Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Heer and Maussner PP 28-41

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Chapter 1

Basic Models

Overview. This chapter introduces you to the framework of dy-


namic general equilibrium models. Our presentation serves two
aims: first, we prepare the ground for the algorithms presented in
subsequent chapters that use one out of two possible characteriza-
tions of a model’s solution. Second, we develop standard tools in
model building and model evaluation used throughout the book.
The most basic DGE model is the so called Ramsey model,
where a single consumer-producer chooses an utility maximiz-
ing consumption profile. We begin with the deterministic, finite-
horizon version of this model. The set of first-order conditions
for this problem is a system of non-linear equations that can be
solved with adequate software. Then, we consider the infinite-
horizon version of this model. We characterize its solution along
two lines: the Euler equations provide a set of difference equa-
tions that determine the optimal time path of consumption; dy-
namic programming delivers a policy function that relates the
agent’s choice of current consumption to his stock of capital. Both
characterizations readily extend to the stochastic version of the
infinite-horizon Ramsey model that we introduce in Section 1.3.
In Section 1.4 we add productivity growth and labor supply to
this model. We use this benchmark model in Section 1.5 to il-
lustrate the problems of parameter choice and model evaluation.
Section 1.6 concludes this chapter with a synopsis of the numer-
ical solution techniques presented in Chapters 2 through 6 and
introduces measures to evaluate the goodness of the approximate
solutions.
Readers who already have experience with the stochastic growth
model with endogenous labor supply (our benchmark model) may
4 Chapter 1: Basic Models

consider to skip the first four sections and to start with Section
1.5 to become familiar with our notation and to get an idea of the
methods presented in subsequent chapters.

1.1 The Deterministic Finite-Horizon Ramsey


Model and Non-Linear Programming

1.1.1 The Ramsey Problem

In 1928 Frank Ramsey, a young mathematician, posed the prob-


lem ”How much of its income should a nation save?”1 and devel-
oped a dynamic model to answer this question. Though greatly
praised by Keynes,2 it took almost forty years and further pa-
pers by David Cass (1965), Tjalling Koopmans (1965), and
William Brock and Leonard Mirman (1972) before Ram-
sey’s formulation stimulated macroeconomic theory. Today, vari-
ants of his dynamic optimization problem are the cornerstones of
most models of economic fluctuations and growth.
At the heart of the Ramsey problem there is an economic agent
producing output from labor and capital who must decide how to
split production between consumption and capital accumulation.
In Ramsey’s original formulation, this agent was a fictitious plan-
ning authority. Yet, we may also think of a yeoman growing corn
or of a household, who receives wage income and dividends and
buys stocks.
In the following we use the farmer example to develop a few
basic concepts. Time t is divided into intervals of unit length and
extends from the current period t = 0 to the farmers planning
1
Ramsey (1928), p. 543.
2
Keynes (1930) wrote:
... one of the most remarkable contributions to mathematical eco-
nomics ever made, both in respect of the intrinsic importance and
difficulty of its subject, the power and elegance of the technical meth-
ods employed, and the clear purity of illumination with which the
writer’s mind is felt by the reader to play about its subject.
1.1 Finite-Horizon Ramsey Model and Non-Linear Programming 5

horizon t = T . Kt and Nt denote the amounts of seed and labor


available in period t, respectively. They produce the amount Yt of
corn according to

Yt = F (Nt , Kt ). (1.1)

The production function F has the usual properties:


1. there is no free lunch: 0 = F (0, 0),
2. F is strictly increasing in both of its arguments,
3. concave (i.e. we rule out increasing returns to scale),
4. and twice continuously differentiable.
At each period the farmer must decide how much corn to produce,
to consume and to put aside for future production. The amount
of next period’s seed is the farmer’s future stock of capital Kt+1 .
His choice of consumption Ct and investment Kt+1 is bounded by
current production:

Ct + Kt+1 ≤ Yt .

