Supplementary Material

Supplement to A Bayesian dynamic stochastic general

equilibrium model of stock market bubbles
and business cycles: Appendices
(Quantitative Economics, Vol. 6, No. 3, November 2015, 599635)

Jianjun Miao
Department of Economics, Boston University, Institute of Industrial Economics, Jinan University, and
AFR, Zhejiang University

Pengfei Wang
Department of Economics, Hong Kong University of Science and Technology

Zhiwei Xu
Antai College of Economics and Management, Shanghai Jiao Tong University

This online material contains six appendices to the paper. Appendix A proves Propo-
sition 1 in the paper. Appendix B derives the stationary equilibrium. Appendix C stud-
ies the bubbly steady state. Appendix D provides the log-linearized equilibrium system
around the bubbly steady state. Appendix E presents a table of business cycle moments.
Appendix F presents a robustness analysis.

Appendix A: Proof of Proposition 1 in the paper

We use a conjecture and verification strategy to find the decision rules at the firm level.
j
We first study the optimal investment problem by fixing the capacity utilization rate ut .
Using (14) and (16), we can write firm js dynamic programming problem as
 j j  j  j j
vt t Kt + bt t vLt t Lt
j
j j j j Lt+1
= max ut Rt Kt Pt It Lt + (A.1)
j
It Lt+1
j Rf t
 j j j j j
+ Qt 1 t Kt + t It + Bt QLt Lt+1

subject to the investment constraint

j
j j j j Lt+1 j
0 Pt It ut Rt Kt Lt + + t Kt  (A.2)
Rf t

Jianjun Miao: miaoj@bu.edu

Pengfei Wang: pfwang@ust.hk
Zhiwei Xu: xuzhiwei09@gmail.com

Copyright 2015 Jianjun Miao, Pengfei Wang, and Zhiwei Xu. Licensed under the Creative Commons
Attribution-NonCommercial License 3.0. Available at http://www.qeconomics.org.
DOI: 10.3982/QE505
2 Miao, Wang, and Xu Supplementary Material

j j j j
For t Pt /Qt , It = 0. Optimizing over Lt+1 yields QLt = 1/Rf t . For t+1 Pt /Qt , the op-
timal investment level must reach the upper bound in the above investment constraint.
We can then immediately derive the optimal investment rule in (18). In addition, the
credit constraint (17) must bind so that

1 j j
L = Qt t Kt + Bt  (A.3)
Rf t t+1

Substituting the optimal investment rule and QLt = 1/Rf t into (A.1) yields
 j j  j  j j
vt t Kt + bt t vLt t Lt
j j  j j j
= ut Rt Kt + Qt 1 t Kt + Bt Lt (A.4)
 j
 j  j j j j L
+ max Qt t /Pt 1 0 ut Rt Kt + t Kt Lt + t+1 
Rf t

j j
Since ut is determined before observing t , it solves the problem
j j  j j j j
max ut Rt Kt + Qt 1 t Kt + Gt ut Rt Kt  (A.5)
j
ut

where Gt is defined by (20). We then obtain the first-order condition

 j
Rt (1 + Gt ) = Qt  ut  (A.6)

j j
Since t = (ut ) is convex, this condition is also sufficient for optimality. From this con-
j
dition, we can immediately deduce that optimal ut does not depend on firm identity so
that we can remove the superscript j.
By defining t (ut ), (A.4) becomes
 j j  j  j j
vt t Kt + bt t vLt t Lt
j j j
= ut Rt Kt + Qt (1 t )Kt + Bt Lt
 j
 j  j j j L
+ max Qt t /Pt 1 0 ut Rt Kt + t Kt Lt + t+1 
Rf t

j
where Lt+1 /Rf t is given by (A.3). Matching coefficients yields

ut Rt + Qt (1 t )

 j  j  j Pt
v t t = + Qt t /Pt 1 (ut Rt + t + t Qt ) if t  (A.7)

Qt

ut Rt + Qt (1 t ) otherwise

  P
 j  Qt tj /Pt 1 Bt  if tj t 
bt t = Qt (A.8)

Bt  otherwise
Supplementary Material 3

and

 j  Qt tj /Pt 1 j Pt
if t 
vLt t = Qt

1 otherwise.

Using (14), we then obtain (21), and (22) and (23).

