Lecture 8

Endogenous Growth and Persistent Effects of Recessions

Maarten De Ridder

London School of Economics

Lent Term 2022

1 / 56
This term

Part I: Shocking theory of the business cycle

• Introduction to business cycles X

• Real Business Cycle (RBC) Model X

• New Keynesian DSGE Models X

Part II: Perspectives on business cycles and steady states

• Heterogeneity versus homogeneity and the effect of policy X

• Endogenous growth and persistent effects of recessions ⇐

• Aggregate shocks? Firm-heterogeneity and the business cycle

2 / 56
Background reading

Barlevy (2004), The Cost of Business Cycles Under Endogenous Growth,

American Economic Review

Acemoglu (2009), Introduction to Modern Economic Growth, Ch 13.1

(LSE library)

Comin (2009), On the Integration of Growth and Business Cycles,

Empirica

Novales, Fernadez, Ruiz (2009), Economic Growth: Theory and

Numerical Solution Methods, Ch 6.1-6.4 (Moodle)

3 / 56
Recall: Macroeconomics
Conceptual division between two sub-fields
• Long-run growth: analyze the determinants of the trend growth

• Short-run business cycles: analyze deviations from the trend

4 / 56
Business Cycles

3
2.5
2
1.5
1

1960q1 1980q1 2000q1 2020q1

Real Gross Domestic Product (log) for the U.S. 1960-2020 (1960 =1)
Source: FRED

5 / 56
Persistent Business Cycles
1.4

1.4

1.4
1.3

1.3

1.3
1.2

1.2

1.2
1.1

1.1

1.1
1

1
2000q1 2005q1 2010q1 2015q1 2000q1 2005q1 2010q1 2015q1 2000q1 2005q1 2010q1 2015q1

(a) United States (b) Euro Area (c) United Kingdom

Real Gross Domestic Product (log index, 2000 =1). Source: OECD

6 / 56
Persistent Business Cycles

Estimates of Potential Real Gross Domestic Product for the U.S.

Source: Summers (2017) based on Congressional Budget Office data

7 / 56
Persistent business cycles

Log Real Gross Domestic Product for the U.S.

Source: Summers (2015), ECB Forum on Central Banking

8 / 56
Persistent effect of large shocks

• Persistent effect of banking crises, currency crises, civil unrest

• Cross-country regressions, ARDL along:
4
X 4
X
git = αi + βj gi,t−j + βs Di,t−s + εit
j=1 s=0

Impulse Responses Systemic Banking Crises on Real GDP

Source: Cerra and Saxena (2008), AER

9 / 56
Today

• Persistent recessions and the costs of business cycles

• Endogenous productivity growth

• Comin and Gertler (2006) Model

10 / 56
Today

• Persistent recessions and the costs of business cycles

• Endogenous productivity growth

• Comin and Gertler (2006) Model

10 / 56
Persistence in DSGE models

Why does the economy return to the old steady state in the model?
• Neoclassical production function: diminishing returns to factors

• Hence: temporary decline in investment raises marginal product

• Policy: investments increase in marginal product of capital

• Eventual return to steady state capital and output

11 / 56
Impulse responses - lecture 3

10-3
Consumption Output 10-3 Capital
4 0.015 5

4
3
0.01
3
2
2
0.005
1
1

0 0 0
0 10 20 30 0 10 20 30 0 10 20 30

Investment 10-3Employment Productivity

0.06 6 0.01

0.008
0.04 4
0.006

0.02 2 0.004

0.002
0 0
0
0 10 20 30 0 10 20 30 0 10 20 30
Quarters

Impulse responses to one standard dev. productivity shock (in log dev. from steady state)

12 / 56
Persistence in DSGE models

• Drop assumption of diminishing returns (‘AK model’):

Yt = θAt Kt

At = Aρt−1 exp (t )

• Representative household:
∞
X Ct1−γ
U = E0 βt
t=0
1−γ

s.t. Ct + Kt+1 = (rt + 1 − δ)Kt

• Firms rent capital from households on competitive markets

13 / 56
Persistence in DSGE models
Equilibrium conditions:
• First order conditions:

Ct−γ −γ 

= βEt (1 + rt+1 − δ)Ct+1
θAt = rt

• Resource constraint:

Ct + Kt+1 = θAt Kt + (1 − δ)Kt

At = Aρt−1 exp (t )

Steady state: balanced growth path with constant growth rate g ,
1/γ
G = G Y = G K = (β(1 + θ − δ))
where GtX ≡ XXt−1
t

⇒ AK model: 1st generation endogenous growth model

14 / 56
Persistence in DSGE models

How do you solve this?