The farmer does not value leisure but works a given number of
hours N each period and seeks to maximize the utility function

U(C0 , C1 , . . . , CT ).

In the farmer example capital depreciates fully, since seed used for
growing corn is not available for future production. When we think
of capital in terms of machines, factories, or, even more generally,
human knowledge, this is an overly restrictive assumption. More
generally, the resource constraint is given by

Yt + (1 − δ)Kt ≥ Ct + Kt+1 ,

where δ ∈ [0, 1] is the rate of capital depreciation. In the follow-


ing, notation will become a bit simpler if we define the production
function to include any capital left after depreciation and drop the
constant N:

f (Kt ) := F (N, Kt ) + (1 − δ)Kt . (1.2)


6 Chapter 1: Basic Models

Since production without seed is impossible, we assume f (0) = 0,


while the other properties of F carry over to f .
We are now in the position to state the finite-horizon deter-
ministic Ramsey problem formally as follows:

max U(C0 , . . . , CT )
(C0 ,...,CT )

s.t.

Kt+1 + Ct ≤ f (Kt ),  (1.3)
0 ≤ Ct , t = 0, . . . , T,

0 ≤ Kt+1 ,
K0 given.

There is no uncertainty in this problem: the farmer knows in ad-


vance how much corn he will get when he plans to work N hours
and has Kt pounds of seed. Furthermore, he is also sure as to how
he will value a given sequence of consumption {Ct }Tt=0 . Therefore,
we label this problem deterministic. Since we assume T < ∞, this
is a finite-horizon problem.

1.1.2 The Kuhn-Tucker Theorem

Problem (1.3) is a standard non-linear programming problem:


choose an n-dimensional vector x ∈ Rn that maximizes the real-
valued function f (x) in a convex set D determined by l constraints
of the form hi (x) ≥ 0, i = 1, . . . , l. The famous Kuhn-Tucker the-
orem provides a set of necessary and sufficient conditions for a
solution to exist:3

Theorem 1.1.1 (Kuhn-Tucker) Let f be a concave C 1 func-


tion mapping U into R, where U ⊂ Rn is open and convex. For
i = 1, . . . , l, let hi : U → R be concave C 1 functions. Suppose
there is some x̄ ∈ U such that

hi (x̄) > 0, i = 1, . . . , l.
3
See, for instance, Sundaram, 1996, Theorem 7.16, p. 187f.
1.1 Finite-Horizon Ramsey Model and Non-Linear Programming 7

Then x∗ maximizes f over D = {x ∈ U|hi (x) ≥ 0, i = 1, . . . , l} if


and only if there is λ∗ ∈ Rl such that the Kuhn-Tucker first-order
conditions hold:
l
∂f (x∗ )  ∗ ∂hi (x∗ )
+ λi = 0, j = 1, . . . , n,
∂xj i=1
∂xj

λ∗i ≥ 0, i = 1, . . . , l,
λ∗i hi (x∗ ) = 0, i = 1, . . . , l.

It is easy to see that problem (1.3) fits this theorem if the utility
function U and the production function f are strictly concave,
strictly increasing, and twice continuously differentiable. Applying
Theorem 1.1.1 to problem (1.3) provides the following first-order
conditions:4
∂U(C0 , . . . , CT )
0= − λt + µt , t = 0, . . . , T, (1.4a)
∂Ct
0 = −λt + λt+1 f ′ (Kt+1 ) + ωt+1 , t = 0, . . . , T − 1, (1.4b)
0 = −λT + ωT +1 , (1.4c)
0 = λt (f (Kt ) − Ct − Kt+1 ) , t = 0, . . . , T, (1.4d)
0 = µ t Ct , t = 0, . . . , T, (1.4e)
0 = ωt+1 Kt+1 , t = 0, . . . , T, (1.4f)