Appendix B: Stationary equilibrium

We define the transformed variables
Ct It Yt Kt
Ct  It  Yt  Kt 
t Zt t t t1 Zt1
Pts Bta Xt Wt
Pts  Bta  Xt  Wt 
t t t Zt t
Qt Qt Zt  Pt = Pt Zt  Rt = Rt Zt  t t t 
/(1)
where t = Zt At . The other variables are stationary and there is no need to
scale them. To be consistent with a balanced growth path, we also assume that K0t =
t1 Zt1 K0 , where K0 is a constant.
The six shocks in the model are given as follows.

1. The permanent TFP shock:

p p
At = At1 at  ln at = (1 a ) ln a + a ln at1 + at  (B.1)

2. The transitory TFP shock:

ln Am m
t = am ln At1 + am t  (B.2)

3. The IST shock:

Zt = Zt1 zt  ln zt = (1 z ) ln z + z ln zt1 + zt  (B.3)

4. The sentiment shock:

ln t = (1 ) + ln t1 + t  (B.4)

5. The labor shock:

ln t = (1 ) ln + ln t1 + t  (B.5)

6. The financial shock:

ln t = (1 ) ln + ln t1 + t 

Here, all innovations are mutually independent and are independently and identically
distributed normal random variables.
4 Miao, Wang, and Xu Supplementary Material

Denote by gt t /t1 the growth rate of t . Denote by g the nonstochastic steady

state of gt , satisfying

ln g ln z + ln a  (B.6)
1
On the nonstochastic balanced growth path, investment and capital grow at the rate
of I g z ; consumption, output, wages, and bubbles grow at the rate of g ; and the
rental rate of capital, Tobins marginal Q, and the relative price of investment goods de-
crease at the rate z .
After the transformation described in Section 3, we can derive a system of 15
equations for 15 transformed variables: {Ct  It  Yt  Nt  Kt  ut  Qt  Xt  Pt  Wt  Rt  mt  Bta 
Rf t  t }.

1. Resource constraint:
  2 
It
Ct + 1 + gzt gt I It = Yt  (B.7)
2 It1

where gzt = Zt /Zt1 .

2. Aggregate investment:
 
 
a 1 t

It = Yt + t Qt Xt + Bt  (B.8)
Pt

where t = Pt /Qt .
3. Aggregate output:

Yt = (ut Xt ) Nt1  (B.9)

4. Labor supply:

Yt
(1 ) t = t  (B.10)
Nt

5. The law of motion for capital:

 
t
Kt+1 = (1 t )Xt + It   (B.11)
1 t

where

 
t d()
>t

6. Capacity utilization:

Yt
(1 + Gt ) = Qt  (ut ) (B.12)
ut Xt
Supplementary Material 5

where
  
  t  
Gt = /t 1 d() = + t 1
>t t

7. Marginal Q:

t+1 Qt+1  
Qt = (1 e )Et ut+1  (ut+1 ) + (1 t+1 ) + t+1 Gt+1  (B.13)
t gzt+1 gt+1

8. Effective capital stock used in production:

1 e
Xt = Kt + e K0  (B.14)
gzt gt

9. Euler equation for investment goods producers:

 2 
It It It
Pt = 1 + gzt gt I + gzt gt I gzt gt

2 It1
It1
It1
  2 (B.15)
t+1 It+1 It+1
Et gzt+1 gt+1 I gzt+1 gt+1 
t
It It
10. The wage rate:

Yt
Wt = (1 )  (B.16)
Nt
11. The rental rate of capital:

Yt
Rt =  (B.17)
ut Xt
12. Evolution of the number of bubbly firms:

mt = mt1 (1 e )t1 + e  (B.18)

13. Evolution of the total value of the bubble:

t+1 mt
Bta = Et a
Bt+1 (1 + Gt+1 )(1 e )t  (B.19)
t mt+1

14. The risk-free rate:

1 t+1 1
= Et (1 + Gt+1 )(1 e ) (B.20)
Rf t t gt+1
15. Marginal utility for consumption:
1 h
t = Et  (B.21)
Ct hCt1 /gt Ct+1 gt+1 hCt
6 Miao, Wang, and Xu Supplementary Material

Appendix C: Steady state

The transformed system presented in Appendix B has a nonstochastic steady state. We
eliminate Wt and Rt , and then obtain a system of 15 equations for 15 steady-state values
Y  N K u Q X P W  R m Ba  Rf  }, where we have removed time subscripts.
{C I
We assume that the function () is such that the steady-state capacity utilization rate is
equal to 1. In addition, we set Q = 1, which pins down G.

