Structural Macroeconometrics

Errata
David N. DeJong with Chetan Dave

Notice Regarding Code

We have found a programming error in the …le stogrow.src posted under the heading
“Solve Stochastic Optimal Growth Model Using Log-Linearization” on the textbook web
site. Code downloaded prior to 9/1/07 should be adjusted as follows: the appearance of
xbar as a local variable in the procedure modelsol.p should be eliminated. xbar must
remain global, so that modi…cations performed within the procedure have global impact.
Several results reported in the text have been tainted by this error. Full details are
provided below. In brief, the a¤ected results are as follows:

The moment comparisons reported in Table 6.4.

The moment-matching exercise in Section 7.4 (Tables 7.1-7.3).
The comparison of asset-pricing models in Section 9.7.
The model simulations reported in Table 11.1.
The maximum likelihood estimates in Table 11.2 (third column).

Chapter 2, Section 2.1.3, Examples

The approximation in equation (2.7) should be

@ log [ 1 ] @ log [ 2 ] c i
(z) (z) = 1 :
@ log(xzt ) @ log(zt ) c+i c+i

Chapter 2, Section 2.2.2, Sims’s Method

On the fourth line of p. 25, the matrix

[I ]

should instead be
[I ]:

1
With z1 being ns 1 and z2 being nc 1, such that ns + nc = 1, then in this case I is ns ns
and is ns nc .
Again on p. 25, there is a problem with equation (2.50). In the text, the equation reads
1 0
0 =Z 11 [ 11 ( 12 22 )]Z :

Instead, de…ning Z:1 as the n ns matrix containing the …rst through ns th columns of Z,
the equation should read
1
0 = Z:1 11 [ 11 ( 12 22 )]Z:

1 Chapter 4, Section 4.1.2, The VAR Model

In equation (4.30), the matrix = E ("t "0t ) should be replaced by e = E (et e0t ), where

et = ["1t 0:::0 "2t 0:::0 :::"mt 0:::0]0 ;

as de…ned on the last line of p. 63. Thanks to Surach Tanboon for pointing out this error.

2 Chapter 4, Section 4.3, The Kalman Filter

Equation (4.57) should read
1
vec(P1j0 ) = (I F F) vec(Q):

Existence requires that the eigenvalues of F be less than 1. Thanks to Michael Hauser for
pointing out this error.
On the right-hand side of equation (4.63), xt 1 should instead be xt 1jt 1 : In the following
line, the term Ptjt 1 on the right-hand side should instead be Pt 1jt 1 : In light of this second
issue, equation (4.64) (the updating equation for Ptjt 1 ) is seen to be incorrect. Instead, it
should read
Ptjt 1 = F Pt 1jt 1 F 0 + Q:
Thanks to Roberto Chang for pointing out this error.
On p. 84, the expression for Jt just above equation (4.71) should read
1
Jt = Ptjt F 0 Pt+1jt :

Thanks to Andres Fernandez for pointing out this error.

Equation (4.72) should read

bT
x 2jT bT
=x 2jT 2 + JT 2 (b
xT 1jT bT
x 1jT 2 ):

Thanks to Michael Hauser for pointing out this error.

2
Chapter 6, Section 6.4
The moment comparisons reported in Table 6.4 are a¤ected by the programming error
involving xbar noted above under “Notice Regarding Code”. Table 6.4 should read as follows:

Table 6.4. Moment Comparison

H-P Filtered Data RBC Model
j j
j j y
'(1) 'j;y (0) 'j;y (1) j y
'(1) 'j;y (0) 'j;y (1)
y 0.0177 1.00 0.86 1.00 0.86 0.0184 1.00 0.79 1.00 0.79
c 0.0081 0.46 0.83 0.82 0.75 0.0085 0.46 0.96 0.77 0.62
i 0.0748 4.23 0.79 0.95 0.80 0.0799 4.34 0.74 0.95 0.75
n 0.0185 1.05 0.90 0.83 0.62 0.0087 0.47 0.73 0.91 0.72
Notes: '(1) denotes …rst-order serial correlation; 'j;y (l) denotes l th order
correlation between variables j and y. Model moments based on the parameterization
=[ ' ]0 = [0:24 0:99 1:5 0:35 0:025 0:78 0:0067]0 .

Substantively, this change casts the performance of the model along the investment di-
mension in a relatively more favorable light. The poor performance noted in the text along
the hours dimension remains evident.

