Scribd
Upload
17 views3 pages

Perturbation Basics

Uploaded by

resat akan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
17 views3 pages

Perturbation Basics

Uploaded by

resat akan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

ECON 6305: Notes On The Numerical Method Used By

Dolo
Aaron Betz
George Washington University

1 Introduction
Recall the simple neoclassical growth model (exogenous labor and no TFP) that we derived in
class. It contains two equations to determine consumption and the evolution of capital. The
first equation is the Euler equation which determines how the household plans consumption
across their infinite future.

u0 (ct+1 )
β −1 = (1 + rk|t − δ) (1)
u0 (ct )
Where rk|t is simply the marginal product of capital in the production function.
 ¯ 1−α
lt
rk|t =α (2)
kt−1

The second equation determines the evolution of capital.


α ¯ 1−α
kt = (1 − δ)kt−1 + kt−1 lt − ct (3)

This document shows how we solve the model recursively. That is, we assume a con-
tinuous solution for control and contemporaneous state variables that only depends on the
previous state variable of the model, kt−1 .

1.1 Formulating a control state model


Given that the growth model has one state, kt−1 , it is possible to express the structural
equations of the model (first order conditions and accounting identities) as equations of
the control function, g(kt−1 ), and the transition rule (or function), h(kt−1 ). For our simple
growth model, with exogenous labor and no TFP, these functions determine consumption
and contemporaneous capital.
ct = g(kt−1 ) (4)

kt = h(kt−1 ) (5)

1
The transition rule, 3, can then be expressed as follows.
α ¯ 1−α
h(kt−1 ) = (1 − δ)kt−1 + kt−1 lt − g(kt−1 ) (6)

And the Euler equation from equation 1 becomes the following.


u0 (g(h(kt−1 )))
β −1 = (1 + rk|t − δ) (7)
u0 (g(kt−1 )
Where g(h(kt−1 ) is simply ct+1 . Equations 6 and 7 now include the functions g(·) and
h(·). But ultimately, these two equations are only functions of the state variable, kt−1 , and
exogenous variables and parameters1 .

2 A Vector Valued Function


2.1 A steady state
Assume h(·) has a fixed point or steady state where the following condition holds.

k ∗ = h(k ∗ ) (8)

It is trivial to show that c∗ exists and is also in steady state when kt = k∗.

2.2 Implicit differentiation at the steady state


Now imagine a vector valued function of our two equations (6 and 7) that equates to the zero
vector. If we can solve for a steady-state we know one point that exists along this function.
" #  
∗ ∗ ∗ α ¯ 1−α ∗
−h(k ) + (1 − δ)k + (k ) l t − g(k ) 0
F (k ∗ , g(k ∗ ), h(k ∗ ), g(h(k ∗ ))) = −1 ∗ u0 (g(h(k∗ ))) = (9)
−β + (1 + rk|t − δ) u0 (g(k∗ ) 0

By construction, our function F should still be zero if we evaluate it at k ∗ + .2 Where  is


a small or even a large number.3
 
∗ ∗ ∗ ∗ 0
F (k + , g(k + ), h(k + ), g(h(k + ))) = (10)
0
Therefore, within some neighborhood of k ∗ , F should always evaluate to zero. If F is
unchanging from zero then its derivative with respect to k ∗ should also be zero.4 Taking the
derivative of F with respect to k ∗ , abstractly, yields the following equations.
 
0 ∗ 0 ∗ 0 ∗ 0 ∗ 0
F1 (·) + F2 (·)g (k ) + F3 (·)h (k ) + F4 (·)g (k )h (k ) = (11)
0
1
Recall that rk|t is only a function of kt−1 , α, and l¯t
2
If a solution exists then for a given stock of capital, kt−1 , the choices of the household, h(kt−1 ) and
g(kt−1 ) should still satisfy the Euler equation and transition rule.
3
Existence of a solution is a topic for a later graduate course.
4
This should be true of higher derivatives too.

2
Recall that F is a vector valued function so equation 11 actually represents two differential
equations that equal zero. Notice that we now have two equations (equating the derivate
of the two rows of F to zero )and two unknowns g 0 (k ∗ ) and h0 (k ∗ ). Solving for these two
unknowns gives us enough to linearly approximate g(·) and h(·) in the following manner.

h(kt ) ≈ h(k ∗ ) + h0 (k ∗ )(kt − k ∗ ) (12)

g(kt ) ≈ g(k ∗ ) + g 0 (k ∗ )(kt − k ∗ ) (13)


Dolo solves any given control state model in this manner. With multiple states (or shocks)
the decision rules (g and h) now have multiple arguments. The problem then generalizes to
solving a system of multiple derivatives for each decision rule.

You might also like