Lecture Notes on Dynamic Macroeconomics

Emilio Espino

July 16, 2021

ii
Contents

Overview and Summary vii

1 Mathematical Preliminaries: An Overview 1

1.1 Basic Concepts for Convex Analysis . . . . . . . . . . . . . . . 1
1.2 Finite Dimensional Programming Problems . . . . . . . . . . . 12
1.3 Uncertainty and Expected Utility (NEW) . . . . . . . . . . . 17
1.3.1 Stochastic processes . . . . . . . . . . . . . . . . . . . 17
1.3.2 Expected Utility . . . . . . . . . . . . . . . . . . . . . 18
1.4 Infinite Dimensional Programming Problems . . . . . . . . . . 18
1.4.1 Sufficiency of Euler Equations and Transversality Con-
ditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.2 On the Necessity of the Transversality Condition . . . 22

2 Optimality and Competitive Equilibrium 25

2.1 The Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Arrow-Debreu Economies . . . . . . . . . . . . . . . . . . . . . 28
2.3 Sequential Markets Economies . . . . . . . . . . . . . . . . . . 34
2.4 Equivalence of Market Structures . . . . . . . . . . . . . . . . 36
2.5 Application: Economies with Government and Ricardian Equiv-
alence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 The One-Sector Neoclassical Growth Model 45

3.1 The Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 PO Allocations: The Planner’s Problem . . . . . . . . . . . . 47
3.3 Competitive Decentralization:
An Arrow-Debreu Framework . . . . . . . . . . . . . . . . . . 49
3.4 Welfare Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.1 The First Welfare Theorem . . . . . . . . . . . . . . . 52

iii
iv CONTENTS

3.4.2 The Second Welfare Theorem . . . . . . . . . . . . . . 52

3.4.3 Existence: On The Negishi’s Approach . . . . . . . . . 55
3.5 On the Dynamics of the Deterministic Version . . . . . . . . . 57
3.6 Recursive Formulation: A Heuristic Introduction . . . . . . . . 62

4 Asset Pricing and Trading 71

4.1 The Lucas Tree Model with Heterogenous Agents . . . . . . . 71
4.2 Competitive Sequential Trading
with Dinamically Complete Markets . . . . . . . . . . . . . . . 72
4.3 Equilibrium Asset Prices . . . . . . . . . . . . . . . . . . . . . 77
4.3.1 Short-Lived Assets . . . . . . . . . . . . . . . . . . . . 77
4.3.2 Long-Lived Asset: Pricing the Firm . . . . . . . . . . . 77
4.3.3 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.4 The Risk-Free Interest Rate, the Term Structure and
the Yield Curve . . . . . . . . . . . . . . . . . . . . . . 79
4.4 The Equity Premium Puzzle . . . . . . . . . . . . . . . . . . . 84
4.4.1 The Market Price of Risk and Hansen-Jagannathan
Bounds (incomplete) . . . . . . . . . . . . . . . . . . . 87
4.5 Aggregation and the Distribution of Wealth . . . . . . . . . . 89
4.6 Firm’s Financial Policy and the Modigliani-Miller Theorem . . 92
4.7 Arbitrage and State Prices . . . . . . . . . . . . . . . . . . . . 94

5 Consumption and Savings 105

5.1 Full Risk-Sharing with Complete markets . . . . . . . . . . . . 105
5.1.1 Consumption Dynamics I:
Different Discount Factors . . . . . . . . . . . . . . . . 106
5.1.2 Consumption Dynamics II:
Identical Discount Factors . . . . . . . . . . . . . . . . 107
5.2 The Permanent Income Hypothesis (PIH) . . . . . . . . . . . 109
5.2.1 Quadratic Utility: Certainty Equivalence and Consump-
tion as a Random Walk. . . . . . . . . . . . . . . . . . 110
5.2.2 Quadratic Utility:
Wealth Dynamics and Borrowing Constraints . . . . . 116
5.3 Precautionary Savings: Prudence and Borrowing Constraints . 118
5.3.1 Prudence: Warming Up in A Two-Period Model . . . . 119
5.3.2 Prudence: A Basic Multiperiod Model . . . . . . . . . 122
5.3.3 Borrowing Constraints . . . . . . . . . . . . . . . . . . 123
5.3.4 A Natural Debt Limit . . . . . . . . . . . . . . . . . . 125
CONTENTS v

6 Overlapping Generation Economies 129

6.1 Pure Exchange OLG Economies . . . . . . . . . . . . . . . . . 130
6.1.1 Competitive Equilibrium . . . . . . . . . . . . . . . . . 130
6.1.2 First Welfare Theorem in OLG Economies . . . . . . . 134
6.2 OLG Economies with Production . . . . . . . . . . . . . . . . 137
6.2.1 Dynamics and Steady State . . . . . . . . . . . . . . . 140
vi CONTENTS
Overview and Summary

These notes were developed as part of the first year Graduate Macroeco-
nomic class. The main goal is to provide methodological tools for studying
macroeconomic questions.
The emphasis is on macroeconomic theory with microfoundations, i.e.,
the economy’s aggregate behavior is the result of the interaction of fully
rational economic agents (households, firms, the government, etc.) taking
as given the primitives describing the economy; i.e., consumers’ preferences,
production technologies, information, and so on.
We will concentrate on what I consider the three main frameworks used
in modern macroeconomics. First, we consider a dynamic pure exchange
economy under uncertainty. This has been the natural setting to study dif-
ferent properties of asset markets. Second, we study the celebrated neoclas-
sical growth model, the main setting used to analyze aggregate quantities in
production economies (output, consumption, investment, etc.). Finally, we
analyze a prototype overlapping generation model, which has been useful to
study different issues concerning distributional aspects.
These notes have benefited from the superb graduate students that have
worked as TA in the corresponding classes; in order: Hernan Seone, Manuel
Macera, David Kohn, Martin Beraja, Nicolas Caramp, Julian Kozlowski,
Andrés Sarto, Samuel Kaplan, Giselle Montamat, Fernando Cirelli, Alvaro
Boitier. Unfortunately, these notes are still too underground, but I plan to
revise them in the near future. Naturally, comments are welcome.
Please do NOT distribute without permission!
© by Emilio Espino.

vii
viii PREFACE
Chapter 1

Mathematical Preliminaries:
An Overview

1.1 Basic Concepts for Convex Analysis

Let S be a set of elements with two operations (+, ·) (addition and scalar
multiplication)

Definition 1 (S, +, ·) is a real vector space if it is closed under addition

and scalar multiplication (i.e. for all x, y ∈ S and all α ∈ R, x + y ∈ S
and α · x ∈ S). Additionally, the following algebraic laws must be satisfied:
∀x, y, z ∈ S and ∀α, 1, 0 ∈ R
a. Commutativity: x + y = y + x.
b. Associativity: (x + y) + z = x + (y + z) .
c. Distributivity: (x + y) · α = α · x + α · y.
d. ∃θ ∈ S such that θ + x = x and 0 · x = θ.
e. 1 · x = x

To get a notion of distance between elements of a set, we have the follow-

ing definition.

Definition 2 A metric space is a set S together with a metric ρ : S ×S →

R such that ∀x, y, z ∈ S :
a. ρ (x, y) ≥ 0 with equality if and only if x = y.
b. ρ (x, y) = ρ (y, x) .
c. Triangular inequality: ρ (x, z) ≤ ρ (x, y) + ρ (y, z) .

1
2 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Definition 3 A normed vector space is a real vector space S together

with a norm (S, k·k) where k·k : S → R such that ∀x, y ∈ S, α ∈ R :

a. kxk ≥ 0 (= 0 iff x = θ).

b.kαxk = |α| kxk .
c.kx + yk ≤ kxk + kyk (Triangular inequality).

Example 1 Let S = RN and consider some examples of norms on RN :

N
!1/2 N
X X
kxk = (xn )2 , kxk = |xn | , kxk = max |xn | .
n
n=1 n=1

The next result shows how to obtain a natural metric in a normed vector
space.

Claim 1 Let (S, k·k) be a
normed vector space. Let ρk·k (x, y) = kx − yk for
all x, y ∈ S. Thus S, ρk·k defines a metric space

Proof. a. ρk·k (x, y) ≥ 0 =0 if f x − y = θ.

b. ρk·k (x, y) = kx + (−y)k = k(−y) + xk = k(−1) (y − x)k = |−1| ky − xk =
ρk·k (y, x) .
c.ρk·k (x, y) = kx − yk = k(x − z) + (z − y)k ≤ kx − zk + kz − yk =
ρk·k (x, z) + ρk·k (z, y) .

Let N = {0, 1, ...} be the set of natural numbers. An infinite sequence of

elements of S (denoted {xn }∞ n=0 ⊂ S) is a mapping x : N → S; that is xn ∈ S
for all n = 0, 1, ...

Definition 4 A sequence {xn }∞ n=0 in a metric space (S, ρ) converges to

x ∈ S if ∀ > 0, there exists some N such that ρ (xn , x) <  ∀n ≥ N .

We use the conventional notation to denote convergence: xn → x, lim

n
xn = x, where x is denoted the limit point.

Remark 1 xn → x if and only if yn = ρ (xn , x) → 0 where {yn } is a sequence

in R+ .
1.1. BASIC CONCEPTS FOR CONVEX ANALYSIS 3

The following concept is very useful when the candidate limit point is not
immediately available.

Definition 5 A sequence {xn }∞ n=0 in S is Cauchy if for each  > 0, there

exists some N such that ρ (xn , xm ) <  for all n, m ≥ N .

In general, N means that the election of N depends on .

Exercise 1 (Limit points are unique) If xn → x and xn → y, then x = y.

Proof. We prove this statement by contradiction. Suppose that xn → x,

xn → y but x 6= y. Then, it follows that ρ (x, y) = δ > 0 by definition of ρ.
Also, there exist (Nx , Ny ) such that:
δ
ρ (xn , x) < for all n ≥ Nx ,
2
δ
ρ (xn , y) < for all n ≥ Ny .
2
But it follows by triangular inequality that:
δ δ
ρ (x, y) ≤ ρ (xn , x) + ρ (xn , y) < + =δ
2 2
for all n ≥ N = max(Nx , Ny ). Consequently, we get the desired contradiction.

Definition 6 Let {xn }∞ n=0 be a sequence in S and nk be increasing in k. We

∞
say that {xnk }k=0 is a subsequence of {xn } .

Claim 2 xn → x if and only if every subsequence at {xn } converges to x.

Proof. (If) Take {xnk } = {xn } .

(Only if) Suppose that xn → x but there is a subsequence not converging
to x. This means that there exists some  such that for all N there is some
nN such that ρ (xnN , x) ≥ . This contradicts that xn → x.
In some cases, we do not need to know x itself but whether a limit point
exists or not.

Definition 7 A metric space (S, ρ) is complete if every Cauchy sequence

converges.
4 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Thus, in complete metric spaces, checking that a sequence is Cauchy

implies convergence.

Important Fact. If ρ (x, y) = |x − y| for all x, y ∈ R, then (R, ρ) is a

complete metric space.

Let F : S → Y be a function where (Y, ρy ) is a metric space.

Definition 8 F is continuous at x ∈ S if for all (xn ) in S such that xn → x,

then it follows that (yn ) in Y such that yn = F (xn ) satisfies yn → y = F (x)

We say that F is real-valued if Y ⊆ Rk . F is continuous over S if F is

continuous for all x ∈ S.

Let A ⊂ S, where (S, ρ) is a metric space and ρ is induced by some norm

k·k on S.

Definition 9 A is closed in S if for all {xn }∞n=0 ⊂ A such that xn → x ∈ S,

it follows that x ∈ A (A includes its limit points).
A is open in S if for all x ∈ A, there exists  > 0 such that B (x) ≡
{y ∈ S : ρ (x, y) < } ⊂ A.
A is bounded if there exists some B < ∞ such that kxk ≤ B for all
x ∈ S.
A is compact if it is closed and bounded.
A is sequentially compact if any {xn }∞ n=0 ⊂ A has a convergent sub-
sequence (i.e. ∃ {xnk } such that xnk → x ∈ A).

Remark 2 It can be shown that compact and sequentially compact are

equivalent definitions in metric spaces. That is, A ⊂ S, is compact if and
only if compact sequentially.

Below we prove one of these implications.

Claim 3 Let S ⊂ Rn , with ρ (x, y) = kx − yk for all x, y ∈ S. If A ⊂ S, is

sequentially compact, then it is closed and bounded.
1.1. BASIC CONCEPTS FOR CONVEX ANALYSIS 5

Figure 1.1: Open Set

Proof. (Closed) Let {xn }∞n=0 ⊂ A and xn → x ∈ S. Then, any subsequence

xnk → x. Since, A is sequentially compact, x ∈ A.
(Bounded) Suppose that A is unbounded. Then, for all k there exists
some xk ∈ A such that kxk k > k. But then kxn k → ∞ and this sequence
{xk }∞
k=0 cannot have a convergent subsequence.

Convexity is an important concept in economic theory. We say that

A ⊂ S is a convex set if ∀α ∈ [0, 1] and ∀x, y ∈ A :

αx + (1 − α) y ∈ A.

x ∈ A is an interior point if ∃ > 0 such that B (x) ⊂ A. We denote

intA as the set of interior points of A.
x is in the boundary point of A if ∀ > 0,

B (x) ∩ A 6= ∅ and B (x) ∩ Ac 6= ∅.

N
 p ∈NR / {0} and c ∈ R, the hyperplane generated by (p, c) is
Given
Hp,c = x ∈ R : p · x = c .

Theorem 1 (Separating Hyperplane Theorem) Suppose that A ⊂ RN is con-

/ A. Then, there exists some p ∈ RN / {0} and c ∈ R
vex and closed and x ∈
such that

p · y < c < p · x for all y ∈ A.

More generally, suppose that A, B ⊂ RN are convex and disjoint. Then,
there exists some p ∈ RN / {0} and c ∈ R such that p · y ≤ c ≤ p · x for all
x ∈ A, y ∈ B.
6 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Proof. (See Mas Colell et al. (1995, p.948))

Let δ = inf{||x − y|| : x ∈ C, y ∈ D}
We argue that there exists (c, d) such that δ = ||c − d||.
Since C, D are non-empty, ∃(x̂, ŷ) ∈ C × D such that δ ≤ ||x̂ − ŷ||.
Let Ŝ = Ĉ × D̂ = {(x, y) ∈ C × D : ||x − y|| ≤ ||x̂ − ŷ||}, where Ĉ = {x ∈
C : ||x − y|| ≤ ||x̂ − y||} and D̂ = {y ∈ D : ||x − y|| ≤ ||x − ŷ||}.
We now show that Ŝ is compact. To do that, we show that it is bounded and
closed.

(i) Bounded:

For x ∈ Ĉ, ||x|| ≤ ||y|| + ||y − x||, by the triangle inequality

||x|| ≤ ||y|| + ||y − x̂||, by the definition of Ĉ

This means that Ĉ is bounded. It can be analogously proved that D̂

is bounded. Since, Ŝ is the cartesian product of Ĉ and D̂, it is also
bounded.

(ii) Closed:
Suppose {xn , yn } ∈ Ŝ and {xn , yn } → (x, y) ∈ C × D.
Now, ||xn − yn || ≤ ||x̂ − ŷ|| → ||x − y|| ≤ ||x̂ − ŷ|| since weak inequalities
are preserved by limits.
So (x, y) ∈ Ŝ.

Since ||.|| is continuous and Ŝ is compact, we can apply Weierstrass’ Theo-

rem:
∃(c, d) ∈ C × D with ||c − d|| = min||x − y|| such that δ = ||c − d|| > 0 ,
where δ > 0 since c 6= d (disjoint).
1.1. BASIC CONCEPTS FOR CONVEX ANALYSIS 7

D̂ ŷ

x̂ Ĉ

Figure 1.2: Ŝ is compact

Now define:

a=d−c
||d||2 − ||c||2
b=
2

a 6= 0 since δ > 0.

We argue that:

a · x < b ∀x ∈ C
a · y > b ∀y ∈ D
8 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

c
C

Figure 1.3: An illustration of the Separating Hyperplane Theorem

Remember that (see exercises):

||x − y||2 + ||x + y||2 = 2||x||2 + 2||y||2

(x − y) · (x + y) = ||x||2 − ||y||2

Note that
 
c+d (d + c)
a· = (d − c) ·
2 2
2 2
||d|| − ||c||
=
2
=b

We show that a · y > b ∀y ∈ D. Showing that a · x < b ∀x ∈ C is identical.

We will prove it by contradiction.
Let ŷ ∈ D and suppose that a · ŷ ≤ b.
Define

g(y) = ||y − c||2

Then,
∇g(y) = 2(y − c)
1.1. BASIC CONCEPTS FOR CONVEX ANALYSIS 9

We claim that (ŷ−d) is a descent direction for g at d (it lowers from ||d−c||).
Recall that for a given point z ∈ Rn and a function f , we say that a direction
t ∈ Rn is a descent direction if there exists ᾱ > 0 such that f (x + αt) <
f (x) ∀α ∈ (0, ᾱ), with α ∈ R.
In order to show that (ŷ − d) is a descent direction for y at d, we will show
that the directional derivative of g at d in the direction (ŷ − d) is negative:

∇g(d) · (ŷ − d) = 2(d − c) · (ŷ − d)

= 2 (d − c) · ŷ − ||d||2 + c · d
 

||d||2 − ||c||2
 
2
≤2 − ||d|| + c · d
2
= − ||d||2 + ||c||2 − 2 c · d
 

= −||d − c||2 < 0

Where the inequality follows from the fact that a · ŷ ≤ b.

Having shown that (ŷ − d) is a descent direction, this means that ∃ α̂ such
that ∀α ∈ (0, α̂),

g (d + α(ŷ − d)) < g(d)

||d + α(ŷ − d) − c||2 < ||d − c||2 = δ 2

But this implies that the point d + α(ŷ − d) is closer to c than d, which
contradicts that (c, d) were the minimizers.

Theorem 2 (Supporting Hyperplane Theorem) Suppose that A ⊂ RN is con-

/ intA. Then, there exists some p ∈ RN / {0} such that:
vex, x ∈

p·y ≤p·x ∀y ∈ A.

Proof. x ∈ ∂(A), where ∂(A) is the boundary of set A. Hence, B1/n (x) ∩
AC 6= ∅.
Take any xn . It can be constructed such that xn → x, xn ∈ int(AC ).
Hence, by the Separating Hyperplane Theorem, there exist (pn , cn ) such that:

pn · xn > cn > pn · y ∀y ∈ A
pn
Moreover, p̂n = also satisfies this for all n (Check).
||pn ||
10 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Figure 1.4: Supporting Hyperplane

Then {p̂n } with ||p̂n || = 1 is in a compact set, and {cn } also.

Hence, it has a convergent subsequence: {p̂n , cn } → (p, c)
Therefore, taking the limit and since weak inequalities are preserved by limits:

p·x≥c≥p·y ∀y ∈ A

Let A ⊂ R and denote ā as an upper bound of A if ā ≥ x, ∀x ∈ A We say

that ā is the least upper bound (the supremum) of A if ā ≥ x, ∀x ∈ A and
@x ∈ A such that x > ā − . Similarly, we say that a is the greatest lower
bound (the infimum) of A if a ≤ x, ∀x ∈ A and @e a ≤ x, ∀x ∈ A.
a > ā, e

Theorem 3 (Weierstrass’ Theorem) Let F : C → R be continuous and

C ⊂ RN compact. Then, there exist x, x ∈ C such that

F (x) ≤ F (x) ≤ F (x), ∀x ∈ C

That is, F (x) = maxF (x) = supF (x) (the global max is attained)
x∈C x∈C
F (x) = minF (x) = inf F (x) (the global min is attained)
x∈C x∈C

Proof. Simon and Blume, pg. 823.

To check the necessity of compactness of the domain C, see the following
counterexamples.
1. F : R → R f (x) = x attains neither its sup nor its inf (closed but
not bounded domain)
2. f (x) = x ∀x ∈ [0, 1) = A. Here, supf (x) = 1 but there is no x ∈ A
x∈A
such that f (x) = 1. A is not closed: 1 ∈
/ A is one of its limit points.
1.1. BASIC CONCEPTS FOR CONVEX ANALYSIS 11

Figure 1.5: Example Non-Continuous Function

3. Continuity.

A function F : A → R, where A ⊂ RN is a convex set, is concave if

∀α ∈ [0, 1] ,
F (αx + (1 − α) y) ≥ αF (x) + (1 − α) F (y) ∀x, y ∈ A.
F is strictly concave if ∀α ∈ (0, 1) ,
F (αx + (1 − α) y) > αF (x) + (1 − α) F (y) ∀x, y ∈ A.
Theorem 4 Let A be a convex set and consider F : A → R.
i) If F is concave, the set of maximizers is convex.
ii) If F is strictly concave, it can have at most a unique maximizer.
Proof. i) Let x, x0 be maximizers (and then F (x) = F (x0 )). Note that
xα = αx + (1 − α)x0 ∈ A because A is convex. Since F is concave
F (xα ) ≥ αF (x) + (1 − α) F (x0 ) ≥ αF (y) + (1 − α) F (y) = F (y) ,
for all y ∈ A. By definition, xα is a maximizer as well.
ii) Let x, x0 be maximizers (and then F (x) = F (x0 )). Consider xα =
αx + (1 − α)x0 ∈ A as before. If F is strictly concave, it follows that F (xα ) >
αF (x) + (1 − α) F (x0 ) = F (x). But this contradicts that x and x0 are both
maximizers.

Some additional properties of concave functions are:

(a) If F and G are concave, then αF + βG is concave for all α, β ≥ 0.
(b) If F is concave and G is concave and increasing, then G ◦ F = G(F )
is concave.
(c) If F and G are concave, then H(x) = min(F (x), G(x)) is concave.
12 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Definition 10 Let F : A → R, where A ⊂ RN is an open subset. Let x ∈ A

and let 4 = (0, ..., 4n , ..., 0) such that x + 4 ∈ A. If it exists,

∂F (x) F (x + 4) − F (x)
= lim ,
∂xn 4→0 4
is called the partial derivative of F with respect to xn at x.
∂F ∂F
Let ∇F (x) = ( ∂x1
(x), ..., ∂xN
(x)) be the vector of partial derivatives of
F. We say that F is continuously differentiable (denoted C 1 ) if ∇f (x) is a
continuous function. It is important to mention that differentiable functions
are continuous.

Remark 3 Suppose that A ⊂ RN is open and convex subset and F : A → R

is C 1 . Very importantly, it can be shown that F is concave if and only if
F (x) − F (x0 ) ≤ ∇F (x) · (x − x0 ) for all x, x0 ∈ A.

Theorem 5 (Brower’s Fixed Point Theorem) Let ∆N −1 ≡ {x ∈ RN + :

PN N −1 N −1
n=1 xn = 1}. Suppose that f : ∆ →∆ is continuous. Then, f has
a fixed point (i.e. there is x∗ ∈ ∆N −1 such that x∗ = f (x∗ )).

Theorem 6 (Implicit Function Theorem) Suppose that A ⊂ RN , B ⊂

RM are open and F : A × B → RM is C k (i.e. k times continuously dif-
ferentiable). Suppose also that the Jacobian matrix ∂F (a, b) of F at some
(a, b) ∈ A × B is nonsingular. If F (a, b) = 0, then there exists  > 0 and
δ > 0 and a C k function G : RN → RM such that if kx − ak < , then
F (x, G(x)) = 0 and kG(x) − bk < δ.

1.2 Finite Dimensional Programming Prob-

lems
Motivating Example. Consider a consumer with a utility function U : RL+ →
R, which is assumed continuous, increasing, concave and differentiable with
∂F
∂xl
(x1 , ..., 0, ..., xL ) = 0 for all l. Suppose that this consumer has an en-
dowment w ∈ RL++ and he takes prices p ∈ RL++ as given. The consumer’s
problem is given by:
1.2. FINITE DIMENSIONAL PROGRAMMING PROBLEMS 13

max U (x), (AP)

x≥0

subject to,
L
X L
X
p·x= p l xl ≤ p · w = pl w l
l=1 l=1

Notice that the budget constraint is linear in x ∈ RL+ . Every linear

function is concave (and convex!).1 This is a prototype concave problem:
both the objective function and the constraint are concave.
In general, we have a concave programming problem if it has the following
structure:

max F (x), (CP)

x

subject to
G1 (x) ≥ 0, ..., GK (x) ≥ 0,
where F : RN N
+ → R, Gj : R+ → R for j = 1, ..., K are all concave functions.
We assume that K < L.
Thus, we can reinterpret our previous example where F = U , K = 1
L
P L
P
with: G1 (x) = p · w − p · x = pl wl − pl xl ≥ 0.
l=1 l=1
Let L(x, λ) = F (x) + λ · G(x) be the Lagrangian associated to the pro-
gramming problem (CPP) above. We say that (x∗ , λ∗ ) is a saddle point of L
if:
L(x, λ∗ ) ≤ L(x∗ , λ∗ ) ≤ L(x∗ , λ),
for all x ∈ RN and all λ ∈ RK
+.

Why are concave structures important? First, if x∗ solves the problem

(CPP) locally, then x∗ solves (CPP) globally. Second, we can find relatively
easy necessary and sufficient conditions to get a global maximizer for these
problems. To do this, consider the following fundamental theorem.
1
Remember that f : RL + → R is linear if:
1. f (x + y) = f (x) + f (y) ∀x, y ∈ RL+ , and;
L
2. f (λx) = λf (x) , x ∈ R+ and λ ∈ R.
Hence, using the definition of dot product, it is clear that a linear function is both
concave and convex.
14 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Theorem 7 Let F : RN N
+ → R, Gj : R+ → R (j : 1, ..., K < L) be contin-
uously differentiable concave functions. Consider the optimization problem
(CP) above.
Then x∗ ∈ Rn+ is a global Maximizer for (CP) if and only if there exists
λ∗ = (λ∗1 , ..., λ∗K ) , given the Lagrangian

L (x1 , ..., xL ; λ1 , ..., λK ) = F (x) + λ1 G1 (x) + ... + GK (x)λK ,

the following conditions are satisfied

(T.1) For all l = 1, ..., L,

∂L (x∗1 , ..., x∗L , λ∗1 , ..., λ∗K ) ∂F (x∗1 , ..., x∗L )

=
∂x∗l ∂x∗l
∗ ∗
∗ ∂G1 (x1 , ..., xL )
−(λ1 + ...
∂x∗l
∂GK (x∗1 , ..., x∗L )
+λ∗K ) = 0,
∂x∗l

(T.2) λ∗j [Gj (x∗ )] = 0 ∀j : 1, ..., K.

(T.3) λ∗j ≥ 0, Gj (x∗ ) ≥ 0 ∀j : 1, ..., K.
∗ ∗
That is, (x , λ ) is a saddle point of L.

Proof. See Simon and Blume: for necessity (”only if”) see pg. 430-1; for
sufficiency (”if”) pg. 532-3.
It is important to mention that we do not need concavity for necessity.
Concavity is important for sufficiency.

Remark 4 In fact, the result is incomplete as we stated the problem. We

need to check that x∗ satisfies some ”constraint qualification condition”. For
example, the so-called Slater Constraint Qualification: ∃ an open ball U about
x∗ ∈ RN 0
+ such that Gj s are concave and ∃ some z ∈ U such that Gj (z) > 0,
for all j : 1, ..., K.

Remark 5 We should mention that this result can be obtained with weaker
assumptions than those we have imposed on F and Gj . It is sufficient that
these functions are quasiconcave (remember that a concave function is qua-
siconcave, but the converse is not true in general).
1.2. FINITE DIMENSIONAL PROGRAMMING PROBLEMS 15

The result above is crucial because it allows us to characterize solutions

through necessary and sufficient Khun-Tucker conditions.
To see its power, consider the example described above for an economy
populated with I agents.
Motivating Example (continued).A competitive equilibrium for the econ-
omy described above is a price vector p̂ ∈ RL and an allocation x̂ = (x̂i )i∈I
I I
such that (i) ∀i ∈ I : x̂i solves AP and (ii) x̂i = wi .
P P
i=1 i=1
Given our assumptions, we can characterize the solution for agent i’s AP
with necessary and sufficient first order conditions as follows:

∂Ui (x̂i ) bi
 
i bi p̂
∇Ui (x̂ ) = λ ⇒ = λ p̂n for all n = 1, ..., N, (1.1)
∂xin
p̂wi − p̂x̂i = 0. (1.2)
i i
Note that since we assume that ∇Ui (x̂ ) >> 0 for all ∀x >> 0 and
i
lim ∂U∂x
i (x̂ )
i = ∞, we have that:
xl →0 l

bin > 0 for all n since λ

(i) x bi and pbn are real number; (ii) λ
bi > 0 from (1.1)
and so from (T2) follows the equality in (1.2).
Thus, this characterization implies that if we have a solution to the agent
i’s problem, then there exists some λ̂i such that (1.1) and (1.2). hold. On
the other hand, if we can find an allocation x̂i and a vector of Lagrange
multipliers λ̂i such that (1.1) and (1.2) are satisfied for agent i, then x̂i is the
solution to the i’s problem.
Now, consider the corresponding so-called Pareto problem for this econ-
omy. Let α = (α1 , ..., αm ) >> 0 be the vector of welfare weights consider:
I
X
max αi Ui (xi ), (PP)
(xi )Ii=1 ≥0
i=1
subject to
I
X
wi − xi ≥ 0.


i=1
PI
Since the Ui0 s
are concave, we know that i
i=1 αi Ui (x ) is concave in
I
(xi )i=1 ≥ 0 for all α ≥ 0. Given that the feasibility constraints are linear, we
have a standard concave programming problem. Let
16 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

I N
" I I
#
X X X X
Lp (x1 , ..., xL , λ1 , ..., λK ) = αi Ui (xi ) + qn wni − xin ,
i=1 n=1 i=1 i=1

be the Lagrangian associated to (PP), where qn is the Lagrange multiplier

for the constraint on good n = 1, .., N . Consider the necessary and sufficient
first order conditions for an interior solution:2
αi ∇ui (x̃i ) = q̃ , i = 1, ..., I, (1.3)

I I
!
X X
q̃n wn − x̃in =0 , n : 1, ..., N, (1.4)
i=1 i=1
I
X
wni − x̃in ≥ 0,


q̃n ≥ 0, , , n : 1, ..., N. (1.5)
i=1
i
Since αi > 0and ∇ui (x̃ ) >> 0 for all i, we have q̃ >> 0 (q̃n > 0 for all
n : 1, ..., N ). Hence (P.2) can be written as:
I
X I
X
wn − x̃in = 0, for all n = 1, ..., N.
i=1 i=1

The celebrated 1 st Welfare Theorem (to be studied in detail below) states

that any competitive equilibrium allocation is a Pareto optimal. Thus, we
should be able to show that a competitive equilibrium allocation solves a
Pareto problem for some vector of welfare weights α bi > 0 for all i. To do
this, we need to find a vector of Lagrange multipliers q̂ = (q̂1 , ..., q̂N ) and
welfare weights α̂ >> 0 such that (1.3)-(1.5) are satisfied. Given (b
x, λ,
b p̂), we
construct (α̂, q̂) as follows:
1
α̂i = ∀i ∈ I,
λ
bi

q̂n = p̂n ∀n : 1, ..., N.

