RBC model:
max Et t ln Ct ln Lt t 0
ut ( Ct , Lt )
Subject to:
Yt Ct It
Nt Lt 1 Kt 1 It (1 ) Kt Yt At Kt Nt1
ln At 1 ln At t 1
Step 1: derive for the equilibrium conditions by Lagrangian
Yt 1 1 L E0 ln Ct ln(1 Nt ) t [Ct Kt 1 (1 ) Kt At Kt Nt ] t 0
t
Take derivatives with respect to Ct , Nt , Kt 1 , and t :
Ct :
L 1 1 0 t t Ct Ct Ct Yt
L 1 0 t At K t N t1 Nt : Nt 1 Nt Nt Wt Yt 1
L 0 t Et [t 1 (1 At 1 Kt1 Nt11 )] Kt 1 : Kt 1 Kt 1 Rt 1 Yt 1
L 0 Ct Kt 1 (1 ) K t At K t N t1 0 t : t
Step 1’: derive for the equilibrium conditions by Bellman equation The problem can be restated in the recursive form by Bellman equation.
V (kt ) max[u(Ct , Lt ) Et [V (kt 1 )]] Substitute Ct and Lt by constraints, and the resulting Bellman equation will have only two endogenous control variables Nt (intratemporal) and Kt 1 (intertemporal), but two state variables At (exogenous state variable) and Kt (endogenous state variable). Take F.O.C. of Bellman equation, and apply envelope theorem to get the same sets of equilibrium conditions. The equations can be restated by defining several intermediate variables like Wt , Rt , and Yt . This revision does not change the problem, but it makes future steps easier. The whole dynamic stochastic system can be characterized by both the equilibrium conditions (E1-E6) as well as the law of motion for the state variable At (E7):
(E1)
1 Nt
Wt Ct
Yt Nt
(E2) Wt (1 )
(E3) Yt At Kt Nt1 (E4) 1 Et [
Ct Rt 1 ] Ct 1
Yt Kt
(E5) Rt 1
(E6) Ct Kt 1 (1 ) Kt Yt 0 (E7) ln At 1 ln At t 1 Step 2: obtain the steady state The steady state of the system can be obtained by dropping the time subscripts, because variables will not change in steady state. (S1)
1 N
W C
(S2) W (1 )
Y N
(S3) Y K N 1 (S4) 1 R (S5) R 1
Y K
(S6) C K Y 0 (S7) A 1
Step 3: linearize the system (1) Direct Approach: use Taylor expansion directly (1.1) Any univariate nonlinear equation f ( X t ) 0 : In steady state, the equation becomes f ( X ) 0 . Take first order Taylor expansion around the steady state:
f X t f X f X X t X f X X
(1.2) Any multivariate nonlinear equation f ( X1t , In steady state, the equation becomes f ( X1 ,
Xt X f X Xxˆt X
, X nt ) 0 :
, X n ) 0 , and so:
n
f X 1t ,..., X nt f X i ( X i ) X i xˆit 0 i 1
(2) Indirect Approach: approximation rules implied by Taylor expansion
ln(1 x) x e x 1 x (basic result from Taylor expansion) Xt X
Xt X X X (1 xˆt ) X
X t X e xˆt X (1 xˆt ) X t Yt X Y e xˆt e yˆt X Y e xˆt yˆt X Y (1 xˆt yˆt )
Linearize the Euler’s equations (S1-S6) as well as the law of motion for the exogenous state variable around the steady state. (L1) 0 (1 N )Wwˆ t WNnˆt Ccˆt
ˆt (L2) 0 yˆt nˆt w (L3) 0 yˆt aˆt kˆt 1 nˆt
(L4) Et cˆt 1 Et [rˆt 1 ] cˆt
Y yˆt kˆt rˆt K K C K (L6) kˆt 1 cˆt (1 ) kˆt yˆt Y Y Y (L5) 0
(L7) aˆt 1 aˆt t 1 Step 4: restate the linear system in matrix form Define the vector of variables in period t for this system:
xt exogenous/endogenous state variables y , endogenous control variables t
zt shocks
Note that expectation is needed for controls if we bring the vector to period t 1 :
