Chapter 3
Contents
I Overview 1
II Solution of Deterministic Models 1
II.1 Finite-Horizon Models . . . 1 II.2 Infinite Horizon Models . . . 3
III Solution of Stochastic Models 4
III.1 An Illustrative Example . . . 4 III.2 The Algorithm in General . . . 5
IV Further Applications 6
IV.1 The Benchmark Model . . . 6 IV.2 A Small Open Economy . . . 7
I Overview
• A solution for the deterministic finite-horizon Ramsey model: a system of non-linear equa- tions.
• A solution for the deterministic infinite-horizon Ramsey model: an infinite number of un- knowns.
• A solution for the stochastic infinite-horizon Ramsey model: “deterministic extended path.”
II Solution of Deterministic Models
II.1 Finite-Horizon Models
The Model. max
{Ct,Kt+1}Tt=0 T
∑
t=0
βtC
1−η t
1 − η
s.t. Kt+1 = Ktα− Ct+ (1 − δ)Kt, K0 : given,
KT = 0,
1
where β ∈ (0, 1), η > 0, α ∈ (0, 1), and δ ∈ [0, 1]. The F.O.C.s can be written as a system of T non-linear equations in the T unknown capital stocks K1, . . ., KT.
0 = [(1 − δ)Kt+ Ktα− Kt+1]−η− β[(1 − δ)Kt+1+ Kt+1α − Kt+2
]−η
(1 − δ + αKt+1α−1), t= 0, 1, . . . , T − 1,
K0 : given, KT = 0,
Non-Linear Equation Solvers.
• The common structure of the algorithms is the iterative scheme: xs+1 = xs+ µ∆xs
• An initial guess of the solution x0
• A direction of change ∆x
• A step length µ
• Find xs so that f (xs) ≃ 0 or xs+1− xs ≃ 0 Solutions.
II.2 Infinite Horizon Models
The Model.
{Ct,Kmaxt+1}∞t=0
∞
∑
t=0
βtC
1−η t
1 − η
s.t. Kt+1 = Ktα− Ct+ (1 − δ)Kt, K0 : given.
Kt is approaching to the steady state level K∗.
K∗ =[ 1 − β(1 − δ) αβ
]α−11
We cannot use non-linear equation solvers in this case. However, in which T is large enough that KT+1 = K∗, we can use non-linear equation solvers again. Shooting algorithm can also be applied to solve this problem.
Solutions.
3
III Solution of Stochastic Models
III.1 An Illustrative Example
{Ct,Kmaxt+1}∞t=0E0
[ ∞
∑
t=0
βtC
1−η t
1 − η ]
s.t. Kt+1 = ZtKtα− Ct+ (1 − δ)Kt, ln Zt= ρ ln Zt−1+ ǫt,
K0 : given,
where ρ ∈ [0, 1) and ǫt∼ N (0, σ2). The F.O.C.s can be written as follows. 0 = [(1 − δ)Kt+ Ktα− Kt+1]−η− βEt
{[(1 − δ)Kt+1+ Kt+1α − Kt+2
]−η
(1 − δ + αKt+1α−1)}, t= 0, 1, . . . ,
K0 : given,
and transversality condition. Under the assumption of the rational expectations, the expected future path of Zt is:
{Z0, Z1, . . .} = {Zt}t=0∞ ={Z0ρt}∞
t=0.
Then we can obtain an approximation of this path from the solution of the system of T non-linear equations
0 =[(1 − δ)Kt+ Z0ρtKtα− Kt+1
]−η
− β[(1 − δ)Kt+1+ Z0ρt+1Kt+1α − Kt+2
]−η
(1 − δ + αKt+1α−1), t= 0, 1, . . . , T − 1,
K0 : given, KT+1 = K∗,
III.2 The Algorithm in General
Notation.