3
Chapter 7, Section 7.4
The moment-matching exercise conducted in this section are a¤ected by the program-
ming error involving xbar noted above under “Notice Regarding Code”. Tables 7.1-7.3
should read as follows:

Table 7.1. Baseline Moment Comparison

Moment Point Estimate Std. Err. Model Moment t stat
y 0.0177 0.0019 0.0184 -0.41
c 0.0081 0.0008 0.0085 -0.47
i 0.0748 0.0070 0.0799 -0.74
n 0.0185 0.0021 0.0087 4.62
'y (1) 0.8642 0.0199 0.7947 3.50
'c (1) 0.8318 0.0293 0.9601 -4.38
. 'i (1) 0.7891 0.0307 0.7367 1.71
'n (1) 0.8992 0.0144 0.7290 11.78
'y;c (0) 0.8237 0.0379 0.7782 1.20
'y;i (0) 0.9502 0.0101 0.9459 0.42
'y;n (0) 0.8342 0.0241 0.9107 -3.18
'y;c (1) 0.7467 0.0436 0.7591 -0.28
'y;i (1) 0.7991 0.0355 0.6790 3.38
'y;n (1) 0.6230 0.0396 0.6312 -0.21
Notes: '(1) denotes …rst-order serial correlation; 'j;y (l) denotes l th order
correlation between variables j and y. Model moments based on the parameterization
=[ ' ]0 = [0:24 0:99 1:5 0:35 0:025 0:78 0:0067]0 .

4
 Table 7.2. Parameter Estimates
Parameter Calibrated Value All Moments C Subset I Subset N Subset
0.24 0.1860 0.1792 0.3224 0.1791
(8:81e 6) (4:89e 4) (0:0013) (2:41e 5)
0.99 0.9983 0.9990 0.9916 0.9990
[0:9877; 0:999] (1:17e 6) (1:12e 4) (7:27e 5) (5:52e 6)
0.025 0.0267 0.0400 0.0100 0.0400
[0:01; 0:04] (3:76e 5) (5:07e 9) (3:20e 9) (4:07e 8)
. 1.5 1.184 2.4999 2.0512 1.0206
[0:5; 2:5] (0:0089) (0:0031) (0:3812) (13:484)
' 0.35 0.3364 0.3345 0.3785 0.3345
(2:24e 6) (1:33e 4) (4:48e 4) (6:56e 6)
0.78 0.8952 0.6000 0.8166 0.9499
[0:6; 0:95] (0:0075) (0:5728) (0:0593) (1:4438)
0.0067 0.0051 0.0099 0.0062 0.0006
(0; 0:01] (2:04e 4) (0:0437) (0:0018) (0:0045)
Notes: Brackets reported under calibrated parameter values indicate range restrictions.
Also, parentheses reported under parameter estimates indicate standard errors.
Finally, moments included under J Subset are y ; j ; 'j (1); 'y;j (0) and 'y;j (1); j = c; i; n.

5
 Table 7.3. Moment Comparisons Using Model Estimates
Moment All Moments C Subset I Subset N Subset
y 0.0195 ( 0:98) 0.0212 ( 1:88) 0.0178 ( 0:056) 0.0028 (7:96)
c 0.0109 ( 3:27) 0.0084 ( 0:36) XX XX
i 0.0739 (0:13) XX 0.0784 ( 0:52) XX
n 0.0079 (4:96) XX XX 0.0001 (8:39)
'y (1) 0.8952 ( 1:56) XX XX XX
'c (1) 0.9801 ( 5:05) 0.8427 ( 0:37) XX XX
. 'i (1) 0.8430 ( 1:76) XX 0.8031 ( 0:46) XX
'n (1) 0.8334 (4:55) XX XX 0.8520 (3:26)
'y;c (0) 0.8400 ( 0:43) 0.8597 ( 2:16) XX XX
'y;i (0) 0.9262 (2:38) XX 0.9589 ( 0:87) XX
'y;n (0) 0.8691 ( 1:45) XX XX 0.7313 (4:27)
'y;c (1) 0.8408 ( 2:16) 0.6961 ( 1:16) XX XX
'y;i (1) 0.7674 (0:89) XX 0.7623 (1:04) XX
'y;n (1) 0.6870 ( 1:87) XX XX 0.59700 (0:66)
Notes: '(1) denotes …rst-order serial correlation; 'j;y (l) denotes l th order
correlation between variables j and y: t statistics indicating discrepancies
with empirical counterparts provided in parentheses. XX denotes unestimated moment.

Based on the incorrect …gures reported in the text, we noted that “... although the
standard RBC model is well-known to have di¢ culties in characterizing the behavior of hours,
we have also found its performance along the consumption dimension to be disappointing.”
In light of these corrected …gures, this conclusion is no longer valid: e.g., only one of the …ve
t statistics reported in the revised Table 7.3 under “C Subset”exceeds 2 in absolute value.

3 Chapter 9, Section 9.4.1

The expression for s:e:(g N )I in equation (9.26) should be modi…ed by taking the square
root of the right-hand side.