• Perturbation: need to log-linearize around the steady state

• Capital, consumption do not have steady state in levels

• Solution: log-linearize around the balanced growth path with:

• Constant growth of consumption, capital
• As well as constant capital/consumption ratio

15 / 56
Log-linearized equations
• Resource constraint and capital accumulation:
θ C∗
k
gbt+1 = K
abt − k cbt∗
G G
where ct∗ is the log consumption to capital ratio

• Euler equation written in terms of Ct∗ :

"  −γ #
Ct+1
1 = βEt (1 + θAt+1 − δ)
Ct
"  #
Ct+1 Kt+1 Kt −γ

1 = βEt (1 + θAt+1 − δ)
Ct Kt+1 Kt
Ct+1 Kt+1 −γ
"  ∗  #
1 = βEt (1 + θAt+1 − δ)
Ct∗ Kt

Log-linearized:
 
h ∗ i
c k θ 1
Et gbt+1 + gbt+1 = Et [b
at+1 ]
1+θ−δ γ
∗
as well as definition gbtc = cbt∗ − cbt−1
∗ and process abt = ρb
at−1 + εt

16 / 56
Persistence in DSGE models

Consumption Path
1.4

1.35

1.3

1.25

1.2

1.15

1.1

1.05
Actual
Trend
1
0 5 10 15

Path of consumption index (year 0 = 1) after 8% negative productivity shock

γ = 4, β = 0.99, ρ = 0.5, θ is such that g = 1.02. Code: Moodle.

17 / 56
Cost of business cycles

Lucas (1987): business cycles have minimal costs

• Assume a simple growth process for stochastic consumption:

Ct = λt (1 + εt )C0
with λ > 1, C0 > 0, εt i.i.d. disturbances with variance σ 2

• Utility:
∞
X Ct1−γ
U = E0 βt
t=0
1−γ

• Risk-averse consumer would prefer certain income (ε = 0 ∀t)

But how much?

18 / 56
Cost of business cycles

Find percentage of consumption household would give up for certainty:

∞ 1−γ ∞ 1−γ
X
t[λt C0 ] X [λt (1 + εt )(1 + χ)C0 ]
β = E0 βt
t=0
1−γ t=0
1−γ

Solve for χ (problem set) to get:

1 2
χ≈ γσ
2

Derive σ 2 from variance of HP deviations of consumption. Lucas (2003):

1
χ = · 4 · (0.022)2
2
= 0.0003872
≈ 0.04%

19 / 56
Business cycles and growth

Lucas (1987) assumes that volatility σ 2 does not affect growth rate λ

• In other words: transitory shocks have transitory effects

• How does the model need to change to match the data?

• Introduce endogenous growth

20 / 56
Our Model

Representative household and firm:

∞
X Ct1−γ
U = E0 βt
t=0
1−γ

Yt = At Kt

Capital accumulation:

Kt+1 = It + (1 − δ)Kt
" t  #
Y Is
Kt = +1−δ K0
s=0
Ks

21 / 56
Path of consumption
Define

Ct It
ct = it =
Yt Yt

Path of consumption:

Ct = cAK
"t t t t #
Y Is
= +1−δ K0 ct At
s=0
Ks
" t #
Y
= λs (1 + t )C0
s=0

Where:
• Growth in potential output: λs = is As + 1 − δ

• Disturbances: 1 + t = ct At
c0 A0
→ deviation of consumption from trend

22 / 56
Cost of business cycles
Repeat the Lucas calculation:
∞  t 1−γ ∞  Qt
1−γ
X λ C0 X
s=0 λs (1 + t )(1 + χ)C0
βt = E0 βt
t=0
1−γ t=0
1−γ