1. Resource constraint:

C + I = Y  (C.1)

where we have used the fact that I = z g .

2. Aggregate investment:
 
 
a 1

I = Y + QX + B  (C.2)
P

where 1 ( ) = > d() and = P/Q.
3. Aggregate output:

Y = X N 1  (C.3)

4. Labor supply:

Y
(1 ) =  (C.4)
N
5. End-of-period capital stock:
 
 
K = 1 (1) X + I   (C.5)
1

where

 
d()
>

6. Capacity utilization:

Y
(1 + G) = Q (1) (C.6)
X
where
  
   
G= / 1 d() = + 1
>
7. Marginal Q:
1   
1 = (1 e ) (1) + 1 (1) + G  (C.7)
z g
Supplementary Material 7

8. Effective capital stock used in production:

1 e
X = K + e K0  (C.8)
z g

9. Euler equation for investment goods producers:

P = 1 (C.9)

10. The wage rate:

Y
W = (1 )  (C.10)
N
11. The rental rate of capital:

Y
R =  (C.11)
X
12. Evolution of the number of bubbly firms:

m = m(1 e ) + e  (C.12)

13. Evolution of the total value of the bubble:

Ba = Ba (1 + G)(1 e ) (C.13)

14. The risk-free rate:

1 1
= (1 + G)(1 e ) (C.14)
Rf g

15. Marginal utility for consumption:

1 h
=  (C.15)
C hC/g Cg hC

For convenience, define t = Pt /Qt = Pt /Qt as the investment threshold. We use a

variable without the time subscript to denote its steady-state value in the transformed
stationary system. The following proposition characterizes the bubbly steady state.1

Proposition C1. Suppose that > 0 and 0 < min < (1 e ) < . Then there exists a
unique steady-state threshold (min  max ) satisfying

  1
/ 1 d() = 1 (C.16)
> (1 e )
1 The bubbleless steady state can be obtained by setting Ba = 0 and m = = 0. In this case, we can remove

(C.13) and (C.12).

8 Miao, Wang, and Xu Supplementary Material

If the parameter values are such that

  
Ba k 1 (1) x
=     x > 0 (C.17)
Y 1/ (1 e )

where we define
 1
1 e K0
k + e  (C.18)
z g K

x     (C.19)
z g 1 (1) (1 e ) 1 (1 e )

then there exists a unique bubbly steady-state equilibrium with the bubbleoutput ratio
given in (C.17). The steady-state growth rate of the bubble is given by = Rf /g , where Rf
is the steady-state interest rate. In addition, if
1
 (1) =  (C.20)
(1 e ) x
then the capacity utilization rate in this steady state is equal to 1.

Proof. In the steady state, (B.15) implies that P = 1. Hence, by definition, we have =
1/Q. Then by the evolution equation (B.19) of the total bubble, we obtain the steady-
state relation

1  
1=G= / 1 d() (C.21)
(1 e ) >

Define the expression on the right-hand side of the last equality as a function of :
G( ). Then we have G(min ) = 1 1 and G(max ) = 0. Given the assumption that
min
min < (1 e ), there is a unique solution to (C.21) by the intermediate value theo-
rem. In addition, by the definition of G, we have
 
  
G=
1 


where ( ) = > d(). Thus, ( ) can be expressed as

    
= G + 1  (C.22)

Suppose that the steady-state capacity utilization rate is equal to 1. The steady-state
version of (B.13) gives (C.7) and the steady-state version of (B.12) gives (C.6). Using these
two equations, we can derive
 
Y Q gz g  
= 1 (1) G  (C.23)
X 1 + G (1 e )
Substituting (C.21) into the above equation yields

QX
= x  (C.24)
Y
Supplementary Material 9

where x is given by (C.19). To support the steady state u = 1, we use (B.12) and (C.24)
to show that condition (C.20) must be satisfied.
From (B.14), the end-of-period capital stock to the output ratio in the steady state
satisfies
K X
= k  (C.25)
Y Y
where k is given by (C.18). Then from (B.11), we can derive the steady-state relation
 
