies Walras’ law, saying that with Walrasian budget constraints the aggregate
value of excess demands is identically zero. Walras’ law reflects that when households
satisfy their Walrasian budget constraint, then as an arithmetic necessity the economy
as a whole has to satisfy an aggregate budget constraint for the period in question. It
follows that the equilibrium condition ̂ =  is ensured as soon as there is clearing in
the output and labor markets. And more generally: if there are  markets and  − 1 of
these clear, so does the ’th market.

Consequently, when (  ) = (     ) all markets clear in this flexwage-flexprice

economy with perfect competition and a representative household with the “endowment”-
pair ( ̄). Such a state of affairs is known as a classical or Walrasian equilibrium.4 A
key feature is expressed by (8) and (10): output and employment are supply-determined,
i.e., determined by the supply of production factors, here labor.

The intuitive mechanism behind this equilibrium is the following adjustment process.
Imagine that in an ultra-short sub-period  − 6= 0 In case  −   0 ( 0)
there will be excess supply (demand) in the labor market. This drives  down (up). Only
when  =  and full employment obtains, can the system be at rest. Next imagine
that  −   6= 0 In case  −    0 ( 0) there is excess supply (demand) in the output
market. This drives  down (up). Again, only when  =   and  =  (whereby
 =   ), so that the output market clears under full employment, will the system be at
rest.

This adjustment process is fictional, however, because outside equilibrium the Wal-
rasian supplies and demands, which supposedly drive the adjustment, are artificial con-
structs. Being functions only of initial resources and price signals, the Walrasian supplies
and demands are mutually inconsistent outside equilibrium and can therefore not tell
what quantities will be traded during an adjustment process. The story needs a consider-
able refinement unless one is willing to let the mythical “Walrasian auctioneer” enter the
scene and bring about adjustment toward the equilibrium prices without allowing trade
until these prices are found.

Anyway, assuming that Walrasian equilibrium has been attained, by comparative stat-
ics based on (11) and (12) we see that in the classical regime: (a)  and  are proportional
to ; and (b) output is at the unchanged full-employment level whatever the level of .
This is the neutrality of money result of classical macroeconomics.
4
To underline its one-period nature, it may be called a Walrasian short-run or a Walrasian temporary
equilibrium.

8
 The neutrality result also holds in a quasi-dynamic context where we consider an actual
change in the money stock occurring in historical time. Suppose the government/central
bank at the beginning of the period brings about lump-sum transfers to the households
in the total amount ∆  0 As there is no taxation, this implies a budget deficit which
is thus fully financed by money issue.5 So (134) is replaced by

   + ̂ =  + ∆ +  ̄ + Π  (15)

If we replace  in the previous formulas by  0 ≡  + ∆ we see that money neutrality

still holds. As saving is income minus consumption, there is now positive nominal private
saving of size   = ∆ +  ̄ + Π −    =  0 −  = ∆ On the other hand
the government dissaves, in that its saving is   = −∆ where ∆ is the government
budget deficit. So national saving is and remains  ≡   +   = 0 (it must be nil as there
are no durable produced goods).

2.3 The case of  and  fixed in the short run

In standard Keynesian macroeconomics nominal wages are considered predetermined in

the short run, fixed in advance by wage bargaining between workers (or workers’ unions)
and employers (or employers’ unions). Those who end up unemployed in the period do
not try to − or are not able to − undercut those employed, at least not in the current
period.

Likewise, nominal prices are set in advance by firms facing downward-sloping demand
curves. It is understood that there is a large spectrum of differentiated products, and 
and  are composites of these. This heterogeneity ought of course be visible in the model
− and it will become so in Section 19.3. But at this point the model takes an easy way
out and ignores the involved aggregation issue.

Let  in the current period be given at the level ̄  Because firms have market power,
the profit-maximizing price involves a mark-up on marginal cost, ̄  = ̄  (which
is also the average cost). We assume that the price setting occurs under circumstances
where the chosen mark-up becomes a constant   0, so that

̄
 = (1 + ) ≡ ̄  (16)

5
Within the model this is in fact the only way to increase the money stock. As money is the only asset
in the economy, a change in the money stock can not be brought about through open market operations
where the central bank buys or sells another financial asset.