! 1
P∞ 1−γ
β t λt(1−γ)
t=0
1+χ= P∞ t  Qt
1−γ
E0 t=0 β s=0 λs (1 + t )

Simulate path of consumption for large number of years (e.g. 200,000), set
ρ = 0, calculate χ using the equation to get:

χ ≈ 2.3%
⇒ cost of the same shocks is two orders of magnitude larger when they have
persistent effect

• Code: Moodle

• Note: we’ve increased b-cycle costs by making shocks more persistent

23 / 56
Cost of business cycles
Barlevy (2004), AER:
• AK model like the one presented here

• Transitory shocks to total factor productivity

• Additional assumption: diminishing returns to investments
 
It
Kt+1 = φ Kt + (1 − δ)Kt
Kt
   
∂φ It
Kt
∂2φ It
Kt
> 0, <0
∂It ∂It2

• Results:

1. Permanent reduction in consumption from transitory shock

2. Lower average growth when volatility is higher

24 / 56
Summary

So far:
• Some recessions seem to have persistent effect on output

• If capital does not have diminishing returns, model can replicate this

• Large increase in predicted costs of business cycles

Empirical evidence: strong diminishing returns to capital

• Macro: capital share in national accounts implies α ≈ 0.3 − 0.4

• Firm-level production data: α ≈ 0.2 − 0.45

(e.g. Ackerberg, Caves & Frazer (2015), ECTA)

Solution: endogenous growth in productivity

25 / 56
Today

• Persistent recessions and the costs of business cycles

• Endogenous productivity growth

• Comin and Gertler (2006) Model

25 / 56
Endogenous productivity growth
 
Ȧt Ẏt wt L t K̇t wt Lt L˙t U̇t
Total factor productivity: At = Yt − 1− Yt Kt − Yt Lt − Ut

1.4
1.3
1.2
1.1
1

1995q1 2000q1 2005q1 2010q1 2015q1 2020q1

Real TFP for the U.S. 1995-2020 (log, 1995 =1)

Red-dashed: raw series. Blue-sold: utilization-adjusted. Source: Fernald (FRBSF)

26 / 56
Endogenous productivity growth

Key insights:
• Productivity growth comes from technology creation and adoption

• Neoclassical production fn: admits exponential growth through A

• Technology creation/adoption require particular investments: R&D

Ȧt
= g (RDt , At , ..)
At
• These investments depend on expected profits, financing costs, etc

RDt = h(rt , At , πt , πt+1 , ..)

27 / 56
Investment in Productivity

2
1.8
1.6
1.4
1.2
1

1995 2000 2005 2010 2015

Intangible investments (R&D, employee training, software development)

in the United States (1995-2016). Source: Intan-Invest Project (2020)

28 / 56
Procyclical R&D

Procyclicality of Research and Development in the United States

Source: Barlevy (2007), Figure 1

29 / 56
Endogenous technological change

Broadly divided into two groups:

1. Models of process innovation
• R&D expands the variety of technologies that used in production
• Seminal reference: Romer (1990)

2. Models of product innovation

• Invention of new or better goods
• Grossman and Helpman (1991), Aghion and Howitt (1992)

Aim of endogenous growth literature: understand growth

• Policies, incentives to maximize welfare on balanced growth path

30 / 56
Romer (1990) model: summary

• Innovation as generating new blueprints or ideas for production

• Three important features:

1. Cost of research and development are paid as fixed upfront costs

2. Ideas and technologies are nonrival :
• Many firms can benefit from the same idea
• Innovator needs a reward: introduce monopolistic competition (CES)

3. Increasing returns to scale?

31 / 56
Preferences

• Time is discrete, representative household is infinitely lived

• Utility function:
∞  t 1−γ
X 1 Ct −1
t=0
1+ρ 1−γ
• Labor is supplied inelastically, measure of labor supply is L.