where λt is the Lagrangean multiplier attached to the resource


constraint of period t,

f (Kt ) − Ct − Kt+1 ≥ 0,

and where µt and ωt+1 are the multipliers related to the non-
negativity constraints on Ct and Kt+1 , respectively. The multipli-
ers value the severeness of the respective constraint. A constraint
4
As usual, a prime denotes the first (two primes the second) derivative of a
function f (x) of one variable x. Condition (1.4c) derives from the budget
constraint of period T , f (KT ) − CT − KT +1 ≥ 0, which has the multiplier
λT , and the non-negativity constraint on KT +1 , which has the multiplier
ωT +1 .
8 Chapter 1: Basic Models

that does not bind has a multiplier of zero. For example, if Ct > 0
then (1.4e) implies µt = 0. If we want to rule out corner solutions,
i.e., solutions where one or more of the non-negativity constraints
bind, we need to impose an additional assumption. In the present
context this assumption has a very intuitive meaning: the farmer
hates to starve to death in any period. Formally, this translates
into the statement
∂U(C0 , . . . , CT )
→ ∞ if Ct → 0 for all t = 0, . . . , T.
∂Ct

This is sufficient to imply Ct > 0 for all t = 1, . . . , T , µt = 0


(from (1.4e)), and the Lagrangean multipliers λt equal the mar-
ginal utility of consumption in period t and, thus, are also strictly
positive:

∂U(C0 , . . . , Ct )
= λt .
∂Ct

Condition (1.4d), thus, implies that the resource constraints al-


ways bind. Furthermore, since we have assumed f (0) = 0, positive
consumption also requires positive amounts of seed Kt > 0 from
period t = 0 through period T . However, the farmer will consume
his entire crop in the last period of his life, since any seed left
reduces his lifetime utility. More formally, this result is implied
by equations (1.4f) and (1.4c), which yield λT KT +1 = 0. Taking
all pieces together, we arrive at the following characterization of
an optimal solution:

Kt+1 = f (Kt ) − Ct , (1.5a)


∂U(C0 , . . . CT )/∂Ct
= f ′ (Kt+1 ). (1.5b)
∂U(C0 , . . . CT )/∂Ct+1

The lhs of equation (1.5b) is the marginal rate of substitution


between consumption in two adjacent periods. It gives the rate
at which the farmer is willing to forego consumption in t for con-
sumption one period ahead. The rhs provides the compensation
for an additional unit of savings: the increase in future output.
1.2 Infinite-Horizon Ramsey Model and Dynamic Programming 9

1.2 The Deterministic Infinite-Horizon Ramsey


Model and Dynamic Programming
In equation (1.5b) the marginal rate of substitution between two
adjacent periods depends on the entire time profile of consump-
tion. For this reason, we must solve the system of 2T − 1 non-
linear, simultaneous equations (1.5) at once to obtain the time
profile of consumption. Though probably difficult in practice this
is, in principle, a viable strategy as long as T is finite. However,
if we consider an economy with indefinite final period, that is, if
T approaches infinity, this is no longer feasible. We cannot solve
for infinitely many variables at once. To circumvent this prob-
lem, we restrict the class of intertemporal optimization problems
to problems that have a recursive structure. Recursive problems
pose themselves every period in the same, unchanged way. Their
solution is not a time profile of optimal decisions determined at
an arbitrary initial period t = 0 but consists in decision rules that
determine the agent’s behavior at each future point in time. The
time additive separable (TAS) utility function, which we introduce
in the next subsection, allows for a recursive formulation of the
Ramsey problem. For this problem we derive first-order conditions
via the Kuhn-Tucker method in Subsection 1.2.2. There is, how-
ever, an alternative approach available: dynamic programming,
which we consider in Subsection 1.2.3. Subsection 1.2.4 provides
a characterization of the dynamics of the infinite-horizon Ramsey
model. We close this section with a brief digression that considers
the few models that admit an analytic solution of the Ramsey
problem.