9
While ̄ is considered exogenous (not determined within the model), ̄ is endogenously
determined by the given ̄   and  There are barriers to entry in the short run.

Because of the fixed wage and price, the distinction between ex ante (also called
planned or intended) demands and supplies and the ex post carried out purchases and
sales are now even more important than before. This is because the different markets may
now also ex post feature excess demand or excess supply (to be defined more precisely
below). According to the principle that no agent can be forced to trade more than desired,
the actual amount traded in a market must equal the minimum of demand and supply.
So in the output market and the labor market the actual quantities traded will be

 = min(     ) and (17)

 = min(     ) (18)

respectively, where the superscripts “” and “” are now used for demand and supply in
a new meaning to be defined below. This principle, that the short side of the market
determines the traded quantity, is known as the short-side rule. The other side of the
market is said to be quantity rationed or just rationed if there is discrepancy between  
and   . In view of the produced good being non-storable, intended inventory investment
is ruled out. Hence, the firms try to avoid producing more than can be sold. In (17) we
have thus identified the traded quantity with the produced quantity, 

But what exactly do we mean by “demand” and “supply” in this context where market
clearing is not guaranteed? We mean what is appropriately called the effective demand
and the effective supply (“effective” in the meaning of “operative” in the market, though
possibly frustrated in view of the short-side rule). To make these concepts clear, we need
first to define an agent’s effective budget constraint:

DEFINITION 1 An agent’s (typically a household’s) effective budget constraint is the

budget constraint conditional on the perceived price and quantity signals from the mar-
kets.

It is the last part, “and quantity signals from the markets”, which is not included in
the concept of a Walrasian budget constraint. The perceived quantity signals are in the
present context the actual employment constraint faced by the household and the profit
expected to be received from the firms and determined by their actual production.6 So
the household’s effective budget constraint is given by (4). In contrast, the Walrasian
6
We assume the perceived quantity signals are deterministic.

10
budget constraint is not conditional on quantity signals from the markets but only on the
“endowment” ( ̄) and the perceived price signals and profit.

DEFINITION 2 An agent’s effective demand in a given market is the amount the agent
bids for in the market, conditional on the perceived price and quantity signals that con-
strains its bidding. By “bids for” is meant that the agent is both able to buy that amount
and wishes to buy that amount, given the effective budget constraint. Summing over all
potential buyers, we get the aggregate effective demand in the market.

DEFINITION 3 An agent’s effective supply in a given market is the amount the agent
offers for sale in the market, conditional on perceived price and quantity signals that
constrains its offering. By “offers for sale” is meant that the agent is both able to bring
that amount to the market and wishes to sell that amount, given the set of opportunities
available. Summing over all potential sellers, we get the aggregate effective supply in the
market.

When  = ̄  the aggregate effective output demand,    is the same as households’

consumption demand given by (6) with  = ̄ , i.e.,

+
  =  = ̄
 (19)
1+
In view of the inelastic labor supply, households’ aggregate effective labor supply is simply

  = ̄

Firms’ aggregate effective output supply is

  =   ≡ ̄ (20)

Indeed, in the aggregate the firms are not able to bring more to the market than full-
employment output ,    And every individual firm is not able to bring to the market
than what can be produced by “its share” of the labor force. On the other hand, because
of the constant marginal costs, every unit sold at the preset price adds to profit. The
firms are therefore happy to satisfy any output demand forthcoming − which is in practice
testified by a lot of sales promotion.

Firms’ aggregate effective demand for labor is constrained by the perceived output
demand,    because the firm would loose by employing more labor. Thus,

 
 =  (21)

11
 By the short-side rule (17), combined with (20), follows that actual aggregate output
(equal to the quantity traded) is

 = min(     ) 5   

So the following three mutually exclusive cases exhaust the possibilities regarding aggre-
gate output:

 =      (the Keynesian regime),

 =      (the repressed inflation regime),
 =   =   (the border case).