• Households hold a portfolio of all firms in the economy

32 / 56
Production

• Consumption good Ct is homogeneous

• Competitively produced with aggregate production function:

"Z #
At
1 1−β
Yt = xit di Lβ
1−β 0

• xit is the amount of input from technology type i at time t

• At : measure of intermediate-input technologies available at time t
• For given At , economy exhibits constant returns to scale

33 / 56
Resource Constraint

Resource constraint of the economy:

Ct + Xt + RDt = Yt

• RDt : spending on research and development

• Innovators spend RDt to develop new intermediate input technologies

• Xt : spending on intermediate input production

• Once invented, inputs can be produced at marginal cost Ψ > 0.

34 / 56
Innovation

• A continuum of innovators invests to generate new inputs

• Aggregate innovation:

At+1 − At = ϕRDt
• Successful innovator receives perpetual patent to produce some xit

• Note: no aggregate uncertainty

• Individual projects by innovators can fail, but overall level of
innovation is deterministic function of aggregate R&D RDt

35 / 56
Optimization: final good sector

• Maximization by competitive final good producers:

"Z #
At Z At
1 1−β β
max xit di L − pitx xit di − wt L
L,xit ;i∈[0,At ] 1 − β 0 0

• Demand for intermediate goods

1
xit = (pitx )− β L

• Note: doesn’t depend on the wage rate or the number of

technologies At

36 / 56
Optimization: technology monopolists

Problem of the owner of patent to produce i obtained at time 0

• Goal: maximize present value of profits
∞ Y
t  
X 1
Vi0 = max πit
xpit
t=1 s=1
1 + rs

• Profits under constant marginal costs Ψ:

πit = (pitx − Ψ)xit

1
• Constrained by demand function xit = (pitx )− β L.

37 / 56
Optimization: technology monopolists
For all i at all t:
• Optimal price is markup over (constant) marginal cost:

Ψ
pitx = = 1 (normalization)
1−β

• Hence profit is:

1
πit = (pitx − Ψ)(pitx )− β L
  − β1
Ψ Ψ
= β L = βL
1−β 1−β

• Output:
1
xit = (pitx )− β L = L

38 / 56
Equilibrium output

It follows that output has increasing returns to scale:

"Z #
At
1 1−β
Yt = xit di Lβ
1−β 0
"Z #
At
1
= L1−β di Lβ
1−β 0
1
= At L
1−β

Hence: a constant growth rate of At will generate constant GDP growth

39 / 56
Equilibrium interest rate
• Free entry condition: profits compensate for investment upon entry

• Recall:
At+1 − At = ϕRDt

• Average cost of entry: 1/ϕ. Benefit: Vt

Vt ϕ = 1

• Combine this with the definition of the value function:

∞  t
X 1
V0 ϕ = ϕ π=1
t=1
1+r
! 
1 1
= ϕβL 1
=1
1 − 1+r 1+r
⇒ rt = r ∗ = ϕβL

40 / 56
Equilibrium growth
Household’s first order condition for consumption:
Ct+1
1/γ
= (1 + ρ)−1 (1 + rt )
Ct

Combined with the transversality condition. Note that:

1/γ
Gc = (1 + ρ)−1 (1 + rt )

1/γ
= (1 + ρ)−1 (1 + ϕβL)

Straightforward to show that model admits a balanced growth path

equilibrium where Y , C and A grow at constant rate

1/γ
G = (1 + ρ)−1 (1 + ϕβL)

(see Acemoglu)

41 / 56
Intuition

1/γ
G = (1 + ρ)−1 (1 + ϕβL)

• Growth falls in rate of impatience ρ

• Growth increases in research-effectiveness parameter ϕ

• Growth increases in population size L (‘scale effect’, see Jones, 1995)

• Growth increases in profitability parameter β

• β raises incentive for research
• Means interest rate must increase for free entry condition
• Household responds to higher interest rate by lowering consumption
• ⇒ More resources to growth