I 1    X
=   k 1 (1)
Y Y
 
1    QX
=   k 1 (1) (C.26)
G+1 Y
    
1 k 1 (1) x
=   
G + 1

where the second line follows from (C.22) and = 1/Q, and the last line follows from
(C.24). After substituting (C.21) into the above equation, we solve for 1 ( ):
 
  1/ (1 e ) 1
1 =     (C.27)
Y )1 k 1 (1) x 1
(I/

From (B.8), the steady-state total value of bubble to GDP ratio is given by

Ba I 1 QX
=   
Y Y 1 Y
Substituting (C.21), (C.26), and (C.24) into the above equation yields (C.17). We require
Ba /Y > 0. By (23) and (34), the growth rate of bubbles of the surviving firms in the steady
state is given by = Rf /g . 

Appendix D: Log-linearized system

We eliminate equations for Wt and Rt . The log-linearized system for 13 variables
{Ct  It  Yt  Nt  Kt  ut  Qt  Xt  Pt  mt  Bta  Rf t  t }, including two growth rates, are summa-
rized as follows.
1. Resource constraint:
C I
Yt = Ct + It  (D.1)
Y Y
2. Aggregate investment:
x
It = Yt + (t + Qt + Xt )
+ x + Ba /Y + x + Ba /Y
(D.2)
Ba /Y
+ Bta t Pt 
+ x + Ba /Y
10 Miao, Wang, and Xu Supplementary Material

where
 

=   t = Pt Qt  (D.3)
1

3. Aggregate output:

Yt = (ut + Xt ) + (1 )Nt  (D.4)

4. Labor supply:

t + Yt Nt = t  (D.5)

5. End of period the capital stock:

 
 (1) 1 (1) 1 (1)
Kt+1 = ut + Xt + 1 It  (D.6)
k k k G t

where
 
1
G 1 (D.7)
G
6. Capacity utilization:

   (1)
Yt Xt + 1 (1 e ) G t = Qt + 1 +  ut  (D.8)
(1)

7. Marginal Q:

Qt = Et (t+1 t ) + Et (Qt+1 gzt+1 gt+1 )

(1 e ) (1)  (1)
+ Et ut+1 (D.9)
z g  (1)
(1 e )G 

+ Et t+1 + G t+1 
z g

8. Effective capital stock:

1 e
Xt = k (Kt gzt gt ) (D.10)
z g

9. Euler equation for investment goods producers:


Pt = Et (1 + )g2 2z It + 2z g2 (gt + gzt )
 (D.11)
2z g2 It1 2z g2 (It+1 + gzt+1 + gt+1 ) 

10. Evolution of the number of bubbly firms:

mt = (1 e )mt1 + (1 e )t1  (D.12)

Supplementary Material 11

11. Evolution of the total value of the bubble:

   
Bta = Et t+1 t + Bt+1
a
+ 1 (1 e ) G Et t+1
(D.13)
1 (1 e )
+ Et mt+1 
(1 e )

12. The risk-free rate:

 
Rf t = Et (t+1 t gt+1 ) + 1 (1 e )Rf /g G Et t+1  (D.14)

13. Marginal utility for consumption:

 
g g h
t = Ct + (Ct1 gt )
g h g h g h
  (D.15)
h g h
Et (Ct+1 + gt+1 ) + Ct 
g h g h g h

14. The growth rate of consumption goods:

 
gt = zt + at + Am m
t At1  (D.16)
1
15. The growth rate of the investment goods price:

gzt = zt  (D.17)

In the above system, G is determined by (C.13),

1
G= 1 (D.18)
(1 e )

(1 ( )) is given by (C.27), and  (1) satisfies (C.20). The log-linearized shock pro-
cesses are listed below.

1. The permanent technology shock:

at = a at1 + at  (D.19)

2. The transitory technology shock:

Am m
t = am At1 + am t  (D.20)

3. The permanent investment-specific technology shock:

zt = z zt1 + zt  (D.21)

4. The labor supply shock:

t = t1 + t  (D.22)
12 Miao, Wang, and Xu Supplementary Material

5. The financial shock:

t = t1 + t  (D.23)

6. The sentiment shock:

t = t1 + t  (D.24)

Appendix E: Business cycle moments

To evaluate our model performance, we present in Table S1 the baseline models predic-
tions regarding standard deviations, correlations with output, and serial correlations of
output, consumption, investment, hours, and stock prices. This table also presents re-
sults for four estimated comparison models discussed in the paper. The model moments
are computed using the simulated data from the estimated model when all shocks are
turned on. We take the posterior modes as parameter values. Both simulated and actual
data are in logs and are HP-filtered.