6
Chapter 9, Section 9.7
The model-estimation exercise conducted in this section are a¤ected by the programming
error involving xbar noted above under “Notice Regarding Code”. In this case, changes
from ‡awed to corrected results are minimal, and the substantive conclusions drawn from
the exercise remain intact. In particular, di¤erences in posterior means reported in Tables
9.3 and 9.4 are less than one half of one posterior standard deviation in all cases. The most
signi…cant change is in the posterior-odds comparison reported in the last full paragraph of
p. 262: the previous odds of 2.4:1 in favor of the self-control speci…cation drop to 1.6:1 given
the correction.
Given the minimal changes in corrected results, Figures 9.3-9.7 are not reproduced here.
Corrected versions of Tables 9.3 and 9.4 are as follows:
Table 9.3. Parameter Estimates
q
d q "d "q "d ;"q d
Means:
Prior 0.960 2.000 0.0000 0.4 0.900 0.900 UN UN UN 10.00
CRRA 0.958 2.914 NA NA 0.877 0.914 0.114 0.096 0.311 10.75
Self-Con. 0.967 2.04 0.00294 0.32 0.879 0.910 0.114 0.149 0.320 7.54
Std. Dev.:
Prior 0.020 1.000 0.01 0.2 0.050 0.050 UN UN UN 5.00
CRRA 0.004 0.826 NA NA 0.033 0.027 0.007 0.034 0.130 4.44
Self-Con. 0.008 0.598 0.00330 0.16 0.029 0.023 0.007 0.043 0.124 3.65
Posterior Correlations
q
; "q ; "d ;"q "q ; "d ;"q "d ;"q ; d
CRRA -0.839 -0.337 0.201 0.557
Self-Con. -0.850 -0.298 0.164 0.556

Notes: UN denotes “uninformative prior”; NA denotes “not applicable”; and x;y denotes
correlation between x and y.
Table 9.4. Summary Statistics
d
p d p p;d
Means:
VAR 0.620 0.268 0.448 0.672
CRRA 0.434 0.244 0.571 0.536
Self-Control 0.418 0.248 0.608 0.564
Std. Dev.:
VAR 0.183 0.065 0.095 0.178
CRRA 0.067 0.041 0.101 0.079
Self-Control 0.069 0.035 0.125 0.101

Notes: VAR statistics summarize ‡at-prior posterior distributions associated with a six-
lag vector autoregression; x denotes the standard deviation of the logged deviation of x
from its steady state value; and x;y denotes the correlation between logged deviations from
steady state values of x and y.

7
Chapter 10, Section 10.2.4, Implementation
In equation (10.33), the roots of Tr (s) should be expressed as

(2j 1)
sbj = cos ; j = 1; 2; :::; r:
r 2

Chapter 10, Section 10.2.6, Application to the Optimal

Growth Model
x c
In equation (10.53), the ratios c
; x = a; k should instead be x
.

Chapter 11, Section 11.1

The model-estimation exercise conducted in this section are a¤ected by the programming
error involving xbar noted above under “Notice Regarding Code”. Table 11.1 should read
as follows:

Table 11.1. Model Simulations.

Linear Approximation Non-Linear Approximation
j j
j j y
'(1) 'j;y (0) j y
'(1) 'j;y (0)
y 0.0153 1.00 0.90 1.00 0.0151 1.00 0.89 1.00
c 0.0110 0.72 0.97 0.95 0.0109 0.72 0.96 0.95
i 0.0330 2.16 0.80 0.94 0.0330 2.19 0.79 0.94
k 0.0189 1.24 0.99 0.86 0.0184 1.22 0.99 0.76
a 0.0112 0.73 0.80 0.93 0.0112 0.74 0.80 0.94
Notes: '(1) denotes …rst-order serial correlation; 'j;y (0) denotes contemporaneous
correlation between variables j and y. Model moments based on the parameterization
=[ ]0 = [0:33 0:96 0:1 2:0 0:8 0:0067]0 .

Substantively, these corrections cast the di¤erences observed across the linear and non-
linear approximation techniques as relatively small in comparison with the di¤erences noted
in the text.

8
Chapter 11, Section 11.3
The estimation exercise conducted in this section are a¤ected by the programming error
involving xbar noted above under “Notice Regarding Code”. Table 11.2 should read as
follows:

Table 11.2. Maximum Likelihood Estimates

Parameter Actual Value Log-Lin. Appx. Non-Lin. Appx. Exact Policy Fcn.
0.33 0:331 0:327 0:327
(1:69e 04) (1:11e 05) (3:77e 05)
0.96 0:957 0:968 0:968
(4:88e 04) (1:04e 04) (3:52e 05)
0.80 0:789 0:794 0:809
(2:18e 03) (5:10e 05) (2:80e 05)
" 0.0067 0:0062 0:0061 0:0061
(5:36e 04) (1:01e 06) (1:08e 06)
u;y 0.00284 0:00485 0:00309 0:00316
(5:45e 04) (2:77e 07) (6:90e 07)
u;i 0.0090 0:00484 0:00086 0:00087
(5:12e 04) (7:81e 09) (6:76e 09)
log L: 782.81 917.74 917.78
Notes: Standard deviations are in parentheses.

The substantive impact of these changes is minimal.