42 / 56
Cycles and growth

Aim of endogenous growth literature: understand growth

• Policies, incentives to maximize welfare along balanced growth path

• Does not consider effect of fluctuations on growth, welfare

Example: we’ve seen that profitability is positive for growth

• But profitability is parameter-determined: it does not vary over time

• Needed: a model where incentives for R&D are subject to shocks

43 / 56
Today

• Persistent recessions and the costs of business cycles

• Endogenous productivity growth

• Comin and Gertler (2006) Model

43 / 56
Comin and Gertler (2006, AER)

• Understand both growth and fluctuations, and how they interact

• Start with the Romer model of expanding varieties

• New layer: consider both innovation and technology adoption

(match evidence)
• Study short and long-run effect of transitory real shocks
(e.g. labor supply)

We’ll study the main mechanism and intuition

44 / 56
Production

Final output is produced competitively along:

γ 1−γ
Yt = Ktα L1−α
t Mt

Materials Mt are produced by combining intermediate goods xit :

!µ
Z At
1/µ
Mt = xit di
0

µ > 1 is the markup charged by monopolist producer of intermediate i.

45 / 56
Innovation

Key assumption: innovation comes with convex costs between periods

• Innovators: equate marginal benefit of innovation to marginal costs

• Optimization: set net-present value of marginal innovation equal to

innovation costs
• Net-present value of innovation comes from profits (procyclical)

46 / 56
Households
Shocks come through labor disutility of the household:
∞ 1+1/η
" !#
X
t+i w Lt+i
Et β ln Ct+i − µt+i
1 + 1/η
i=0

s.t. Ct + Kt+1 = (Rt − δ)Kt + wt Lt

Note: these are “balanced growth preferences” (see lecture 1)
• Euler equation:
−1
Ct−1 = βEt (Rt+1 − δ)Ct−1
 

• Static consumption-labor decision:

 η
wt 1
Lt =
Ct µw
t

47 / 56
Mechanism

Say there is a positive shock to µw

t :

• Contraction in labor supply

• Reduction in aggregate output (recession)
• Reduction in profitability of owning intermediate input
• Reduction in effort to adopt new intermediate input technology
• Reduction in R&D to develop new technologies

48 / 56
Impulse Responses

Impulse Response to Unit Shock to Wage Markup

Solid: full model. Dashed: no endogenous TFP. Source: Comin and Gertler (2006)

49 / 56
Comin and Gertler (2006)

The actual model contains much more

• Variable capital utilization with endogenous depreciation

• Endogenous markups and profits through entry and exit

• A goods-producing and capital-producing sector

• Careful calibration and comparison with data

50 / 56
Persistent business cycles are in fashion

Embed the Comin and Gertler structure in a New Keynsian model:

• Anzoategui, Comin, Gertler, Martinez (2019), AEJ: Macroeconomics

• Ikeda and Kurozumi (2019), Journal of Monetary Economics

Endogenous growth in New Keynesian models:

• Benigno and Fornaro (2017), Review of Economic Studies

• Bianchi, Kung and Morales (2019), Journal of Monetary Economics

• Garga and Singh (2019), Journal of Monetary Economics

• Queralto and Moran (2018), Journal of Monetary Economics

• Queralto (2019), Journal of Monetary Economics

51 / 56
Lesson Learned?

GDP Forecast: November 2020 versus March 2020

Source: UK Office of Budget Responsibility

52 / 56
Lesson Learned?

Contributors to change in GDP Forecast: November 2020 versus March 2020

Source: UK Office of Budget Responsibility (OBR)

53 / 56
Lesson Learned?

“We will continue our V-shaped recovery and launch a record-smashing

economic boom" - Donald J Trump, October 2020

54 / 56
Trumped

.1
.05
0
-.05
-.1

1960q1 1980q1 2000q1 2020q1

Real Gross Domestic Product (log deviations from HP Trend) for the U.S. 1960-2021
Source: FRED

55 / 56
What have we done?

• Evidence on persistent effect of recessions X

• First generation endogenous growth model, cost of business cycles X

• Second generation endogenous growth models X

• Growth models with business cycles X

56 / 56