Beruflich Dokumente
Kultur Dokumente
George McCandless
UCEMA
Spring 2007
Simple
Added indivisible labor to gain persistence and covariance with output
Set rules for RBC game
Match second moments
Newer rule: match impulse response functions
Hansens basic model
Robinson Crusoe maximizes the discounted utility function
max
1
X
t=0
u(ct ; lt )
ht ) = ln ct + A ln(1
ht )
with A > 0.
The production function is
f ( t ; kt ; h t ) =
t
1
t kt ht
t+1
for 0 <
+ "t+1
=) E
is 1 and
t+1
> 0.
)kt + it
ct + it
Bellmans equation
The basic Bellmans equation
V (kt ;
t)
ht ) + Et [V (kt+1 ;
ct ;ht
t+1 )
subject to
1
t kt ht
t+1
kt+1
ct + it ;
=
t + "t+1 ; and
= (1
)kt + it :
Simpler to write as
V (kt ;
t)
max ln
kt+1 ;ht
+A ln(1
1
t kt ht
+ (1
)kt
ht ) + Et [V (kt+1 ;
kt+1
t+1 )
t ]]
t ]]
1
+ (1
)kt
+ Et [Vk (kt+1 ; t+1 ) j t ]
0=
1
t kt ht
kt+1
and
@V (kt ;
@ht
t)
0 = (1
A
1
+ (1
1
t kt ht
)kt
t kt ht
kt+1
1
1
ht
t)
1
+ (1
1
t kt ht
)kt
t kt
kt+1
1 1
ht
+ (1
kt+2
1
t kt ht
"
Et
) (1
ht )
)kt
kt+1
1 1
t+1 kt+1 ht+1
1
t+1 kt+1 ht+1 + (1
+ (1
)kt+1
and
(1
t kt ht
1
t kt ht
=A
+ (1
In equilibrium,
ct =
1
t kt ht
+ (1
)kt
kt+1
t kt
1 1
ht
and
wt = (1
t kt ht
ht ) wt = Act
Stationary states
3
)kt
kt+1
A
(1
1
1
(1
i;
"
(1
#11
G(xt )
H(xt )
ln(H(xt ))
gives
F 0 (x)
(xt
F (x)
ln(F (x)) +
G0 (x)
(xt x)
G(x)
H 0 (x)
ln(H(x))
(xt x)
H(x)
x)
ln(G(x)) +
ln(H(x)))
x)
G0 (x)
(xt
G(x)
H 0 (x)
(xt
H(x)
x)
x)
t Kt
Ht1
Take logs
ln Yt = ln
+ ln Kt + (1
) ln Ht
1
Yt
Y
ln +
+ ln K +
+ (1
) ln H +
(1
Ht
Kt
) ln H
get
1
Yt
Y
1
t
Kt
K +
(1
)
H
That reduces to
Yt
+1
Y
(1
) Ht
Kt
+
H
K
Ht
Xt = XeXt
or
et = ln Xt
X
ln X
At Bt
AeAt B e
=
Ct
C e Cet
becomes
AB
eAt +
et
B
et
B
et
C
Reference:Uhlig, Harald, (1999) "A toolkit for analysing nonlinear dynamic stochastic models easily", in Ramon Marimon and Andrew Scott,
Eds., Computational Methods for the Study of Dynamic Economies, Oxford University Press, Oxford, p.30-61.