Appendix F: Robustness

F.1 Extended model with consumer sentiment index

Table S2 reports the prior and posterior distributions of estimated parameters in the
extended model of Section 5, with the consumer sentiment index as one of the observa-
tion series. The parameters {aj  bj }5j=1 in the table are coefficients in the equation for the
sentiment shock and in the observation equation of the consumer sentiment index. The
variable err represents the standard deviation of the measurement error.

F.2 Priors

In our baseline model of the paper, we choose 10 percent as the prior mean of because
we know that the stock market volatility is very high. To see if our result is robust to a
smaller prior mean of , we set the prior as Inv-Gamma with mean 001 and standard
deviation infinite. We redo Bayesian estimation and report estimation results in Table S3.
We find that these results are very similar to those in the baseline estimation.

F.3 A hybrid model

Our baseline model has abstracted away from many other potentially important shocks
such as news shocks or uncertainty shocks. Thus, it is possible that the sentiment
shock is not important at all in explaining stock prices and real variables if other
shocks are taken into account. To examine this possibility, we follow the methodol-
ogy of Ireland (2004) and combine the DSGE model with the VAR model. We then
Supplementary Material 13

Table S1. Business cycles statistics.

Y C I N SP P

Standard Deviations (%)

U.S. data 170 093 419 179 1082 111
Baseline model 184 146 429 130 1058 106
No stock price 115 095 304 111 132 122
No sentiment 150 140 332 157 1020 249
No bubble 177 165 410 187 1028 261
Extended 246 194 535 125 1220 112

Standard Deviations Relative to Y

U.S. data 100 055 247 105 636 065
Baseline model 100 079 232 070 574 058
No stock price 100 083 263 096 115 106
No sentiment 100 093 221 104 678 165
No bubble 100 093 232 106 581 148
Extended 100 079 217 051 496 045

First-Order Autocorrelations
U.S. data 090 090 087 093 077 086
Baseline model 089 093 079 078 076 085
No stock price 083 089 073 077 072 088
No sentiment 091 091 083 074 072 081
No bubble 094 094 087 078 072 075
Extended 091 094 084 084 076 086

Correlation With Y
U.S. data 100 093 097 082 042 013
Baseline model 100 094 088 061 039 007
No stock price 100 088 080 068 045 008
No sentiment 100 085 074 056 006 014
No bubble 100 090 071 052 008 007
Extended 100 096 091 064 050 008

Note: The model moments are computed using the simulated data (20,000 periods) from the estimated model at the poste-
rior mode. All series are logged and detrended with the HP filter. The columns labeled Y , C , I , N , SP , and P refer, respectively,
to output, consumption, investment, hours worked, the stock price, and the relative price of investment goods. No bubble
corresponds to the model without bubbles. No sentiment corresponds to the baseline model without sentiment shocks. No
stock price corresponds to the baseline model without using the stock price data in estimation. Extended corresponds to
the model in Section 5.

estimate this hybrid model using Bayesian methods.2 Following Ireland (2004), we
now shut down all the shocks in the baseline model except the sentiment shock
and introduce four measurement errors into the measurement equations for the data
{PtsData  CtData  ItData  ln N Data }. Specifically, let
P sData ln(g )
t Pts

C Data C ln(g )
t
+
t
= + t  (F.1)
ItData It ln(g )
ln N Data Nt ln(N)

2 We thank Tao Zha for suggesting that we conduct this analysis.

14 Miao, Wang, and Xu Supplementary Material

Table S2. Priors and posteriors of estimated parameters in the extended model.