2.3.1 The Keynesian regime:  =      

In this regime we can substitute  =   into (19) and solve for  :

 
 = = ≡  ≡ 
=   (22)
 ̄ 
where we have denoted the resulting output   (the superscript “” for “Keynesian”). The
inequality in (22) is required by the definition of the Keynesian regime, and the identity
comes from (11). Necessary and sufficient for the inequality is that ̄    ≡    In
view of (16), the economy is thus in the Keynesian regime if and only if

̄    (1 + ) (23)

Since     in this regime, we may say there is “excess supply” in the output market or,
with a perhaps better term, there is a “buyers’ market” situation (sale less than desired).
The reservation regarding the term “excess supply” is due to the fact that we should
not forget that  −    0 is a completely voluntary state of affairs on the part of the
price-setting firms.

From (1) and the short-side rule now follows that actual employment will be
 
 =  = =  ̄ =    (24)
  ̄
Also the labor market is thus characterized by “excess supply” or a “buyers’ market”
situation. Profits are Π = ̄  − ̄  = (1 − ̄ (̄ ))̄  = (1 − (1 + )−1 ) −1   0
where we have used, first,  = , then the price setting rule (16), and finally (22).

This solution for ( ) is known as a Keynesian equilibrium for the current period.
It is named an equilibrium because the system is “at rest” in the following sense: (a)

12
agents do the best they can given the constraints (which include the preset prices and
the quantities offered by the other side of the market); and (b) the chosen actions are
mutually compatible (purchases and sales match). The term equilibrium is here not used
in the Walrasian sense of market clearing through instantaneous price adjustment but in
the sense of a Nash equilibrium conditional on perceived price and quantity signals. To
underline its temporary character, the equilibrium may be called a Keynesian short-run
(or temporary) equilibrium. The flavor of the equilibrium is Keynesian in the sense that
there is unemployment and at the same time it is aggregate demand in the output market,
not the real wage, which is the binding constraint on the employment level. A higher
propensity to consume (lower discount factor ) results in higher aggregate demand,   
and thereby a higher equilibrium output,   . In contrast, a lower real wage due to either
a higher mark-up,  or a lower marginal (= average) labor productivity,  does not
result in a higher   . On the contrary,   becomes lower, and the causal chain behind
this goes via a higher ̄  cf. (16) and (22). In fact, the given real wage, ̄ ̄ = (1+)
is consistent with unemployment as well as full employment, see below. It is the sticky
nominal price at an excessive level, caused by a sticky nominal wage at an “excessive”
level, that makes unemployment prevail through a too low aggregate demand,    A lower
nominal wage would imply a lower ̄ and thereby, for a given  stimulate   and thus
raise   

In brief, the Keynesian regime leads to an equilibrium where output as well as em-
ployment are demand-determined.

The “Keynesian cross” and effective demand The situation is illustrated by the
“Keynesian cross” in the (   ) plane shown in Fig. 19.1, where   =   = (1 +
)−1 ( + ̄ ) We see the vicious circle: Output is below the full-employment level
because of low consumption demand; and consumption demand is low because of the low
employment. The economy is in a unemployment trap. Even though at   we have Π  0
and there are constant returns to scale, the individual firm has no incentive to increase
production because the firm already produces as much as it rightly perceives it can sell
at its preferred price. We also see that here money is not neutral. For a given  = ̄ 
and thereby a given  = ̄  a higher  results in higher output and higher employment.

Although the microeconomic background we have alluded to is a specific “market

power story” (one with differentiated goods and downward sloping demand curves), the
Keynesian cross in Fig. 19.1 may turn up also for other microeconomic settings. The key
point is the fixed ̄    and fixed ̄  ̄ 

13
   d
Y  

M
Y
Yd  P  
1 

 

M
P  
1 

45  
O  Y 
Yk   Yf


Figure 1: The Keynesian regime (̄   (1 + );  and   given, ̄ fixed).

The fundamental difference between the Walrasian and the present framework is that
the latter allows trade outside Walrasian equilibrium. In that situation the households’
consumption demand depends not on how much labor the households would prefer to sell
at the going wage, but on how much they are able to sell, that is, on a quantity signal
received from the labor market. Indeed, it is the actual employment,  that enters the
operative budget constraint, (4), not the desired employment as in classical or Walrasian
theory.