eAt +
et
B
et
C
eA+
e
C
e
B
+ eA+
=
So
e
B
e
C
et + B
et
1+A
eAt +
The approximation is
At Bt
Ct
et
B
+ eA+
et
C
AB
C
et
B
e
B
et ;
C
e
C
e
B
et
A
e
A
eA+
et + B
et
1+A
et
C
et + B
et
1+A
et
C
eXt +aYt
et Yet
X
h
i
e
Et aeXt+1
Et [Xt+1 ]
et + aYet ;
1+X
0;
h
i
et+1
a + aEt X
h
i
et+1
= X 1 + Et X
6
e
B
e
C
et
C
e
C
ACt
Ct
Yt
=
=
rt
Ct
(rt+1 + (1
Ct+1
Yt
(1
) (1 Ht )
Ht
Yt + (1
)Kt Kt+1
1
K
H
t t
t
Yt
Kt
Et
))
1 = Et
Et
"
CeCt
CeCet+1
r
et+1
re
+ (1
))
CeCt
CeCet+1
i
h
e
e
e
e
) eCt Ct+1
Et reCt Ct+1 +ert+1 + (1
h
i
h
et C
et+1 + ret+1 + (1
et
rEt 1 + C
) 1+C
i
h
et C
et+1 + re
rt+1 ;
= Et 1 + C
et+1
C
et
C
ACeCt = (1
et
AC 1 + C
et
AC C
(1
"
(1
) (1
Y Yet
e
H
Y
1 + Yet
H
Ht )
et
H
et
H
#
1 H Y e
)
Yt
H
7
Yt
Ht
e
) Y eYt
(1
(1
(1
) Y 1 + Yet
Y e
Ht
H
H Y
H
Yet
Y
)H
(1
(1
(1
H )Y
H
so
et
0=C
Yet +
et
H
1 H
e t = Yet
H
et
H
1 H
et + K
e t + (1
0
where r = Y =K
Yet
et
K
et
)H
ret
Yet
+ "t+1
et+1
et
+ "t+1
1 + et + "t+1
et+1 = et +
t+1
et + K
e t + (1
e t ret
= Yet K
0
and
et
)H
et+1 = et +
t+1
Yet
e t+1
K
et
C
et
H
ret
h i
et
xt = K
2
3
Yt
6 Ct 7
7
yt = 6
4 Ht 5
rt
Dene the stochastic variable as
zt = [ t ]
Solving the log-linear version of the model
The model can be written as
0 = Axt + Bxt 1 + Cyt + Dzt ;
0 = Et [F xt+1 + Gxt + Hxt 1 + Jyt+1 + Kyt + Lzt+1 + M zt ] ;
zt+1 = N zt + "t+1 , Et ("t+1 ) = 0:
Where
6
A=6
4
3
0
K 7
7
0 5
0
6 K(
B=6
4
9
+ 1) 7
7
5
1
1
6 Y
C=6
4 1
1
3
0
0 7
7
0 5
1
1
1 H
1
C
0
0
0
1
0
3
7
7
5
J=
K=
r
0
;
;
L = [0]
M = [0]
N = [ ]
Solving the linear version of the model
We look for a solution of the form
xt
yt
= P xt
= Rxt
+ Qzt
1 + Szt
1
Note that here C is of full rank and has a well dened inverse C
(F JC 1 A)P 2
C 1 (AP + B);
N0
(F
= vec JC
JC
1
(JC
A) + Ik
L)N + KC
G + KC
(AQ + D)
= P xt
= Rxt
10
A)P
(JR + F P + G
and
S=
+ Qzt ;
1 + Szt :
1
KC
KC
B+H
A) vec(Q)
We begin by substituting the laws of motion into the two equations of the
model
Reduce each equation to one in which there are only two variables:
xt
and zt :
= P xt
= Rxt
+ Qzt ;
+
Szt :
1
1
+ Qzt ] + Bxt
+ C [Rxt
+ Szt ] + Dzt
= Et [F [P [P xt 1 + Qzt ] + Q [N zt + "t+1 ]] + G [P xt
+Hxt 1 + J [R [P xt 1 + Qzt ] + S [N zt + "t+1 ]]
+K [Rxt 1 + Szt ] + L [N zt + "t+1 ] + M zt ]
+ Qzt ]