It follows by construction that (1.1) and (1.2) imply that (b x, q̂) solves
i
PP given the welfare weights α̂ and consequently (x̂ )i∈I is a Pareto optimal
allocation.
2
Note that for each i ∈ I, we need to choose xin , n : 1, ..., N.
1.3. UNCERTAINTY AND EXPECTED UTILITY (NEW) 17

1.3 Uncertainty and Expected Utility (NEW)

This section provides a brief introduction to the theory of stochastic processes
in discrete time. Probability theory is the branch of mathematics that deals
with uncertainty.3 Then, we also provide an introduction to expected utility.

1.3.1 Stochastic processes

A stochastic process is a model of uncertainty that unfolds over time; i.e. a
sequence {st : t = 0, 1, ..., T ≤ ∞} of random variables on a probability space
{Ω, F, P } where st (ω) ∈ Rn .
A probability space is a triple {Ω, F, P } where Ω is a set called the sample
space, with typical element ω ∈ Ω. Subsets of Ω are called events, and F is
a collection of such subsets satisfying the following properties:

(i) ∅ ∈ F

(ii) If A ∈ F, then Ac ∈ F

(iii) If An ∈ F for all n ∈ N, then ∪∞

n=1 An ∈ F

F is called a σ-algebra of Ω. The third element in the triple {Ω, F, P },

P , is called a probability measure. It satisfies the following properties:

(i) P (∅) = 0

(ii) P (A) ≥ 0 for all A ∈ F

(iii) Let An for all n ∈ N be a collection of disjoint subsets of F, then

∞
X
P (∪∞
n=1 An ) = P (An )
n=1

(iv) P (Ω) = 1
3
A.N. Kolmogorov is considered the founding father of modern probability theory. In
1933, he provided an axiomatic basis that it is now universally accepted. Thus, he invented
so many things in this field and so his name will appear repeatedly. If you study probability
theory seriously, you will soon detect that this is the reason why, as usual if you invent
too many things, he will share some results with other names to avoid confusion.
18 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

We consider an environment where stochastic events evolve as follows.

Information is revealed each period t = 0, 1, ... and represented by st . We
assume that at date 0, s0 is known. We denote St = {ζ1 , ..., ζn } for all
t ≥ 0 as the exogenous state space such that st ∈ St = S for all t. Let
st = (s0 , ..., st ) ∈ S t = ×tj=0 St represent the partial history of events up to
date t. Thus, Ft = {ω ∈ S ∞ : ω = {st , ...}, st ∈ S t } denotes the set of events
corresponding to the information available at time t and Ω = S ∞ (i.e. the
set of infinite sequences). If an event B ∈ Ft , then at time t this event is
known to be true or false. The filtration {Ft } satisfies Ft ⊂ Fs ⊂ F for all
t ≤ s (i.e. events are never ”forgotten”) and it represents how information is
revealed through time.
For any random variable Y , we let Et (Y ) = E(Y | Ft ) denotes the con-
ditional expectation of Y given Ft . An adapted process is a sequence of
random variables {Xt } such that for each t, Xt is a random variable with
respect to {Ω, Ft }. Informally, this means that at time t, the outcome of Xt
is known. An adapted process is a martingale if, for any time t and s > t,
we have Et (Xs ) = Xt .
We say that the stochastic process {st } has the Markov property if ∀k ≥ 2
and ∀t:
P r{st+1 | st , st−1 , st−2 , ..., st−k } = P r{st+1 | st }
A time-invariant Markov chain is defined by:
1) A vector s̄ = (ζ1 , ..., ζn ) containing all possible state values.

2) An n×n transition matrix P indicating the probability that the system

moves from one state to another in one period:
 
P11 P12 . . . P1n
 P21 P22 . . . P2n 
P =  ..
 
.. ... .. 
 . . . 
Pn1 Pn2 . . . Pnn
P r{st+1 = s̄j | st = s̄i }.
where Pij = P
Notice that nj=1 Pij = 1 ∀ i = 1, ..., n

3) An n × 1 column vector π0 indicating the probabilities of being in each

state i in period t = 0:

π0i = π(s0 = s̄i ) = P r{s0 = s̄i }

1.3. UNCERTAINTY AND EXPECTED UTILITY (NEW) 19
Pn
This vector satisfies i=1 π0i = 1.
We will call πt = π(st ) to the n × 1 vector indicating the probabilities of
being in each state i = 1, ..., n in period t (i.e., the probability of st taking a
particular value). In other words, πt indicates the unconditional probability
of st .
Using the transition matrix P and the vector π0 we can calculate π1 in
the following way:
π10 = π00 P

π1 = P 0 π0
   0  
π1 (s̄1 ) P11 P12 . . . P1n π0 (s̄1 )
 π1 (s̄2 )   P21 P22 . . . P2n   π0 (s̄2 ) 
 ..  =  ..
     
.. . . ..   .. 
 .   . . . .   . 
π1 (s̄n ) Pn1 Pn2 . . . Pnn π0 (s̄n )
    
π1 (s̄1 ) P11 P21 . . . Pn1 π0 (s̄1 )
 π1 (s̄2 )   P12 P22 . . . Pn2   π0 (s̄2 ) 
 ..  =  ..
    
.. .. ..   .. 
 .   . . . .   . 
π1 (s̄n ) P1n P2n . . . Pnn π0 (s̄n )
   
π1 (s̄1 ) P11 π0 (s̄1 ) + P21 π0 (s̄2 ) + ... + Pn1 π0 (s̄n )
 π1 (s̄2 )   P12 π0 (s̄1 ) + P22 π0 (s̄2 ) + ... + Pn2 π0 (s̄n ) 
 ..  = 
   
.. 
 .   . 
π1 (s̄n ) P1n π0 (s̄1 ) + P2n π0 (s̄2 ) + ... + Pnn π0 (s̄n )
Similarly, for π2 :
π20 = π10 P
= (π00 P )P
= π00 P 2

And thus πk0 = π00 P k . We can summarize this by saying that the probability
distributions follow the following law of motion:
0
πt+1 = πt0 P
20 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

A distribution is said to be stationary if it remains unaltered when time

passes: πt+1 = πt = π.
Stationary distributions must also follow the law of motion. Then:

π0 = π0P
π = P 0π
(P 0 − I)π = 0
P
This means that π is an eigenvector (normalized to satisfy πi = 1)
a unit eigenvalue P 0 . Since none of the elements of P are
associated to P
negative and nj=1 Pij = 1, we know that P 0 has at least one unit eigenvalue
and then there is some π that satisfies the equation (P 0 − I)π = 0.
We want to know if given some initial distribution π0 , the distributions πt
approach a stationary distribution lim πt = π∞ , and wether the stationary
t→∞
distribution π∞ will depend on the initial distribution π0 .

Definition 11 Let π∞ be the only vector satisfying (P 0 − I)π∞ = 0. If for

every possible initial distribution π0 it is true that (P t )0 π0 converges to the
same distribution π∞ (called ”stationary distribution” or ”invariant distribu-
tion of P”), we say that the Markov chain is ”asymptotically stationary with
a unique invariant distribution”.

Theorem 8 Let P be a stochastic matrix with Pij > 0 ∀ i, j. Then P has a

unique stationary distribution, and the process is asymptotically stationary.

Throughout this notes, we mostly assume that {st }∞ t=0 is a finite state first-
order stationary Markov chain. Recalling our definitions above, this means
that Pr{st+1 | st } = Pr{st+1 | st } for all (st+1 , st ). Transition
P probabilities
are denoted by Pr{st+1 = s | st = s} = π(s, s ) > 0 where s0 ∈S π(s, s0 ) = 1
0 0

for all s ∈ S. We write st+1 /st to denote that st+1 is an immediate successor
of st . A simple application of the so-called Chapman-Kolmogorov equations
implies that the probability of st is constructed from π as follows:

π(st ) = Pr{st } = π(s0 , s1 )....π(st−1 , st ), (1.6)

where st π(st ) = 1 by definition.

P
We can work backwards in this case given the definition of a probability
measure on finite events, (1.6), denoted by π above. If we denote Ω =
1.4. INFINITE DIMENSIONAL PROGRAMMING PROBLEMS 21

{ω : ω = {s0 , s1 , ..., st , ...}, st ∈ St }, the Kolgomorov’s Extension Theorem

implies that there exists a unique probability measure P on the probability
space (Ω, F) induced by π, where F is the σ − f ield determined by the Borel
sets.

1.3.2 Expected Utility

Throughout these notes we will work with agents that maximize their ex-
pected utility when making decisions in an environment with uncertainty.
This means that the objective function of optimizing agents will be a weighted
sum of the utility attained by each individual outcome, where the weights
are the probabilities assigned to the occurrence of each outcome.

1.4 Infinite Dimensional Programming Prob-

lems
Let X ⊂ RL be the endogenous state space. We assume that actions at t are
taken after st has been observed and, in general, x(st ) ∈ X denotes the value
of x at the node st .

Definition 12 Let 2X the power set of X (i.e. the set of all the subsets of
X.). For each st ∈ S, we define Γ(st , ·) : X → 2X as a correspondence.
That is, given (s, x) ∈ S × X, Γ assigns a particular subset of X.

Let

Π (s0 , x0 ) = {{xt }∞ t t
t=0 : xt+1 : S → X such that xt+1 (s ) ∈ Γ(st , xt (s
t−1
))
for all st−1 , st , given (s0 , x0 )},

be the set of feasible plans with typical element x

e (s0 , x0 ) .
Denote F : S ×X ×X → R as the return function. Consider the following
sequential programming problem:

∞ X
X
∗
β t π st F st , xt (st−1 ), xt+1 (st ) . (SP)



V (s0 , x0 ) ≡ sup
x
e(s0 ,x0 )∈Π(s0 ,x0 ) t=0
st
22 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

Suppose that F is uniformly bounded and β ∈ (0, 1). Consequently,

∗
V (s0 , x0 ) < ∞, for all (s0 , x0 ) ∈ S × X (why? convince yourself!).
Now consider a variational approach to characterize a solution for (SP).
First, suppose that xe∗ (s0 , x0 ) solves (SP). Then, given the Markovian struc-
ture of shocks, a necessary condition for an optimum is that ∀st :

x∗t+1 (st ) = arg max{F st , x∗t (st−1 ), y

,
X
π (st , s0 ) F s0 , y, x∗t+2 (st , s0 ) },


+β
s0

subject to y ∈ Γ (st , x∗t (st−1 )) and x∗t+2 (st , s0 ) ∈ Γ (s0 , y) for all s0 .
To see this, note that otherwise there exists some set such that y ∗ 6=
x∗t+1 (e
st ). To get a contradiction, construct an alternative feasible plan as
follows:
xt+1 (st ) = x∗t+1 (st ) ∈ Γ st , x∗t (st−1 ) ∀st 6= set

st ) = y ∗ ∈ Γ st , x∗t (st−1 )


xt+1 (e
e∗ (s0 , x0 )
This alternative plan obtains a larger value, but this violates that x
was an optimum for (SP).

Necessity of Euler Equation for Interior Solutions

Suppose that F (s, x, y) is strictly concave and continuously differentiable
in (x, y) for each s. Suppose also that x e∗ (s0 , x0 ) is an interior optimal so-
lution. That is, x∗t+1 (st ) ∈ int Γ (st , x∗t (st−1 )) for all t and all st . Then, it
follows from the concave programming arguments detailed above for interior
solutions that:

∂F st , x∗t (st−1 ), x∗t+1 (st ) ∗ ∗ t 0

t


0 ∂F st , xt+1 (s ), xt+2 (s , s )
X
+β π (st , s ) =0
∂yj s0
∂xj
(EE)
∀j = 1, ..., L.

1.4.1 Sufficiency of Euler Equations and Transversality

Conditions
To simplify, assume that X ∈ RL+ and then xt+1 (st ) ≥ 0 for all st .
1.4. INFINITE DIMENSIONAL PROGRAMMING PROBLEMS 23

Theorem 9 Let F : s × X × X be continuous, bounded, differentiable, in-

creasing in x and strictly concave in (x, y) for each s. Suppose that there
e∗ (s0 , x0 ) ∈ Π (s0 , x0 ) such that it satisfies the
exists an interior feasible plan x
Euler Equations and the following transversality condition:

lim E β T Fx sT , x∗T , x∗T +1 · x∗T



= (T C)
T →∞
( )
X
βT π sT F x sT , x∗T sT −1 , x∗T +1 sT x∗T sT −1






lim = 0.
T →∞
sT

e∗ (s0 , x0 ) solves the sequential problem (SP).

Then, x
Remark 6 F bounded can be replaced by |V ∗ (s0 , x0 )| < +∞. Also, Fx ≥ 0
and x ≥ 0 imply that TC is satisfied if and only if

( )
X
βT T
Fxj sT , x∗T s T −1
, x∗T +1 s T
x∗jT s T −1






lim π s =0
T →∞
sT

∀j = 1, ..., L.
Proof. Suppose that x e∗ (s0 , x0 ) is an interior feasible plan that satisfies (EE)
and (TC). Let x e (s0 , x0 ) be any arbitrary alternative feasible plan. Define
T X
X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F st , xt (st−1 ), xt+1 (st ) .




DT ≡
t=0 st

The statement is proved if lim DT ≥ 0. Now note that it follows by concavity

T →∞
and differentiability of F :
F st , x∗t (st−1 ), x∗t+1 (st ) − F st , xt (st−1 ), xt+1 (st )

≥ Fx st , x∗t (st−1 ), x∗t+1 (st ) · x∗t (st−1 ) − xt (st−1 )

+Fy st , x∗t (st−1 ), x∗t+1 (st ) · x∗t+1 (st ) − xt+1 (st ) ,

∀t, ∀st . Then, it is easy to check that:

T X
X
β t π st {Fx st , x∗t (st−1 ), x∗t+1 (st ) · x∗t (st−1 ) − xt (st−1 )




DT ≥
t=0 st
+Fy st , x∗t (st−1 ), x∗t+1 (st ) · x∗t+1 (st ) − xt+1 (st ) },



24 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

and since x∗0 = x0 , we have that:

T X
X
β t π st Fx st , x∗t (st−1 ), x∗t+1 (st ) · x∗t (st−1 ) − xt (st−1 )




=
t=1 st
T −1 X
X
β t π st Fy st , x∗t (st−1 ), x∗t+1 (st ) x∗t+1 (st ) − xt+1 (st )




+
t=0 st
X
β T π sT Fy sT , x∗T (sT −1 ), x∗T +1 (sT ) x∗T +1 (sT ) − xT +1 (sT )




+
ST

We use the EE’s in the second line to write the last expression as follows:

T X
X
β t π st Fx st , x∗t (st−1 ), x∗t+1 (st ) · x∗t (st−1 ) − xt (st−1 )




=
t=1 st
T
XX−1

X
β t π st β π (st , s0 ) Fx s0 , x∗t+1 (st ), x∗t+2 (st , s0 ) · x∗t+1 (st ) − xt+1 (st )



−
t=0 st s0
X
+β T T
Fy sT , x∗T (sT −1 ), x∗T +1 (sT ) · x∗T +1 (sT ) − xT +1 (sT )




π s
sT
T X
X
β t π st Fx st , x∗t (st−1 ), x∗t+1 (st ) · x∗t (st−1 ) − xt (st−1 )




=
t=1 st
T −1 X
X
β t+1 π st+1 Fx st+1 , x∗t+1 (st ), x∗t+2 (st , st+1 ) · x∗t+1 (st ) − xt+1 (st )




−
t=0 st+1
X
X
−β T π sT β π (sT , s0 ) Fx s0 , x∗T +1 (sT ), x∗T +2 (sT , s0 ) · x∗T +1 (sT ) − xT +1 (sT ) .

sT s0

Since Fx · x ≥ 0, we can conclude that

X
DT ≥ −β T π sT +1 Fx sT +1 , x∗T +1 (sT ), x∗T +2 (sT +1 ) · x∗T +1 (sT ).

sT +1

We take the limit with T → ∞ and apply the TC on the RHS to get the
desired result since lim DT ≥ 0.
T →∞
1.4. INFINITE DIMENSIONAL PROGRAMMING PROBLEMS 25

1.4.2 On the Necessity of the Transversality Condition

Consider problem (SP) described above and assume that
∞ X
X
β t π st F st , xt (st−1 ), xt+1 (st ) ≤ B,



t=0 st

e (s0 , x0 ) ∈ Π (s0 , x0 ) .
for all x
Note that this implies that
∞ X
X
β t π st π sT F xt (z t−1 ), xt+1 (z t ) = 0,




lim (LC)
T →∞
t=T +1 st

for all x e (s0 , x0 ) ∈ Π (s0 , x0 ) .4

We make the following additional assumptions.
(i). Γ(st , ·) is convex valued for all st and 0 ∈ Γ (st , 0) for all st .
(ii) F is C 1 and concave and Fx (st , x, y) ≥ 0, Fy (st , x, y) ≤ 0 for all
(st , x, y) .
We have already shown that any interior optimal plan (i.e. x e∗ (s0 , x0 ) ∈
intΠ (s0 , x0 )) satisfies the EE’s where:

X
Fy st , x∗t st−1 , x∗t+1 st + β π (st , s0 ) Fx s0 , x∗t+1 (st ), x∗t+2 (st , s0 ) = 0





s0
(1.7)
t
for all s .

Theorem 10 (Necessity of the TC, Kamihigashi (2002)) Under our as-

sumptions, any interior optimal path x
e (s0 , x0 ) must satisfy:
X
lim β T π sT Fy sT, x∗T (sT −1 ), x∗T +1 (sT ) · x∗T +1 (sT ) = 0



(TC)
T →∞
sT

or, equivalently using the Euler Equations,

4
P∞
Remember that if {zt } is a sequence of positive number, whenever t=0 zt ≤ B < ∞
PT
then the sequence yP T = t=0 zt must converge (i.e., a bounded increasing sequence).
∞
Therefore, the tail, t=T +1 zt , goes to 0.
26 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW

X
lim β T π sT Fx sT, x∗T (sT −1 ), x∗T +1 (sT ) x∗T (sT −1 ) = 0



(TC’)
T →∞
sT

Proof. Note first that if f : [0, 1] → R is concave, then ∀γ, λ such that
0 ≤ γ ≤ λ < 1 it follows that
f (1) − f (λ) f (1) − f (γ)
≤ .
1−λ 1−γ
Let xe∗ (s0 , x0 ) be an interior optimal path and T ∈ Z+ . Given that x e∗ is
interior and (0, 0) is feasible ∀st ( that is, 0 ∈ Γ (st , 0) , ∀st ), it follows that
there exists λ ∈ [0, 1), sufficiently close to 1, such that
 ∗ t−1
λ t−1 xt (s ) ∀t ≤ T, ∀st
xt (s ) =
λx∗t (st−1 ) ∀t ≥ T + 1, ∀st−1
e∗ , we have that
defines a feasible plan. By optimality of x
∞ X
X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F st , xλt (st−1 ), xλt+1 (st )




0 ≤
t=0 st
X
β T π sT F sT , x∗T (sT −1 ), x∗T +1 (sT ) − F sT , x∗T (sT −1 ), λx∗T +1 (sT )




=
sT
∞ X
X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F st , λx∗t (st−1 ), λx∗t+1 (st ) ,




+
t=T +1 st

It follows by the initial remark that if we put δ = 0, then

P∞ P t t
 ∗ t−1 ∗ t

∗ t−1 ∗ t


t=T +1 s t β π (s ) F s t , xt (s ), xt+1 (s ) − F s t , λx t (s ), λxt+1 (s )
1−λ
∞ X
X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F (st , 0, 0) .



≤
t=T +1 st

On the other hand, note that for all sT = sT −1 , sT it follows by concavity

that
F sT , x∗T (sT −1 ), λx∗T +1 (sT ) − F sT , x∗T (sT −1 ), x∗T +1 (sT )

≥ Fy sT , x∗T (sT −1 ), λx∗T +1 (sT ) · (λ − 1)x∗T +1 (sT )

= (1 − λ) −Fy sT , x∗T (sT −1 ), λx∗T +1 (sT ) · x∗T +1 (sT ).



1.4. INFINITE DIMENSIONAL PROGRAMMING PROBLEMS 27

Therefore, since
X
β T π sT F sT , x∗T (sT −1 ), x∗T +1 (sT ) − F sT , x∗T (sT −1 ), λx∗T +1 (sT )




−
sT
∞
X X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F st , λx∗t (st−1 ), λx∗t+1 (st ) ,




≤
t=T +1 st

all this implies that

X
β T π sT −Fy sT , x∗T (sT −1 ), x∗T +1 (sT ) · x∗T +1 (sT )



0 ≤ (1 − λ)
sT
∞
X X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F (st , 0, 0) .



≤
t=T +1 st

The first inequality is due to the fact that −Fy (s, x, y) · y ≥ 0 for all
(s, x, y) by assumption. We claim that when T → ∞, then the (RHS) goes
to 0 by (LC). To see this, notice that
∞ X
X
β t π st F st , x∗t (st−1 ), x∗t+1 (st ) − F (st , 0, 0)



t=T +1 z t
X∞ X ∞ X
X
t t
st , x∗t (st−1 ), x∗t+1 (st ) β t π st |F (st , 0, 0)|




≤ βπ s F +
t=T +1 st t=T +1 z t

This last two expressions must go to 0 whenever (LC) holds. Therefore,

since weak inequalities are preserved in the limit, the proof is concluded.

Exercises
Exercise 1 In text.
Exercise 2 Show that if a sequence converges, then it is Cauchy.
Exercise 3 Show that if a sequence is Cauchy, then it is bounded. That is,
supn,m d (xn , xm ) < ∞.
Exercise 4 Show that xn → x if and only if every subsequence of {xn }
converges to x.
Exercise 5 Show that if for every n, xn+1 ≥ xn , and the sequence {xn }∞
n=1
is bounded, then xn converges.
28 CHAPTER 1. MATHEMATICAL PRELIMINARIES: AN OVERVIEW
Chapter 2

Optimality and Competitive

Equilibrium

We first study pure exchange dynamic economies as a benchmark to

analyze risk-sharing, asset prices and consumption allocation across agents.
A pure exchange economy is an economy where there are no production op-
portunities. The primitives that define the economy are the set of agents
(or, indistinctly, consumers, households), the commodity space, the consump-
tion set, preferences and endowments. In a private ownership economy
we need to define the distribution of the aggregate endowment across the
agents. That is, we need to define property rights on aggregate resources.
Finally, agents will interact through markets to trade goods and securities.
Below, we will consider two alternative dynamically complete markets struc-
tures.

2.1 The Economy

We consider an economy populated by I (types of) agents where i ∈
I = {1, ..., I} denotes a typical consumer and I denotes the set of agents.
Agents live T ≤ ∞ periods. Time is discrete and denoted by t = 0, 1, 2, ..., T .
There are N consumption goods every period which cannot be stored. This
economy is subject to random shocks that can affect either preferences or
endowments. These shocks are represented by st ∈ St = {ζ1 , ..., ζS } at date
t and s0 is known at date 0. Let st = (s0 , ..., st ) ∈ S t = {s0 } × S1 × ... × St
represent the partial history of events up to date t. We write st+1 /st to denote

29
30 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

that st+1 is an immediate successor of st . These histories are observed by all

the agents.
We assume that {st } is a finite state first-order stationary Markov
process. Roughly speaking, this means that Prt {st+1 k st } = Pr{st+1 k st }
for all (st+1 , st ) and for all t. Transition probabilities are denoted by π(s, s0 ) >
0 for s, s0 ∈ {ζ1 , ..., ζS }. The probability of st is constructed from π using
(1.6) as follows:

π(st ) = Pr{st } = π(s0 , s1 )....π(st−1 , st ).

In period t, actions are taken after st has been observed and, in general,
x(s ) denotes the value of x at the node st . Agents value consumption at
t

each node st and, thus, the commodity space (i.e., the real vector space
containing all the relevant objects) is given by

X = {{xt }Tt=0 | xt : S t → RN , sup x(st ) < ∞}.

(t,st )

A consumption bundle for agent i is a bounded adapted non-negative

consumption processes. That is, the consumption set for all the agents,
C, is given by

C = X+ = {{ct }Tt=0 | ct : S t → RN
+ , sup c(st ) < ∞}.
(t,st )

Notice that this means that the household will continuously update the infor-
mation set, conditioning on all information available at the time the decision
is made.
Preferences are represented by expected, time-separable, discounted util-
ity. That is , if ci ∈ C then:

T X
X
Ui (ci ) = βit π(st )ui (ci (st )), (UF)
t=0 st ∈S t

where, for all i, βi ∈ (0, 1) and ui : RN+ → R+ is strictly increasing, strictly

concave and continuously differentiable. Additionally, we assume that
∂ui
∂xn
(..., 0, ...) = +∞ for all n to rule out corner solutions; i.e. the so-called
Inada conditions.
2.1. THE ECONOMY 31

Let yi (st ) ∈ RN
++ be the endowment process for agent i at s . That
t

is, if the history of events st has been observed at the beginning of period t,
agent i receives from heaven yin (st ) units of the nth − consumption good as
endowment. We assume that sup(t,st ) kyi (st )k = y < ∞ and denote y(st ) =
PI t N
i=1 yi (s ) ∈ R++ as the vector of aggregate endowment. Note the
individual endowment process is an element of the consumption set.
As you have already noticed, we assume here that these stochastic events
do not affect preferences (i.e. preferences are state independent) but only
endowments. This assumption was made only to simplify notation. State
dependent preferences can be incorporated without any inconvenient.
We say that (ci )Ii=1
PI is an tallocation if ci ∈ C for all i. An allocation
(ci )i=1 is feasible if i=1 ci (s ) ≤ y(s ) for all st .
I t

Definition 13 A feasible allocation (c∗i )Ii=1 is Pareto Optimal (PO) if

ci ) ≥ Ui (c∗i )
ci )Ii=1 such that Ui (b
there exists no alternative feasible allocation (b
for all i, with strict inequality for some agent.

Let αi > 0 denote the welfare weight for agent i. Given the vector
(αi )Ii=1 , the corresponding Pareto problem is given by
I
( T )
X XX
sup αi βit π(st ) ui (ci (st )) , (PP)
(ci )Ii=1 i=1 t=0 st ∈S t

subject to
I
X
ci (st ) ≤ y(st ) for all st , (2.1)
i=1
ci ∈ C for all i.

Exercise 6 1 Suppose that T < ∞ and show that, under our assumptions,
there exists a unique (c∗i )Ii=1 solving (PP). Show also that any solution to (PP)
I
is
PIhomogeneous of degree zero in (αi )i=1 . Conclude that we can normalize
i=1 αi = 1.

Now we are ready to show that there exists a one-to-one relationship

between the set of Pareto optimal allocations and the solutions for the Pareto
problem parametrized by the vector of welfare weights α.
1
When T = ∞ the reasoning is similar but some additional technical details must be
taken into account.
32 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

Theorem 11 If a feasible allocation (c∗i )Ii=1 solves (PP) for some (αi )Ii=1 ,
then (c∗i )Ii=1 is a PO allocation. Furthermore, if (c∗i )Ii=1 is a PO allocation,
then there exists some vector (αi∗ )Ii=1 such that (c∗i )Ii=1 solves PP correspond-
ing to (αi∗ )Ii=1 .
You are asked to prove this theorem in Exercise 7.

In the next sections we consider two alternative market structures to

study competitive exchange. Then, we show in which sense both market
structures are equivalent.

2.2 Arrow-Debreu Economies

In an Arrow-Debreu Economy households trade a full set of contin-
gent goods through markets. This market structure assumes that all these
trades are done at date 0. That is, agents meet only once at date 0
and trade fully enforceable contracts contingent upon realizations of partial
histories st .
More precisely, let Pn (st ) be the price of one unit of the nth −consumption
good to be delivered at st . We assume that P (st ) =P t
(P1 (sP ) , ..., PN (st )) ∈
T
RN t
+ and denote P = (P (s )) as a price system if t=0
t
st P (s ) < ∞.
This last condition will guarantee bounded valuation and consequently the
problems below will be well-defined.
Since demand functions are homogeneous of degree zero, we normalize
P1 (s0 ) = 1. This implies that all prices are in units of the consumption good
1 at date 0.
We assume that markets are competitive and thus households take the
price system as given. Given a price system P , agent i’s solves the following
problem:
XT X
sup βit π(st )ui (ci (st )), (AD-AP)
ci ∈C
subject to t=0 s t

XT X
P st · ci st − yi st ≤ 0.





(BC-AD)
t=0 st
The interpretation in this general stochastic context is that, given the
restrictions imposed by (BC-AD), agent i chooses ex-ante the best contingent
plan (i.e., a consumption bundle that is a sequence of random variables in
the consumption set).
2.2. ARROW-DEBREU ECONOMIES 33

If T < ∞ and P (st ) >> 0 for all st we can apply Weiestrass’ Theorem to
show that the sup is attained in (AP-AD) and it can be replaced by a max
since preferences are continuous and the budget set is compact. If T = ∞,
the same result can be obtained but some additional technical details must
be handled. Also, note that the budget constraint will hold with equality in
any solution since the utility function is strictly increasing.
The equilibrium concept for an economy with this market structure is the
following.

Definition 14 An Arrow-Debreu Competitive Equilibrium (ADCE)

is a price system Pb and an allocation (b ci )Ii=1 such that
(AD.1) Given Pb, b ci solves (AD-AD) for all i.
I
(AD.2) (bci )i=1 is a feasible allocation.