2.3.2 The repressed-inflation regime:  =      

This regime represents the “opposite” case of the Keynesian regime and arises if and only
if the opposite of (23) holds, namely

̄    (1 + )

In view of (16), this inequality is equivalent to ̄     ≡   . Hence ( ̄ ) 

(  ) =   = ̄ In spite of the high output demand, the shortage of labor hinders
the firms to produce more than   . With  =    output demand, which in this model
is always the same as consumption demand,    is, from (6),


+
 = ̄
  =   =   (25)
1+

As before, effective output supply,    equals full-employment output,   

14
  
Yd   M
Y
M P
Y 
d
Y f  
P 1 
 
1   
Cf    
M
P  
1 

45  
O  f Y 
Y  


Figure 2: The repressed inflation regime (̄   (1 + );  and   given, ̄ fixed).

The new element here in that firms perceive a demand level in excess of    As the
real-wage level does not deter profitable production, firms would thus prefer to employ
people up to the point where output demand is satisfied. But in view of the short side
rule for the labor market, actual employment will be

 =   = ̄    = 


So there is excess demand in both the output market and the labor market. Presum-
ably, these excess demands generate pressure for wage and price increases. By assumption,
these potential wage and price increases do not materialize until possibly the next period.
So we have a repressed-inflation equilibrium ( ) = (   ̄) although possibly short-
lived.

Fig. 19.2 illustrates the repressed-inflation regime. In the language of the microeco-
nomic theory of quantity rationing, consumers are quantity rationed in the goods market,
as realized consumption =  =      = consumption demand. Firms are quantity
rationed in the labor market, as     . This is the background for the parlance that in
the repressed inflation regime, output and employment are not demand-determined but
supply-determined. Both the output market and the labor market are sellers’ markets
(purchase less than desired). Presumably, the repressed inflation regime will not last long
unless there are wage and price controls imposed by the government, as for instance may
be the case for an economy in a war situation.7
7
As another example of repressed inflation (simultaneous excess demand for consumption goods and

15
2.3.3 The border case between the two regimes:  =   =   

This case arises if and only if ̄ =   (1 + ) which is in turn equivalent to ̄ =
(1 + )̄  =    ≡   ≡ (  ). No market has quantity rationing and we may
speak of both the output market and the labor market as balanced markets.

There are two differences compared with the classical equilibrium, however. The first
is that due to market power, there is a wedge between the real wage and the marginal
productivity of labor. In the present context, though, where labor supply is inelastic,
this does not imply inefficiency but only a higher profit/wage-income ratio than under
perfect competition (where the profit/wage-income ratio is zero). The second difference
compared with the classical equilibrium is that due to price stickiness, the impact of
shifts in exogenous variables will be different. For instance a lower  will here result
in unemployment, while in the classical model it will just lower  and  and not affect
employment.

2.3.4 In terms of effective demands and supplies Walras’ law does not hold

As we saw above, with Walrasian budget constraints, the aggregate value of excess de-
mands in the given period is zero for any given price vector, (  1) with   0 and
  0 In contrast, with effective budget constraints, effective demands and supplies,
and the short-side rule, this is no longer so. To see this, consider a pair (  ) where
    and  6=   ≡ (  ) Such a pair leads to either the Keynesian regime or
the repressed-inflation regime. The pair may, but need not, equal one of the pairs (̄  ̄ )
considered above in Fig. 19.1 or 19.2 (we say “need not”, because the particular -markup
relationship between  and  is not needed). We have, first, that in both the Keynesian
and the repressed-inflation regime, effective output supply equals full-employment output,

  =   (26)

The intuition is that in view of    , the representative firm wishes to satisfy any
output demand forthcoming but it is only able to do so up to the point of where the
availability of workers becomes a binding constraint.

Second, the aggregate value of excess effective demands is, for the considered price
labor) we may refer to Eastern Europe before the dissolution of the Soviet Union in 1991. In response to
severe and long-lasting rationing in the consumption goods markets, households tended to decrease their
labor supply (Kornai, 1979). This example illustrates that if labor supply is elastic, the effective labor
supply may be less than the Walrasian labor supply due to spillovers from the output market.

16
vector (  1) equal to

 (  −   ) +  (  −   ) + ̂ − 
=  (  − ̄) +    + ̂ −  −   
=  (  − ̄) +   + Π −    (by (4))
=  (  − ̄) +   −    (by (2))
½
 0 if   (  ) and
=  (  − ̄) +  ( −   ) (27)
 0 if   (  ) and    

The aggregate value of excess effective demands is thus not identically zero. As expected, it
is negative in a Keynesian equilibrium and positive in a repressed-inflation equilibrium.8
The reason that Walras’ law does not apply to effective demands and supplies is that
outside Walrasian equilibrium some of these demands and supplies are not realized in the
final transactions.