1.2.1 Recursive Utility

The TAS utility function is defined recursively from


Ut = u(Ct ) + βUt+1 , β ∈ (0, 1). (1.6)
In this definition β is a discount factor and β −1 − 1 is known as
the pure rate of time preference. The function u : [0, ∞) → R
10 Chapter 1: Basic Models

is called the one-period, current-period, or felicity function. We


assume that u is strictly increasing, strictly concave and twice
continuously differentiable.
The solution to the finite-horizon Ramsey model depends upon
the chosen terminal date T . Yet, in as far as we want to portray
the behavior of the economy with Ramsey type models there is
no natural final date T . As a consequence, most models extend
the planning horizon into the indefinite future by letting T → ∞.
Iterating on (1.6) we arrive at the following definition of the utility
function


Ut = β s u(Ct+s ). (1.7)
s=0

If we want to rank consumption streams according to this crite-


rion function, we must ensure that the sum on the rhs is bounded
from above, i.e., Ut < ∞ for every admissible sequence of points
Ct , Ct+1 , Ct+2 , . . . . This will hold, if the growth factor of one-
period utility u, gu := u(Ct+s+1 )/u(Ct+s ), is smaller than 1/β
for all s = 0, 1, 2, . . . . Consider the Ramsey problem (1.3) with
infinite time horizon:


max U0 = β t u(Ct )
C0 ,C1 ,...
t=0
s.t.
 (1.8)
Kt+1 + Ct ≤ f (Kt ), 
0 ≤ Ct , t = 0, 1, . . . ,

0 ≤ Kt+1 ,
K0 given.
In this model we do not need to assume that the one-period util-
ity function u is bounded. Since u is continuous, it is sufficient
to assume that the economy’s resources are finite. In a dynamic
context this requires that there is an upper bound on capital ac-
cumulation, i.e., there is K̄ such that for each K > K̄ output is
smaller than needed to maintain K:
∃K̄ so that ∀Kt > K̄ ⇒ Kt+1 < Kt . (1.9)
1.2 Infinite-Horizon Ramsey Model and Dynamic Programming 11

For instance, let f (K) = K α , α ∈ (0, 1). Then:


K ≤ K α ⇒ K̄ = 11/(α−1) = 1.
Given condition (1.9), any admissible sequence of capital stocks is
bounded by K max := max{K̄, K0 } and consumption in any period
cannot exceed f (K max ). Figure 1.1 makes that obvious: consider
any point to the left of K̄ such as K1 and assume that consumption
equals zero in all periods. Then, the sequence of capital stocks
originating in K1 approaches K̄. Similarly, the sequence starting
in K2 approaches K̄ from the right.

f (K)

45o
K
K1 K̄ K2
Figure 1.1: Boundedness of the Capital Stock

1.2.2 Euler Equations

There are two approaches to characterize the solution to the Ram-


sey problem (1.8). The first is an extension of the Kuhn-Tucker
method5 and the second is dynamic programming.6 According
5
See, e.g., Chow (1997), Chapter 2 and Romer (1991).
6
Here, the standard reference is Chapter 4 of Stokey and Lucas with
Prescott (1989).
12 Chapter 1: Basic Models

to the first approach necessary conditions may be derived from


maximizing the following Lagrangean function with respect to
C0 , C1 , . . . , K1 , K2 , . . . :

 
L = β t u(Ct ) + λt (f (Kt ) − Ct − Kt+1 )
t=0

+ µt Ct + ωt+1 Kt+1 .