Prior Distribution Posterior Distribution

Parameter Distr. Mean Std. Dev. Mode Mean 5% 95%

h Beta 033 024 057 056 050 063

Gamma 2 2 003 004 001 006
 / Gamma 1 1 1566 1557 1164 1943
Beta 03 01 025 025 021 029
Gamma 2 2 264 274 234 315

f1 Gamma 1 1 008 007 001 013

f2 Gamma 1 1 526 613 372 850
f3 Gamma 1 1 067 062 000 110
a1 Gamma 10 3 616 668 345 972
a2 Gamma 10 3 1348 1475 906 1996
a3 Gamma 10 3 820 884 499 122
a4 Gamma 10 3 536 558 329 797
a5 Gamma 10 3 389 398 223 556
b1 Gamma 2 2 025 025 019 030
b2 Gamma 2 2 291 265 062 452
b3 Gamma 2 2 379 363 169 536
b4 Gamma 2 2 166 219 058 381
b5 Gamma 2 2 038 093 008 183

a Beta 05 02 068 067 052 080

am Beta 05 02 081 080 073 087
z Beta 05 02 040 038 026 051
Beta 05 02 096 095 094 096
Beta 05 02 096 096 095 097
Beta 05 02 097 096 095 098

a (%) Inv-Gamma 1 Inf 074 075 059 091

am (%) Inv-Gamma 1 Inf 066 067 056 077
z (%) Inv-Gamma 1 Inf 059 060 053 067
(%) Inv-Gamma 1 Inf 1370 1380 1028 1705
(%) Inv-Gamma 1 Inf 080 081 071 090
(%) Inv-Gamma 1 Inf 076 070 039 097
err (%) Inv-Gamma 1 Inf 863 876 787 971

where t is the vector that contains four measurement errors, g is the gross growth rate
of output, and N is the average hours in the data. Following Ireland (2004), we assume
that the measurement errors t follow a VAR(1) process

t = At1 + B t  (F.2)

where A is the coefficient matrix and B is assumed to be lower triangular such that the
innovations in t are orthogonal to each other.
The measurement errors in (F.2) can be considered as a combination of all omit-
ted structural shocks in our baseline model and allow for potential model misspecifi-
Supplementary Material 15

Table S3. Prior and posterior distributions of parameters.

Prior Distribution Posterior Distribution

Parameter Distr. Mean Std. Dev. Mode Mean 5% 95%

h Beta 033 024 054 054 049 061

Gamma 2 2 003 003 001 006
 / Gamma 1 1 1144 1192 833 1549
Beta 03 01 029 030 022 036
Gamma 2 2 257 260 212 319

f1 Gamma 1 1 005 004 001 007

f2 Gamma 1 1 473 482 254 708
f3 Gamma 1 1 042 032 000 056

a Beta 05 02 096 097 094 099

am Beta 05 02 097 096 095 098
z Beta 05 02 036 034 022 046
Beta 05 02 093 092 090 095
Beta 05 02 099 098 096 099
Beta 05 02 088 087 081 094

a (%) Inv-Gamma 001 Inf 023 023 018 029

am (%) Inv-Gamma 001 Inf 103 104 093 116
z (%) Inv-Gamma 001 Inf 059 060 054 067
(%) Inv-Gamma 001 Inf 17 85 1946 1165 2638
(%) Inv-Gamma 001 Inf 081 082 072 093
(%) Inv-Gamma 001 Inf 077 084 042 121

cations. We allow the measurement errors to be flexible enough so that the data are
not necessarily driven by the sentiment shock. The idea is that if the sentiment shock
is not the driving force, then (F.1) and (F.2) form a first-order Bayesian VAR system and
the measurement errors should be important in explaining fluctuations in the data of
{PtsData  CtData  ItData  ln N Data }. On the other hand, if the baseline model is correctly
specified and the sentiment shock is the main source of fluctuations, then the estimated
measurement errors will be unimportant.
The variance decomposition shows that the sentiment shock remains the single
most important factor accounting for the stock price variation, although its importance
is somewhat reduced. It explains about 82 percent of the variation in the stock prices. It
still accounts for significant fractions of fluctuations in investment, consumption, and
output, explaining about 26, 38, and 35 percent, respectively. As in the baseline model,
the sentiment shock is not important in explaining the fluctuation in hours. We also
find that the estimates of the common parameters in the hybrid model are very similar
to those in the baseline model. The smoothed sentiment shock is still highly correlated
with the consumer sentiment data; the correlation is about 073. These results suggest
that the importance of the sentiment shock is robust to the model variation and specifi-
cation of different shocks.
16 Miao, Wang, and Xu Supplementary Material

Additional reference
Ireland, P. N. (2004), A method for taking models to the data. Journal of Economic Dy-
namics and Control, 28, 12051226. [12, 13, 14]

Submitted October, 2014. Final version accepted May, 2015.