This takes us to Keynes’ refutation of Say’s law and thereby what Keynes and others
have regarded as the core of his theory.

2.3.5 Say’s law and its refutation

The classical principle “supply creates its own demand” (or “income is automatically
spent on products”) is named Say’s law after the French economist and business man
Jean-Baptiste Say (1767-1832). In line with other classical economists like David Ricardo
and John Stuart Mill, Say maintained that although mismatch between demand and
production can occur, it can only occur in the form of excess production in some industries
at the same time as there is excess demand in other industries.9 General overproduction
is impossible. Or, by a classical catchphrase:

Every offer to sell a good implies a demand for some other good.

By “good” is here meant a produced good rather than just any traded article, including
for instance money. Otherwise Say’s law would be a platitude (a simple implication of
the definition of trade). So, interpreting “good” to mean a produced good, let us evaluate
8
At the same time, (27) together with the general equations   = ̄ and   =    shows that we have
̂ =  in a Keynesian equilibrium (where  =   ) and ̂   in a repressed-inflation equilibrium
(where  =   ).
9
There were two dissidents at this point, Thomas Malthus and Karl Marx, two classical economists
that were otherwise not much aggreeing.

17
Say’s law from the point of view of the result (27). We first subtract  (  −   )
=  (  − ̄) on both sides of (27), then insert (26) and rearrange to get

 (  −  ) + ̂ −  = 0 (28)

for any   0 Consider the case    In this situation every unit produced and
sold is profitable. So any  in the interval 0   ≤   is profitable from the supply side
angle. Assume further that  = ̄    ≡ (  ) This is the case shown in Fig. 19.1.
The figure illustrates that aggregate demand is rising with aggregate production. So far
so well for Say’s law. We also see that if aggregate production is in the interval 0  
    then   (=   )   This amounts to excess demand for goods and in effect, by
(28), excess supply of money. Still, Say’s law is not contradicted. But if instead aggregate
production is in the interval     ≤    then   (=   )   ; now there is general
overproduction. Supply no longer creates its own demand. There is a general shortfall of
demand. By (28), the other side of the coin is that when     then ̂   which
means excess demand for money. People try to hoard money rather than spend on goods.
Both the Great Depression in the 1930s and the Great Recession 2008- can be seen in this
light.10

The refutation of Say’s law does not depend on the market power and constant markup
aspects we have adhered to above. All that is needed for the argument is that the agents
are price takers within the period. In addition, the refutation does not hinge on money
being the asset available for transferring purchasing power from one period to the next.
We may imagine an economy where  represents land available in limited supply. As
land is also a non-produced store of value, the above analysis goes through − with one
exception, though. The exception is that ∆ in (15) can no longer be interpreted as a
policy choice. Instead, a positive ∆ could be due to discovery of new land.

We conclude that general overproduction is possible and Say’s law thereby refuted.
It might be objected that our “aggregate reply” to Say’s law is not to the point since
Say had a disaggregate structure with many industries in mind. Considering an explicit
disaggregate production sector makes no essential difference, however, as a simple example
will now show.
10
Paul Krugman stated it this way:

“When everyone is trying to accumulate cash at the same time, which is what happened
worldwide after the collapse of Lehman Brothers, the result is an end to demand [for output],
which produces a severe recession” (Krugman, 2009).

18
Many industries Suppose there is still one labor market, but  industries with pro-
duction function  =   where  and  are output and employment in industry 
respectively,  = 1 2      Let the preferences of the representative household be given
by
X ̂
=   ln  +  ln     0  = 1 2      0    1


In analogy with (4), the budget constraint is
X X X X
  + ̂ =  ≡  +   + Π =  +   
   

where the last equality comes from

Π =   −   

Utility maximization gives   =   (1 + ).