• The system of stochastic difference equations that governs this model is generally described as
0 = Et[gi(xt, yt, zt, xt+1, yt+1, zt+1)] , zt= Πzt−1+ σΩǫt,
i= 1, 2, . . . , n(x) + n(y), t= 0, 1, . . . ,
where ǫt ∼ N(0n(z), In(z))
• xt ∈ Rn(x): the vector of state variables with x0.
• yt ∈ Rn(y): the vector of control and co-state variables (sometimes these are so called “jump variables”).
• zt∈ Rn(z): the vector of purely exogenous variables which are logarithmic. The Algorithm.
Deterministic Extended Path: Simulation of the stochastic DGE model
✓ ✏
Step 1. Initialize: Let p denote the number of periods to consider and (x0, z0) the initial state of the model.
Step 1.1 Draw a sequence of shocks {ǫt}pt=0.
Step 1.2 Compute {zt}pt=1 from zt = Πzt−1+ σΩǫt. Step 1.3 Choose large T .
Step 2. For t = 0, 1, . . . , p repeat the following steps to obtain decisions for period t: Step 2.1 Given zt, compute {zt}pt=1 from zt+s= Πszt.
Step 2.2 Solve the system of T (n(x) + n(y)) equations by using non-linear equations solver or shooting method,
0 = Et[gi(xt+s, yt+s, zt+s, xt+s+1, yt+s+1, zt+s+1)] , i= 1, 2, . . . , n(x) + n(y),
s= 0, 1, . . . , T − 1, x∗ = xt+T.
Keep xt+1 and yt.
Step 2.3 Use xt+1 as starting value for next period t + 1.
✒ ✑
5
IV Further Applications
IV.1 The Benchmark Model
{Ct,Nmaxt,Kt+1}∞t=0E0
[ ∞
∑
t=0
βtC
1−η
t (1 − Nt)θ(1−η)
1 − η
]
s.t. Kt+1 = ZtKtα(AtNt)1−α+ Ct+ (1 − δ)Kt, At+1 = aAt, (a ≥ 1)
ln Zt= ρ ln Zt−1+ ǫt, K0 : given.
When we define the detrended variables as kt ≡ AKtNtt and ct ≡ ACtNtt, we obtain the F.O.C.s as follows.
0 =c−ηt (1 − Nt)θ(1−Nt)− λt,
0 =θc1−ηt (1 − Nt)θ(1−η)−1− (1 − α)λtZtNt−αktα, 0 =akt+1− (1 − δ)kt+ ct− ZtNt1−αktα,
0 =λt− βa−ηEtλt+1(1 − δ + αZt+1Nt+11−αkt+1α−1).
• xt ≡ kt
• yt ≡ (ct, Nt, λt)
• T = 150. Therefore, a system contains 600 unknown variables.
However, we can eliminate ct and λt by using ct = 1−αθ (1 − Nt)ZtNt−αktα. Then, we obtain 2T equations as follows.
0 =ZtρsNt+s1−αkt+sα + (1 − δ)kat+skt+s+1− 1 − α
θ (1 − Nt+s)Z
ρs
t Nt+s−αkt+sα ,
0 =
(Ztρs+1Nt+s+11−α kt+s+1α + (1 − δ)kt+s+1− akt+s+2
ZtρsNt+s1−αkt+sα + (1 − δ)kt+s− akt+s+1
)η
( 1 − Nt+s
1 − Nt+s+1
)θ(1−η)
,
− βa−η(1 − δ + αZtρs+1Nt+s+11−α kα−1t+s+1), s =0, . . . , T − 1,
kT =kT+1 = k∗.
IV.2 A Small Open Economy
• Since a small open economy model has often many points of steady state, the log-linearized approach in chapter 2 is not suitable to solve the model. The deterministic extended path approach can address this problem. (Omitted)
• To use the log-linearized approach in chapter 2, Schmit-Grohe and Uribe (2003) “Closing small open economy models” shows several approach to stabilize a small open economy model.
http://public.econ.duke.edu/~grohe/research/closing_jie.pdf
7