We define the Lagrangian for (AD-AP), denoted LAD , as follows. Let

λi > 0 be the Lagrange multiplier for agent i corresponding to (BC-AD).
Thus, given the price system P we obtain
T X
X T X
X
βit π(st ) ui (ci (st ))+λi P s t · y i s t − ci s t .





LAD (ci , λi ) =
t=0 st t=0 st

Necessary and sufficient conditions to characterize an ADCE are given

by a price system Pb and Lagrange
 multipliers coupled with a consumption
bundle for each agent i λi , b
b ci such that

∂ui
βit π st ci st = λ
bi Pbn st for all n, all st and all i,




b (2.2)
∂cin
T X
X T X

X
Pb st · b
ci s t = Pb st · yi st ,




for all i, (2.3)
t=0 st t=0 st
X
X
cin st = yin st for all n and all st .


b (2.4)
i i

On other hand, we define the Lagrangian for the Pareto problem (PP),
denoted LP P , as follows. Let µn (st ) > 0 be the Lagrange multiplier corre-
sponding to the nth −feasibility constraint (4.11) at st . Thus, for a given α
34 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

we have
I
( T )
X X X
LP P (c, µ, α) = αi βit π(st ) ui (ci (st ))
i=1 t=0 st ∈S t+1
T X I
!
X X
t t t




+ µ s · y s − ci s .
t=0 st i=1

Thus, given (α1 , ..., αI ), necessary and sufficient conditions characterizing

a solution for (PP) are given by a process of Lagrange multipliers µ∗ ∈ X+
and an allocation (c∗i )Ii=1 such that

∂ui ∗ t

αi βit π st (ci s )(α) = µ∗n st (α) for all n, all st and all i,


∂cin
(2.5)
X X
c∗in (st )(α) = yin st for all n and all st .


(2.6)
i i

Now we are ready to use the power of necessary and sufficient conditions
characterizing solutions to show the two fundamental theorems in welfare
economics. Again, we assume that T < ∞ to avoid technical details.

ci )Ii=1 is
Theorem 12 (First Welfare Theorem) An ADCE allocation (b
Pareto optimal.

ci )Ii=1 is a ADCE allocation, there exist a price system Pb and a

Proof. If (b
 I
vector of Lagrange multipliers λ bi such that (2.2)-(2.4) are satisfied. It
i=1
is sufficient to show that there exist µb ∈ X+ and a vector of welfare weights
α
b such that (4.12) and (4.13) are satisfied. We proceed constructively and
define
bn (st ) = Pbn st for all n and all st ,


µ
1
α
bi = for all i.
λ
bi
   
ci )Ii=1 , Pb, λ
Thus, given that (b ci )Ii=1 , µ
b satisfy (2.2)-(2.4), (b b, α
b sat-
ci )Ii=1 solves (PP) for
isfy (4.12)-(4.13) by construction and consequently (b
2.2. ARROW-DEBREU ECONOMIES 35
 I
α
bi = 1
λ
bi ci )Ii=1 is Pareto optimal (Theorem 9).
. We can conclude that (b
i=1

The converse is true if transfers among agents are considered.

Theorem 13 (Second Welfare Theorem) Any Pareto Optimal allocation

(c∗i )Ii=1 can be decentralized as an Arrow Debreu Competitive Equilibrium
(ADCE) allocation with transfers.

Proof. Suppose that (c∗i )Ii=1 is PO allocation. Since we assume that T < ∞,
it follows by Theorem 10 that there exists a vector of welfare weights α∗ such
that (c∗i )Ii=1 solves (PP) given α∗ . Necessary and sufficient conditions for an
interior solution imply that there exists a process of Lagrange multipliers
µ∗ ∈ X+ such that (c∗i (α∗ ), µ∗ (α∗ ))Ii=1 satisfy (4.12) and (4.13) given welfare
weights α∗ . It is necessary and sufficient to show that there exists a price
system P ∗ and a vector of Lagrange multipliers (λ∗i )Ii=1 such that (c∗i , λ∗i )Ii=1
satisfy (2.2)-(2.4) with transfers, given P ∗ . We proceed constructively again
and define
µ∗n (st )(α)
Pn∗ t ∗
for all n and all st ,


s ,α = ∗
µ1 (s0 )(α)
1
λ∗i = for all i.
αi∗
Finally, define transfers for agent i as follows
∞ X
X
∗
P ∗ (st )(α) · c∗i (st )(α) − yi st



Ai (α ) = .
t=0 st

Note that feasibility implies that i Ai (α∗ ) = 0 and then some transfers
P
are negative (i.e. those agents are ”taxed away”). Thus, if agents are entitled

to transfers (Ai (α∗ ))Ii=1 , the budget constraints are satisfied and (c∗i )Ii=1 , P ∗
constitutes an ADCE (with transfers) by construction.

Note that if any αi equals 0, we are not able to construct the Lagrange
multiplier λ∗i as before. In that case, we set c∗i (st ) = 0 for all st and the
equivalence still holds, since αi = 0 implies that the planner will not asign
consumption to houshold i (we need to assume that 0 ∗ ui (0) = 0).
According to our analysis, there is still an open question regarding time
consistency of agents’ choices. More precisely, in this framework agents are
36 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

only allowed to trade and choose consumption bundles ex-ante at date 0.

Would they choose to change their consumption plans
if they were allowed
∗ I ∗
to re-trade at some date t? That is, let (ci )i=1 , P constitute an ADCE
and suppose that markets are ”reopened” at the node st where agents are
allowed to ”retrade” again after having honored their contracts signed at
T −t
date 0. Thus, (c∗i (st+j | st ))j=0 determines agent i’s ”endowment” from node
T −t
st onwards and define (b ci (st+j | st ))j=0 as the ”continuation” consumption
path that they could choose at st with the corresponding new price system
 T −t
Pb (st+j | st ) .
j=0
In Exercise 8 you are asked to show that agents would not choose to
re-trade.

Digression: On the Existence of a Competitive Equilibrium

It can be shown that the functions (Ai (α))Ii=1 are continuous with respect
to α. We can then apply the Brouwer Fixed Point theorem to show that there
exists some α b such that Ai (b
α) = 0 for all i. This implies that there exists
an ADCE without initial transfers. This is the celebrated Negishi’s
approach to compute competitive equilibria. To illustrate this methodology,
consider the following example based on Kehoe (1989).
Suppose that N = 1, I = {1, 2}, T = +∞ and St = 1 for all t (i.e. there
is no uncertainty). Preferences are represented by (UF) where ui (c) = ln c
for all i ∈ I, 0 < β1 < β2 < 1 and yi (t) = 1 for all i ∈ I and all t = 0, 1....
In this case, (2.2)-(2.4) reduce to

1
βit =λ
bi Pb (t) for all t and all i, (2.7)
ci (t)
b
∞
X ∞
X
Pb (t) b
ci (t) = Pb (t) for all i, (2.8)
t=0 t=0

c1 (t) + b
b c2 (t) = 2 for all t. (2.9)
On the other hand, given α = (α1 , α2 ), we have that (4.12) and (4.13)
reduce to

1
αi βit = µ∗ (t) for all n, all t and all i, (2.10)
c∗i (t)(α)
2.2. ARROW-DEBREU ECONOMIES 37

c∗1 (t) (α) + c∗2 (t) (α) = 2 for all t. (2.11)

 t
c∗ (t)(α) c∗ (t)(α)
Note that (2.10) implies that αα12 ββ12 = c∗1 (t)(α) .Note that c1∗ (t)(α) goes to
2 2
0 as t goes to infinity since agent 1 is more impatient and then agent 2 is
compensated in the future. When this equation is coupled with feasibility,
we obtain
α1 β1t
c∗1 (t) (α) = 2 ,
α1 β1t + α2 β2t
α2 β2t
c∗2 (t) (α) = 2 ,
α1 β1t + α2 β2t
α1 β1t + α2 β2t
µ∗ (t) (α) = .
2
∗
As before, let P (t) (α) = µµ∗ (0)(α)
(t)(α)
for all t and note that the last expression
P∞
implies that t=0 P (t, α) < ∞ for all α since 0 < β1 < β2 < 1. This is
important because otherwise agents would be infinitely rich and there would
be no competitive equilibrium.
The transfer function is given by
∞
X
A1 (α) = P (t) (α) (c∗1 (t) (α) − 1) ,
t=0
∞
1 X
= (α1 β1t − α2 β2t ),
α1 + α2 t=0
 
1 α1 α2
= − .
α1 + α2 1 − β1 1 − β2

where A2 (α) = −A1 (α).

Note that Ai (α)0 s are continuous and, very importantly, homogenous of

degree 0 with respect to α and then we can normalize α
b1 = 1. Thus, let α
b1 = 1
1−β2
b2 = 1−β1 to conclude that the corresponding allocation determines an
and α
ADCE allocation since in that case both transfers are 0.

A market structure consisting of a sequence of spot markets for trading in

real goods and a system of financial markets for trading income across time
38 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

and state of nature provides a more realistic description of observed markets

than the stylized structure described in the Arrow-Debreu economy. Now we
consider such an alternative market structure.

2.3 Sequential Markets Economies

Suppose that every period agents meet in spot real and financial markets
to trade consumption goods and S fully enforceable, one-period contingent
securities. We will consider a particular set of securities with a peculiar payoff
matrix known as Arrow securities.2
At every st , there are S different securities, one for each possible shock
realization in period (t + 1). Traded at st , agents can buy or sell an Arrow
security called s0 ∈ S, which pays one unit of consumption good 1 at t + 1
if st+1 = (st , s0 ) and 0 otherwise. Therefore, the payoff matrix is the S × S
identity matrix.
Let p (st ) ∈ RN ++ be the vector of prices for N consumption goods at s .
t

Let q (st ) (s0 ) be the price of the s0 −security at st where ai (st , s0 ) denotes
holdings of this security chosen by agent i at st . We normalize p1 (st ) = 1 for
all st and therefore all prices are in units of consumption good 1. We denote
(q, p) = ((q (st ))st , (p (st ))st ) as the price system.
To rule out Ponzi schemes, we restrict agents to bounded trading
strategies. That is, there exists a large negative number A such that
ai (st , s0 ) ≥ A for all (st , s0 ).3 We assume that A is large enough such that
these constraints are not binding in equilibrium. These securities are all in
zero net supply and the initial endowment is ai (s0 ) = 0 for all i.
Given (q, p), agent i’s problem in an economy with this market structure
is given by
XT X
max βit π(st ) ui (ci (st )), (SM-AP)
(ci ,ai )
t=0 st

subject to
2
Yes, Arrow invented too many things in economics and, then, he also had to share
some names to avoid our confusion. Here, we build on his insight, Arrow (1964), that
sequential trading of one-period securities are sufficient to implement PO allocations and
consequently equivalent to the set of ADCE allocations.
3
Below we define a more intuitive bound, the so-called natural debt limit.
2.3. SEQUENTIAL MARKETS ECONOMIES 39

X
p s t · ci s t + q st (s0 ) ai st , s0 ≤ p st · yi st + ai st for all st ,







s0
(BC-SM)

ci ∈ C, and ai (s0 ) = 0,
T


ai s , sT +1 = 0 if T < +∞, for all sT +1 ∈ S,
ai st , s0 ≥ A, for all st , s0 if T = +∞.

Definition 15 An allocation (b ai )Ii=1 and a price

ci )Ii=1 , portfolio holdings (b
system (b
q , pb) constitute a Sequential Markets Competitive Equilib-
rium (SMCE) if:
(SM.1) Given (b q , pb) , (b
ci , b
ai ) solves (SM-AP) for each i.
(SM.2) All markets clear. That is,

I
X
ci (st ) = y(st )
b for all st ,
i=1
X
ai st , s0 for all st and all s0 .


b = 0
i

We define the Lagrangian for (SM-AP), denoted LSM , as follows. Let

γi (st ) > 0 be the Lagrange multiplier for agent i corresponding to (BC-SM)
at st . Thus, given the price system (q, p)

T X
X
βit π(st ) ui (ci (st ))

LSM (ci , ai , γi ) =
t=0 st
+γi st p st · yi st + ai st − ...





#)
t

t

X t

0 t 0


... − p s · ci s − q s (s ) ai s , s .
s0

Necessary and sufficient conditions to characterize a SMCE are given by

a price system (b γi )Ii=1 and consumption
q , pb), a vector of Lagrange multiplier (b
bundles (bci )Ii=1 such that
40 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

∂ui
βit π st ci st = βit π st γbi st pbn st for all st and for all n,






b
∂cn
(2.12)

βit+1 π st , s0 γbi st , s0 = qb st (s0 ) βit π st γbi st for all st , (2.13)

X
ci s t +
pb st ·b qb st (s0 ) b
ai st , s0 = pb st · yi st +b
ai s t for all st ,







s0

(2.14)
ai sT , snT +1 = 0
(
b o for all sT +1 ∈ S
if T < +∞,
0= (2.15)
lim E0 (βi )t ∂ui
∂c1
ci (st )) b
(b ai (st ) if T = +∞
t→∞
I
X
ci (st ) = y(st )
b for all st , (2.16)
i=1
I
X
ai st , s0 = 0 for all st and all s0 .


b (2.17)
i=1

The first three conditions characterize a solution for (SM-AP) given (b

q , pb).
The last N + S conditions represent market clearing. Notice that Walras’
Law implies that one of these conditions is redundant.

2.4 Equivalence of Market Structures

Now we show that, under our assumptions, both sets of equilibrium allo-
cations coincide. This establishes the equivalence between these alternative
market structures.

Proposition 1 Suppose that (b c, b

a, qb, pb) constitute
 a SMCE.
 Then, there ex-
ists an Arrow-Debreu price system P such that b
b c, P constitute an ADCE.
b

Proof. If (b c, b γi )Ii=1 such that (2.12)-

a, qb, pb) is a SMCE, then there exist (b
(2.16) are satisfied. Construct the price system Pb and the vector of Lagrange
 I
multipliers λ bi as follows:
i=1
2.4. EQUIVALENCE OF MARKET STRUCTURES 41

bi = ∂ui (b
λ ci (s0 )) , for all i,
∂c1

Pb1 (s0 ) = 1 and Pbn (s0 ) = pbn (s0 ) for all n 6= 1.

Then, recursively, let

Pbn st = pbn st qb st−1 (st )...b for all (st−1 , st ) and all n.




q (s0 ) (s1 )

i.e. the price of the n − th consumption good at st is reexpressed in units of

consumption good 1 at t = 0.
We only need to check that (2.2) and (2.3) are satisfied since (2.4) is
satisfied by (2.16). By construction, note first that (2.2) is satisfied for all n
at s0 . Then, it follows by construction that for all st

βit π st γbi st pbn st = pbn st qb st−1 )(st βit−1 π st−1 γ bi st−1 ,

....
= pbn st qb st−1 (st )...b



q (s0 ) (s1 ) γ
bi (s0 ) ,
= Pbn st γ


bi (s0 ) ,
= Pbn st λ


bi .

where the last equality holds because pb1 (s0 ) = 1.

To check (2.3), multiply (2.14) at st by Pb1 (st ) = qb (st−1 ) (st )...b
q (s0 ) (s1 ).
t
Summing them up for all s and for t = 0, ..., k we get
42 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

k X
X k X
X
t t t
Pb1 st bai s t






P s · b
b ci s − yi s =
t=0 st t=0 st
k
XX
X
Pb1 st qb st (s0 ) b
ai (st , s0 ),


−
t=0 st s0
k
XX
Pb1 st bai s t



=
t=0 st
k XX
X
Pb1 st , s0 bai (st , s0 ),


−
t=0 st s0

= Pb1 (s0 ) b
ai (s0 )
X
Pb1 sk , sk+1 bai (sk , sk+1 ),


−
(sk ,sk+1 )
X
Pb1 sk+1 bai (sk , sk+1 ).


= 0−
(sk+1 )

Note first that if T < ∞, this holds for k = T and therefore b ai (sT , sT +1 ) =
T
0 for all (s , sT +1 ) and for all i. Thus, (2.3) is immediately satisfied.
On the other hand, if T = +∞ we have that

X 1 X k+1
∂ui
Pb1 sk+1 bai (sk+1 ) = π sk+1 ci sk+1 b ai (sk+1 )




β b
λ
bi T +1 ∂ci1
sk+1 s
 
1 k+1 ∂ui
= E β (b
ci,k+1 ) b
ai,k+1 .
bi 0
λ ∂c1

It follows by (2.15) that

 
1 k+1 ∂ui k+1


k+1
E β ci s ai (s ) → 0,
bi 0
b b
λ ∂cn
and therefore (2.3) is satisfied for all i.

Now we show that the converse is also true.

2.4. EQUIVALENCE OF MARKET STRUCTURES 43

Proposition 2 Suppose that (b c, Pb) constitute an ADCE. Then, there exists

(b
a, qb, pb) such that (b
c, b
a, qb, pb) constitute a SMCE.

Proof. If (b
c, Pb) constitute an ADCE, then there exists Lagrange multipliers
 I
λ
bi such that (2.2)-(2.4) are satisfied. It is sufficient to find (b
a, qb, pb) and
i=1
γi )Ii=1 such that (2.12)-(2.16) are satisfied. We proceed by construction as
(b
follows:
∂ui
bi st = ci s t for all st and for all i,




γ b
∂c1
0 0 b1 (st , s0 )
γ
t
for all (st , s0 ),


qb s (s ) = βi π (st , s ) t
γ
b1 (s )
1 ∂u 1
pbn st = t
for all st and for all n.




c 1 s
b1 (st ) ∂cn
b
γ
Very importantly, note that (2.2) implies that the last two conditions hold
for all i > 1 and then agent 1 was chosen without loss of generality. Finally,
we need to construct portfolios supporting this allocation. To do that, define
for each agent and for all st
XT X P1 (st+j )
ai s t t+j t+j t+j





= p s · c i s − y i s
P1 (st )
b b b
j=0 t+j t
s |s
T X P1 (st+j )
t
t t

 X
pb st+j · b
ci st+j − yi st+j






= pb s · bci s − yi s + t
j=1 t+j t
P1 (s )
s |s

Note that
XT X P1 (st+j )
pb st+j · b
ci st+j − yi st+j




t
P1 (s )
j=1 t+j t
s |s
T X
X X P1 (st+j )
q st (st+1 ) pb st+j · b
ci st+j − yi st+j





= ... t t
j=1 st+1 st+j
P1 (s ) q (s ) (st+1 )
T X
X X X P1 (st+j )
q st (st+1 ) t+j t+j t+j





= ... p s · c i s − y i s
P1 (st+1 )
b b
j=1 st+1 st+2 st+j
X
q st (st+1 ) b
ai st , st+1 ,



=
st+1
44 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

where the last equality follows by construction of b ai (please, convince your-

T T
self). Clearly, if T < ∞ b
ai (s , sT +1 ) = 0 for all (s , sT +1 ) and for all i.
We define a natural bound for security trading as follows

XT X P1 (st+j )
Ai (st ) = − pb st+j · yi st+j .



t
(2.18)
j=0 j+t t
P (s )
s /s

This is the so-called natural debt limit. It is the value of the maximal amount
that agent i can repay starting from that period t, assuming that his future
consumption is 0 forever; i.e. it will not be feasible for any agent to repay
more than Ai (st ).
Finally, if T = +∞ we need to show that the transversality condition
holds. To do this, notice that t st kP (st )k < ∞ implies that lim P1 (st ) =
P P
t→∞
0 (more technically, st -almost surely). Thus, when this coupled with (2.2)
we have that
bi P1 (st ) = β t π st ∂ui b ci s t → 0




λ i
∂ci1
as t goes to infinity (again, st almost surely) and consequently (2.15) is
satisfied since the supporting asset holdings are uniformly bounded by (2.18).

Corollary 1 Any SMCE allocation is PO.

2.5 Application: Economies with Government

and Ricardian Equivalence
Assume that there is only one consumption good for simplicity (i.e.,
N = 1). Suppose that there is a government that needs to finance a given
stochastic sequence {gt } of government expenditures, where g (st ) is the real-
ization of this process at st . Government consumption is not explicitly valued
and then we can assume that these units are thrown away. We assume that
the government has two instruments to finance its expenditures: (i) lump-
sum taxes, denoted by {τt }, where τt = {τi }Ii=1 ; (ii) government contingent
bonds denoted by {bt }, where bt > 0 represents net debt. We call (g, τ, b) a
fiscal policy.
Consider first an economy with sequential markets.
2.5. APPLICATION: ECONOMIES WITH GOVERNMENT AND RICARDIAN EQUIVALENCE45

Definition 16 A Sequential Market Competitive Equilibrium for this econ-

omy with government is an allocation b c = (b ci )Ii=1 , portfolio holdings b a =
I
ai )i=1 and a price system {b
(b q } such that, given a fiscal policy (b g , τb, b)
b
0
(SMG.1) Given {b q } and (b
g , τb, b), {b
b ai } solves the agent i s problem:
ci , b
T X
X
max βit π(st ) ui (ci (st )), (SMG-AP)
(ci ,ai )
t=0 st

subject to

X
ci s t + q st (s0 ) ai st , s0 ≤ yi st − τbi st + ai st for all st ,






s0
(BC-SMG)

ci ∈ C, and ai (s0 ) = 0,
T


ai s , sT +1 = 0 if T < +∞, for all sT +1 ∈ S,
b
ai st , s0 ≥ A, for all st , s0 if T = +∞.

(SMG.2) Given qb, (b

g , τb, bb) satisfy the government budget constraint:
I
X X
gb (st ) + bb st qb st (s0 ) bb st , s0 ∀st




= τbi (st ) + (GBC)
i=1 s0
X
Yt
b (s0 ) = 0 and lim eb st qb(sj−1 )(sj ) = 0 for all st .
t→∞ j=1
st

(SMG.3) Markets clearing. For all st ,

X
ci st + gb st = y st ,




b
i
X
ai st , s0 = bb st , s0 ∀s0 .



b
i

Consider now an economy where there is contingent trading only at date

0. Government bonds are not used now.

Definition 17 An Arrow-Debreu Competitive Equilibrium for this economy

with government is an allocation {e
ci } and a price system Pe such that given
a fiscal policy (b
g , τb):
46 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

ci } solves the agent i0 s problem

g , τb), {e
(ADG.1) Given Pe and (b
XX
max βit π(st )ui (ci (st )),
ci ∈C
t st
subject to XX
Pe st ci st − yi st − τbi st






≤ 0.
t st

(ADG.2) Given Pe, (b

g , τb) satisfy the government budget constraint:
" #
XX X
Pe st gb st − τbi st = 0




t st i

(ADG.3) Markets clearing. For all st ,

X
ci (st ) + gb (st ) = y st .


e
i

Remark 7 As before, there is a one-to-one relationship between SMCE and

ADCE in this economy with government. You must convince yourself of
these equivalence results. In Exercise 9 you are asked to prove the one-to-one
relationship between SMCE and ADCE in this economy with government.

Given qb (st ), construct the corresponding (implicit) Arrow-Debreu prices

recursively as follows:
Pe (s0 ) = 1,
Pe st = Pe st−1 qb st−1 )(st ∀ st−1 , st .

and denote τ (st ) = τi (st ) for all st .

P
i
 
Proposition 3 (Ricardian Equivalence) Given gb, τb, bb , let {b a, qb} be
c, b
n o
a SMCE. Let τe, eb be any alternative policy such that:

X
∀ t, st gb st + eb st = τe st + qb st (st+1 )eb st , st+1 ,

st+1
(2.19)
bb (s0 ) = eb (s0 ) = 0,
X
Yt
lim eb st qb(sj−1 )(sj ) = 0,
t→∞ j=0
st
2.5. APPLICATION: ECONOMIES WITH GOVERNMENT AND RICARDIAN EQUIVALENCE47

and XX
Pb st τbi st − τei st = 0, for all i.





(2.20)
t st
n o
a such that {b
Then, there exists e a, qb} constitutes a SMCE given gb, τe, b .
c, e e

Within this class of government policies, the way the government finances
its expenditures is irrelevant and, consequently, it does no have any equilib-
rium implications. Two assumptions are critical to obtain this result. First,
taxes are lump sum and, thus, they do not distort incentives at the margin.
Secondly, (2.20) restrict the set of policies to those with same net income
effect for each agent.

Proof. We use the one-to-one equivalence between ADCE and SMCE with
government provedQtin Exercise 9.
Let P (s ) = j=0 qb(s )(sj ) = Pb (st−1 ) qb(st−1 )(st ) be defined as before,
b t j

which are the same Arrow-Debreu prices with a single good, and Pb (s0 ) = 1.
Multiplying equation (2.19) at st by Pb (st ), we can write

X
gb st Pb st +eb st Pb st = τe st Pb st + Pb st , st+1 eb st , st+1 ∀t, st ,








st+1

Adding them up for all st and up to T , we obtain

T X T X
!
X
t t t t

 X
t t
X
Pb st , st+1 eb st , st+1








gb s Pb s + eb s Pb s = τe s Pb s + .
t=0 st t=0 st st+1

P e t b t
Look that for 0 < t < T + 1, we can simplify st b (s ) P (s ) on both sides.
Hence
T X
X T X
t t

X
X
τe st Pb st + Pb st , st+1 eb st , st+1 .





gb s P s =
b
t=0 st t=0 st sT +1

Taking limit T → ∞ and using the transversality condition

∞ X
X ∞ X
tt

X
τe st Pb st




gb s P s =
b
t=0 st t=0 st
48 CHAPTER 2. OPTIMALITY AND COMPETITIVE EQUILIBRIUM

Therefore, the Arrow-Debreu budget constraint for the government does

not change. Since the Arrow-Debreu budget constraint for the individuals
does
 not
 change by construction (see condition (2.20)), we can conclude that
bc, P constitute an ADCE given the fiscal policy (b
b g , τe).
Exercise 9 completes the proof.

Exercises
Exercise 6 In text.

Exercise 7 Prove Theorem 10 for T < ∞.

HINT. Define U ={(ui )Ii=1 ∈ RI : there exists some feasible allocation
(ci )Ii=1 such that Ui (ci ) ≥ ui for all i}. U is denoted as the Utility Possibility
Set. Show that U is convex and closed. Show that any PO allocation is at
the boundary of U. Apply the Supporting Hyperplane Theorem properly.
T −t T −t
Exercise 8 Show that (b ci (st+j | st ))j=0 = (c∗i (st+j | st ))j=0 . That is, under
our assumptions agents would not choose to re-trade.
(Hint: show that if agents choose to re-trade, then the original ADCE
allocation can be ex-ante Pareto dominated. This would contradict the First
Welfare Theorem).
 
Exercise 9 Prove that if e c, Pe constitutes an ADCE given {b g , τb}, then
n o  
there exists b a, qb, b such that (b
b c, b
a, qb) constitutes a SMCE given gb, τb, b . b

 Prove  also that if, on the other hand,{b c, b, qb} constitutes a SMCE given
a
gb, τb, bb , then there exists Pe such that b c, Pe constitutes an ADCE given
(b
g , τb).
Chapter 3

The One-Sector Neoclassical

Growth Model

Now we extend the previous setting to include production opportuni-

ties and, thus, we obtain the celebrated Neoclassical Growth Model. This
is an intertemporal general equilibrium model where agents undertake both
consumption and production decisions.
The primitives that define this economy are the set of agents, the com-
modity space, the consumption set, preferences, production opportunities
and endowments. This is a private ownership economy and, then, we need
to define property rights properly.

3.1 The Economy

We consider an economy populated by I (types of) infinitely-lived agents
where i ∈ I = {1, ..., I} represents a typical consumer. Time is discrete and
denoted by t = 0, 1, 2, ...where T = ∞ (infinite horizon). There is only one
consumption good which can be either consumed or invested. Goods invested
transform one-to-one in new capital next period. There is a constant returns
to scale aggregate technology to produce the consumption good operated by
a representative firm. This technology is subject to productivity shocks and
st ∈ St = {ζ1 , ..., ζS } represents the realization of this shock at date t. We
assume that {st } is a finite state first-order stationary Markov process. Let
st = (s0 , ..., st ) ∈ S t = {s0 } × S1 × ... × St represent the partial history of
events up to date t. We write st+1 /st to denote that st+1 is an immediate

49
50CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

successor of st . These histories are observed by all the agents. Transition

probabilities are denoted by π(s, s0 ) > 0. The probability of st is constructed
from π in the standard way using (1.6) and x(st ) denotes the value of x at
the node st .
Each agent is endowed with one unit of time every period. They can use
that time either to consume leisure or to work. Thus, li (st ) represents labor
supplied by agent i at st (and thus 1 − li (st ) is the leisure). Aggregate labor
at st is given by L (st ) = i li (st ) .
P
A consumption bundle for agent i is a sequence of functions {ct , lt }∞ t=0 .
Agent i’s consumption set, C, is defined by

C = {{ct , lt }∞ t t
t=0 : (ct , lt ) : S → R+ × [0, 1], sup c(s ) < ∞}.
(t,st )

Agent i’s preferences on C are represented by expected, time-separable,

discounted utility. That is , if (ci , li ) ∈ C then:
∞ X
X
β t π(st )ui (ci (st ), 1 − li st ),


U (ci , li ) =
t=0 st

where β ∈ (0, 1) and ui : R+ × [0, 1] → R+ is strictly increasing, strictly

concave, differentiable such that uic (0, 1 − l) = uil (c, 0) = ∞ for all (c, 1 − l) >
0.1

Production Possibilities

Let K(st ) ≥ 0 denote the aggregate stock of capital determined at the

node st and available in period t + 1 to produce with the aggregate technol-
ogy. The depreciation rate is given by δ ∈ (0, 1). Let sF (K, L) represent
this technology, where K is the stock of capital available, L is the level of
labor/employment and s is the productivity shock. F : R2+ → R+ is ho-
mogeneous of degree 1, strictly concave, strictly increasing and continuously
differentiable in both arguments.2 We also assume that for all (K, L) > 0 (a)
∂F (0, L)/∂K = ∂F (K, 0)/∂L = ∞, (b) limK→∞ ∂F (K, L)/∂K = 0 and (c)
F (K, 0) = F (0, L) = 0.
1
Note that uil is the partial derivative of ui with respect to its second argument (i.e.
leisure).
2
Remember that under constant returns to scale, any price taking industry behaves as
a single representative price taking firm.
3.2. PO ALLOCATIONS: THE PLANNER’S PROBLEM 51

An allocation (c, l, L, K) is feasible if (ci , li ) ∈ C for all i, 0 ≤ K(st ) for

all st and
I
X
ci (st ) + X(st ) = st F (K(st−1 ), L(st )),
i=1
X
L(st ) = li st .


i

Condition (a) above rules out corner solutions for capital accumulation.
Very importantly, condition (b) guaranties that there exists some (0, K) such
that 0 ≤ K(st ) ≤ K for all st for all feasible allocations since consumption
must be bounded from below and δ ∈ (0, 1). To see this, let s = max S and
define K such that δK = sF (K, I), where I is the upper bound of labor
supply. Note that δK > sF (K, I) for all K > K given strict concavity of
F and condition (b) above. Thus, without loss of generality, we can restrict
K(st ) ∈ X ≡ [0, K] for all st whenever the initial stock of capital satisfies
0 ≤ K0 ≤ K. On the other hand, if K0 > K it can be shown that in finite
time 0 ≤ K(st ) ≤ K with probability 1.
In any case, in general we can define X ≡ [0, max{K0 , K}] as the set of
sustainable levels of capital and thus guarantee that K(st ) ∈ X for all st . We
denote K = (K(st )st ) as a capital accumulation path.