As a special case, consider   = 1 and  =  ,  = 1 2      Then


 =  (29)
(1 + )
and
X
 = +  ≡  +  


Substituting into (29), we thus find demand for consumption good  as


+  
 = 
≡    for all 
1+
£ ¤
Let   min  (  )  where   ≡ ̄ It follows that every unit produced and
sold is profitable and that


 + +
 = ≤ 

  
1+ 1+
where the weak inequality comes from  ≤   (always) and the strict inequality from
  (  )

Now, suppose good 1 is brought to the market in the amount 1 , where    1

    Industry 1 thus experiences a shortfall of demand. Will there in turn necessarily
be another industry experiencing excess demand? No. To see this, consider the case  
      for all  All these supplies are profitable from a supply side point of view,
and enough labor is available. Indeed, by construction the resource allocation is such that
X
    ≡  ≤ ̄     (30)

19
where ̄ = max [1       ]    . This is a situation where people try to save (hoard
money) rather than spend all income on produced goods. It is an example of general
overproduction, thus falsifying Say’s law.

In the special case where all  =   the situation for each single industry can be
illustrated by a diagram as that in Fig. ??. Just replace           and  in Fig.
?? by         ≡ ( )    and  respectively.

Could the evaluation of Say’s law be more favorable if we allow for the existence of
interest-bearing assets? The answer is no, as we shall see in Chapter ??.

2.4 Short-run adjustment dynamics

We now return to the aggregate setup. Apart from the border case of balanced markets,
we have considered two kinds of “fix-price equilibria”, repressed inflation and Keynesian
equilibrium. Most macroeconomists consider nominal wages and prices to be less sticky
upwards than downwards. So a repressed inflation regime is typically regarded as having
little durability (unless there are wage and price controls imposed by a government). It is
otherwise with the Keynesian equilibrium. A way of thinking about this is the following.

Suppose that up to the current period full-employment equilibrium has applied: 

=   = ( ̄ ) =   and ̄ = (1 + )̄  =    ≡   ≡ (  ) Then, for
some external reason, at the start of the current period a rise in the patience parameter
0
occurs, from  to  0  so that the new propensity to save is  (1 +  0 )  (1 + ). We
may interpret this as “precautionary saving” in response to a sudden fall in the general
“state of confidence”.

Let our “period” be divided into  sub-periods, indexed  = 0 1 2      − 1 of length

1, where  is “large”. At least within the first of these sub-periods, the preset ̄ and
̄ are maintained and firms produce without having yet realized that aggregate demand
will be lower than in the previous period. After a while firms realize that sales do not
keep track with production.

There are basically two kinds of reaction to this situation. One is that wages and
prices are maintained throughout all the sub-periods, while production is scaled down
to the Keynesian equilibrium   = ( 0 ̄ ). Another is that wages and prices adjust
downward so as to soon reestablish full-employment equilibrium. Let us take each case
at a time.

20
Wage and price stay fixed: Sheer quantity adjustment For simplicity we have
assumed that the produced goods are perishable. So unsold goods represent a complete
loss. If firms fully understand the functioning of the economy and have model-consistent
expectations, they will adjust production per time unit down to the level   as fast as
possible. Suppose instead that firms have naive adaptive expectations of the form


−1 = −1   = 0 1 2  

This means that the “subjective” expectation, formed in sub-period  − 1 of demand next
sub-period is that it will equal the demand in sub-period  − 1 Let the time-lag between
the decision to produce and the observation of the demand correspond to the length of
the subperiods. It is profitable to satisfy demand, hence actual output in sub-period 
will be
  ̄ −1
 = −1 = −1 = 0 + 
1+ 1 + 0
in analogy with (19). This is a linear first-order difference equation in  , with constant
coefficients. The solution is (see Math Tools)
µ ¶
∗01 
 = (0 −  ) +  ∗0   ∗0 = =      (31)
1 + 0 0
 ̄

Suppose  0 = 09 say. Then actual production,   converges fast towards the steady-state
value   . When  =    the system is at rest. Fig. 19.x illustrates. Although there is
excess supply in the labor market and therefore some downward pressure on wages, the
Keynesian presumption is that the workers’s side in the labor market generally withstand
the pressure.11

Fig. 19.x about here.

The process (31) also applies “in the opposite direction”. Suppose, starting from the
Keynesian equilibrium  = ( 0 ̄ ) a reduction in the patience parameter  0 occurs,
such that ( 0 ̄ ) increases, but still satisfies ( 0 ̄ )     Then the initial condition
in (31) is 0   ∗0  and the greater propensity to consume leads to an upward quantity
adjustment.
11
Possible explanations of downward wage stickiness are discussed in Chapter ??.