xt 1 exogenous/endogenous state variables E y , endogenous control variables t t 1
zt 1 shocks
The state vector xt has a dimension of N Nexogenous Nendogenous , the control vector yt has a dimension of M . In our case, N 1 1 2 and M 5 , and the vector zt has a dimension of 1. The 7 linear equations can then be rewritten in the matrix form:
0 0 0 0 0 0 0
0 0 K 0 Y 0 0 0 1 0 0 0 0 0 0 A
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
1 0 0 1 0 0 0
ˆt 1 a ˆ kt 1 E cˆ t t 1 E n ˆ t t 1 ˆt 1 Et y E w ˆ t t 1 Et rˆt 1
0 0 0 1 0 0
0 1 K Y Y K 0
0 0
0 C Y
0
0
0
0
1
0
0
0
Y
0
1 0 0
0 1 1
C WN B
Hence, it can be written as:
K 0 1 1 0
0 0 1 1 N W
0
0 1 0 0 0 0
ˆt a kˆ t cˆt ˆt n y ˆ t w ˆ t rˆt
1 0 0 0 0 0 0 C
t 1 zt 1
x x A t 1 B t Czt 1 Et yt 1 yt If the matrix A is nonsingular, we can multiply A1 on both sides directly. If the matrix A is singular, we can apply Uhlig (1997). The system becomes: x xt 1 1 t 1 A B E y y A C zt 1 F t G t t 1 x F t Gzt 1 yt
Apply Jordan decomposition to the matrix F :
1 F P P P 0 N M N M 0 1
0
1
0
0 0 P N M
Note that the diagonal elements of are the eigenvalues of F , while the column vectors of P 1 are the corresponding eigenvectors, based on:
det F I 0 Denote the number of unstable eigenvalues ( 1 ) as u , and the number of stable eigenvalues ( 1 ) as s . Note that u s M N . Blanchard-Kahn condition is: (i)
If u M , there is unique solution to the system (saddle path);
(ii)
If u M , there is no solution to the system;
(iii)
If u M , there are infinite solutions to the system.
We only focus on the saddle path case. Models that do not satisfy this condition cannot be solved. Hence, check this condition always! Now we can rewrite in two blocks, with a stable block and an unstable block:
s 0
0 u
Similarly, we can partition P and G similarly:
P P 11 P21
P12 Gs , G P22 Gu
The original system becomes:
xt 1 P11 P12 E y P P t t 1 21 22 P11 P12 xt 1 s P P E y 0 21 22 t t 1
1
s 0 P11 P12 xt Gs 0 P P y G zt 1 u 21 22 t u 0 P11 P12 xt P11 P12 Gs z u P21 P22 yt P21 P22 Gu t 1
x t 1 Et yt 1
x t yt
x t 1 s Et yt 1 0
G s Gu
0 xt Gs z u yt Gu t 1
(1) The unstable block (red part) of the system can be solved forward by iteration:
Et yt 1 u yt Gu zt 1 yt u1Et yt 1 u1Gu zt 1 yt u1Gu zt 1 Recover the solution for yt using the definition of yt :
yt u1Gu zt 1 xt P11 y P t 21 Gs P11 Gu P21
P12 xt P22 yt yt P21 xt P22 yt P12 Gs Gu P21Gs P22Gu P22 Gu
P21 xt P22 yt u1 P21Gs P22Gu zt 1
yt P221 P21 xt P221 u1 P21Gs P22Gu zt 1
This is the solution (policy/transition function) for endogenous control variables yt in terms of both endogenous states and exogenous states as well as pure shocks. (2) The stable block (red part) of the system can be solved by substituting in the solu-
tion of yt , based on the original form of the system:
xt 1 xt E y F y Gzt 1 t t 1 t xt 1 F11 F12 xt Gs E y F zt 1 t t 1 21 F22 yt Gu xt 1 F11 xt F12 yt Gs zt 1
xt 1 F11 F12 P221P21 xt Gs P221 u1 P21Gs P22Gu zt 1 This is the solution (law of motion) for endogenous state variables xt 1 in terms of both endogenous states and exogenous states as well as pure shocks.