3.2 PO Allocations: The Planner’s Problem

Define the utility possibility set as follows

(ui )Ii=1 ∈ RI : ∃ a f easible allocation (c, l, L, K),


U(s0 , K0 ) ≡
Ui (ci , li ) ≥ ui ∀i, given (s0 , K0 )} .

Under our concavity assumptions on both utility and the production func-
tions, it can be shown that U(s0 , K0 ) is convex and closed for all (s0 , K0 ), the
supporting hyperplane theorem can be applied and, consequently, the set of
Pareto optimal allocations can be parametrized by welfare weights.
Suppose that αi is the welfare weight assigned by the planner to agent i.
Given a vector of welfare weights α = (αi )Ii=1 , the planner’s problem is given
by:
52CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

I
(∞ )
X XX
t
β t π s ui ci s , 1 − li s
t t





max αi , (PP-NGM)
( (ci ,li )Ii=1 ,K ) i=1 t=0 st

subject to
I
X
ci (st ) + X(st ) = st F (K(st−1 ), L(st )) for all st ,
i=1
(ci , li ) ∈ C, K(st ) ∈ X for all st and all i,
K 0 ∈ X given.
PI
where X (st ) = K(st ) − (1 − δ)K(st−1 ) and L (st ) = i=1 li (st ) .

Characterizing the Solution

To abridge notation, we write

uin st = uin ci st , 1 − li st





for n = c, l,
Fj st = Fj K st−1 , L st





for j = K, L.

We define the Lagrangian for (PP-NGM), LN GM t

P P , as follows. Let µ (s )
be the Lagrange multiplier corresponding to the feasibility constraint at st
and then

∞ X
I
!
X X
LN GM
β t π s ui ci s , 1 − li s
t t t





P P (c, l, K, µ, α) = αi
i=1 t=0 st
∞ X
X I
X
µ st (st F (K(st−1 ), li st )



+
t=0 st i=1
I
X
+(1 − δ)K(st−1 ) − ci st − K(st )).


i=1

A natural combination of results from Chapter 1 shows that the intratem-

poral first order conditions, the Euler Equations and the transversality con-
dition are necessary and sufficient to characterize an interior solution for
(PP-NGM). Those are given by
3.3. COMPETITIVE DECENTRALIZATION:AN ARROW-DEBREU FRAMEWORK53

ci s t : αi β t π st uc st = µ st ,





(3.1)

li st : αi β t π st u l st = µ st F L st ,






(3.2)

X
k st : µ st = µ st , s0 s0 Fk st , s0 + (1 − δ) ,




(3.3)
s0
( )
X
T T T −1





lim µ s sT F k s + (1 − δ) K s =0 (3.4)
T →∞
sT

plus the feasibility conditions.

3.3 Competitive Decentralization:

An Arrow-Debreu Framework
Suppose that all trade is done at date 0. That is, agents can sign (and
fully enforce) contingent contracts for each st at date 0. Let P (st ) ∈ R+
be the price of the consumption good to be delivered at the node st . We
normalize P (s0 ) = 1 (all prices are in units of the date 0-consumption PI good).
Agent i is endowed at date 0 with ki0 > 0 units of capital where i=1 ki0 =
K0 . In this economy agents accumulate capital to rent it to the firms. Given
that we assume constant returns to scale and competitive factor markets,
we can consider without loss of generality that there is only a representative
firm operating the aggregate technology.
Let (w (st ) , r(st )) denote the wage and the rental price of capital at st ,
respectively, in units of the st -consumption good. Thus, given the factor
prices (w (st ) , r(st )) at st , the representative firm rents labor and capital to
maximize profits. Given (w (st ) , r(st )), the firm’s problem at st is

π(st ) = max st F K d , Ld − r(st )K d − w st Ld ,




(FP-NGM)
(k,l)≥0

where we denote K d (st ), Ld (st ) the solutions for this problem at st . Observe
that the representative firm does not face any intertemporal decision and,
54CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

thus, maximization of the present discount value of profits is equivalent to

period-by-period (static) profit maximization given by (FP-NGM). Given
that F is homogeneous of degree 1, equilibrium profits will always be zero
(i.e. π(st ) = 0 for all st ).
Given (P, w, r) , agent i’s problem is given by

∞ X
X
β t π st ui ci st , 1 − li st ,





max (AP-NGM)
(ci ,li ,ki )
t=0 st

subject to

∞ X
X
P st ci st + ki st − (1 − δ)ki (st−1 )





t=0 st
X∞ X
P st (r(st )ki (st−1 ) + w st li st ).




≤
t=0 st

such that (ci , li ) ∈ C, ki (st ) ≥ 0 for all st and ki0 > 0 given.

Definition 18 An Arrow-Debreu Competitive Equilibrium (ADCE)

is an allocation (b
c, b
l, b b d, L
k, K bd ) and a price system (Pb, w, b rb) such that
 
(ADCE.1) Given (P , w,b b rb), b ci , li , ki solves the agent i’s problem above.
b b
 
(ADCE.2) Given (Pb, w, b rb), K b d, L bd solves the firm’s problem.
(ADCE.3) All markets clear. That is, for all st

X
b d st = li st ,


L b
i

I
d t

X
ki st−1 ,


K s =
b b
i=1

I 
X
 
d t

ci s t + b
ki st − (1 − δ) b
ki st−1 = st F Kb d st , L



b b s .
i=1
3.4. WELFARE THEOREMS 55

Characterizing an ADCE

First note that (FP-NGM) is a standard static concave problem and thus
it can characterized with necessary and sufficient first order conditions. For
an interior solution, those are given by

st Fk k d (st ), ld (st ) = r(st ),

(3.5)
st Fl k d (st ), ld (st ) = w st .

for all st . As already mentioned, these conditions imply that equilibrium

profits are zero given that F is HOD 1.
Let λi represent agent i’s Lagrange multiplier for the budget constraint.
Then, the necessary and sufficient conditions to characterize an interior so-
lution for (AP-NGM) are given by

ci s t : β t π st uic st = λi P st ,





(3.6)

li st : β t π st uil st = λi P st w st ,






(3.7)

X
k st : P st = P st , st+1 r(st , st+1 ) + (1 − δ) ,




(3.8)
st+1

∞ X
X
P st [ci st + xi st − r(st )ki (st−1 ) − w st li st ] = 0






(3.9)
t=0 st
( )
X
T T
+ (1 − δ) k sT −1





lim λi P s sT Fk s =0 (3.10)
T →∞
sT

where xi (st ) = ki (st ) − (1 − δ)ki (st−1 ) is individual investment.

3.4 Welfare Theorems

Let us briefly analyze the welfare properties of the neoclassical growth
model.
56CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

3.4.1 The First Welfare Theorem

Proposition 4 Suppose that [(b c, b
l, b b d, L
k, K bd ), (Pb, w,
b rb)] determine an ADCE.
d bd
Then, (b
c, l, k, K , L ) is a PO allocation.
b b b

Proof. Check that any allocation (b c, b

l, b b d, L
k, K bd ) satisfying (3.5)-(3.10) will
also satisfy (3.1)-(3.4), for some vector of welfare weights.

3.4.2 The Second Welfare Theorem

Here we describe how to decentralize any Pareto optimal allocation parametrized
by α as an ADCE with transfers. As before, we will proceed constructively.
Let (c(α), l(α), K(α)) be the solution to the planner’s problem (PP-NGM)
corresponding to α where µ (α) µ(α) is the corresponding vector of Lagrange
multipliers. For each st , define the implicit price system by
P (st )(α) = µ(st )(α)/µ(s0 )(α),
r(st )(α) = st Fk K(st−1 )(α), L(st )(α) ,

w(st )(α) = st Fl K(st−1 )(α), L(st )(α) .

The value of individual consumption for agent i is

∞ X
X
VCi (s0 , K0 , α) = P (st )(α)ci (st )(α).
t=0 st

In the same way, the value of individual human wealth for agent i is
∞ X
X
VWi (s0 , K0 , α) = P (st )(α)w(st )(α)li (st )(α).
t=0 st

We need to define transfers to decentralize the PO allocation correspond-

ing to α. First, given an aggregate initial condition (s0 , K0 , α), define the
implicit value of the firm as follows
XX
P (st )(α)[st F K(st−1 )(α), l(st )(α)


VF (s0 , K0 , α) = (3.11)
t st
+(1 − δ)K(st−1 )(α) − K(st )(α) − w(st )(α)l(st )(α)],
XX
= P (st )(α)[r(st )(α)K(st−1 )(α) + (1 − δ)K(st−1 )(α) − K(st )(α)].
t st
3.4. WELFARE THEOREMS 57

where the last line follows by homogeneity of degree 1.

Finally, let θi (α) be defined such that

VF (s0 , K0 , α)θi (α) = VCi (s0 , K0 , α) − VWi (s, K, α),

for all i ∈ I. Defines agent i’s transfer at date 0 by

Ai (s0 , K0 , α) = θi (α)VF (s0 , K0 , α). (3.12)

The planner owns the stock of capital in this setting. Agents’ Euler equa-
tions corresponding the capital accumulation (3.8) will hold by construction.
Thus, agents are indifferent about the level of capital they want to accu-
mulate. Remember that capital accumulation is redundant for insurance
purposes: markets are complete at date 0 and then they can trade any af-
fordable contingent consumption bundle they wish. Thus, we need to make
the corresponding allocation affordable for each agent and define a distribu-
tion of the aggregate stock of capital across agents.

 
Proposition 5 Let (b c, b
l), K
b be a PO allocation. Then, there exist a price
b and individual capital holdings (ki )Ii=1 such that constitute an
system (Pb, w)
ADCE with transfers given by 3.12.

Proof. Note that by definition

Ai (s0 , K0 , α) = θi (α)VF (s0 , K0 , α),

= VCi (s0 , K0 , α) − VWi (s0 , K0 , α),

and, then, agent i0 s budget constraint is satisfied. Agent i uses this transfer
to generate additional wealth through capital accumulation and thus make
this allocation affordable. With this purpose, the planner assigns capital as
follows:
ki (st )(α) = θi (α)K(st )(α) for all st .

Note that following this individual capital accumulation plan, agent i

58CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

obtains
XX
P (st )(α)[r(st )(α)ki (st−1 )(α) + (1 − δ)ki (st−1 )(α) − ki (st )(α)]
t st
XX
= P (st )(α)[r(st )(α)θi (α)K(st−1 )(α) + (1 − δ)θi (α)K(st−1 )(α) − θi (α)K(st )(α)]
t st
XX
= θi (α) P (st )(α)[r(st )(α)K(st−1 )(α) + (1 − δ)K(st−1 )(α) − K(st )(α)],
t st
= θi (α)VF (s0 , K0 , α) = Ai (s0 , K0 , α).

Thus, it makes the consumption allocation affordable while agents have in-
centives to keep these capital holdings.

Capital Accumulation by the Representative Firm

Alternatively, we can consider the following setting. At dateP

0 agents are
endowed with a share θi0 > 0 of the representative firm where Ii=1 θi0 = 1.
Assume that firms do not issue more shares. This firm accumulates capital
and hires labor to maximize its vale, given by (3.11).

Exercise 10 (i) Carefully define an ADCE for this economy.

(ii) Show that under constant returns to scale, any ADCE in this setting
is equivalent to the economy previously described where capital accumulation
was carried out by the individuals such that ki0 = θi0 K0 .

Exercise 6 implies that to decentralize a PO allocation α, we can define

transfers by:
Ai (s0 , K0 , α) = [θi (α) − θi0 ]VF (s0 , K0 , α),
and, therefore, the actual share of the firm owned by agent i is simply
θi (α)VF (s0 , K0 , α) as desired.
Note that (K(st , α)) satisfies (3.3)-(3.4) and consequently this capital
accumulation path maximizes thevalue of the firm givenα.
It follows that the allocation (b li (α))Ii=1 , K(α)
ci (α), b b can be decentral-
ized as an alternative ADCE with transfer as discussed above in Exercise 6.
It is critical to understand that constant returns to scale play a prime role
to obtain this result.
3.4. WELFARE THEOREMS 59

3.4.3 Existence: On The Negishi’s Approach

Let (c(α), l(α), K(α)) be the solution to the planner’s problem (PP-NGM)
corresponding to α where µ (α) is the vector of Lagrange multipliers. We
take as given the initial distribution of shares of the representative firm (or,
equivalently, P
the date 0 distribution of the initial stock of capital), denoted
θi > 0 with Ii=1 θi0 = 1. Given (s0 , K0 , θ0 ), define the net transfers needed
0

to decentralize the PO allocation α as follows

τi (s0 , K0 , θ0 )(α) = VCi (s0 , K0 , α) − VWi (s, K, α) − θi0 VF (s0 , K0 , α).

We follow Negishi’s [1960] approach to prove the existence of an ADCE

without transfers in this setting.

Proposition 6 (Negishi) Given (s0 , K0 , θ0 ), there exists α0 = α(s0 , K0 , θ0 ) ∈

RI+ such that
τi (s0 , K0 , θ0 )(α0 ) = 0, for all i.

Proof. Note first that if αi = 0 then ci (s, K, α) = 0 and consequently

ViC (s, K, α) = 0 for all (s, K) in this case. This implies that

τi (s0 , K0 , θ0 )(α) = −VWi (s, K, α) − θi0 VF (s, K, α), (3.13)

and thus τi (s0 , K0 , θ0 )(α) < 0 for all α. Note also that
I
X I
X I
X I
X
0
τi (s0 , K0 , θ )(α) = VCi (s0 , K0 , α) − VWi (s, K, α) − θi0 VF (s0 , K0 , α)
i=1 i=1 i=1 i=1
I
!
X
= VC (s0 , K0 , α) − VWi (s, K, α) + VF (s0 , K0 , α) ,
i=1

where VC (s0 , K0 , α) denotes the value of aggregate consumption. It follows

by the definition of VF and feasibility that
I
X
VF (s0 , K0 , α) + VWi (s, K, α) = VC (s0 , K0 , α),
i=1

PI 0
and thus i=1 τi (s0 , K0 , θ )(α) = 0 for all α.
60CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

We can establish the following properties.

(1) Since (c(α), l(α), K(α)) are HOD zero in α, it follows from the FOC’s
corresponding to the planner’s problem that µ (α) is HOD one in α. Thus,
by definition, τi (s0 , K0 , θ0 )(α) is HOD zero in α for each i.
(2) (Informal - Part I) Due to the celebrated Theorem of the Maximum,
it can be shown that (c(α), l(α), K(α)) are continuous in α. It follows from
the FOC’s corresponding to the planner’s problem that µ (α) is continuous
in α. Thus, we can conclude that τi (s0 , K0 , θ0 )(α) is continuous in α for each
i.
(3) It follows by strict concavity of both utilities functions and the pro-
duction function that τi (s0 , K0 , θ0 )(α) is uniquely determined.
Normalize welfare weights in the I − 1 dimensional unit simplex, ∆I−1 .
Given (s0 , K0 ), define the vector-valued function g as follows:
max[αi − τi (s0 , K0 , θ0 )(α), 0]
gi (α) = PI ,
h=1 max[αh − τh (s0 , K0 , θ0 )(α), 0]
for each i. Note that
I
X I
X I
X
0
H(α) ≡ max[αh −τh (s0 , K0 , θ )(α), 0] ≥ αi − τh (s0 , K0 , θ0 )(α) = 1,
h=1 h=1 h=1

for all α ∈ ∆I−1 . Also, observe that gi (α) ∈ [0, 1] and

P
i gi (α) = 1 for
I−1
all α. Consequently, g is a continuous function mapping ∆ into itself. It
follows from the Brower’s fixed point theorem that there exists some α0 =
α(s0 , K0 , θ0 ) ∈ ∆I−1 such that α0 = g(α0 ).
Suppose now that αi0 = 0 for some i. By definition of gi (α0 ), this implies
that −τi (s0 , K0 , θ0 )(α0 ) ≤ 0. But (3.13) implies that −τi (s0 , K0 , θ0 )(α0 ) > 0
if αi0 = gi (α0 ) = 0. This would lead to a contradiction and, hence, αi0 > 0 for
each i. This implies that αi0 − τi (s0 , K0 , θ0 )(α0 ) > 0 for all i and consequently

H(α0 )αi0 = max[αi0 − τi (s0 , K0 , θ0 )(α0 ), 0] = αi0 − τi (s0 , K0 , θ0 )(α0 ),

for each i. Note that this implies that

X X X
H(α0 ) αi0 = αi0 − τi (s0 , K0 , θ0 )(α0 ) = 1 = H(α0 ).
i i i

Therefore, αi0 = αi0 −τi (s0 , K0 , θ0 )(α0 ), which immediately shows that τi (s0 , K0 , θ0 )(α0 ) =
0 for each i.
3.5. ON THE DYNAMICS OF THE DETERMINISTIC VERSION 61

The importance of Negishi’s strategy to prove the existence of a com-

petitive equilibrium is that it changes the dimension of the problem in a
fundamental way: instead of looking for an equilibrium price system (an infi-
nite dimensional object), it looks for a particular vector of welfare weights (a
finite dimensional object) to back up equilibrium prices. This can be done,
however, imposing stronger assumptions on preferences and technologies.

3.5 On the Dynamics of the Deterministic

Version
We will consider a rather simplified version of the previous model to study its
dynamic properties. Suppose that I = 1 to consider an economy populated
by identical agents with total measure 1 and names in the unit interval [0, 1].
Also, we assume that S = {1} to get a deterministic setting. Representative
agent’s preferences are restricted such that leisure is not valued; i.e., u (c, l) =
u (c) for all l.
Since both welfare theorems hold in this setting, consider the conditions
characterizing the unique PO allocation given by

u0 (ct ) = µt , (3.14)

µt = βµt+1 (f 0 (kt+1 ) + (1 − δ)) , (3.15)

lim µT (f 0 (kT ) + (1 − δ)) kT = 0, (3.16)

T →∞

ct + kt+1 − (1 − δ) kt = f (kt ) = F (kt , 1) . (3.17)

where µt is the Lagrange multiplier corresponding to the feasibility constraint

in period t. Note that (3.14) determines an implicit function c (µ) , where
lim c (µ) = 0, lim c (µ) = +∞, and c (µ) is strictly decreasing.
µ→∞ µ→0
62CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Figure 3.1: Locus A

Thus, the unique interior solution will be characterized by a system of

difference equations

kt+1 = f (kt ) + (1 − δ) kt − c (µt ) ,

µt = βµt+1 (f 0 (kt+1 ) + (1 − δ)) .

To get the standard writing, we could also write this last equation as

−1
µt+1 = µt [β (f 0 (f (kt ) + (1 − δ) kt − c (µt )) + (1 − δ))] .
Once we obtain (c0 , k0 ) , we can determine the solution through itera-
tion. But c0 (and µ0 ) is a choice variable and, indeed, the second necessary
condition to determine the solution will be given by (3.16).
Consider first the locus (A) such that k = f (k) + (1 − δ) k − c (µ) , and
thus c (µA (k)) = f (k) − δk. We can obtain

∂µA f 0 (k) − δ
= 0 .
∂k c (µA (k))
| {z }
<0


Let k be such that f k = δk. Note that when k → k, c goes to zero and
thus µ → ∞.
3.5. ON THE DYNAMICS OF THE DETERMINISTIC VERSION 63

Figure 3.2: Locus B

Note that if µ > µA (k), then c (µ) < c (µA (k)) and thus kt+1 − kt > 0. If
µ < µA (k), then c (µ) > c (µA (k)) and thus kt+1 − kt = k < 0. The arrows in
the graph indicate these properties.
Consider now the other locus (B) where µt+1 = µt = µ. Hence,

β (f 0 (k ∗ ) + (1 − δ)) = 1,

where k ∗ is denoted the golden rule steady state capital level. Here, µ satisfies
k ∗ = f (k) + (1 − δ) k − c (µ) and then µB (k) is defined by

c (µB (k)) = (f (k) − δk) + (k − k ∗ )

Several properties can be established. First, if k = k ∗ , then µA (k ∗ ) =
0
µB (k ∗ ). Also, note that ∂µ∂k
B (k)
= f c(k)+(1−δ)
0 (µ (k))
B
< 0. Very importantly, µB (k) is
not well defined when k is ”too low”. More precisely, we need that

f (k) + (1 − δ) k ≥ k ∗
Thus, let k be defined by f (k) + (1 − δ) k = k ∗ and note that µB (k) = +∞.
To see the dynamics note that if µ > µB (k) , then c is relatively low, k t+1
is relatively high (f 0 (kt+1 ) is low) and thus µt+1 > µt .
We need to show that the only solution satisfying (3.14)-(3.17) will assign
µ0 such that (k0 , µ0 ) belongs to the stable arm given by the arrows.
We need to show that regions I and III are absorbent. That is, once the
system gets in there, it does stay there for all consecutive periods. Regions
II and IV are transient; i.e. possibly except for one path, the dynamics will
make the system to get into either I or III. Note that there is at least one
path in region II and IV converging to (k ∗ , µ∗ ).
64CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Figure 3.3: Phase Diagram Deterministic NGM

Claim 4 No PO allocation can take values in region I.

Proof. Note that any solution that enters in region I moves in a southwest
direction. This implies that µ(t) → 0 and then c (µ(t)) → +∞. Also, kt
decreases monotonically. But then there exists some Te such that c µ(Te) ≥
f kTe + (1 − δ) kTe . But this implies that the solution {c∗t } imposes c∗t = 0

∀t ≥ Te, and this leads to a contradiction since the Inada conditions hold.

The idea is that the aggregate state of capital is 0 in finite time and,
then, consumption is 0 from that point in time on. No solution can satisfy
that property since the Inada Conditions implies that the unique solution is
interior.

Claim 5 No PO allocation can take values in region III.

Proof. Assume that the system is in region III at date t where:

−1
β t+1 µt+1 = β t µt [1 − δ + f 0 (kt+1 )]
The sequence {kt } is monotonically increasing and thus it either converges
to some point or goes to infinity.
3.5. ON THE DYNAMICS OF THE DETERMINISTIC VERSION 65

Since there is a maximum sustainable level of capital defined by k, it

follows that lim kj = k (it is a monotonically increasing sequence bounded
t→∞
from above). Thus, for any  there exists a Te such that for all t ≥ Te, then
kt − k ≤ .
Since, 1 − δ + f 0 k < 1, it follows by continuity that it is possible to

pick  such that

1 − δ + f 0 (kt ) ≤ 1 − a < 1, (3.18)
for all t ≥ Te for some a ∈ (0, 1).Then, it follows for all j ≥ 0
h  i−1 1
β T +j+1 µTe+j+1 = β T +j µTe+j 1 − δ + f 0 kTe+j+1 ≥ β T +j µTe+j
e e e
1−a

given (??) above. Therefore, β T +j µTe+j is a strictly increasing sequence,

e

violating the transversality condition (3.16).

To complete the proof, note that it is impossible that the solution goes
from region III into region IV. To see this, assume that k > ka since that is
the only way to get into region IV from region III. When (k, µ) reach µA (k),
capital in the next period is k 0 = k and thus
−1 −1
µ0 = µ {β [1 − δ + f 0 (k)]} > µ {β [1 − δ + f 0 (ka )]} > µ,

where this last inequality follows because β [1 − δ + f 0 (ka )] < 1. Conse-

quently, (k 0 , µ0 ) must necessarily be in region III.

The two previous results imply that any solution path representing the
evolution of the PO allocation must be in region II and IV. Thus, it must
converge to (k ∗ , µ∗ ) because otherwise the optimal path gets into regions I
or III.

Claim 6 There is a unique PO allocation.

a a b
 b
Proof. Suppose that
there are two solutions c = {c t } and c = ct .
Then U (ca ) = U cb ≡ U 0 . Then, given the concavity of the production
function, the feasible set is convex and thus any convex combination of those
allocations belongs to the set. By strict concavity of the utility function, ba
a a b
convex combination should give more utility than
a a b
 bc = {ct } and c = ct
which contradicts that both c = {ct } and c = ct are PO allocations.

Claim 7 {kt } converges monotonically to k ∗ .

66CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Proof. We only need to check that the system cannot jump from one side
of the converging stable arm to the other. To see this, suppose that (k, µ) is
in region II and thus µ < µB (k). Given that
k ∗ = f (k) + (1 − δ) k − c (µB (k)) ,
it follows that c (µ) > c (µB (k)) and consequently k 0 = f (k) + (1 − δ) k −
c (µ(k)) < k ∗ implying that the system stays in region II.

To conclude, the stable arm represents the unique solution to the planner’s
problem and the system convergence monotonically. To determine these
properties, we had to use both k0 and also the transversality condition (3.16)
as boundary conditions for this nonlinear system of 2 first order difference
equations.

3.6 Recursive Formulation: A Heuristic In-

troduction
Consider an economy populated by identical agents with total measure
1 and names in the unit interval. Consequently, the planner’s problem (PP-
NGM) reduces to:
∞ X
X
β t π st u C st , 1 − L st





v(s0 , K0 ) = max (PPR-NGM)
(C,L,K)
t=0 st

subject to
C(st ) + K(st ) − (1 − δ)K(st−1 ) = st F (K(st−1 ), L(st )) for all st ,
C ∈ C, K(st ) ∈ X for all st ,
s0 and K0 ∈ X are given.
That is, v(s0 , K0 ) is the maximum utility attained for an economy with
(s0 , K0 ) as initial values. That is, the level of utility evaluated at the solution
of the Pareto problem given (s0 , K0 ). As you might properly remember, this
is object is the indirect utility function in the classical static setting studied
in micro. In a dynamic framework, in general it is denoted the value function.
Note that PPR-NGM can be rewritten as follows. Let
Γ(s, K) = {(K 0 , L) ∈ X × [0, 1] : 0 ≤ K 0 ≤ sF (K, L) + (1 − δ)K} ,
3.6. RECURSIVE FORMULATION: A HEURISTIC INTRODUCTION67

and define

Π(s0 , K0 ) = {(K, L) : K(st ), L(st ) ∈ Γ(st , K(st−1 )), ∀st , given (s0 , K0 )},

as the set of feasible allocations given (s0 , K0 ). Note that by definition (K, L) ∈
Π(s0 , K0 ) if and only if (K(s0 ), L(s0 )) ∈ Γ(s0 , K0 ) and (K(st ), L(st ))t≥1 ∈
Π(s1 , K(s0 )) for all s1 (where (K(st ), L(st ))t≥1 is called the continuation of
(K, L)).
Notice that

(∞ )
XX
β t π s u C(st ), 1 − L s
t t




v(s0 , K0 ) = max ,
(K,L)∈Π(s0 ,K0 )
t=0 st
= max {u (C(s0 ), 1 − L (s0 ))
K(s0 ), L(s0 ) ∈ Γ(s0 , K 0 ),
(K(st ), L(st ))t≥1 ∈ Π(s1 , K(s0 ))
∞ X
)
X
β t π st u C(st ), 1 − L st




+
t=1 st

where C(st ) = st F (K(st−1 ), L(st )) + (1 − δ)K(st−1 ) − K(st ).

Under our assumptions, it can be shown that we can proceed to split the
max operator to obtain

v(s0 , K0 ) = max max {u (C(s0 ), 1 − L (s0 ))
(K(s0 ),L(s0 ))∈Γ(s0 ,K0 ) (K(st ),L(st ))t≥1 ∈Π(s1 ,K(s0 )), ∀s1
∞ X
))
X
β t π st u C(st ), 1 − L st




+
t=1 st

While this operation will be in general valid in our setting but it will NOT
be necessarily valid in ANY setting. But now it is easy to observe that we
could write

v(s0 , K0 ) = max {u (C(s0 ), 1 − L (s0 ))

(K(s0 ),L(s0 ))∈Γ(s0 ,K0 )
∞ X
)
X
β t π s u C(st ), 1 − L s
t t




max +
(K(st ),L(st ))t≥1 ∈Π(s1 ,K(s0 )), ∀s1
t=1 st
68CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Finally, notice that

∞ X
X
β t π st u C(st ), 1 − L st }




t=1 st
X
= βπ (s0 , s1 ) u (C (s0 , s1 ) , 1 − L (s0 , s1 )) +
s1
∞
X X
β t π (s0 , s1 ) ...π (st−1 , st ) u C(st ), 1 − L st



t=2 (s1 ,...,st )

where
X∞ X
β t π (s0 , s1 ) ...π (st−1 , st ) u C(st ), 1 − L st



t=2 (s1 ,...,st )
 
X ∞
X X
β t−1 π (s1 , s2 ) ...π (st−1 , st ) u C(st ), 1 − L st 



= β π (s0 , s1 ) 
s1 t=2 (s2 ,...,st )

Therefore,
∞ X
X
β t π st u C(st ), 1 − L st }




t=1 st
X
= β π (s0 , s1 ) {u (C (s0 , s1 ) , 1 − L (s0 , s1 ))
s1

∞
X X


+ β t−1 π (s1 , s2 ) ...π (st−1 , st ) u C(st ), 1 − L st

t=2 (s2 ,...,st )

and then we could proceed to get

v(s0 , K0 ) = max {u (C(s0 ), 1 − L (s0 ))
(K(s0 ),L(s0 ))∈Γ(s0 ,K0 )
X
+β π (s0 , s1 ) t t
max {u (C (s0 , s1 ) , 1 − L (s0 , s1 ))
(K(s ),L(s ))t≥1 ∈Π(s1 ,K(s0 ))
s1

X X 
t−1 t t



+ β π (s1 , s2 ) ...π (st−1 , st ) u C(s ), 1 − L s

t≥2 (s2 ,...,st )
( )
X
= max u (C(s0 ), 1 − L (s0 )) + β π (s0 , s1 ) v(s1 , K(s0 )) ,
(K(s0 ),L(s0 ))∈Γ(s0 ,K0 )
s1
3.6. RECURSIVE FORMULATION: A HEURISTIC INTRODUCTION69

where the last equality follows by definition of v.