21
Downward wage and price adjustment Several of Keynes’ contemporaries, among
them A. C. Pigou, maintained that the Keynesian state of affairs with  =     
could only be very temporary. Pigou’s argument was that a fall in the price level would
take place and lead to higher purchasing power of  The implied stimulation of ag-
gregate demand would bring the economy back to full employment. This hypothetically
equilibrating mechanism is known as the “real balance effect” or the “Pigou effect” (after
Pigou, 1943).

Does the argument go through? To answer this, we imagine that the time interval
between different rounds of wage and price setting is as short as our sub-periods. We
imagine the time interval between households’ decision making to be equally short. Given
the fixed markup , an initial fall in the preset ̄ is needed to trigger a fall in the preset
̄  The new classical equilibrium price and wage levels will be

 0 = and  0 =  0 
0 
Both will thus be lower than the original ones − by the same factor as the patience
parameter has risen, i.e., the factor  0  In line with “classical” thinking, assume that
soon after the rise in the propensity to save, the incipient unemployment prompts wage
setters to reduce ̄ and thereby price setters to reduce ̄  Let both ̄ and ̄ after a few
rounds be reduced by the factor  0  Denoting the resulting wage and price ̄ 0 and ̄ 0 
respectively, we then have
 0 ̄ 0  0 
̄ 0 =  ̄ 0 = (1 + ) = ≡  0 ≡ 0  
1+   
Seemingly, this restores aggregate demand at the full-employment level   = ( 0 ̄ 0 )
=  .

While this “classical” adjustment is conceivable in the abstract, Keynesians question

its practical relevance for several reasons:

1. Empirically, it seems to be particularly in the downward direction that nominal

wages are sticky. And without an initial fall in the nominal wage, the downward
wage-price spiral does not get started.

2. A downward wage-price spiral, i.e., deflation, increases the implicit real interest
rate, ( − +1 )+1  thus tending to dampen aggregate demand rather than the
opposite.

3. If we go outside our simple model, there are additional objections:

22
 (a) the monetary base is in reality only a small fraction of financial wealth, and so
the real balance effect can not be powerful unless the fall in the price level is
drastic;
(b) many firms and households have nominal debt, the real value of which would
rise dramatically, thereby leading to bankruptcies and a worsening of the con-
fidence crisis, thus counteracting a return to full employment.

One should be aware that there are two distinct kinds of “price flexibility”. It can
be “imperfect” or “perfect” (also called “full”). The first kind relates to a gradual price
process, for instance generated by a wage-price spiral as at item 2 above. The latter kind
relates to instantaneous and complete price adjustment as with a Walrasian auctioneer,
cf. Section 2. It is the first kind that may be destabilizing rather than the opposite.

2.5 Digging deeper

As it stands the above theoretical framework has many limitations:

(a) The wage and price setting should be explicitly modelled and in this connection
there should be an explanation of the wage and price stickiness.

(b) It should be made clear how to come from the existence of many differentiated
goods and markets with imperfect competition to aggregate output and income which in
turn constitute the environment conditioning individual agents’ actions.

(c) To incorporate better the role of asset markets, including the primary role of money
as a medium of exchange rather than a store of value, at least one alternative asset should
enter, an interest-bearing asset.

(d) The model should be truly dynamic with forward-looking endogenous expectations
and gradual wage and price changes depending on the market conditions, in particular
the employment situation.

We now comment briefly on these points.

3 Price adjustment costs

The classical theory of perfectly flexible wages and prices and neutrality of money seems
contradicted by overwhelming empirical evidence. At the theoretical level the theory

23
ignores that the dominant market form is not perfect competition. Wages and prices
are usually set by agents with market power. And there may be costs associated with
changing prices and wages. Here we consider such costs.

The literature has modelled price adjustment costs in two different ways. Menu costs
refer to the case where there are fixed costs of changing price. Another case considered
in the literature is the case of strictly convex adjustment costs, where the marginal price
adjustment cost is increasing in the size of the price change.

The most obvious examples of menu costs are of course costs associated with

1. remarking commodities with new price labels,

2. reprinting price lists (“menu cards”) and catalogues.