= F [P [P xt 1 + Qzt ] + QN zt ] + G [P xt 1 + Qzt ]
+Hxt 1 + J [R [P xt 1 + Qzt ] + SN zt ]
+K [Rxt 1 + Szt ] + LN zt + M zt
11
+ [AQ + CS + D] zt ;
and
0
[F P P + GP + H + JRP + KR] xt 1
+ [F P Q + F QN + GQ + JRQ + JSN + KS + LN + M ] zt :
0 = AP + B + CR
0 = AQ + CS + D
0 = F P P + GP + H + JRP + KR
0 = F P Q + F QN + GQ + JRQ + JSN + KS + LN + M
Explaining the solution
The third equation is
0 = F P 2 + GP + JRP + H + KR
and the rst is (if the inverse of C exists)
R=
AP
= F P 2 + GP
+H
0
0
J C
1
K C
AP + C
AP + C
B P
= FP
JC AP + GP JC 1 AP 2 JC 1 BP
+H KC 1 AP KC 1 B
= F JC 1 A P 2
JC 1 B + KC 1 A G P
KC
B+H
1 matrix (a scalar)
F
KC
JC
1
A P2
JC
B + KC
B+H
G P
AP
AQ
AQN
A Q+ F
JC
A QN
JC
A QN
= F P Q + F QN + GQ + JRQ JC
KC 1 AQ KC 1 D + LN + M
JC
DN
Rearrange to get
F P + G + JR
= JC
DN + KC
KC
LN + M
DN + KC
KC
LN + M
A Q+ F
KC
JC
A Q
A QN
A) vec(B)
= JC
DN + KC
KC
LN + M
A Q+ F
JC
A QN
as
W QI + XQN = Z
(notice that we added I) where
W
X
Z
= F P + G + JR KC
= F JC 1 A
= JC 1 DN + KC 1 D
A
LN + M
W ) vec (Q) + (N 0
or
(I 0
If (I 0
W + N0
W + N0
X) is invertible
vec (Q) = (I 0
W + N0
X)
14
vec (Z)
First vec
a11
a21
vec
a12
a22
6
6
6
=6
6
6
4
a13
a23
a11
a21
a12
a22
a13
a23
7
7
7
7:
7
7
5
B =
6
6
6
= 6
6
6
4
a11 b11
a11 b21
a11 b31
a21 b11
a21 b21
a21 b31
3
b12
b22 5 =
b32
b11
4 b21
b31
a12
a22
a11 b12
a11 b22
a11 b32
a21 b12
a21 b22
a21 b32
a12 b11
a12 b21
a12 b31
a22 b11
a22 b21
a22 b31
a12 b12
a12 b22
a12 b32
a22 b12
a22 b22
a22 b32
a11 B
a21 B
3
7
7
7
7:
7
7
5
Calibration
Solution to model is numerical
Need values for parameters
Some we borrow from literature (quarterly)
= :99
= :025
= :36
Need a value for A
Choose A so that H = 1=3
Use stationary state equation for H
H=
1+
A
(1
1
1
(1
a12 B
a22 B
r = 1= = 1:0101
From data for US use
= :95
3
0
7
6 12: 353 7
7
7
B=6
5
4 0:36 5
1
3
1
1:5004 0
0:9186
0
0 7
7
0
:64
0 5
0
0
1
2 3
0
6 0 7
7
D=6
4 1 5
0
F
G
H
J=
= [0]
= [0]
= [0]
K=
1
0
0 :0348
1
L =
M =
N =
[0]
[0]
[:95]
16
3
7
7
5
3
1: 452 3
6 0:392 7
7
S= 6
4 0:706 7 5
1: 452 3
and
ret
e t + 0:1132et ;
0:9537K
e t + 1:4523et ;
= 0:2045K
e t + 0:3920et ;
= 0:5691K
e t + 0:7067et ;
=
0:2430K
e t + 1:4523et :
=
0:7955K
=
et = :95et
= :0032
et
H
1:640 "
29:9 0%
et
H
94:3 2%
ret
3:492 "
63:6 7%
Iet
4 88:6 4%
Iet
11:742 "