Again, the tricky part is that we are splitting the ”max” operator with-
out justification and this might not be right in any general setting. In
our framework, this follows by the so-called Principle of Optimality, due
to Richard Bellman.3 The key idea behind this principle is the following.
Let (K ∗ , L∗ ) ∈ Π(s0 , K0 ) be the optimal plan chosen at date 0. Once (t, st )
is reached, the set of all feasible plan is fully described by Π(st , K ∗ (st−1 )).
If the principle of optimality holds, there is no feasible alternative plan
(K, e ∈ Π(st , K ∗ (st−1 )) such that more utility from (t, st ) can be attained.
e L)
That is why (K ∗ , L∗ ) will be called time consistent. However, there is a large
number of relevant economic settings where optimal plans are time incon-
sistent and, then, a different approach is needed to attack each particular
problem.4
Notice now that period 0 has nothing in particular. More precisely, once
we know the current shock and the aggregate stock of capital, the problem
is exactly the same every period. That is, intuitively any solution should
satisfy

v(st , K(st−1 )) = t t



max {u C(s ), 1 − L s
(K,L)(st )∈Γ(st ,K(st−1 ))
X
+β π (st , s0 ) v(s0 , K(st ))}.
s0

More generally, we denote (s, K) as the aggregate state variables. These

variables are sufficient statistics to summarize the economic setting: prefer-
ences, technology, endowment and the distribution of property rights (where
this last will appear in economies with private ownership).
Let K 0 be the aggregate stock of capital available next period and write
"
v(s, K) = max u (sF (K, L) + (1 − δ)K − K 0 , 1 − L) (3.19)
(K 0 ,L)∈Γ(s,K)
#
X
+β π (s, s0 ) v(s0 , K 0 )
s0

3
See Stokey, Lucas and Prescott (1989, Chapter 4 and 9) for a superb discussion of this
issue.
4
A seminal contribution is Kydland, Finn E., and Edward C. Prescott. 1977. “Rules
Rather than Discretion: The Inconsistency of Optimal Plans.” Journal of Political Econ-
omy, Vol. 85(3), pp. 473–491.
70CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

The solutions to the problem above will be denoted K 0 (s, K) and L(s, K)
and represent the optimal level of next period capital and labor, respectively.
At a glance, (3.19) looks much better: instead of dealing with a infinite di-
mensional programming problem where we need to find the optimal contin-
gent infinite sequence {K(st ), L(st )}st , we just look for the optimal level of
investment and labor, (K 0 , L), given the function v. But that is, indeed, the
problem: the function v is unknown and it needs, if possible, to be found and
characterized. Under some standard assumptions, it can be shown that it ex-
ists and it is unique. Furthermore, some properties about the function v can
be established under mild assumptions, in particular its partial derivative.
There is a solution method to find v that works only under very restrictive
assumptions: first, guess a particular function v; then, verify that it satisfies
the functional equation (3.19) given above. The next example illustrates this
alternative.

A Rudimentary Solution Method: Guess and Verify

Suppose that u(C, 1 − L) = ln C + γ ln(1 − L) where γ > 0. Also, assume

that F (K, L) = K α L1−α , δ = 1 and α ∈ (0, 1). Finally, assume that st+1 =
(st )ρ εt+1 where ρ ∈ [0, 1] and {εt }∞
t=0 is an i.i.d. process with probability
π(ε) > 0 and E(ln ε) = 0. Therefore, (3.19) reduces to
(
α 1−α 0


v(s, K) = 0
max ln sK L − K ) + γ ln(1 − L (3.20)
0≤K ≤sF (K,L) ; L∈(0,1)
)
X
+β π (ε0 ) v(sρ ε0 , K 0 ) .
ε0

Once the value function itself v is known, this optimization problem would
be standard and it could be solved using well-known arguments. Then, the
strategy is the following. We guess a particular functional form for v, solve
the corresponding problem and then verify if effectively v satisfies the oper-
ator defined by (3.20).
To do so, we guess that v(s, K) = A1 + A2 ln s + A3 ln K. We then need
to find out for which set of parameters (A1 , A2 , A3 ) this function satisfies the
operator above. Hence, we first solve
X
0
α 1−α
π (ε0 ) (A1 +A2 ln (sρ ε0 )+A3 ln K 0 )}.


max
0
{ln sK L − K ) + γ ln(1 − L +β
K ,L
0
3.6. RECURSIVE FORMULATION: A HEURISTIC INTRODUCTION71

subject to 0 ≤ K 0 ≤ sF (K, L) and L ∈ (0, 1). Necessary and sufficient first

order conditions for an interior solution imply that

1 βA3
K0 : = ,
sK α L1−α − K0 K0
(1 − α)sK α L−α γ
L : α 1−α 0
= ,
sK L −K 1−L
and thus
βA3
K0 = sK α L1−α ,
1 + βA3
(1 − α) (1 + βA3 )
L = .
γ + (1 − α) (1 + βA3 )

In order to compute (A1 , A2 , A3 ), notice that if our guess is correct these

parameters must satisfy
 (1−α) !
1 (1 − α) (1 + βA 3 )
A1 + A2 s + A3 ln K = ln sK α
1 + βA3 γ + (1 − α) (1 + βA3 )
 
γ
+γ ln
γ + (1 − α) (1 + βA3 )
 
βA3 α 1−α
+βA1 + βA2 ρ ln s + βA3 ln sK L
1 + βA3
= constant + (1 + βA2 ρ + βA3 ) ln s + (α + βA3 α) ln K.

Thus, our conjecture was correct where in particular

α
A3 =
1 − βα

Very importantly, the optimal policy functions are given by

K 0 (s, K) = βαsK α L1−α .

(1 − α)
L = .
(1 − α) + γ(1 − βα)

Very importantly, notice that labor is a constant fraction of time, inde-

pendent of any fluctuation generated by (s, K).
72CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Exercises
Exercise 10 In text.

Exercise 11 (The AK Growth Model) Suppose that there is neither la-

bor nor leisure in the economy. Assume that {st } is an i.i.d. process, u(c) =
c1−σ and F (K) = AK. Show that the value function satisfies v(s, K) =
g(s) (K)1−σ for some function g(s).

Exercise 12 (Lucas-Uzawa Endogenous Growth Model) Consider an

economy populated by a continuum of infinitely-lived identical agents with
names in the unit interval (i.e. i ∈ [0, 1] represents a typical consumer).
Time is discrete and denoted by t = 0, 1, 2, .... There is only one consump-
tion good which can be either consumed or invested. The economy is sub-
ject to shocks and st represents its realization at date t. We assume that
{st } is a finite state first-order stationary Markov process. Transition prob-
abilities are denoted by π(s, s0 ) > 0 where st , s, s0 ∈ S = {1, ..., s}. Let
st = (s0 , ..., st ) ∈ S t+1 represent the partial history of aggregate shocks up to
date t.
The consumption set, C, is defined by

C = {{ct }∞
t=0 : (ct ) : S
t+1
→ R+ , sup c(st ) < ∞}.
(t,st )

Agent i’s preferences on C are represented by expected, time-separable,

discounted utility where if c ∈ C then:
∞ X
X c(st )1−σ
U (c) = β t π(st ) ,
1−σ
t=0 st

where β ∈ (0, 1) and σ > 0.

Each agent is endowed with one unit of time every period. They can
use that time either to work or to accumulate human capital. Let H (st )
be the stock of human capital of the representative agent at st and available
in period t + 1. At date t, an agent with human capital H (st−1 ) who works
u (st ) hours at the firm supplies u (st ) H (st−1 ) units of effective labor. On
the other hand, 1 − u (st ) represents the units of time devoted to accumulate
human capital according to the linear technology

H st − H st−1 = A(1 − u st )H st−1 ,

3.6. RECURSIVE FORMULATION: A HEURISTIC INTRODUCTION73

where the initial stock of human capital is given by H0 = H (s−1 ) > 0 and A
is a positive parameter.
There is a CRS aggregate technology to produce the consumption good
operated by a representative firm and subject to productivity shocks. Let
K(st ) ≥ 0 denote the aggregate stock of physical capital determined at the
node st and available in period t + 1 to produce with the aggregate technology.
Goods invested transform one-to-one in new physical capital next period and
the depreciation rate is given by δ ∈ (0, 1). Therefore, the stock of physical
capital evolves according to

K(st ) = I(st ) + (1 − δ)K(st−1 )

where I(st ) is investment in units of the consumption good and K0 > 0 is

given at date 0.
Let sK α L1−α represent the technology to produce the consumption good,
where K is the stock of physical capital available, L are units of effective
labor and s is the productivity shock. Assume that α ∈ (0, 1).
Note that an allocation is feasible if (c) ∈ C, 0 ≤ K(st ), u (st ) ∈ [0, 1] for
all st and

α
1−α
c(st ) + I(st ) = st K(st−1 ) L(st ) , ∀st
L(st ) = u st H st−1 , ∀st

Pareto Optimality

(a) Carefully define the Planner’s Problem for this economy. Characterize
the solution justifying your answer.

Competitive Decentralization: Arrow-Debreu Markets

(b) Carefully define a Competitive Equilibrium for an Arrow-Debreu econ-

omy.
Suppose that the accumulation of physical capital is carried out by the
representative firm and dividends are distributed back to the represen-
tative agent every period.
Characterize an Arrow-Debreu Competitive Equilibrium justifying your
answer.
74CHAPTER 3. THE ONE-SECTOR NEOCLASSICAL GROWTH MODEL

Welfare Theorems

(c) Show that both Welfare Theorems hold.

Balanced Growth Path

(d) Consider a deterministic version of the economy described above where

S = {1}.
Characterize a balanced growth path where consumption, investment,
physical capital and human capital grow at constant rates (but not nec-
essarily the same).

Hint: Use the FOC’s characterizing the Planner’s problem to answer this
question

Bonus. Discuss conditions under which there is positive growth and, at

the same time, lifetime utility of the representative agent is finite.
Chapter 4

Asset Pricing and Trading

Security markets enable the risk-averse members of the economy to share

their risk in order to smooth their consumption. By trading securities, indi-
viduals can reduce and even eliminate many of the risks they face. On the
other hand, they are rewarded for bearing the unavoidable risks. As we will
see, if markets are ”efficient”, that kind of risk arises from fluctuations in
aggregate output faced by the economy as a whole. It seems natural that the
price of a security reflects how well its income stream is adapted to meet the
needs of the individuals, taking into account the other securities in the mar-
ket. The object of this chapter is to study asset prices and their interaction
with consumption dynamics.

4.1 The Lucas Tree Model with Heteroge-

nous Agents
We consider an infinite horizon pure exchange economy under uncertainty
similar to that the framework introduced in Chapter 2 where there is only
one consumption good every period (i.e. N = 1). At the node st , y(st )
still represents the aggregate endowment but this output now is produced by
different trees. That is, there are J trees producing the unique consumption
good where dj (st ) is the fruit produced at the j th -tree at st and y(st ) =
PJ t t
j=1 dj (s ) for all s . It is assumed that the fruit is not storable but the tree
is perfectly durable. This setting is a version of the celebrated Lucas’ (1978)
tree model.
We assume that {st } is a finite state first-order stationary Markov process

75
76 CHAPTER 4. ASSET PRICING AND TRADING

where the transition probabilities are denoted by π(s, s0 ) > 0 and st , s, s0 ∈

{ζ1 , ..., ζN }. The probability of st is constructed from π as before:

π(st ) = Pr{st } = π(s0 , s1 )....π(st−1 , st ).

Additionally, we assume that dj (st ) = dj (st−1 , st ) for all st−1 and for all
j. This means that both output of the J trees and the aggregate endowment
depend only on the current shock st .
Consider the Pareto problem (PP) defined in Chapter 2 given some vector
of welfare weights α. Under our assumptions, we can characterize the unique
solution with the necessary and sufficient FOC given by,
 t
βi αi u0i (ci (st )(α))
=1 (4.1)
βh αh u0h (ch (st )(α))
I
X
ci (st )(α) = y(st ) (4.2)
i=1

Note that it follows from (4.1)

" t #
−1 βh αh 0
ci (st )(α) = (u0i ) u (ch (st )(α)) (4.3)
βi αi h

where (u0i )−1 , the inverse function of u0i , is well defined since ui is strictly
concave.

4.2 Competitive Sequential Trading

with Dinamically Complete Markets
Here we begin our analysis with competitive sequential trading and dynam-
ically complete markets. We first present the general framework regarding
trading opportunities.
Every period t and after having observed st , agents meet in spot markets
to trade the consumption good and different assets. The possibility of trading
J different tree’s shares implies that there are J long-lived assets available.
Let θij (st ) be the number of tree j 0 s shares chosen t
P by tagent i at s . There is
t
one outstanding share for each firm and then i θij (s ) = 1 for all s and j.
4.2. COMPETITIVE SEQUENTIAL TRADINGWITH DINAMICALLY COMPLETE MARKETS77

Firm j pays dj (st ) as dividends at each date t with realization st . That is,
this is a fixed dividend policy. Below we allow for alternative firm’s financial
policies to show that real equilibrium allocations are immune to these changes
and remain unaltered. Let θij (s−1 ) denote agent i0 s endowment of tree j 0 s
shares at date 0.
Let pj (st ) be the one share of firm j at the node st after dividends have
been paid (i.e., the ex-dividend price of the firm, since there is only one fully
divisible outstanding share).
Agents can also trade a full set of Arrow securities in zero net supply.
The Arrow security s0 traded at st pays one unit of consumption next period
if st+1 = s0 and 0 otherwise. Let q(st )(s0 ) be the price of this security at st .
Denote ai (st , s0 ) the holdings of this security. As usual, we restrict agents
to bounded trading strategies to rule out Ponzi schemes. These (implicit)
bounds are assumed to be sufficiently large such that they do not bind in
equilibrium. All prices are in units of the st -consumption good.
From the consumer side, given a price system (p, q) agent i0 s problem is
given by:
XX
max β t π(st )u(ci (st )),
(ci ,ai ,θi )
t st

subject to

X X
ci (st ) + pj (st )θij (st ) + q(st )(s0 )ai (st , s0 )
j s0
X
= ai (st−1 , st ) + [pj (st ) + dj (st )]θij (st−1 ),
j

where ci ≥ 0 and (ai , θi ) are bounded.

Definition 19 A Sequential Markets Competitive

n o Equilibrium (SMCE)
is a price system (b
p, q
b) and an allocation (b
ci , θi , b
b ai )i such that:

(STCE 1) Given (b
p, q
b), (b
ci , θbi , b
ai ) solves agent i’s problem for each i.
78 CHAPTER 4. ASSET PRICING AND TRADING

(STCE 2) All markets clear. For all st

I
X J
X
t
ci (s ) =
b dj (st ) = y(st ),
i=1 j=1
X
ai (st , s0 ) = 0 for all s0 ,
b
i
X
θbij (st ) = 1 for all j.
i

If we define γi (st ) > 0 as the Lagrange multiplier for agent i at st , then

given the price system (q, p) the corresponding Lagrangian is

∞ X
X
β t π(st ) u(ci (st )) + ...

LSM (ci , ai , γi ) =
t=0 st
"
X
t
ai (st−1 , st ) + [pj (st ) + dj (st )]θij (st−1 )


+γi s
j
#)
X X
−ci (st ) − pj (st )θij (st ) − q(st )(s0 )ai (st , s0 ) .
j s0
4.2. COMPETITIVE SEQUENTIAL TRADINGWITH DINAMICALLY COMPLETE MARKETS79

Necessary and sufficient conditions that characterize a SMCE are

bi st = u0 b ci s t




γ
bi st , s0 β t+1 π(st , s0 ) = qb st (s0 ) β t π(st )b
γi st




γ

X X
ci (st ) +
b pbj (st )θbij (st ) + qb(st )(s0 )b
ai (st , s0 )
j s0
X
ai (st−1 , st ) +
= b pj (st ) + dj (st )]θbij (st−1 ),
[b
j

X
β t π(st )b
γi st pbj (st ) = bi st , st+1 β t+1 π(st , st+1 )[b
pj (st , st+1 )+dj (st , st+1 )]



γ
st+1

where this last equation can be written

X u0 (b
ci (st , st+1 ))
pbj (st ) = βπ(st , st+1 ) pj (st , st+1 ) + dj (st , st+1 )]
[b
st+1
u0 (b
ci (st ))
X
qb st (st+1 ) [b
pj (st , st+1 ) + dj (st , st+1 )]


=
st+1

The corresponding transversality conditions are

X
β T π(sT )b
γi sT bai s T



lim =
T →∞
sT
X
β T π(sT )u0 b
ci s T ai s T




lim b =
T →∞
sT
lim E0 β T u0 (b

ci,T ) b
ai,T = 0
T →∞
80 CHAPTER 4. ASSET PRICING AND TRADING
X
β T π(sT )b
γi sT dj (sT ) + pj (sT ) θbij (sT −1 )



lim
T →∞
sT
X
= lim β T −1 π(sT −1 )θbij (sT −1 ) × ...
T →∞
sT −1
!
X
γi (sT −1 , sT ) dj (sT ) + pj (sT −1 , sT )


βπ(sT −1 , sT )b
sT
X
= lim β T −1 π(sT −1 )b
γi (sT −1 )pj (sT −1 )θbij (sT −1 )
T →∞
sT −1
X
β T −1 π(sT −1 ) u0 b
ci sT −1 pj (sT −1 )θbij (sT −1 )



= lim
T →∞
sT −1
n o
T −1 0
= lim E0 β u (b
ci,T −1 ) pj,T −1 θij,T −1 = 0
b
T →∞

plus (STCE 2). The transversality conditions have the following heuristic
interpretation. If any of these expressions were strictly positive, the agent
would be overaccumulating assets so that a higher expected life-time utility
could be achieved by, for example, increasing consumption today. The coun-
terpart to such nonoptimality in a finite horizon model would be that the
agent dies with positive asset holdings.
Before proceeding to price different assets, let’s a close look at this ex-
pression
X u0 (b
ci (st , st+1 ))
pbj (st ) = βπ(st , st+1 ) pj (st , st+1 ) + dj (st , st+1 )]
[b
st+1
u0 (b
ci (st ))
X
qb st (st+1 ) [b
pj (st , st+1 ) + dj (st , st+1 )]


=
st+1

The first line makes clear that agents value units of the consumption
good at date t + 1, state st+1 , using their intertemporal marginal rate of
u0 (b
ci (st ,st+1 ))
substitution (IMRS), βπ(st , st+1 ) u0 (bci (st )) . The presence of a full set of
(fully enforceable) Arrow securities makes the IMRS of each agent in the
economy equal to the price of the corresponding Arrow security, qb (st ) (st+1 );
i.e., all the intertemporal individual valuations coincide. More about this
later.
This framework is particularly relevant because it allows to price any
security.
4.3. EQUILIBRIUM ASSET PRICES 81

4.3 Equilibrium Asset Prices

4.3.1 Short-Lived Assets
Consider a short-lived security n in zero net supply and payoffs at t given
by hn (st ). Denote qn (st ) as the price of this security at st . We price this
asset using non-arbitrage arguments.1 That is, this asset can be perfectly
replicated holding hn (s0 ) Arrow securities tomorrow in state s0 . Therefore,
since in equilibrium agents cannot arbitrate to make infinite profits, it follows
by non-arbitrage that
X
qn (st ) = q(st )(s0 )hn (s0 ). (4.4)
s0

4.3.2 Long-Lived Asset: Pricing the Firm

Standard non-arbitrage conditions imply that the following condition will be
satisfied for all j and all st :
X
pj (st ) = q(st )(s0 )[dj (s0 ) + pj (st , s0 )]. (4.5)
s0

Stock Prices without Bubbles

Let P (st+j kst ) be the Arrow-Debreu prices for one unit of the consump-
tion good to be delivered at st+j conditional on st , which are constructed
using the prices of the Arrow securities as before. Iterating on (4.5) we
obtain
∞ X
X X
t
pj (s ) = P (st+j kst )dj (st+j ) + lim P (st+k kst )pj (st+k ).
k→∞
j=1 st+j kst st+k kst

It follows by the necessary transversality condition that the last term, an

1
Cochrane’s Happy-Meal Theorem: the price of a happy meal (in a frictionless market)
should be the same as the price of one hamburger, one small fries, one small drink and
the toy.
82 CHAPTER 4. ASSET PRICING AND TRADING

asset bubble, goes to zero. To see this, note that

X
lim θij (st+k )P (st+k kst )pj (st+k )
k→∞
st+k kst
X β t+k π(st , st+1 , ..., st+k ) u0 (b
ci (st , st+1 , ..., st+k ))
= lim t π(st ) 0 (b t ))
pj (st+k )θij (st+k )
k→∞
t+k t
β u c i (s
s ks
1 X
β t+k π(st , st+1 , ..., st+k )u0 b
ci st , st+1 , ..., st+k pj (st+k )θij (st+k



= t t 0 t
lim
β π(s )u (bci (s )) k→∞ t+k t
s ks

for all i, j and for all st due to the TC P (you t+kwill need to convince yourself
of one more step). And then, since i θij (s ) = 1 for all st+k , we can
conclude that X
lim P (st+k kst )pj (st+k ) = 0,
k→∞
st+k

for all j, as required. This kind of anomaly can be ruled out in general in
economies where the transversality condition is a necessary condition. When
markets are dynamically complete, this is always true. More about this later.
Therefore, the expression above reduces to
∞ X
X
pj (st ) = P (st+j kst )dj (st+j ),
j=1 st+j kst

and consequently the stock price is the present discounted value of the divi-
dends’ process.

4.3.3 Derivatives
Here we give examples of so-called derivative assets, that is of assets whose
returns are somehow derived from the return of another asset.
European Call Option
At st , consider a security that in T periods, after the realizations st+T
has been observed and dividends have been paid gives the right, but not
the obligation, to buy a share of the tree j at the strike price b > 0. Let
EOption t
qj,T (s ) be the price at st of an European call option with maturity T
and thus a specific expiration date t + T .2
2
An European option can only be exercised on the expiration date. If the option can
4.3. EQUILIBRIUM ASSET PRICES 83

Note that replicating arguments imply that by non-arbitrage,

EOption t
X EOption t 0
qj,T (s ) = q(st )(s0 )qj,T −1 (s , s ),
s0

where X
EOption t
qj,1 (s ) = q(st )(s0 ) max(pj (st , s0 ) − b, 0).
s0

That is, the option will be executed at T only its the price of the j tree is
higher that the strike price b (i.e. pj (st+T ) > b). Consequently,
EOption t
X
qj,T (s ) = P (st+T kst ) max(pj (st+T ) − b, 0).
st+T kst

Thus, the option price will value these payoffs properly.

4.3.4 The Risk-Free Interest Rate, the Term Structure

and the Yield Curve
Here we consider a simple example of fixed income securities.
The n risk free bond at st is a zero coupon bond which in n periods
from today (that is, in period t + n) and after the realizations st+n has been
observed, will pay one unit of the consumption good no matter which state
has realized.3 Let qn (st ) be its price at st and define its price recursively as
follows X
qn (st ) = q(st )(s0 )qn−1 (st , s0 ), (4.6)
s0

where qn−1 (st , s0 ) is the price of the n − 1 risk free bond’s price at (st , s0 ).
The intuition is the standard. By non arbitrage, today’s price must equal the
right to have tomorrow in all the states a n − 1 risk free bond. For example,
if n = 1 the risk-free interest rate at st , Rrf (st ), is defined by
X
−1 t
Rrf (s ) = qrf (st ) = q(st )(s0 ).
s0

exercised any time as well as on the expiration date is named an American option. On
the other hand, a put option gives to the owner the right, but not the obligation, to sell.
More about this later.
3
Zero coupon means that it will pay nothing in between.
84 CHAPTER 4. ASSET PRICING AND TRADING

It is key to understand that even though safe bonds represent sure claims
to future consumption, these assets are subject to price risk prior to maturity.
To see this more precisely, let Rn−1 (st ) = qn (st ) and note that (4.6) for n = 2
implies that
X
R2−1 (st ) = −1 t 0
q(st )(s0 )Rrf (s , s ), (4.7)
s0
X u0 (ci (s0 )) −1 t 0
= π (st , s0 ) β R (s , s ),
0 (c (s )) rf
s 0
u i t
 0
u (ci (s0 )) −1 t 0

= E β 0 Rrf (s , s ) k st ,
u (ci (st ))
−1 t −1 t 0
 
= Rrf (s )E Rrf (s , s ) k st +
 0
u (ci (s0 )) −1 t 0

+Cov β 0 , Rrf (s , s ) k st
u (ci (st ))

by definition of (conditional) covariance.

As we will discuss in more detail later, when markets are dynamically
complete equilibrium allocations are PO and thus individual consumption
depends only on the current shock. Given the Markovian structure of the
−1 t −1
shocks, it follows by definition that Rrf (s ) = Rrf (st ) for all st . Therefore
we can express (4.7) as

R2−1 (s) = Rrf

−1
 −1 0 
(s)E Rrf (s ) k s +
 0
u (ci (s0 )) −1 0

+Cov β 0 , R (s ) k s
u (ci (s)) rf

Roughly speaking, if the intertemporal marginal rate of substitution (IMRS),

0 0
β uu0(c(cii(s(s)))) , is negatively correlated with the price of the risk free bond tomor-
−1 0
row, Rrf (s ), then R2−1 (s) is lower than the (conditional) expected value of
−1 −1 0 −1 0
Rrf (s)Rrf (s ). In that case, qrf (s0 ) = Rrf (s ) is ”on average” lower pre-
0 0
cisely when consumption is more valued (i.e., high β uu0(c(cii(s(s)))) ) and, then, this
t
drives down the price of the 2 period bond at s to provide incentives to hold
this asset. Thus, the covariance term is a risk premium that arises because
investors are not risk neutral.
4.3. EQUILIBRIUM ASSET PRICES 85

Define the yield to maturity as follows

en (st ) = (Rn (st ))1/n = (qn (st ))−1/n ,
R (4.8)
1/n
u0 (ci (st ))

= ,
β n E [u0 (ci (st+n )) k st ]
 1/n
−1
= β −1 u0 (ci (st )) E [u0 (ci (st+n )) k st ]

This is the implicit yield per period. The term structure of the interest
rate (TSIR) is the collection of yields to maturity for bonds with different
maturities (also called the ”yield curve”).
It is already apparent that understanding what moves bond yields might
be important for forecasting. Indeed, yields on long-maturity bonds are
expected values of average future short yields, after an adjustment for risk.
This means that the current yields curve contains information about the
future path of the economy. Therefore, we could proceed as in (4.7) to
decompose (4.8) and obtain the corresponding risk premium represented by
covariance terms. But casual inspection of (4.8) suggests that this general
investigation might get quite complicated, as we have an implicit covariance
term involving nonlinear functions of consumption in at least three periods.
Since discretion is the better part of valor, it makes sense to examine some
simplified versions of the general problem.

A Deeper Understanding of the Yield Curve:

a Stylized Example4
Suppose that the economy is populated by a representative agent with
momentary utility function
c1−σ
u (c) = , σ ≥ 0.
1−σ
Below we show that, indeed, the existence of a representative agent with these
preferences is guaranteed under certain conditions if markets are dynamically
complete.
Assume now that output (and, thus, aggregate consumption) is deter-
ρ
mined by y(st ) = st (y(st−1 )) where ρ ∈ [0, 1] and {st } is an i.i.d. process
with ln s ∼ N (µs , ξs2 ) (here (µs , ξs2 ) represent the population mean and vari-
ance, respectively).
4
Vox populi, vox dei.
86 CHAPTER 4. ASSET PRICING AND TRADING

Define the net risk-free interest rate at st , rRF (st ), is defined rRF (st ) =
−1
ln Rrf (st ) = ln (qrf (st )) where now
ρ −σ
st+1 (y(st ))
Z 
t −1 t
qrf (s ) = Rrf (s ) = β F (dst+1 )
y(st )

−σ(ρ−1) 
= β y(st ) E (st+1 )−σ .

Remember that for any random variable x normally distributed, x ∼ N (µ, ξ 2 ),

1 2
E(ex ) = e(µ+ 2 ξ ) . Observe that
1 2 2
E (st+1 )−σ = E e−σ ln st+1 = e(−σµs + 2 σ ξs ) ,
 

since −σ ln st+1 ∼ N (−σµs , σ 2 ξs2 ). Therefore,

1 1
rRF (st ) = − ln qrf (st ) = ln − σ (1 − ρ) ln y(st ) + σµs − σ 2 ξs2 .