Note that in this expression the Lagrangean multipliers λt , µt ,


and ωt+1 refer to period t values. Period t = 0 values are given by
β t λt , β t µt , and β t ωt+1 . The first-order conditions for maximizing
L are given by:

u′ (Ct ) = λt − µt , (1.10a)
λt = βλt+1 f ′ (Kt+1 ) + ωt+1 , (1.10b)
0 = λt (f (Kt ) − Ct − Kt+1 ), (1.10c)
0 = µ t Ct , (1.10d)
0 = ωt+1 Kt+1 . (1.10e)

We continue to assume that the farmer hates starving to death,

lim u′ (C) = ∞, (1.11)


C→0

so that the non-negativity constraints never bind. Since u is


strictly increasing in its argument, the resource constraint always
binds. Therefore, we can reduce the first-order conditions to a
second order difference equation in the capital stock:
u′ (f (Kt ) − Kt+1 )
− βf ′ (Kt+1 ) = 0. (1.12)
u′ (f (Kt+1 ) − Kt+2 )
This equation is often referred to as the Euler equation, since
the mathematician Leonhard Euler (1707-1783) first derived it
from a continuous time dynamic optimization problem. To find
the unique optimal time path of capital from the solution to this
functional equation we need two additional conditions. The period
t = 0 stock of capital K0 provides the first condition. The second
1.2 Infinite-Horizon Ramsey Model and Dynamic Programming 13

condition is the so called transversality condition, which is the


limit of the terminal condition λT KT +1 = 0 from the finite-horizon
Ramsey problem (1.3). It requires

lim β t λt Kt+1 = 0, (1.13)


t→∞

that is, the present value of the terminal capital stock must ap-
proach zero. In the Ramsey model (1.8), condition (1.13), is a
necessary condition,7 as well as conditions (1.10).

1.2.3 Dynamic Programming

We now turn to a recursive formulation of the Ramsey problem.


For this purpose we assume that we already know the solution
(denoted by a star) {K1∗ , K2∗ , . . . } ≡ {Kt∗ }∞
t=1 so that we are able
to compute the life-time utility from


v(K0 ) := u(f (K0 ) − K1∗ ) + β t u(f (Kt∗ ) − Kt+1

).
t=1

Obviously, the maximum value of life-time utility v(K0 ) depends


upon K0 directly – via the first term on the rhs of the previ-
ous equation – and indirectly via the effect of K0 on the opti-
mal sequence {Kt∗ }∞t=1 . Before we further develop this approach,
we will adopt the notation that is common in dynamic program-
ming. Since K0 is an arbitrary initial stock of capital, we drop the
time subscript and use K to designate this variable. Furthermore,
we use a prime for all next-period variables. We are then able to
define the function v recursively via:

v(K) := max

u(f (K) − K ′ ) + βv(K ′). (1.14)
0≤K ≤f (K)

The first term to the right of the max operator is the utility of
consumption C = f (K) − K ′ as a function of the next-period
capital stock K ′ . The second term is the discounted optimal value
7
See Kamihigashi (2002).
14 Chapter 1: Basic Models

of life-time utility, if the sequence of optimal capital stocks starts


in the next period with K ′ . Suppose we know the function v so
that we can solve the optimization problem on the rhs of equation
(1.14). Obviously, its solution K ′ depends upon the given value of
K so that we may write K ′ = h(K). The function h is the agent’s
decision rule or policy function. Note that the problem does not
change with the passage of time: when the next period has arrived,
the agent’s initial stock of capital is K = K ′ and he has to make
the same decision with respect to the capital stock of period t = 2,
which we denote by K ′′ . In this way he can determine the entire
sequence {Kt∗ }∞ t=1 .
Yet, we may also view equation (1.14) as an implicit defini-
tion of the real-valued function v and the associated function h.
From this perspective, it is a functional equation,8 named Bell-
man equation after its discoverer the US mathematician Richard
Bellman (1920-1984). His principle of optimality states that the
solution of problem (1.8) is equivalent to the solution of the Bell-
man equation (1.14). Stokey and Lucas with Prescott(1989),
pp. 67-77, establish the conditions for this equivalence to hold. In
this context of dynamic programming v is referred to as the value
function and h as the policy function, decision rule, or feed-back
rule. Both functions are time invariant. The mathematical theory
of dynamic programming deals with the existence, the properties,
and the construction of v and h. Given that both u(C) and f (K)
are strictly increasing, strictly concave and twice continuously dif-
ferentiable functions of their respective arguments C and K, and
that there exists a maximum sustainable capital stock K̄ as de-
fined in (1.9), one can prove the following results:9
1. The function v exists, is differentiable, strictly increasing, and
strictly concave.
2. The policy function g is increasing and differentiable.
3. The function v is the limit of the following sequence of steps
s = 0, 1, . . . :
8
As explained in Section 12.1, a functional equation is an equation whose
unknown is a function and not a point in Rn .
9
See, e.g., Harris (1987), pp. 34-45 or Stokey and Lucas with Prescott
(1989), pp. 103-105.
1.2 Infinite-Horizon Ramsey Model and Dynamic Programming 15