214:1%
18
1
X
u(ct ;
t)
t=0
subject to
ct + it = wt ht + rt kt
t
Expected utility
u(ct ;
t)
u(ct ;
t)
A ln(1 h0 )
+ A(1
h0
A ln(1 h0 )
ln ct + ht
h0
ln ct + ht
ht
) ln(1)
h0
Household problem
Maximization problem becomes
max
1
X
[ln ct + Bht ]
t=0
with
B=
A ln(1 h0 )
h0
subject to constraints
1
t kt ht
= ct + kt+1
(1
)kt
and
ln
t+1
ln
+ "t+1
Household problem
First order conditions
0
1
(1
) t kt ht + B;
ct
1
1
1 1
+ Et
ht
t+1 kt
ct
ct+1
+ (1
Ct
Ct
(rt+1 + (1
Ct+1
(1
) Yt
:
BHt
Et
)) ;
Full model
Ct
Ct
(rt+1 + (1
Ct+1
(1
) Yt
BHt
Et
20
))
Ct + Kt+1
rt
Yt
= Yt + (1
)Kt
1 1
=
K
H
t t
t
1
=
t K t Ht
Stationary state
Equations
= r + (1
(1
)Y
BH
r
Y
C
= K
H
1
= K H
= Y
K
solve to give
(1
H=
B 1
)
1
and
(1
K=
1 + (1
Stationary state
Comparing to basic Hansen model
To get same stationary state, need H the same in both cases
Then other variables will be the same
Old stationary state equation
H=
1+
A
(1
1
1
(1
1+
A
(1
1
1
We replaced B with
(1
i=
(1
A ln(1 h0 )
h0
)
1
A ln(1 h0 )
h0
;
(1
Stationary state
Solve to get
h0
=
ln(1 h0 )
A
(1
1+
)
A
(1
1
h
(1
1
)
(1
i
)
i =G
G is a constant
To nd h0
= :573, and H = :3335
Get h0 = :583,
Log-linear model
et Et C
et+1 + rEt ret+1
C
et + H
e t Yet
C
et + (1
et K K
e t+1
Y Yet C C
)K K
e t (1
et
Yet et
K
)H
Yet
et
K
ret
Solve for
xt
yt
2
6
A=6
4
1
6 Y
C=6
4 1
1
J=
+ Qzt
+
Szt
1
1
3
0
K 7
7
0 5
0
1
C
0
0
= P xt
= Rxt
6 K(
B=6
4
1
0
(1
+ 1) 7
7
5
1
2
3
0
6 0 7
7
D=6
4 1 5
0
3
0
0 7
7
0 5
1
)
0
K=
and
et + S
yt = R K
where
3
1: 941 8
6 0:470 3 7
7
S=6
4 1: 471 5 5
1: 941 7
2
3
0:055
6 0:531 6 7
7
R=6
4 0:476 6 5
0:945
2
Results
Yet
6:431
100%
"
et
C
4:081 "
63:4 6%
et
H
3:444 "
53:5 5%
ret
4:514 "
70:1 9%
Iet
15:722 "
244:5%
0.01
technology
0.008
0.006
0.004
0.002
0
10
20
30
40
50
periods
60
70
80
90
100
= :95
1
+ "t
e 1 = 0 using
Then calculate the time path of capital with K
e t+1 = P K
e t + Qet
K
24
-3
20
x 10
15
Y
K
10
C
5
H
0
r
-5
10
20
30
40
50
periods
60
70
80
90
100
-3
20
x 10
15
K
10
C
5
H
0
r
-5
10
20
30
40
50
periods
60
70
80
90
100
25
0.02
K
Indivisible labor model
0.015
0.01
0.005
0
100
-0.005
50
-0.01
-0.01
0
-0.005
0.01
0.005
0.015
0.02
Basic model
-3
20
x 10
15
H
45
Y
10
K
5
C
0
r
-5
-5
5
10
response of model w ith divisible labor
15
20
-3
x 10
26