β 2
Similarly, the yield of the n − th risk free bond at st , rn (st ), is defined by
 
en (st ) = ln Rn (st ) 1/n = ln qn (st ) −1/n
rn (st ) = ln R

where
−σ
y(st , st+1 , ..., st+n )
Z Z 
t
qn (s ) = ... β F (dst+1 )...F (dst+n ).
st+1 st+n y(st )

Notice that future output can be represented as

ρ
y(st , st+1 , ..., st+n ) = st+n y(st , st+1 , ..., st+n−1 )

ρ2
= st+n (st+n−1 )ρ y(st , st+1 , ..., st+n−2 )
(n−1)
ρn
= st+n (st+n−1 )ρ ... (st+1 )ρ y(st ,

and, therefore, it follows that

Z Z  −σ
(n−1)
t −σ(ρn −1)
t n
qn (s ) = β y(s ) ... st+n (st+n−1 )ρ ... (st+1 )ρ F (dst+1 )...F (dst+n ).
st+1 st+n

Since {st } is an i.i.d. log-normally distributed process and

 −σ
= e−σ(ln st+n +ρ ln st+n−1 +...+ρ t+1 )
ρ ρ(n−1) (n−1) ln s
st+n (st+n−1 ) ... (st+n−1 ) ,
4.3. EQUILIBRIUM ASSET PRICES 87

we can conclude that

1
rn (st ) = − ln qn (st )
n
1 1
= ln + σ(ρn − 1) ln y(st )
β n
1 1
1
+ 1 + ρ + ... + ρn−1 σµs − 1 + ρ2 + ... + ρ2(n−1) σ 2 ξs2


n n 2
n 2n
1 1 1 (1 − ρ ) 1 (1 − ρ ) 1 2 2
= ln − σ(1 − ρn ) ln y(st ) + σµs − σ ξs
β n n (1 − ρ) n (1 − ρ2 ) 2

Observe that the slope of the yield curve can be obtained from the fol-
lowing expression

1 (1 − ρn )
 
t RF t
rn (s ) − r (s ) = 1− σ (1 − ρ) ln y(st )
n (1 − ρ)
1 (1 − ρn ) 1 (1 − ρ2n ) 1 2 2
   
− 1− σµs + 1 − σ ξs
n (1 − ρ) n (1 − ρ2 ) 2
1 (1 − ρn ) y(st )
   
= 1− σ ln µs t ρ
n (1 − ρ) e y(s )
2n
 
1 (1 − ρ ) 1 2 2
+ 1− σ ξs .
n (1 − ρ2 ) 2

The first term is key to understand what is behind the yield curve in this
setting and it can interpreted as follows. If the economy is in an expansion
and current output is above its trend (i.e., y(st ) > eµs y(st )ρ ), the first term
is increasing in n. Since the second term is always increasing in n, the yield
curve is upward sloped.
Roughly speaking, a high level of output lowers short term rates rela-
tively more. Income today is relatively high, so people want to save the
extra income; consequently, they drive up the price of risk free bonds and
correspondingly drive down the interest rate. In addition, the term structure
steepens because short term rates fall more than long term rates. This is
because the effect of the positive shock dies off as long as ρ < 1 , so that
if income is high today, it is also expected to be higher than average next
period, but not quite so high. The size of the effect, and thus the incentive
to save, diminishes, leading to a smaller increase in long rates. Importantly,
observe that this description is also valid for the case of i.i.d. shocks if we
88 CHAPTER 4. ASSET PRICING AND TRADING

impose ρ = 0. On the other hand, consider the extreme persistency level

with ρ = 1. In this case the yield curve has cero slope and, importantly,
independent of output.

4.4 The Equity Premium Puzzle

A risk premium depends on how much risk must be borne and how much
compensation a risk averse agent requires to bear it. Now we study the equity
premium puzzle alerted by Mehra and Prescott (1985).
Consider an economy populated by individuals with preferences repre-
1−σ
sented by ui (c) = c1−σ for all i with σ > 0. In that case, (5.8) implies that

ci (s0 )−σ C (s0 )−σ

q(st )(s0 ) = π (st , s0 ) β = π (s t , s0
) β ,
ci (st )−σ C (st )−σ
where C represents aggregate consumption. Consequently, asset price prices
will be identical to those in an economy with a representative agent consum-
1−σ
ing the aggregate endowment with preferences u(c) = C1−σ .
Let y(st ) = st be the aggregate consumption where st+1 = st t+1 . We
assume that {t+1 } is an i.i.d. process and ln  is normally distributed with
N (µ , ξ2 ) (here (µ , ξ2 ) represent the population mean and variance, respec-
tively).
Let p(st ) be the price of one share of the representative tree in this econ-
omy. Standard FOC’s imply that
C (st+1 )−σ
Z
t t
p(s ) = β −σ [p(s , st+1 ) + st+1 ]π (st , dst+1 ) (4.9)
st+1 C (st )
Z  −σ
st+1
= β [p(st , st+1 ) + st+1 ]π (st , dst+1 )
st
= E β (t+1 )−σ (p(st , st t+1 ) + st t+1 ) ,


where the expectation is taken with respect to t+1 .

Given the structure of the endowment process and the fact that p(st ) is
the value of the dividends’ stream, we conjecture that p(st ) = wst . To check
our guess, we proceed as usual to observe that in this case (4.9) reduces to
wst = E β (t+1 )−σ (wst t+1 + st t+1 )


= (w + 1)st βE 1−σ ,

4.4. THE EQUITY PREMIUM PUZZLE 89

and thus
(w + 1) 1 βE {1−σ }
= ⇒ w= ,
w βE {1−σ } 1 − βE {1−σ }
and our conjecture is verified to be correct. Define the equity return in state
st+1
p(st , st+1 ) + st+1 (w + 1)
RE (st+1 ) = t
= t+1 ,
p(s ) w
E{}
and thus its expected value is given by E {RE (t + 1)} = βE{ 1−σ } .

On the other hand, the risk-free interest rate is determined by

Z
−1 t t
q(st )(st+1 )π (st , dst+1 ) = E β (t+1 )−σ .

Rrf (s ) = qrf (s ) =
st+1

If x is has normal distribution N (µx , ξx2 ), then αx ˜ N (αµx , α2 ξx2 ) and

E {exp(x)} = exp(µx + 21 ξx2 )). For any x, we have that x = exp(ln x) by
definition and then
1
qrf (st ) = βE (t+1 )−σ = β exp(−σµ + σ 2 ξ2 )),

2
and also
E {} exp(µ + 12 ξ2 ))
E {RE (t + 1)} = =
βE {1−σ } β exp((1 − σ)µ + 12 (1 − σ)2 ξ2 ))
1
= β −1 exp(σµ − σ 2 ξ2 + σξ2 ).
2
Let rRF and rE be the net risk free interest rate and the equity return,
respectively. Then, we have that
ln E(RE ) − ln Rrf (st ) ≈ rRF − rE = σξ2 .
as the equity premium predicted by an economy satisfying our assumptions.
In their celebrated contribution, Mehra and Prescott (1985) proposed to
compare this predicted equity premium to the one observed in the data. To
discipline the exercise, they argue that the model parameters should meet the
criteria of cross-model verification. That is, not only they must be consistent
with the observations under consideration but they should not be grossly
inconsistent with other observation in growth theory, business-cycle theory,
labor market behavior and so on.
90 CHAPTER 4. ASSET PRICING AND TRADING

Data 1889 - 1978 (Mehra and Prescott (1985))

Mean risk-free interest rate Rrf 1.008

Mean return on equity RE 1.069
Mean Consumption Growth 1.018
Standard Deviation of Consumption (Variance) 0.036 (0.00129)

Note that this implies that the average equity premium in the data is
0.0618 (approximately 6%).
Now we need to impose some restrictions to the choice of β and σ. Var-
ious studies suggest that σ is a small number, certainly less than 10. The
parameter β is calibrated such that it matches the risk-free interest rate in
a deterministic version of the economy in steady state. This implies that β
is approximately equal to 0.99(≈ (1.008)−1 ).
Thus using σ = 10 (the upper bound) and β = 0.99, we obtain the
predicted risk free interest rate
1
ln R
brf = − ln β + σµ − σ 2 ξ2 = 0.120
2
⇒ R
brf = 1.127,

that is, 12.7%! This inconsistency between the predicted and the observed
risk-free interest rate (which in the US averages less than 1%) has been
dubbed ”the risk-free rate puzzle”.
On the other hand, the predicted return on equity is 1.141 (that is, 14.1%)
and consequently the predicted equity premium amounts 1.4%, far lower than
the 6.18% historically observed in the data. This happens even imposing an
arbitrarily high number for σ, which most studies indicate to be close to 3.
This empirical fact has been known as ”the equity premium puzzle”.
It is important to mention that even though a higher σ might help to
explain the last puzzle, it would impose an even higher risk free interest rate,
leaving the first puzzle mentioned more unexplained.

The Equity Premium Puzzle in Production Economies

Expanding the set of technologies in a pure exchange economy with com-
plete sequential markets to admit capital accumulation and production as
in, for instance, Brock (1979) does not increase the set of joint equilibrium
processes on consumption and asset prices. Since the set of equilibria in a
production company is a subset of those in an exchange economy, it follows
4.4. THE EQUITY PREMIUM PUZZLE 91

immediately that if the equity premium cannot be accounted for in an ex-

change economy, modifying the technology to incorporate production will
not alter this conclusion.
To see this, let  denote preferences, τ technologies, E the set of the
exogenous processes on the aggregate consumption good, P the set of tech-
nologies with production opportunities, and m(, τ ) the set of equilibria for
economy (, τ ).

Theorem 14 For all 

∪ m(, τ ) ⊃ ∪ m(, τ ).
τ ∈E τ ∈P

Proof. Given  for some τ0 ∈ P , let (q0 , c0 ) be a joint equilibrium process

on asset prices and consumption, respectively. A necessary condition for
equilibrium is that the asset prices q0 be consistent with c0 , the optimal
consumption for the household with preferences . Thus, if (q0 , c0 ) is an
equilibrium, then q0 = g(c0 , ), where g is defined by the first-order necessary
conditions for household maximization.
This functional relation must hold for all equilibria, regardless of whether
they are for a pure exchange or a production economy. Let (q0 , c0 ) be an equi-
librium for some economy (, τ0 ) with τ0 ∈ P . Consider the pure exchange
economy with preferences  and τ1 = c0 . Our contention is that (q0 , c0 )
is a joint equilibrium process for asset prices and consumption for the pure
exchange economy (, τ1 ). For all pure exchange economies, the equilibrium
consumption process is τ , so c1 = τ1 = c0 , given that more is preferred to
less. If c0 is the equilibrium process, the corresponding asset price must be
g(c0 , ), which is q0 . Hence, q0 is the equilibrium price system for the pure
exchange economy (, τ1 ), proving the theorem.

4.4.1 The Market Price of Risk and Hansen-Jagannathan

Bounds (incomplete)
The market price of risk is defined in terms of asset prices and their one-
period payoffs. Hansen and Jagannathan (1991) interpret the equity pre-
mium puzzle in terms of the high “market price of risk” implied by time-series
data on asset returns.
Let hn (st+1 ) be h’s one-period payoffs at t+1 and qn (st ) is its correspond-
ing price at st
92 CHAPTER 4. ASSET PRICING AND TRADING

−σ
t
X c (st , st+1 )
qn (s ) = βπ (st , st+1 ) hn (st+1 )
st+1
c (st )−σ
 −σ  
ct+1
= Et β −σ hn,t+1
c
 −σt 
c
= Et β t+1 Et (hn,t+1 )
c−σ
t
 −σ 
ct+1
+Covt β −σ , hn,t+1
ct

The Cauchy-Schwarz inequality is

 −σ   −σ 
ct+1 c
Covt β −σ , hn,t+1 ≤ stdt β t+1 stdt (hn,t+1 )
ct c−σ
t

 −σ 
c
and hence, q RF (st ) = Et β ct+1
−σ , we have
t

c−σ
 
t+1
qn (st ) stdt β c−σ
t
RF t
≥ E t (hn,t+1 ) − RF t
sdt (hn,t+1 ) (4.10)
q (s ) q (s )

The LHS denotes the price of the security in terms of the price of the risk
free bond. The bound () holds with equality for those asset that are!
on the
c−σ
t+1
stdt β
c−σ
t
efficient mean-standard deviation frontier. The expresion is called
q RF (st )
the market price of risk. It provides an estimate of the rate at which the price
of a security falls with an increase in the conditional standard deviation of
its payoff.
Hansen and Jagannathan (1991) used data on asset prices and returns
alone to estimate the market price of risk
Equity premium puzzle: the market price of risk implied by the asset
market data alone is much higher than can be reconciled with the aggregate
consumption data.
c−σ
 
That is, aggregate consumption is not volatile enough to make stdt β ct+1
−σ
t
high enough for the reasonable values of σ.
4.5. AGGREGATION AND THE DISTRIBUTION OF WEALTH 93

4.5 Aggregation and the Distribution of Wealth

Let αi > 0 denote the welfare weight for agent i. Given the vector (αi )Ii=1 ,
the corresponding Pareto problem is given by
I
(∞ )
X XX
max αi β t π(st ) ui (ci (st )) , (PP)
(ci )Ii=1
i=1 t=0 st ∈S t

subject to
I
X
ci (st ) = y(st ) for all st , (4.11)
i=1
ci ∈ C for all i.

Thus, given (α1 , ..., αI ), necessary and sufficient conditions characterizing

a solution for (PP) are given by a process of Lagrange multipliers µ∗ ∈ X+
and an allocation (c∗i )Ii=1 such that

αi β t π st u0i c∗i st (α ) = µ∗ st for all st and all i,

(4.12)

X
X
c∗i st )(α = yi (st ) for all st . (4.13)
i i

Under our assumption, we can characterize the unique solution with the
necessary and sufficient FOC given by
αi u0i (ci (st )(α))
= 1, (4.14)
αh u0h (ch (st )(α))
I
X
ci (st )(α) = y(st ). (4.15)
i=1

Note that it follows from (5.2)

 
−1 αh 0
t
ci (s )(α) = (u0i ) t
u (ch (s )(α)) (4.16)
αi h

where (u0i )−1 , the inverse function of u0i , is well defined since ui is strictly
concave.
94 CHAPTER 4. ASSET PRICING AND TRADING

Then, it follows from feasibility that ch (st ) is the unique solution to

I  
X X −1 αh 0
t
ci (s )(α) = (u0i ) u (ch (s )(α)) + ch (st )(α) = C(st ) = y(st ).
t

i=1 i6=h
αi h
(4.17)
where C(st ) stands for aggregate consumption at st .
Hence, given the distributional parameters α (which are fixed in this
artificial device called ”the Planner’s problem”), the only determinant of
individual consumption is the realization of the aggregate endowment. Very
importantly, this implies that at any competitive equilibrium, any kind of
individual risk will be perfectly traded away across agents. Additionally,
ch (st−1 , st )(α) = ch (st )(α) for all h; i.e., individual consumption depends
only on the current shock and thus the distribution of consumption is history
independent.
On the other hand, although the intertemporal margianl rates of sub-
stitutions are equalizaed acorss agents, these will in general depend on the
distribution parameter α. To see this consider a setting with a full set of
Arrow securities. Iin that economy, the competitive equibilibrium allocation
is PO for some α, (c∗i (α))Ii=1 . Notice that IMRS are equalized and thus, for
all agent i, we have that

u0 (c∗ (s0 ) (α))

q st (s0 ) = β π (st , s0 ) 0i ∗i


ui (ci (st ) (α))

But then, Arrow security prices and the corresponding state prices will
both depend on α in general.
Aggregation form in the sense of Gorman are possible. That is, under
specific assumptions, a fictitious representative aggregate consumer can be
constructed to deliver competitive equilibrium prices as IMRS at aggregate
quantities. Suppose that

(c−γi )1−σ

if σ 6= 1 and σ > 0
ui (c) = 1−σ ,
ln(c − γi ) if σ = 1

The parameter γi ≥ 0 is interpreted as the minimum consumption level

required by agent i.
4.5. AGGREGATION AND THE DISTRIBUTION OF WEALTH 95

First order conditions reduce to

αi [ci (α) − γi ]−σ = αh [ch (α) − γh ]−σ ,

1 1
(αi )− σ [ci (α) − γi ] = (αh )− σ [ch (α) − γh ],
1 1
(αh ) σ [ci (α) − γi ] = (αi ) σ [ch (α) − γh ]

Adding up across h, we get

(αi )1/σ
[ci (α) − γi ] = PI 1/σ
[C − γ], (4.18)
h=1 (αh )

where C = Ii=1 ci stands for aggregate consumption and γ = Ii=1 γi de-

P P
notes the aggregate minimum consumption requirement.
Given any vector α, observe that the objective function of the planner’s
problem reduces to

I
(∞ )
t 1−σ
X XX (c(s ) − γi )
αi β t π(st ) (4.19)
i=1 t=0 st ∈S t
1−σ
(∞ I
)
XX X (c(st ) − γi )1−σ
= β t π(st ) αi
t=0 st ∈S t i=1
1−σ

where for any (st ), condition (4.18) implies

I
X (c(st ) − γi )1−σ
αi
i=1
1−σ
1−σ 1−σ
I
X αi σ (C(st ) − γ)
= αi  1−σ
PI 1/σ 1−σ
i=1
h=1 (αh )
I 1−σ
1 X (C(st ) − γ)
1
=  1−σ αi σ
PI 1/σ 1−σ
h=1 (αh ) i=1

I
!σ 1−σ
X 1/σ (C(st ) − γ)
= (αh )
h=1
1−σ
96 CHAPTER 4. ASSET PRICING AND TRADING

And thus, for any α (4.19) transforms into

I
!σ ( ∞ 1−σ
)
X 1/σ
XX (C(st ) − γ)
(αh ) β t π(st )
h=1 t=0 st ∈S t
1−σ

Since choices are immune to affine linear transformations, we can conclude

that there exists a ficticious representative aggregate consumer with IMRS
−σ −σ
0 (C(s0 ) − γ) 0 (ci (s0 ) − γi )
β π (st , s ) −σ = β π (st , s ) −σ
(C(st ) − γ) (ci (st ) − γi )
for all i = 1, ..., I.

4.6 Firm’s Financial Policy and the Modigliani-

Miller Theorem
The basic question here is the following: can a firm manipulate its financial
policy to affect its total value (i.e. the total value of its stock plus the total
value of its debt). The celebrated Modigliani-Miller Theorem asserts that,
under certain conditions regarding the asset market structure, this value is
completely independent of the firm’s financial Q policy.
Given a price system (p, q), define P (st ) = th=1 q(sh−1 )(sh ) and Pb(st ) =
P (st )/π(st ). Thus, a financial policy (Bj , Θj , Dj ) is feasible for the firm j if

Θj (st−1 )Dj (st ) + Bj (st ) = dj (st ) + pj (st ) Θj (st ) − Θj (st−1 ) (4.20)

X
+ q(st )(s0 )Bj (st , s0 ),
s0
 
lim E0 Pb(st )Bj (st ) = 0 (4.21)
t→∞
 
lim E0 Pb(st )pj (st )Θj (st ) = 0, (4.22)
t→∞

where Dj (st ) is the dividend distributed among share owners holding Θj (st−1 )
shares issued at st−1 , Bj represents contingent debt and (Θj (st ) − Θj (st−1 ))
is the amount of new shares issued at st . At date t, the left-hand side in
(4.20) describes the firm’s obligations whereas the right-hand side describes
its sources of revenues. Conditions (4.21) and (4.22) rule out Ponzi schemes.
4.6. FIRM’S FINANCIAL POLICY AND THE MODIGLIANI-MILLER THEOREM97

A SMCE for this setting is defined as usual taking into account that the
market clearing condition for asset markets are now given by
X X
ai (st , s0 ) = Bj (st , s0 ) for all s0 ,
i j
X
θij (s ) = Θj (st ) for all j.
t

Proposition 7 (Modigliani-Miller) The total value of the firm j is inde-

pendent of its financial policy (Bj , Θj , Dj ).
Proof. Take the price of the Arrow securities (q) as given. The ex-dividend
price of each share, (4.5), in this setting is given by
X
pj (st ) = q(st )(st+1 )[Dj (st , st+1 ) + pj (st , st+1 )], (4.23)
st+1

Then, multiply (4.23) by Θj (st ) to obtain

X
Θj (st )pj (st ) = q(st )(st+1 ) Θj (st ) Dj (st , st+1 ) + pj (st , st+1 ) ,



st+1

i.e., the total value of j stocks after dividends and debt obligations have been
paid at st whenever Θj (st ) is the total number of shares issued at st .
From (4.20), we have that (4.23) satisfies
X X
Θj (st )pj (st ) = q(st )(st+1 )dj (st+1 ) − q(st )(st+1 )Bj (st , st+1 )(4.24)
st+1 st+1
XX
+ q(s )(st+1 )q(s , st+1 )(st+2 )Bj (st , st+1 , st+2 )
t t

st+1 st+2
X
+ q(st )(st+1 )pj (st , st+1 )Θj (st , st+1 ).
st+1

Let P (st+h k st ) = q(st )(st+1 )q(st , st+1 )(st+2 )...q(st , st+1 , ..., st+h−1 )(st+h )
be the price of one unit of the consumption good delivered at st+h in st −units.
Thus, using (4.20)-(4.22) repeatedly in (4.24), we cancel the debt terms to
obtain
X X∞ X
t t t t
Θj (s )pj (s ) + q(s )(st+1 )Bj (s , st+1 ) = P (st+h k st )dj (st+h ).
st+1 h=1 st+h kst
(4.25)
98 CHAPTER 4. ASSET PRICING AND TRADING

Since the right-hand side of this last expression is independent of the firm’s
financial policy (i.e., the total value of the dividend process, purely techno-
logical), its total value itself is unaffected by any alternative (Bj , Θj , Dj ).

Note that adding up (4.25) across firms j, we obtain

X X X ∞
X X X
t t t t
Θj (s )pj (s ) + q(s )(st+1 ) Bj (s , st+1 ) = P (st+h k st ) dj (st+h )
j st+1 j h=1 st+h kst j

X∞ X
= P (st+h k st )y(st+h ),
h=1 st+h kst

i.e., the total wealth at the economy, unaffected by firms’ financial plans.
This theorem can be extended in several directions not discussed here.
However, we should mention two important properties that make the theorem
work in more general settings:
1. Consumers and firms face the same financial market opportunities.
Thus, consumers can repackage the financial component of dividends and,
in particular, if needed they can undo whatever firms do in the financial
markets.
2. In production economies, the total value of dividends depends only on
production plans, independently of financial plans.

4.7 Arbitrage and State Prices

Consider an economy with competitive sequential trading where agents meet
in spot markets to trade the consumption good and different securities.
To
 simplify the analysis below, suppose that T < ∞, and denote L =
{xt }Tt:1 | xt : S t → R as the linear space of adapted processes. Formally,
a security x is an adapted payoff process, Dx ∈ L, where Dx,t denotes the
random variable that represents the payoff paid by the security x at time
t (thus, Dj (st ) denotes a particular realization at date t if the event st has
happened). Each security has an adapted security price process, Px ∈ L,
so that Px,t is the price of the security, ex-dividend (i.e., traded after the
dividend has been paid at date t), at date t (again, a random variable).5
5
Alternatively, asset can be assume to be traded cum-dividend. This means that they
are traded with an entitlement to the coming dividend payment attached. This dividend
4.7. ARBITRAGE AND STATE PRICES 99

In what follows, we assume the following asset market structure. There

are J ≤ S short-lived securities in zero net supply. Security j ∈ {1, ..., J}
pays hj (s) at the state s ∈ S and, thus, hj = (hj (1), ..., hj (S))0 denotes the
j-payoff vector. Let qj (st )and aj (st ) denote the price and holdings of the
j-security at st , respectively. Agents’ initial endowments are normalized to

J
zero; i.e., aj (s−1 ) = 0 for all j. We denote a = {aj (st )}∀st j:1 as a trading
strategy and, as usual, we restrict agents to bounded trading strategies to
rule out Ponzi schemes. Since we assume that if T < ∞, then aj (sT ) = 0
for all j. All prices are in units of the st -consumption good. Assume that
the J vectors h0j s are linearly independent and denote H = [h1 , ..., hJ ] as the
J × S payoff matrix. The dividend process generated by a trading strategy
a, denote δ a , is defined by

δ a (st ) = h(st ) · a(st−1 ) − q(st ) · a(st ),

for all t and all st .

Given (q, H), a trading strategy a is an arbitrage if δ a > 0 (that is, if
δ a (st ) ≥ 0 for all st with strict inequality for some set ); it is a portfolio of-
fering ”something for nothing”. More precisely, in an environment where
investment opportunities have constant returns to scale, absence of arbi-
trage means that there is no investment portfolio which gives a positive net
payoff in some state with non-negative net payoffs in all remaining states.
Since there are constant returns to scale, any portfolio with an arbitrage
opportunity can generate an arbitrarily large payoff in some state with no
corresponding sacrifice in any other state. But this is problematic because it
is essentially like having a good with zero price -in such a situation there is
no solution to the agents’ problem and hence no equilibrium.
Let A be the space of trading strategies and observe that for any a and b
in A, we have δ αa+λb = αδ a + λδ b for all α, λ ∈ R. Thus, the market subspace
M = {δ a : a ∈ A} of dividend processes generated by trading strategies is a
linear subspace of the space L of adapted processes.
A state-price m is a strictly positive adapted process (i.e., m ∈ L++ ),
such that m(st ) ∈ RS++ for all st and
X
qj (st ) = Et (hj (s0 ) m(st )(s0 )) = π(st , s0 )(hj (s0 ) m(st )(s0 )), (4.26)
s0

has already been declared but not paid, so the market knows how much it is worth and
the corresponding price will reflect this. We will not consider this case below.
100 CHAPTER 4. ASSET PRICING AND TRADING

for each j ∈ {1, ..., J}. That is, the price of any security is the state-
price weighted sum (or expectation) of the security’s state-contingent payoffs.
Identifying the state prices is the major task at hand. To prove the next re-
sult we use the following version of the Separating Hyperplane Theorem for
cones.
Theorem 15 (Linear Separation of Cones) Suppose Q and K are closed
convex cones in RN such that Q ∩ K = {0}. If K does not contain a lin-
ear subspace other than {0}, then there is nonzero vector p ∈ RN such that
p · x < p · y for each x ∈ Q and each y ∈ K/{0}.
Now we show that the no-arbitrage condition is equivalent to the existence
of a state-price vector. To keep technical issues simple, we assume that
T < ∞.
Theorem 16 There is no arbitrage if and only if there exists a state-price
vector.
Proof. Let L+ be non-negative orthant of L and notice that the only linear
subspace of L+ is {0}. Observe that there is no arbitrage if and only if
M ∩ L+ = {0}.
Step 1. Suppose first that there is no arbitrage, i.e. M ∩ L+ = {0}.
Theorem 14 implies the existence of p ∈ L/{0} such that p · z < p · x for all
z ∈ M and for all x ∈ L+ /{0}. Suppose that there exists some zb ∈ M such
that p · zb 6= 0. Since M is a linear space, λz ∈ M for all λ ∈ R. Thus, if you
fix some x ∈ L+ /{0} (and then fix p · x), there exists some λ ∈ R such that
p · λb
z ≥ p · x, and we get a contradiction. Hence, p · z = 0 for all z ∈ M .
This implies that 0 < p · x for all x ∈ L+ /{0} and we claim that this implies
that p ∈ L++ . To see this, note that otherwise, if p(st ) ≤ 0 for some st , we
can choose x e such that it is equal to 0 for every node except at st and, then,
p·xe ≤ 0, a contradiction. Therefore, since p · z = 0 for any arbitrary z ∈ M ,
then
XT X
p(st )δ a (st ) = 0 (4.27)
t=0 st

for any arbitrary trading strategy a ∈ A. Given an arbitrary node st and j,

let aj (st ) = 1 while 0 for all the other nodes and j0s. Thus, (4.27) implies
that X
qj (st ) = p(st , s0 )hj (s0 )
st+1
4.7. ARBITRAGE AND STATE PRICES 101

The state-price process is constructed making m(st )(s0 ) = π(s1t ,s0 ) p(st ) for
each s0 .
Step 2. Suppose now that there exists a state-price. Denote

P m (st ) = m(s0 )(s1 )...m(st−1 )(st ) > 0

for all st (i.e., the shadow date 0 price implied by state prices). Consider any
trading strategy a ∈ A to compute
T X
X
m a
E (P ·δ ) = π(st )P m (st )δ a (st )
st
t=0
T
XX
π(st )P m (st ) h(st ) · a(st−1 ) − q(st ) · a(st ) .


=
t=0 st

Note that if a is an arbitrage, then E (P m · δ a ) > 0 by definition.

On the other hand, since m is a state price, (4.26) implies

T X
X
E (P m · δ a ) = π(st )P m (st ) h(st ) · a(st−1 ) − q(st ) · a(st ) .


t=0 st
T XX
X
π(st−1 )π(st−1 , st )P m (st−1 )m(st−1 )(st ) h(st ) · a(st−1 )


=
t=0 st−1 st

XT X
− π(st )P m (st )q(st ) · a(st )
t=0 st
T X
X X X
P m (st−1 ) aj (st−1 ) m(st−1 )π(st−1 , st )(st )hj (st )


=
t=0 st−1 j st
T X
X
− π(st )P m (st )q(st ) · a(st )
t=0 st
T
XX
P m (st−1 ) q(st−1 ) · aj (st−1 )


=
t=0 st−1

XT X
P m (st ) q(st ) · a(st )


−
t=0 st
102 CHAPTER 4. ASSET PRICING AND TRADING

Since aj (sT ) = aj (s−1 ) = 0 for all j, we can conclude that for any trading
strategy a ∈ A
T X
X
E (P m · δ a ) = P m (st−1 ) q(st−1 ) · aj (st−1 )


t=1 st−1

XT X
P m (st ) q(st ) · a(st )


−
t=0 st
= −P (s ) q(sT ) · a(sT ) = 0
m T

Therefore, there cannot be any arbitrage opportunity.

Exercises
Exercise 13 Exercise 13.3 [Ljungqvist, Sargent (2004)] Growth slowdowns
and stock market crashes, donated by Rodolfo Manuelli.6
Consider a simple one-tree pure exchange economy. The only source of
consumption is the fruit that grows on the tree. This fruit is called dividends
by the tribe inhabiting this island. The stochastic process for dividend dt is
described as follows: If dt is not equal to dt+1 , then dt+1 = γdt with probabil-
ity π, and dt+1 = dt with probability (1 − π). If in any pair of periods j and
j + 1, dj = dj+1 , then for all t > j , dt = dj .

In words, the process if not stopped grows at a rate γ in every period.

However, once it stops growing for one period, it remains constant forever
on. Let d0 equal one.

Preferences over stochastic processes for consumption are given by:

∞
X
U = E0 β t u(ct )
t=0

c(1−σ)
where u(c) = (1−σ)
. Assume that σ > 0 , 0 < β < 1 , γ > 1, and
(1−σ)
βγ < 1.