v s+1 (K) = max u(f (K) − K ′ ) + βv s (K ′ ),


0≤K ′ ≤f (K)

with v 0 = 0.
We illustrate these findings in Example 1.2.1

Example 1.2.1
Let the one-period utility function u and the production function f
be given by

u(C) := ln C,
f (K) := K α , α ∈ (0, 1),

respectively. In Appendix 1 we use iterations over the value function


to demonstrate that the policy function Kt+1 = h(Kt ) that solves the
Ramsey problem (1.8) is given by

Kt+1 = αβKtα .

Furthermore, the value function is linear in ln K and given by

v(K) = a + b ln K,

1 αβ α
a := ln(1 − αβ) + ln αβ , b := .
1−β 1 − αβ 1 − αβ

The dynamic programming approach also provides the first-order


conditions (1.12). It requires two steps to arrive at this result.
First, consider the first-order condition for the maximization prob-
lem on the rhs of equation (1.14):

u′ (f (K) − K ′ ) = βv ′(K ′ ). (1.15)

Comparing this with condition (1.10a) (assuming µt = 0) reveals


that the Lagrange multiplier λt ≡ βv ′ (Kt+1 ) is a shadow price for
newly produced capital (or investment expenditures): it equals
the current value of the increase in life-time utility obtained from
an additional unit of capital. Second, let K ′ = h(K) denote the
solution of this implicit equation in K ′ . This allows us to write
the Bellman equation (1.14) as an identity,
16 Chapter 1: Basic Models

v(K) = u(f (K) − h(K)) + βv(h(K)),

so that we can differentiate with respect to K on both sides. This


yields

v ′ (K) = u′ (C) (f ′ (K) − h′ (K)) + βv ′(K ′ )h′ (K),

where C = f (K) − h(K). Using the first-order condition (1.15)


provides

v ′ (K) = u′ (C)f ′ (K). (1.16)

Since K is an arbitrarily given stock of capital, this equation re-


lates the derivative of the value function v ′ (·) to the derivative
of the one-period utility function u′ (·) and the derivative of the
(net) production function f ′ (·) for any value of K. Thus, letting
C ′ = f (K ′ )−K ′′ denote next period’s consumption, we may write

v ′ (K ′ ) = u′ (C ′ )f ′ (K ′ ).

Replacing v ′ (K ′ ) in (1.15) by the rhs of this equation yields


u′ (f (K ′ ) − K ′′ ) ′ ′
1=β f (K ).
u′(f (K) − K ′ )
This equation must hold for any three consecutive stocks of capital
(K, K ′ , K ′′ ) that establish the optimal sequence {Kt∗ }∞
t=1 that
solves the Ramsey problem (1.8). Thus, it is identical to the Euler
equation (1.12), except that we used primes instead of the time
indices.

1.2.4 The Saddle Path

To gain insights into the dynamics of the Ramsey model (1.8) we


use the phase diagram technique to characterize the solution of the
Euler equation (1.12). Substituting the resource constraint Ct =
f (Kt ) − Kt+1 into (1.12) yields a first-order, non-linear system of
difference equations that governs the optimal time path of capital
accumulation:

You might also like