But the term menu costs should be interpreted in a broader sense, including pecuniary
as well non-pecuniary costs of:

3. information-gathering,

4. recomputing optimal prices,

5. conveying the new directives to the sales force,

6. the risk of offending customers by frequent and/or large price changes,

7. search for new customers willing to pay a higher price,

8. renegotiating contracts.

Menu costs induce firms to change prices less often than if no such costs were present.
And some of the points mentioned in the list above, in particular point 7 and 8, may be
relevant also in the different labor markets.

The menu cost theory is one of the microfoundations provided by modern Keynesian
economics for the presumption that nominal prices and wages are sticky in the short run.
The main theoretical insight of the menu cost theory is the following. There are menu
costs associated with changing prices. Even small menu costs can be enough to prevent
firms from changing their price. This is because the opportunity cost of not changing
price is only of second order, i.e., “small”; this is a reflection of the envelope theorem (see

24
Appendix). But owing to imperfect competition (price  MC), the effect on aggregate
output, employment, and welfare of not changing prices is of first order, i.e., “large”.

The menu cost theory provides the more popular explanation of nominal price rigidity.
Another explanation rests on the presumption of strictly convex price adjustment costs.
In this theory the price change cost for firm  is assumed to be  =  ( − −1 )2 
  0 Under this assumption the firm is induced to avoid large price changes, which
means that it tends to make frequent, but small price adjustments. This theory is related
to the customer market theory. Customers search less frequently than they purchase. A
large upward price change may be provocative to customers and lead them to do search in
the market, thereby perhaps becoming aware of attractive offers from other stores. The
implied “kinked” demand curve can explain that firms are reluctant to suddenly increase
their price.

4 Adding interest-bearing assets

To incorporate the key role of financial markets for the performance of the macroeconomy,
at least one extra asset should enter in a short-run model, an interest-bearing asset. This
gives rise to the IS-LM model that should be familiar from Blanchard, Macroeconomics.

An extended IS-LM model is presented in the recent editions of the mentioned text
by Blanchard (alone) and in Blanchard et al., Macroeconomics: A European Perspective,
2010, Chapter 20. The advantage of the extended version is that the commercial banking
sector is introduced more explicitly so that the model incorporates both a centralized
bond market and decentralized markets for bank loans.

5 Adding dynamics and a Phillips curve

Adding dynamics, expectations formation, and a Phillips curve leads to a medium-run

model. An introduction is provided in the first-mentioned Blanchard textbook, chapters 8
and 14. Medium-run models describe fluctuations in production and employment around
a trend, often considered related to the “natural rate of unemployment”. Adding capital
accumulation, technical progress, and growth in the labor force to the model, GDP gets
a rising trend.

Roughly speaking, this course, Macroeconomics 2, can be interpreted as dealing with

an economy moving along this trend. We have more or less ignored the fluctuations,

25
simply by assuming flexible prices and perfect competition. In a realistic model with
imperfect competition and price stickiness in both output and labor markets the natural
rate of unemployment is likely to be higher than in an economy with perfect competition.
And hump-shaped deviations from trend GDP, that is, business cycles, are likely to arise
when the economy is hit by large shocks, for instance a financial crisis.

The third macro course, Macroeconomics 3, deals with short and medium run theory
and emphasizes issues related to monetary policy.

6 Appendix

ENVELOPE THEOREM Let  =  ( ) be a continuously differentiable function of

two variables, of which one, , is conceived as a parameter and the other,  as a control
 
variable. Let () be a value of  at which 
( ) = 0, i.e., 
( ()) = 0. Let
 () ≡  ( ()). Provided  () is differentiable,


 0 () = ( ())

where   denotes the partial derivative of (·) w.r.t. the first argument.

   
Proof  0 () = 
( ()) + 
( ()) 0 () = 
( ()), since 
( ()) = 0 by
definition of (). ¤

That is, when calculating the total derivative of a function w.r.t. a parameter and
evaluating this derivative at an interior maximum w.r.t. a control variable, the envelope
theorem allows us to ignore the terms that arise from the chain rule. This is also the case
if we calculate the total derivative at an interior minimum.12

12
For extensions and more rigorous framing of the envelope theorem, see for example Sydsaeter et al.
(2006).

26