6
See also Joseph Zeira (1999)
4.7. ARBITRAGE AND STATE PRICES 103

(a) Define a competitive equilibrium in which shares to this tree are traded.
(b) Display the equilibrium process for the price of shares in this tree pt as
a function of the history of dividends. Is the price process a Markov
process in the sense that it depends just on the last periodś dividends?
(c) Let T be the first time in which dT +1 = dT = γ (T −1) . Is pT −1 > pT
? Show conditions under which this is true. What is the economic
intuition for this result? What does it say about stock market declines
or crashes?
(d) If this model is correct, what does it say about the behavior of the ag-
gregate value of the stock market in economies that switched from high
to low growth (e.g., Japan)?
Exercise 14 (Exercise 13.4 [Ljungqvist, Sargent (2004)) The term struc-
ture and consumption, donated by Rodolfo Manuelli]
Consider an economy populated by a large number of identical households.
The (common) utility function is:
∞
X
U = E0 β t u(ct )
t=0
(1−θ)
where 0 < β < 1 , and u(x) = x(1−θ) , for some θ > 0. (If θ = 1, the
utility is logarithmic.) Each household owns one tree. Thus, the number of
households and trees coincide. The amount of consumption that grows in a
tree satisfies:
ct+1 = c∗ cϕt εt+1
where 0 < ϕ < 1, and εt is a sequence of i.i.d. log normal random
variables with mean one, and variance σ 2 . Assume that, in addition to
shares in trees, in this economy bonds of all maturities are traded.
(a) Define a competitive equilibrium.
(b) Go as far as you can calculating the term structure of interest rates,
R̃jt , for j = 1, 2, ....
(c) Economist A argues that economic theory predicts that the variance
of the log of short-term interest rates (say one-period) is always lower
than the variance of long-term interest rates, because short rates are
“riskier”. Do you agree? Justify your answer.
104 CHAPTER 4. ASSET PRICING AND TRADING

(d) Economist B claims that short-term interest rates, i.e., j = 1, are

“more responsive” to the state of the economy, i.e., ct , than are long-
term interest rates, i.e., j large. Do you agree? Justify your answer.

(e) Economist C claims that the Fed should lower interest rates because
whenever interest rates are low, consumption is high. Do you agree?
Justify your answer.

(f ) Economist D claims that in economies in which output (consumption in

our case) is very persistent (ϕ ≈ 1), changes in output (consumption)
do not affect interest rates. Do you agree? Justify your answer and, if
possible, provide economic intuition for your argument.

Exercise 15 (Asset Pricing with Two States) Consider an infinite hori-

zon economy populated by a representative agent. Time is discrete where
t = 0, 1, 2..., ∞. Agent’s preferences are represented by:
∞ X
X
β t π st ln c st ,




t=0 st

where β ∈ (0, 1).

Here π (st ) denotes the unconditional probability of reaching node st =
(s0 , s1 , ..., st ). Assume that st is a first-order stationary Markov process with
st ∈ {1, 2}.
This is an endowment economy. The agent is entitled to a process:

yt = λt λt−1 ...y0

for t ≥ 0 and y0 > 0.

λt is a function of st :

 1 if st = 1
λt =
1 + ζ if st = 2


where ζ > 0.
Assume that in period t = 0 we have:

s0 = 1
4.7. ARBITRAGE AND STATE PRICES 105

Also, suppose that:

Pij = Pr [st+1 = j|st = i] ≥ 0

for all (i, j) pairs.

Sequential Markets Competitive Equilibrium

Suppose that there is a full set of Arrow securities and a risk-free bond
that can be traded. Let pbt be the time t of a risk-free claim to one unit of
consumption at time t + 1.

(a) Define a Sequential Markets Competitive Equilibrium (SMCE)

for this economy. Be as precise as possible.

(b) Go as far as you can characterizing the SMCE.

Asset Pricing

−1
(c) Let Rt = pbt be the one period risk-free gross interest rate. Give a
formula for Rt and tell how it depends on the history st .

(d) Suppose that β = 0.95, ζ = 0.02 and:

P11 = 1
P12 = 0
P21 = 0.5
P22 = 0.5

Compute Rt when st = 1 and when st = 2. Interpret.

Exercise 16 (Derivates) (Exercise from the September 2012 final exam in

Graduate Macroeconomics at UTDT).
106 CHAPTER 4. ASSET PRICING AND TRADING

(a) Coupon Bonds.

Suppose that a coupon bond issued in period t that matures in τ periods
at which time it pays one unit of the consumption good. It also pays a
coupon equal to h in periods t + 1, ..., t + τ − 1.
Show that the price of such a bond, qcτ , satisfies
τ −1
X
qcτ (st ) = h q k (st ) + q τ (st )
k=1

i = 1, ..., τ and all st . Here q k (st ) denotes the zero coupon bond with
maturity k = 1, ..., τ at st .

(b) Pricing a Forward Contract on Equity.

The owner of an n−period forward contract purchases one share of the
equity at the forward price z where the forward price has been set such
that the initial value of the contract is equal to zero. The ex-dividend
price of the equity at time t + n is p(st+n ). Notice that this is NOT
an option. Derive an expression for the forward price on the forward
contract.

Exercise 17 Suppose that there are N consumption goods every period. Agent
i’s preferences are represented by expected, discounted time-separable utility
for which the instantaneous utility function is
N
!
X (ci,n − γi,n )1−σ
u(ci,1 , ..., ci,n , ..., ci,N ) = φn
n=1
1−σ

where γi,n is the minimum consumption required by agent i of commodity

n = 1, ..., N .
Consider an economy where a full set of Arrow securities can be traded
every period (very importantly, the corresponding competitive equilibrium al-
location is a PO parametrized by some welfare weight α). Assume that dis-
count factors are identical (i.e., βi = β for all i = 1, ..., I). Show that there
exists a representative aggregate consumer with preferences represented by
1−σ
(∞ N
)
t
XX X (C n (s ) − γn )
β t π(st ) φn
t=0 t t n=1
1−σ
s ∈S
4.7. ARBITRAGE AND STATE PRICES 107

where Cn = Ii=1 ci,n stands for aggregate consumption of the n−th commod-
P

ity and γ = Ii=1 γi denotes the aggregate minimum consumption requirement

P
of the n − th commodity. In particular, show that the intertemporal marginal
rates of substitution of the aggregate representative agent coincides with the
IMRS’s of any agent i = 1, ..., I.
108 CHAPTER 4. ASSET PRICING AND TRADING
Chapter 5

Consumption and Savings

Throughout this chapter we will use the standard notation for the case of
only 1 consumption good.
P Let yi (st ) be the agent i realization of his endow-
ment at date t, with i∈I yi (st ) = y(st ) denoting the aggregate endowment.
As a starting point, consider the most rudimentary asset market struc-
ture to analyze consumption dynamics: an endowment economy where at
any period t markets for state contingent securities are completely absent
(i.e. there is no way to trade contingent to future states st+1 ) and, addition-
ally, there is no storage technology to transfer resources across periods (i.e.,
the consumption good is perishable). In this economy, an individual i who
receives an income process {yit }∞t=0 has no other choice than consuming her
income every period, i.e.
ci (st ) = yi (st ), (5.1)

for each st . This is the most trivial setting where individual consumption ab-
sorbs all the idiosyncratic fluctuations in individual income. That is, agents
cannot trade away any kind of individual risk.

5.1 Full Risk-Sharing with Complete markets

Consider an endowment economy with dynamically complete markets
where households can trade state contingent goods through a full set of en-
forceable contracts. Under those assumptions, competitive equilibrium allo-
cations are PO and, then, these allocations can be parametrized by welfare
weights.

109
110 CHAPTER 5. CONSUMPTION AND SAVINGS

5.1.1 Consumption Dynamics I:

Different Discount Factors
Consider the Pareto problem (PP) defined in Chapter 2 given some vec-
tor of welfare weights α corresponding the (efficient) competitive equilibrium
allocation. Under our assumption, we can characterize the unique solution
with the necessary and sufficient FOC given by
t
αi u0i (ci (st )(α))

βi
= 1, (5.2)
βh αh u0h (ch (st )(α))

I
X
ci (st )(α) = y(st ). (5.3)
i=1

Note that it follows from (5.2)

" t #
−1 βh αh 0
ci (st )(α) = (u0i ) u (ch (st )(α)) (5.4)
βi αi h

where (u0i )−1 , the inverse function of u0i , is well defined since ui is strictly
concave.
The limit distribution of consumption (and, consequently, equilibrium
wealth) will concentrate mass only on those agents with the highest discount
factors.

Proposition 8 Suppose that there two agents i and h such that βi < βh .
∞
Then, limt→∞ ci (st ) = 0 for all {st }t=0 .

Proof. Consumption is bounded from above by maxs y(s) and u0i (0) = +∞.
The result follows from inspecting condition (5.2) above.
In a market economy with complete markets, equilibrium asset prices will
adjust such that impatient agents borrow from patient agents to anticipate
consumption. Meanwhile, patient agents accumulate wealth such that they
end up consuming the aggregate endowment.
5.1. FULL RISK-SHARING WITH COMPLETE MARKETS 111

5.1.2 Consumption Dynamics II:

Identical Discount Factors
Now assume that all agents are equally patient; i.e., βi = βh for all i, h.
Notice that (5.2) implies that for all i, h

u0i (ci (st )(α)) αh

0
= , (5.5)
uh (ch (st )(α)) αi

for all st . Sometimes this is the definition of full risk-sharing (or full insur-
ance).1
Furthermore, it follows from (5.4) and (4.2) that individual consumption
for agent h depends only upon y(st ) for all st given a vector of welfare weights
(αi )Ii=1 . To see this, notice that
 
t 0 −1 αh 0 t
ci (s )(α) = (ui ) u (ch (s )(α)) , (5.6)
αi h

and then it follows from feasibility that ch (st ) is the unique solution to
 
0 −1 αh 0
X
(ui ) uh (ch (s )(α)) + ch (st )(α) = y(st ).
t
(5.7)
i6=h
α i

The only determinant of individual consumption is the realization of the

aggregate endowment. Very importantly, this implies that at any competitive
equilibrium, any kind of individual risk will be perfectly traded away across
agents.
1
This will also imply that individual consumption must be bounded from below for all
i. To see this, note that for any feasible allocation there exists some agent i such that
ci (st )(α) ≥ inf s y(s) t
I for all s . If for some agent h, consumption is not uniformly bounded
∞
from below, there exists an infinite subsequence {stk }k=1 such that ch (stk )(α) → 0. But,
on the other hand, note that (5.5) implies that:
αi 0 αi 0
u0h (ch (stk )(α)) = u (ci (stk )(α)) ≤ sup ui (ci (stk )(α))
αh i tk
(i,ci (s )) αh
αi 0
≤ sup u (inf y(s)/I) < ∞,
αh i i s

for all stk . But this expression contradicts that ch (stk ) → 0 since in that case u0h (ch (stk )) →
∞.
112 CHAPTER 5. CONSUMPTION AND SAVINGS

For the particular case where ui (c) = c1−σ for all i with σ > 0, (5.2)
implies that
 1/σ
ci (st )(α) αi
t
= ,
ch (s )(α) αh
and then (5.7) reduces to
1/σ 1/σ
α α
ci (s )(α) = P i 1/σ y(st ) = P i 1/σ C(st ),
t
(5.8)
h αh h αh

where C(st ) is aggregate consumption. Hence, individual consumption is

perfectly linearly correlated with aggregate consumption. For instance, if
there is no aggregate risk (i.e., y(st ) = y for all st ), individual consumption is
constant and thus independent of individual income fluctuations; individual
risk is completely smoothed out.

Empirical Implications
Expression (5.7) has strong empirical implications and, indeed, it has
been intensively tested empirically. Under CRRA preferences, for instance,
equation (5.8) implies that the log-change in individual consumption should
equal the log-change in aggregate consumption, for every individual, in every
period. If we estimate from micro-data the relationship
∆ log ci,t = β1 ∆ log Ct + β2 ∆ log yi,t + εi,t , (5.9)
where yi,t is current individual income. Then, the full risk-sharing hypothesis
implies to test β1 = 1, β2 = 0. On the other hand, autarky has predictions
that imply to test the alternative hypothesis: β1 = 0, β2 = 1; i.e., con-
sumption follows perfectly individual current income. In general, evidence
suggests that both hypothesis should be rejected, albeit the data seem to be
much closer to full risk-sharing in many contexts.2
Remark 8 From the empirical point of view, a reasonable model for con-
sumption must lie between autarky and full risk-sharing; i.e., agents should
have access to ”partial” consumption insurance.
2
Mace, Barbara (1991), ”Full Insurance in the Presence of Aggregate Uncertainty,”
Journal of Political Economy.
Cochrane, John (1991), ”A Simple Test of Consumption Insurance,” Journal of Political
Economy.
Parker, Jonathan and B. Preston (2005), ”Precautionary Saving and Consumption Fluc-
tuations,” American Economic Review, Vol 95 No 4 (September), 1119-43.
5.2. THE PERMANENT INCOME HYPOTHESIS (PIH) 113

5.2 The Permanent Income Hypothesis (PIH)

To study the Permanent Income Hypothesis, we need to consider an
alternative incomplete asset market structure. The key difference between
the Permanent Income Hypothesis model (PIH) and the complete markets
model is in the set of securities that households can trade. Under the PIH,
agents are restricted to trade only a non-contingent asset that pays one unit
of consumption next period, independently of the realization of the state
st+1 ; i.e., a one-period risk free bond.3 Agent i’s budget constraint reduces
to
ci (st ) + q(st )ai (st ) = yi (st ) + ai (st−1 ), (5.10)
where (q(st ), ai (st )) denote the bond price and bond holdings at st , respec-
tively.
In other words, the agent is cut-off of any state-contingent security and
has only access to a simple financial instrument to transfer resources over
time. We are interested in studying under what circumstances the absence
of insurance opportunities will induce the consumer to hold a certain amount
of the bond in order to smooth consumption.
To grasp the intuition, we study a partial equilibrium setting assuming
away fluctuations in the price of the risk-free bond; that is

1
q(st ) = q ≡ , (5.11)
1+r

for all st , where r is the (implicit) interest rate. Thus, we can impose a
No-Ponzi scheme condition as follows
" t+τ #
1
lim Et at+τ (5.12)
τ →∞ 1+r
 t+τ !
X X 1
= lim ... π(st , st+1 , ..., st+τ ) at+τ ≥ 0.
τ →∞
s s
1+r
t+1 t+τ

1

t+τ
where 1+r is the implicit price of one unit of the consumption good
delivered at date t + τ .
3
It is important to mention that the coming results depend upon the particular asset
market structure under study.
114 CHAPTER 5. CONSUMPTION AND SAVINGS

At date t, write (5.10) as follows

1
ai (st−1 ) = ci (st ) − yi (st ) + ai (st ),


1+r
and note that for each st+1 we have
1
ai (st ) = ci (st , st+1 ) − yi (st , st+1 ) + ai (st , st+1 ).
1+r
1
Multiply this expression by π(st , st+1 ) 1+r , add them up for all st+1 and
observe that
 
1 t 1 X t t t 1 t
ai (s ) = π(s , st+1 ) ci (s , st+1 ) − yi (s , st+1 ) + ai (s , st+1 ) .
1+r 1+r s 1+r
t+1

Denoting xt+τ to any random variable with realization x(st , st+1 , ..., st+τ ),
proceed repeatedly as before, use (5.12) and take the limit to obtain
∞  τ
t−1
X 1
ai (s ) = Et (ci,t+τ − yi,t+τ ) , (5.13)
τ =0
1 + r

where Et denotes the expectation conditional on st . Thus, individual finan-

cial wealth at date t equals the difference between expected discounted future
streams of consumption and income.

5.2.1 Quadratic Utility: Certainty Equivalence and Con-

sumption as a Random Walk.
The most strict version of the PIH makes two key additional assump-
tions. First, households have quadratic utility functions; i.e.
1
u(c) = b1 c − b2 c2 , (5.14)
2
b1 , b2 > 0. Secondly, the risk free interest rate equals the discount rate; i.e.,
β(1 + r) = 1. Indeed, this last assumption is less important; it helps to make
the computations below less cumbersome.
With these two additional assumptions, agent i’s Euler equation reads

1 X
b1 − b2 ci (st ) =β π(st , st+1 )(b1 − b2 ci (st , st+1 )) (5.15)
(1 + r) s t+1
5.2. THE PERMANENT INCOME HYPOTHESIS (PIH) 115

and thus
ci (st ) = Et (ci,t+1 ) (5.16)
i.e. individual consumption is a martingale.45
Dropping the index i, it is useful to note that from the law of iterated
expectations and the martingale property (5.16) imply that

Et (ct+2 ) = Et (Et+1 ct+2 ) = Et (ct+1 ) = ci (st ),

and therefore, more in general, we obtain

Et (ct+τ ) = ct , for any τ ≥ 0. (5.17)

Using (5.17) in expression (5.13), we obtain

∞  τ ∞  τ
t−1
X 1 X 1
a(s )+ Et (yt+τ ) = Et (ct+τ )
τ =0
1+r τ =0
1+r
∞  τ
t
X 1
= c(s )
τ =0
1+r
1+r
= c(st ) .
r
and therefore we express consumption
" ∞  τ #
r X 1
c(st ) = a(st−1 ) + Et (yt+τ ) (5.18)
1+r τ =0
1+r
r  t−1
a(s ) + H(st ) ,

=
1+r

where H(st ) denotes human wealth at st ; i.e. the expected discounted value
of future individual earnings. Since financial wealth is at = a(st−1 ), define
r
permanent income as the annuity value (i.e. 1+r ) of total (human and fi-
t t−1 t
nancial) wealth W (s ) = a(s ) + H(s ). Therefore, we can summarize this
important result for this setting.
4
We say that the stochastic process {xt } is a martingale if, E[|xt |] < ∞ for all t, and
if for any τ > 0, Et [xt+τ ] = x(st ).
5
Observe that without the assumption that β(1 + r) = 1, (5.15) generalizes to ci (st ) =
α0 + α1 Et (ci,t+1 ) (that is, β(1 + r) = 1 implies that α0 = 0 and α1 = 1).
116 CHAPTER 5. CONSUMPTION AND SAVINGS

Remark 9 If preference are quadratic and β(1 + r) = 1, then individual

consumption follows a martingale process and equal permanent income, i.e.
the annuity value of human and financial wealth.

Hence, under the assumption that utility is quadratic and that the real
interest rate is a constant and equal to the consumer’s rate of subjective
time preference, optimal consumption is proportional to permanent income
or wealth. One implication of this representation is that permanent changes
in income lead to permanent changes in consumption.

Certainty Equivalence
Notice that if one solves the non-stochastic version of the PIH problem
stated earlier, the Euler equation (5.15) implies immediately that ct+1 = ct .
Thus, proceeding as before in the budget constraint we can conclude that
" ∞  τ #
r X 1
ct = at + yt+τ . (5.19)
1+r τ =0
1 + r

If one compares equation (5.18) with (5.19), one can observe that individ-
ual consumption satisfies a certainty equivalence property. That is, one can
obtain the solution of the stochastic problem as follows. First, solve the de-
terminist problem and, then, instead of the individual income (yt+τ ) plug its
conditional expectations (Et (yt+τ )). This property is a direct consequence of
linear-quadratic preferences: the variance and higher moments of the income
process do not matter for consumption choices.

Consumption Dynamics
Since Et−1 ct = ct−1 for all t, it follows from (5.18), the change in con-
sumption at time t equals
r 
∆ct = c(st ) − c(st−1 ) = c(st ) − Et−1 (ct ) = W (st ) − Et−1 (Wt ) .

1+r
(5.20)
t−1
Since agents decide asset holdings for period t at s , we have that
Et−1 (at ) = a(st−1 ). Thus, since W (st ) = a(st−1 ) + H(st ), we have that
Et−1 (Wt ) = a(st−1 ) + Et−1 (Ht ) and, then, an innovation to human wealth
5.2. THE PERMANENT INCOME HYPOTHESIS (PIH) 117

at t reduces to
W (st ) − Et−1 (Wt ) = H(st ) − Et−1 (Ht ) (5.21)
∞   τ
X 1
= [Et (yt+τ ) − Et−1 (Et (yt+τ ))]
τ =0
1+r
∞  τ
X 1
= (Et − Et−1 ) (yt+τ )
τ =0
1+r

since the law of iterated expectations implies that Et−1 (Et (yt+j )) = Et−1 (yt+j ).
Using (5.21) in (5.20) we obtain
∞  τ
r X 1
∆ct = (Et − Et−1 ) (yt+τ ) (5.22)
1 + r τ =0 1 + r

Equation (5.22) leads us to conclude the following important result in this

setting.

Remark 10 Under the PIH, changes in individual consumption between t−1

and t are proportional to the revised value of expected individual earnings
driven by the arrival of new information at the current period.

Example with a Specific Income Shock

To make further progress, we make some additional assumptions on the
individual income process. Indeed, we choose a specification that is very
common in labor economics: individual income (imagine labor income) is
the sum of two orthogonal components, a permanent component ytp which
follows a martingale, and a transitory component ut , that is i.i.d. over time.
That is, income is determined by
yt = ytp + ut , (5.23)
ytp = yt−1
p
+ vt
where vt is the innovation to the permanent component and the process
{vt } i.i.d. over time. Additionally, we assume that {ut }, {vt } are linearly
independent i.i.d. processes with E(ut ) = E(vt ) = 0 for all t.
p p
Since yt−1 = yt−1 − ut−1 and yt = yt−1 + vt + ut , individual income can be
expressed as
yt = yt−1 + vt + (ut − ut−1 ) (5.24)
118 CHAPTER 5. CONSUMPTION AND SAVINGS

Now we show that we can express changes in individual consumption as

a function of the permanent innovation, vt , and the transitory innovation,
ut , only. To see this, observe first that for any τ ≥ 0

Et−1 (yt+τ ) = Et−1 (yt+τ −1 + vt+τ + ut+τ − ut+τ −1 )

= Et−1 (yt+τ −1 ) + Et−1 (ut+τ ) − Et−1 (ut+τ −1 ) + Et−1 (vt+τ )
= Et−1 (yt+τ −1 ) − Et−1 ut+τ −1 ,

since both Et−1 (ut+τ ) = E(ut+τ ) = 0 and Et−1 (vt+τ ) = E(vt+τ ) = 0 for all
τ ≥ 0. Similarly, for any τ ≥ 0 we have that

Et (yt+τ ) = Et (yt+τ −1 − ut+τ −1 + vt+τ + ut+τ )

= Et (yt+τ −1 ) − Et (ut+τ −1 ) + Et (ut+τ ) + Et (vt+τ )

Hence, for τ = 0 we have that

(Et − Et−1 ) (yt ) = vt + ut

i.e., the unexpected change in income yt is the sum of the permanent and
the transitory innovations at time t. For τ = 1 we have that

(Et − Et−1 ) (yt+1 ) = (Et − Et−1 ) (yt ) − (Et − Et−1 ) (ut )

+ (Et − Et−1 ) (vt+1 + ut+1 )
= vt + ut − ut = vt

Now we complete our argument by induction. We argue that (Et − Et−1 ) (yt+τ +1 ) =
vt if (Et − Et−1 ) (yt+τ ) = vt for all τ ≥ 1. To see this, observe that

(Et − Et−1 ) (yt+τ +1 ) = (Et − Et−1 ) (yt+τ ) − (Et − Et−1 ) (ut+τ )

+ (Et − Et−1 ) (vt+τ +1 + ut+τ +1 )
= vt .

Then, our claim is true by induction since (Et − Et−1 ) (yt+1 ) = vt and,
thus, the forecast revision of individual income between t − 1 and t at any
period beyond t just equals the permanent innovation at time t, vt . Going
5.2. THE PERMANENT INCOME HYPOTHESIS (PIH) 119

back to expression (5.22), the innovation to permanent income is

∞  τ
r X 1
∆ct = (Et − Et−1 ) (yt+τ )
1 + r τ =0 1 + r
" ∞  τ #
r X 1
= (Et − Et−1 ) (yt ) + (Et − Et−1 ) (yt+τ )
1+r τ =1
1+r
" ∞  τ #
r X 1
= (ut + vt ) + vt
1+r τ =1
1+r
" ∞  τ #
r X 1
= ut + vt
1+r τ =0
1 + r
r
= ut + vt .
1+r
Therefore, households adjust their consumption responding to the annu-
alized changes in income. This means that they will respond only weakly to
transitory shocks (ut ), whereas they will respond one by one to permanent
income shocks (vt ). This is so because the former shocks have only a small
effect on the permanent income, while the latter change permanent income
one by one, by definition.
We summarize the most important implications of this analysis.

1. The agent’s optimal consumption decisions obey certainty equivalence;

comparing the rules for optimally allocating consumption over time in
the certainty and the uncertainty case, we see that these rules are iden-
tical (of course, the realized consumption path in the uncertainty case
will ex-post differ from the consumption path without uncertainty).

2. For the general case (that is, without the assumption β(1 + r) = 1),
consumption should obey the regression

ct+1 = α0 + α1 ct + εt+1

where εt+1 is a random variable satisfying Et (εt+1 ) = 0. The main im-

plication of the PIH model in its standard, certainty equivalence form
is that period t + 1 consumption, ct+1 , is perfectly predicted by period
t consumption, ct , and no other variables that are in the households’
120 CHAPTER 5. CONSUMPTION AND SAVINGS

information set at period t should help predict its dynamics. In par-

ticular, once ct is included in the regression, current income yt , current
assets holdings, past consumption or other variables should not enter
with significant coefficients.

3. The income realization in period yt+1 does affect ct+1 , but only that part
that is an unexpected change from income in period t. The component
of income yt+1 that is already predicable by yt or any other variable
that is in the period t information set should not affect consumption
in period t. Note that with complete markets not even unexpected
changes in individual income should affect consumption, since agents
can perfectly insure against these unexpected idiosyncratic changes.

The permanent income hypothesis has been subject to much empirical

testing. Indeed in the data consumption typically responds strongly to per-
manent income shocks and this response tends to be much stronger than
the response of consumption to transitory shocks. However, the response
to transitory shocks is much larger than we would predict from the model
leading to suspect of binding liquidity/borrowing constraints.

5.2.2 Quadratic Utility:

Wealth Dynamics and Borrowing Constraints
So far, we have ignored the presence of borrowing constraints. Indeed, we
imposed a No-Ponzi scheme condition, but never checked whether it is ac-
tually binding. We show below that the validity of this strategy depends on
the income process.
Wealth dynamics with borrowing constraints
First, note that from the agent’s budget constraint we have

1
a(st−1 ) = c(st ) − y(st ) + a(st ),


1+r
and thus rearranging, we obtain an expression for wealth changes

∆at+1 = (1 + r)y(st ) + ra(st−1 ) − (1 + r)c(st ). (5.25)

Using the optimal consumption choice from (5.18) in this expression, we

5.2. THE PERMANENT INCOME HYPOTHESIS (PIH) 121

obtain
" ∞  τ #
X 1
∆at+1 = (1 + r)y(st ) + ra(st−1 ) − r a(st−1 ) + Et (yt+τ )
τ =0
1+r
∞  τ !
X 1
= (1 + r)y(st ) − ry(st ) − r Et (yt+τ ) .
τ =1
1+r

Observe that for all τ

 τ −1  τ
r 1 1
= −
(1 + r)τ 1+r 1+r

and thus
" ∞  #
X 1 1
∆at+1 = y(st ) − Et (yt+τ ) − Et (yt+τ ) (5.26)
τ =1
(1 + r)τ −1 (1 + r)τ
∞  τ −1 !
X 1
=− Et (∆yt+τ )
τ =1
1+r

This last expression has strong implications. For instance, suppose that
the individual income process follows a random walk, i.e. yt = yt−1 + εt ,
{εt } is an i.i.d. process with E(εt ) = 0. Then it is easy to see that ∆yt+τ =
εt+τ for all τ and, thus, (5.26) implies that ∆at+1 = 0 for all t. Therefore,
individual financial wealth is constant at the initial level (i.e. at = a0 for all
t) and, consequently, any properly chosen borrowing constraint will never be
binding. The reason for this result is that financial wealth changes only if
the individual is consuming just part of its income changes (and saving the
remaining part) in order to smooth consumption. With permanent shocks,
however, all income shock is consumed in every period since (5.18) reduces
to
r
c(st ) = a0 + y(st ).
1+r
However, this result depends on the assumption that the individual in-
come process follows a random walk. To see this, suppose instead that the
individual income process is i.i.d. and, thus, we have that ∆yt+1 = εt+1 − εt ,
122 CHAPTER 5. CONSUMPTION AND SAVINGS

∆yt+2 = εt+2 − εt+1 , and so on. In this case, (5.26) reduces to

∞  τ −1 !
X 1
∆at+1 = − Et (∆yt+τ ) (5.27)
τ =1
1+r
∞  τ −1
X 1
= − Et (εt+τ − εt+τ −1 )
τ =1
1 + r
= εt .

This means that financial wealth follows a random walk, and since i.i.d shocks
compound with probability one, as a result we obtain that any constraint on
asset holding will be binding sooner or later.
To conclude, in this stylized setting whether ignoring borrowing con-
straint is troublesome or not depends on the specific individual income pro-
cess assumed. Thus, this result highlights the fact that borrowing constraint
cannot be simply ignored in general.

5.3 Precautionary Savings: Prudence and Bor-

rowing Constraints
In this section we look at partial equilibrium extensions of the basic PIH
model. That is, we will continue to assume that the interest rate process that
the agent faces is exogenously given (partial equilibrium) and that the agent
can only self-insure against income fluctuations by trading one-period risk
free bonds (i.e. by borrowing and lending at a risk free rate).
We will relax two other crucial assumptions underlying the martingale
hypothesis from last section. First, in order to incorporate a precaution-
ary savings motive into the model, we will relax the assumption of linear
marginal utility. So agents will reduce current consumption (and hence in-
crease saving) as a reaction to an increase in future income risk while with
linear marginal utility, certainty equivalence implies that a mean preserv-
ing spread in individual income risk has no impact on saving (see equation
(5.26)).
Secondly, so far we assumed that agents can always borrow as much as
desired to smooth consumption over time (of course, subject to a non-binding
no Ponzi condition). This may be an assumption with little empirical appeal
and its relaxation (i.e. the imposition of some form of potentially binding
5.3. PRECAUTIONARY SAVINGS: PRUDENCE AND BORROWING CONSTRAINTS123

borrowing constraint, a so-called liquidity constraint) will indeed invalidate

the martingale property of consumption.

5.3.1 Prudence: Warming Up in A Two-Period Model

Given an interest rate R, consider a simple two-period consumption-
saving problem
max u(c0 ) + βE [u(c1 )]
{c0 ,c1 ,a1 }

subject to

c0 + a1 = y0
c1 = Ra1 + ye1
c0 , c1 ≥ 0,

where y0 is initial income given at 0 and ye1 denotes next period random
income. Under the simplifying assumption βR = 1, the FOC implies that

u0 (y0 − a1 ) = E [u0 (Ra1 + ye1 )] (5.28)

The LHS in this expression is increasing in a1 since u00 < 0. On the other
hand, the RHS is decreasing for the same reason and, hence, a∗1 is uniquely
determined. Note that current consumption c0 is determined by

c0 = y0 − a∗1 .

Mean-preserving Spreads and Prudence

Rothschild and Stiglitz (1970) have introduced mean-preserving spreads
as a convenient way of characterizing the riskiness of two distributions with
the same mean. Mas-Colell et al (1995, Proposition 6.D.2) show that for
any two distributions with the same mean F and G, G is a mean preserving
spread of F if and only if F second order stochastically dominates G.6
6
For any two distributions with the same mean F and G, F second order stochastically
dominates G if for every nondecreasing concave function v : R+ → R, we have
Z Z
v(x)dF (x) ≥ v(x)dG(x).
124 CHAPTER 5. CONSUMPTION AND SAVINGS

Let us see what happens to optimal consumption at t = 0 if the uncer-

tainty over income next period ye1 rises, i.e. as future income becomes riskier.
Consider a mean-preserving spread of ye1 . Define
≈
y 1 = ye1 + ε1 (5.29)

where ε1 is the stochastic component with E (ε1 ) = 0 and var(ε1 ) = σε and

≈ ≈ ≈
thus, E(y 1 ) = E (e
y1 ) and var(y 1 ) > var (e
y1 ) (evidently, y 1 = ye1 if σε = 0).
Equation (5.28) becomes

u0 (y0 − a1 ) = E [u0 (Ra1 + ye1 + ε1 )] (5.30)

We claim that if u0 is convex (i.e., u000 > 0), then a mean-preserving

spread of ye1 (i.e. σε > 0) will increase the value of the RHS of (5.30), which
induces a rise in a∗1 (and a fall in c∗0 ) to keep the equality in (5.30). To see
this, observe that −u0 is nondecreasing since u00 < 0 and concave if u000 > 0.
Denote a1 (σε ) to the unique solution of equation (5.30) given σε . Therefore,
a mean-preserving spread of ye1 , given a1 (0) at (5.30), will make the expected
value of −u0 decrease. This means that

u0 (y0 − a1 (0)) < E [u0 (Ra1 (0) + ye1 + ε1 )] ,

and thus a1 (σε ) > a1 (0) to recover the equality at (5.30). Roughly speaking,
we can say that if u000 > 0, individual savings are increasing in the riskiness
of her individual income.
The convexity of the marginal utility (or u000 > 0) is called prudence and
refers to a property of preference, like risk aversion. Indeed, risk aversion
refers to the curvature of the utility function, whereas prudence refers to the
curvature of the marginal utility function. More precisely, Kimball (1990)
defines the index of absolute prudence as the ratio −u000 (.)/u00 (.), so in a
similar vein to the Arrow-Pratt index of absolute risk-aversion, −u00 (.)/u0 (.).

Conclusion 17 If the marginal utility is convex (or u000 > 0), then the risk
averse individual is prudent and, consequently, a rise in future income uncer-
tainty leads to a rise in current saving and a decline in current consumption.

It can be easily seen that any utility function with decreasing absolute
risk aversion (DARA class), which includes CRRA, displays positive third
5.3. PRECAUTIONARY SAVINGS: PRUDENCE AND BORROWING CONSTRAINTS125

derivative. To see this, let α(c) be the coefficient of absolute risk aversion.
Then,
u00 (c)
α(c) = − 0 , (5.31)
u (c)
and therefore
u000 (c)u0 (c) − [u00 (·)]2
α´(c) = −
[u0 (c)]2
With DARA α0 (c) < 0, then we have that
2
−u000 (c)u0 (c) + [u00 (·)] < 0,
and thus this implies that
[u00 (c)]2
u000 (c) > >0
u´(c)
Intuitively, a rise in uncertainty reduces the certainty equivalent for next
period income and, with DARA, this effectively increases the degree of risk-
aversion of the agent, which induces agents to save more (other things equal).
Prudence is a motive for additional saving in order to take precaution
against possible negative realizations of future income shocks. In this sense,
saving induced by prudence are called precautionary savings or self-insurance.
In this simple two-period partial equilibrium model, one can define precau-
tionary savings due to income uncertainty with variance σε , as the difference
between optimal asset holdings under uncertainty, a∗1 (σε ), and the optimal
asset holdings under certainty, a∗1 (0).7
Whenever Rβ > 1, there are other saving motives; i.e. intertemporal mo-
tives. In this case agents have incentives to postpone consumption and save
more since they are patient relative to the market. On the other hand, the
saving motive of the pure PIH with quadratic utility (i.e. uncertainty has ba-
sically no role) and Rβ = 1 (i.e. intertemporal motives are inactive) is called
smoothing motive. The agent wants to smooth consumption through income
shocks. That is why the saving motives just analyzed above are associated to
future income uncertainty and called precautionary or self-insurance motive.
Finally, in a life-cycle models individuals face a retirement period. Then, dur-
ing the working stage of their life-cycle, they would have a life-cycle motive
for savings associated to the desire of smoothing consumption.
7
It should be underscored that the term prudence characterizes preferences, whereas
precautionary saving characterizes agent’s optimal behavior: prudence leads to precau-
tionary saving, but both concepts should be kept separately.
126 CHAPTER 5. CONSUMPTION AND SAVINGS

5.3.2 Prudence: A Basic Multiperiod Model

Here we generalize the two period model to a multiperiod model with i.i.d
income shocks and finite-horizon (i.e., T < ∞). This last assumption lets
us isolate the impact of income uncertainty on savings from that derived
from potentially binding borrowing constraints (see below). In this case the
household’s problem can be written as
T X
X
max β t π(st ) u(c(st ))
(ci ,ai )
t=0 st

subject to

c st + a st ≤ y (st ) + R a st−1 for all st ,

c st


≥ 0, and a (s0 ) given,
T


a s = 0.

Note that since income process {yt } is i.i.d, we can define a sufficient
statistics for the agent’s choice problem, the so-called cash in hand, denoted
by w(st ) = y(st )+R a(st−1 ). This is so because (at , yt ) always enter additively
and, very importantly, the current levels of yt do not provide any information
about the future realization of income shocks since shocks are i.i.d. Hence
the agent’s problem can be reformulated as
T X
X
β t π(st ) u(c st ),


max
(c)
t=0 st

subject to

w st , st+1 = y(st+1 ) + R w st − c st for all st , st+1 ,

c st


≥ 0, and w (s0 ) given,
T
= c sT for all sT .



w s
5.3. PRECAUTIONARY SAVINGS: PRUDENCE AND BORROWING CONSTRAINTS127

Thus, FOC’s with respect to ct imply that for all st

X
u0 (c st ) = βR π(st+1 ) u0 y(st+1 ) + R w st − c st






st+1

= βR Et u0 yt+1 + R w st − c st





.

i.e., the precautionary saving results of the two-period model go through as

long as the agent is prudent (i.e., u000 > 0). Now it is convenient to make the
analysis considering the reaction of current consumption to increases in the
riskiness of the agent’s income.

5.3.3 Borrowing Constraints

Let us briefly discuss what can happen in the presence of potentially
binding borrowing constraints. To isolate the role of borrowing constraints,
we abstract from prudence altogether and assume that households have
quadratic utility so that certainty equivalence holds. Assume that infinitely-
lived agents (i.e. T = ∞) cannot borrow at all, i.e. they face the constraint
a(st ) ≥ 0 for all st . It should be underscored that this particular class
of borrowing constraint is just for expositional purposes and, indeed, all the
analysis below goes through for borrowing constrains of the form a(st ) ≥ −A
for all st for some A > 0. To see this, define b(st ) = (ai (st ) − A) for all st
and rewrite the budget constraint as follows

c st + b st ≤ y (st ) + R b st−1 + (R − 1)A

where

b (s0 ) given and b st ≥ 0.

Depending upon the individual income process, these constraints may be

binding or not and, thus, they must be taking into account explicitly. It is
more transparent to go back to our formulation in previous sections where
the risk-free bonds can be traded at the price q = R1 and pays one unit of
consumption next period. In this case, it is easy to see that (5.15) generalizes
to 
Et (ct+1 ) if a(st ) > 0
c(st ) = (5.32)
y(st ) + a(st−1 ) if a(st ) = 0.
128 CHAPTER 5. CONSUMPTION AND SAVINGS

The first line is just the previous agent’s FOC’s when the constraint is not
binding. The second line descends directly from the reformulated budget
constraint
1
a(st ) = (y(st ) + a(st−1 ) − c(st )),
R
when the borrowing constraint is binding, i.e. a(st ) = 0. In this case, the
agent would like to borrow to finance current consumption further, but she
is not allowed to do so and so she must limit herself to consume as much as
she has at hand, all her resources.
Will this borrowing constraint be ever binding? For instance, suppose
that income is a random walk. Inspecting equation (5.27) we had concluded
above that financial wealth would follow a random walk, i.i.d shocks com-
pound with probability one and the non-negativity constraint on asset hold-
ing would be binding sooner or later.
Now notice that (5.32) can be reformulated as follows

c(st ) = min{y(st ) + a(st−1 ), Et (ct+1 )} (5.33)

t−1
= min{y(st ) + a(s ), Et (min{yt+1 + at+1 , Et+1 (ct+2 )})}

This expression has two immediate implications. First, remember that in

the absence of binding borrowing constraints at any period, we have that

c(st ) = Et (ct+τ +1 )

for all τ ≥ 0. Suppose that there exists some future period t + τ such that
for some realization of the income shock, st+τ , the borrowing constraint is
binding and thus y(st+τ ) + a(st+τ −1 ) < Et+τ (ct+τ +1 ). Since all probabilities
are positive, this implies that for any st+τ −1

Et+τ −1 y(st+τ ) + a(st+τ −1 ) < Et+τ −1 (Et+τ (ct+τ +1 )) .

Using this fact and the law of iterated expectations in equation (5.33), we can
conclude that c(st ) < Et (ct+τ +1 ). But this makes evident that even though
the liquidity constraint is not necessarily binding in period t, future potential
binding borrowing constraints affect current consumption.8
8
Thus, studies that provide support to the hypothesis that borrowing constraints are
not currently binding do not demonstrate that borrowing constraints are absent in general
or do not affect consumption decisions.
5.3. PRECAUTIONARY SAVINGS: PRUDENCE AND BORROWING CONSTRAINTS129

Secondly, consider a mean-preserving spread in income risk. This makes

the states with low income more likely and, consequently, the corresponding
borrowing constraints are also more likely to bind in the future. More pre-
cisely, if the set of income realizations such that the borrowing constraint
binds enlarges, Et (min {yt+1 + at+1 , Et (ct+2 )}) becomes smaller since the
min operator takes the first value in more states. This reduces the value
of Et (ct+1 ) and so does current consumption ct . Intuitively, saving increases
in reaction to increases in future income risk because risk-averse agents fear
future contingencies with low income and, aware of their inability to smooth
low income shocks via higher borrowing, they increase their precautionary
savings to prevent this situation (i.e. they save for self-insurance). Very
importantly, these precautionary savings shows up without prudence (i.e.
without a precautionary saving motive) but purely from the existence of
binding liquidity constraints (and risk aversion). Hence, when observing in-
creases in saving as reaction to increased income uncertainty, it may have a
preference based interpretation (i.e. agents are prudent, u000 > 0) or due to
credit markets imperfections preventing or limiting uncontingent borrowing.

Conclusion 18 Even if prudence is assumed away (e.g. quadratic utility

function), a rise in future income uncertainty in the presence of potentially
binding borrowing constraints can lead to a rise in current saving for precau-
tionary reasons, with a corresponding decline in current consumption.9

5.3.4 A Natural Debt Limit

The previous analysis has shown how critical is to model debt limits. We
imposed an exogenous borrowing constraint like at+1 ≥ −A, where A > 0
is a parameter (in our previous case A = 0). Now we can move forward to
model a natural debt limit.
First suppose that the income process {yt }∞
t=0 is deterministic. Since
consumption is assumed bounded from below, in particular non-negative,

9
It is important to emphasize that even though we have showed these result for
quadratic utility, it is a general result that hold for concave utility functions.
130 CHAPTER 5. CONSUMPTION AND SAVINGS

i.e. ct ≥ 0 for all t, we can iterate the budget constraint to obtain

at+1 at+1
at = −yt + ct + ≥ −yt +
1+ r 1 + r
1 at+2 yt+1 at+2
≥ −yt + −yt+1 + ≥ −yt − +
1+r 1+r 1+r 1+r
 j
X 1
≥ − yt+j
j=0
1 + r

Thus, this constraint restricts the household to accumulate more debt

than the present discounted valued of total individual income; i.e. the max-
imum level that can be repaid by an agent consuming zero every period.
Very importantly, observe that assuming away uncertainty makes the asset
market structure dynamically complete; roughly speaking, there is an asset
for a unique future state.
Now suppose that we come back to the uncertainty case where income is
stochastic. The requirement in this case is more sophisticated: households
must repay their debt almost surely (i.e. with probability 1). With that
purpose, substitute yt at each t with the lowest possible realizations of the
income shock with positive probability, denoted ymin = mins {y(s)}. The
so-called natural debt limit when there is only an uncontingent risk-free asset
becomes
1+r
−A = − ymin (5.34)
r
This expression equals the (absolute value of) present discounted value of the
lowest income realization with positive probability and, thus, no exogenous
borrowing constraint can ever be looser than this limit (see Aiyagari (1994)).
Very importantly, note that an agent borrowing up to this borrowing limit
(5.34) faces a positive probability next period that consumption is zero. We
argue that the natural debt limit (5.34) will not be binding as long as the
utility function satisfies the Inada conditions; i.e. u´(0) = ∞. To see this,
suppose that an agent has borrowed up to this limit; i.e. a(st−1 ) = − 1+r
r
ymin .
Consider the positive probability event such that income is precisely ymin .
Observe that, since a(st ) ≥ − 1+r
r
ymin , the budget constraint implies that
1 1+r 1
c(st ) = a(st−1 ) + ymin − a(st ) = − ymin + ymin − a(st )
1+r r  1 +
 r
1 1 t 1 1 1+r
= − ymin − a(s ) ≤ − ymin − − ymin = 0
r 1+r r 1+r r
5.3. PRECAUTIONARY SAVINGS: PRUDENCE AND BORROWING CONSTRAINTS131

This shows that the agent will have to consume zero with positive probability.
This cannot be an optimal choice since it is ruled out by the Inada condition
in that state. Thus, preferences by their own will insure that the natural
borrowing limit will never bind. Therefore, while you can safely assume
interior solutions if the natural debt limit is imposed, this is not true for
ad-hoc debt limits. Importantly, this makes evident that the peculiar case
studied above (see equation (5.27)) where financial wealth followed a random
walk (and, consequently, any finite borrowing limit would be violated with
probability one) draws heavily on the assumption of quadratic utility, where
the Inada condition is not satisfied.

Exercises
Exercise 18 (Wages and Consumption) A household solves
∞
X
max β t U (ct , lt )
{ct ,lt ,St }
t=0

where ct is consumption and lt is leisure, subject to the time constraint

lt + nt = 1

and the budget constraint

wt nt + (1 + r) St = ct + St+1

where nt is the labor supply, {wt } is a known sequence of wage rates, r is the
real and constant interest rate, St is savings, and S0 is given. The function
U is continuous, increasing, and strictly concave in both arguments. Con-
sumption and leisure are normal goods. Households maximize the discounted
present value of utility by choosing consumption and leisure.

(a) Set up the maximization problem and derive the first-order conditions
using the sequential approach. Demonstrate that optimal labor supply
can be expressed as a function:

nt = H(ct , wt ).

Describe the properties of the function H.

132 CHAPTER 5. CONSUMPTION AND SAVINGS

(b) Assume that β (1 + r) = 1. Describe the behavior of consumption and

savings over time, given the sequence of wages.

(c) Suppose that the utility function takes the form

α1 ln ct + α2 ln lt .

Find the optimal consumption, savings, leisure, and labor supply ex-
pressed as a function of wages and past savings.

(d) Suppose that the wage rate follows a specific pattern: in even periods
let wt = wh and in odd periods wt = wl where wh > wl . Describe how
consumption, saving, and labor move over time.
Chapter 6

Overlapping Generation
Economies

In this chapter, we study overlapping generations economies of the type intro-

duced by Samuelson (1958). The main feature of this model is that an agent
“lives” for two periods. Each period, a new generation is borned and enters
into trading so that there are always two types of agents, young and old. The
overlapping generations model has a special type of friction where agents of
different generations are unable to commit to certain transactions over time.
Bilateral borrowing and lending arrangements are not available because of
incomplete participation stemming from the demographic structure.
The overlapping generations model has the critial property that com-
petitve equilibirum allocations cannot be Pareto optimal, implying that a
reallocation of resources would make some agents better off and no agent
worse off. The dynamic inefficiency arises because there is a double infinity
of agents and time periods, according to Shell [30]. We explore the implica-
tions of dynamic efficiency in the overlapping generations model.
In a deterministic model with production and capital accumulation, Di-
amond [12] showed that there may be an over-accumulation of capital in an
OLG economy, leading to a dynamic inefficiency. By decreasing the capital
stock and redistributing the good, the dynamic inefficiency can be elimi-
nated. The presence of the dynamic efficiency can be determined from a
simple relationship between the rate of growth of the population and the
marginal productivity of capital, or equivalently, the real interest rate.

133
134 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

6.1 Pure Exchange OLG Economies

Consider an economy where at every period t = 1, 2, 3...., a generation with
Nt = (1 + n)t N0 identical agents are born. Each agent born at t lives two
periods and we index generations by the date of birth. At date t = 1, there
is an initial old generation born at date t = 0 with N0 agents. Without loss
of generality, we assume N0 = 1. An agent born at date t at date t and old
at date t + 1.
There is only one consumption good every period. Consequently, the
commodity space is X = R2 and the consumption set is given by Xt = R2+
for all t ≥ 1, and X0 = R+ . Preferences are represented by a separable utility
function
u (c1t ) + βu (c2t+1 ) ∀ (c1t , c2t+1 ) ∈ Xt , ∀t ≥ 1
where β ∈ (0, 1), u : R+ → R+ is continuously differentiable, strictly increas-
ing, strictly concave and satisfies Inada conditions. Assume u (c10 ) is strictly
increasing and continuous. Let ct = (c1t , c2t+1 ) denote a consumption bundle
for an agent born at t.
Each agent born at t ≥ 1 is endowed with ωt = (ω1t , ω2t+1 ) ∈ Xt units
of the consumption good. Assume that ωt  0 for all t. The initial old
generation is endowed with ω01 ≥ 0.

6.1.1 Competitive Equilibrium

Definition 20 c = {c10 , (c1,1 , c2,1 ) , ..., (c1t , c2t+1 ) , ...} is an allocation if (c1t , c2t+1 ) ∈
Xt for all t.
An allocation c is feasible if

Nt−1 c2t + Nt c1t ≤ Nt−1 ω2t+1 + Nt ω1t

for all t ∀t ≥ 1.
c = {b
A feasible allocation b c10 , (b
c1,1 , b
c2,1 ) , ..., (b
c1t , b
c2t+1 ) , ...} is PO if there
is no alternative feasible e
c such that

u (e c2t+1 ) ≥ u (b
c1t ) + βu (e c1t ) + βu (b
c2t+1 )

with some strict inequality for some generation t.

Suppose that each agent can buy and sell the consumption good in each
period at perfectly foreseen prices. Suppose that p1 = 1; i.e., the consumption
6.1. PURE EXCHANGE OLG ECONOMIES 135

good at date 1 is the numeraire. Hence, pt is the price of the consumption

good at date t in terms of the consumption good at date 1.
Let P = {p = {p1 , ..., pt , ...} : pt ≥ 0, ∀t ≥ 1 and p1 = 1} be the set of se-
quences of present (date 1) prices.

Definition 21 A (Perfect Foresight) Competitive Equilibrium for this econ-

c = {b
omy is an allocation b c10 , (b
c1,1 , b
c2,1 ) , ..., (b
c1t , b
c2t+1 ) , ...} and a price system
pb ∈ P such that
OLG 1. ∀t ≥ 0, (b c1t , b
c2t+1 ) solves the problem for the representative
agent of generation t
max u (c1t ) + βu (c2t+1 )
(c1t ,c2t+1 )∈Xt

subject to
pbt c1t + pbt+1 c2t+1 = pbt ω1t + pbt+1 ω2t+1
c10 = ω01 .
The initial old generation simply consumes b
OLG 2. Feasibility: for all t ≥ 1
Nt−1b
c2t + Ntbc1t = Nt−1 ω2t+1 + Nt ω1t
c2t ωt
⇐⇒ c1t = t−1 + ωtt
b
+b
1+n 1+n
Remark. Notice that an agent at date t is trading both ctt and ct+1 t (and
t t+1
also selling ωt and ωt ), given the price system. Also, we are not assuming
that agents in different generations are identical since both preferences and
endowments can be different. However, all agents in the same generation are
identical (ut and ωt are the same for all agents born at date t).
We analyze the agent t’s problem. Given p ∈ P, the constraint set is
compact, non-empty (i.e., it contains the endowment point) and the objective
function is continuous. Thus there exists a solution to the problem. Since
the utility function is strictly concave this solution is unique. Suppose for
simplicity that (ω1t , ω2t+1 ) = (ω1 , ω2 ) for all t and let c1t = c1 (pt , pt+1 ) and
c2t+1 = c2 (pt , pt+1 ) be the unique solution to the agent t’s problem. Since
demand functions are homogeneous of degree zero we can write them
 
pt+1
c1t = c1
p
 t 
pt+1
c2t+1 = c2
pt
136 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

where qt ≡ pt+1
pt
is the price of the consumption good t + 1 in terms of
1
consumption good t. So, qt ≡ 1+r t
where rt is the implicit interest rate.

Proposition 9 There exists a unique competitive equilibrium where

c10 = ω01 ,
b
(b
c1t , b
c2t+1 ) = (ω1 , ω2 ) for all t

i.e., there is no trade in equilibrium.

Proof. To see why, notice that since any agent in the initial old generation
only cares about goods dated at t = 1 and has only endowment at time t = 1,
then
c10 = ω01
b
Using market clearing for t = 1 gives

c10 + N1b
b c11 = ω01 + Nt ω1

and thus b c11 = ω1 . The budget constraint for agent t = 1 implies that
c22 = ω2 . The result follows by induction.
b

Equilibrium prices
We now describe the equilibrium price vector for this economy. Using the
previous proposition, and since the first order conditions are necessary and
sufficient for the agent maximization problem, it follows that

pt+1 u0 (ω2 )
qt = = 0
pt u (ω1 )

for all t (i.e., the budget set is tangent to the indifference curve evaluated at
the endowment point). Notice that these relative prices, qt , determine the
net interest rate, r, such that
pt+1 1
qt = = ,
pt 1+r
1

t
for all t. This implies that pt = 1+r for all t.

Offer Curves
6.1. PURE EXCHANGE OLG ECONOMIES 137

Consider the agent t’s problem

max u (c1t ) + βu (c2t+1 ) ,

(c1t ,c2t+1 )

subject to

c1t + qt c2t+1 = ω1 + qt ω2 .
λt : Lagrange multiplier corresponding to agent t. Nec. and Suff. condi-
tions for an interior solution

u0 (c1t ) = λt , (6.1)

βu0 (c2t+1 ) = λt qt , (6.2)

c1t + qt c2t+1 = ω1 + qt ω2 . (6.3)

From (6.3)

(ω1 − c1t )
qt = . (6.4)
(c2t+1 − ω2 )
Then, (6.2) and (6.1) coupled with (6.4) gives

βu0 (c2t+1 ) (ω1 − c1t )

= q t = . (6.5)
u0 (c1t ) (c2t+1 − ω2 )

Define excess demands for generation t when young (yt ) and old (zt )

yt = c1t − ω1 ,
zt = c2t+1 − ω2 .

And then write

βu0 (zt + ω2 ) yt
0
=− .
u (yt + ω1 ) zt
This last expression determines the Offer Curve, yt = f (zt ): the tangency
of the budget constraint and the indiference curve.
138 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

The feasibility constraint at t can be expressed

Nt−1 zt−1 + Nt yt = 0,
and then
1
yt = − zt−1 .
1+n
Autarky: (yt , zt ) = 0, while the autarky equilibrium price is
βu0 (ω2 )
qA = . (6.6)
u0 (ω1 )

6.1.2 First Welfare Theorem in OLG Economies

Here we analyze whether competitive equilibrium allocations are Pareto Op-
timal or not. If not, there is room for policy intervention. For simplicity,
assume that

ωt = (ω1t , ω2t+1 ) = (ω1 , ω2 )

with w1 > w2 ≥ 0.
There are two cases to determine optimality.
1 1
Proposition 10 Classical Case. If qC = 1+r C
< 1+n , the Competitive equi-
librium allocation, autarky, is a PO allocation.
1 1
Samuelson Case. If qS = 1+r S
> 1+n , the Competitive equilibrium alloca-
tion, autarky, is a not PO allocation.
Compute first the stationary PO allocation
max u(c1 ) + βu(c2 ),
(c1 ,c2 )

subject to (1 + n) c1 + c2 = (1 + n) ω1 + ω2 .
µ: Lagrange multiplier for feasibility
u0 (c∗1 ) = µ (1 + n)
βu0 (c∗2 ) = µ
(1 + n) c1 + c2 = (1 + n) ω1 + ω2
And thus
βu0 (c∗2 ) 1
∗
=
u0 (c1 ) 1+n
Observe that we are NOT considering the IOG’s welfare. This will be
critical.
6.1. PURE EXCHANGE OLG ECONOMIES 139

A Closer Look at the Samuelson Case

On the Failure of the 1st Welfare Theorem
Observe why autarky cannot be PO.
1
If qA > 1+n , each generation will be strictly better off with the station-
ary PO allocation (btw, autarky is a particular trivially feasible stationary
allocation).
What about the IOG?
The planner can transfer ω1 − c∗1 to the IOG. That’s, the young will
give up consumption conditional on the planner’s commitment to give him
c∗2 − ω2 when he’s old. Those resources, in turn, will come from the new
young generation, and so on....

The implications of the overlapping generations model under different

trading arrangements lie at the heart of many debates in the literature. Hence,
it is important to understand the structure of the trading arrangements in
this model. Suppose, in particular, that there is no government or clear-
ing house or other agency that facilitates trade between agents of different
cohorts. Then, as we argue below, there is no trade.
Consider the possible kinds of trades.
Trade with agents within a cohort: If agents within a cohort (from the
same generation) are heterogeneous, then these agents can trade privately is-
sued IOUs. They essentially can borrow and lend among themselves since
they share the same lifetimes. Hence within a generation, the marginal
rates of substitution across agents will be equalized, as will the intertem-
poral marginal rate of substitution. Hence we can work with a representative
member of generation t. Trade between agents of different cohorts: At time
t, there are two types of agents present in the economy: old agents born in
period t - 1 and young agents t. A young agent at period t will never lend
to an old agent at time t because the old agent will expire by period t + 1
when the loan is to be repaid. Similarly the old agent at time t will never
lend to a young agent at time t because the old agent will be dead next period
and unable to use the repayment. A young agent generally would like to shift
some consumption from youth to old age, but to do so would require entering
into an agreement with a young agent born at time t + 1. Such trades are not
allowed if we consider sequential trading because the time t +1 young agent is
not available to negotiate trades at time t. But suppose for the moment that
140 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

such trades could be negotiated. Suppose that a young agent at time t could
sign a contract with a young agent at time t+1.Would there be an incentive
for the agents to enter into a contract? If the young agent would like to shift
some of his wealth forward to time t + 1, that would require that the young
agent at time t + 1 transfer some resources to the old generation at time t.
But what can the young agent at time t offer to a young agent at time t +
1? The young agent at time t is unable to transfer resources to period t + 2,
which is the time period that young agents born in period t+1 would like to
receive the resources.Hence, even if contingent claims among young agents at
time t and t +1 could be negotiated, no one would enter into these contracts.
Hence, the OLG model is not a model of incomplete markets, because these
markets exist, it is just that no one wishes to participate in them. Hence
for there to be any trading among different cohorts, there must be an outside
asset or some institution that clears the trades. A convenient outside asset
is nominal money.
6.2. OLG ECONOMIES WITH PRODUCTION 141

6.2 OLG Economies with Production

Each period t = 1, 2, 3...., a generation with Nt = (1 + n)t N0 identical agents
is born. Each agent lives two periods, index generations by the date of birth.
At date t = 1, there is an initial old generation born at date t = 0 with
N0 = 1 agents.
There is one consumption good every period. Let ct = (c1t , c2t+1 ) denote
a consumption bundle for an agent born at t. Preferences are represented by
the utility function
u (c1t ) + βu (c2t+1 )
for all (c1t , c2t+1 ), where β ∈ (0, 1), u : R+ → R+ is standard. Assume u (c10 )
is strictly increasing.
Endowments. Each agent born at t ≥ 1 is endowed with 1 unit of time
when young, supplied inelastically (i.e., leisure is not valued).
The initial old generation is endowed with the initial stock of capital k1 ,
supplied in the capital market inelastically (Mr. 0 has strongly monotone
preferences).
Technology. Production possibilities described by
Yt = F (Kt , Nt ) ,
F is C 2 , HOD 1, str. increasing and str. concave for all Kt , Nt > 0.
Kt denotes capital available at t and Nt is labor.
Technology to accumulate capital given by
Kt+1 = Xt + (1 − δ) Kt
Xt is investment in the consumption good. For simplicity, assume that δ = 0
(the depreciation rate).
Representative firm.
wt real wage at t, rt rental price at t.
Firm’s Problem. At t ≥ 1, the firm faces (wt , rt ) and solves the static
problem
F Kt , Ntd − wt Ntd − rt Kt .


max
(Kt ,Ntd )≥0
In equilibrium, Ntd = Nt for all t.
Characterize an interior solution with nec. and suff. FOC’s
Kt : FK (Kt , Nt ) = rt
Nt : FN (Kt , Nt ) = wt
142 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

F (Kt ,Nt ) Kt
F is HOD 1, denote Nt
= F (Kt /Nt , 1) = f (kt ) where kt = Nt
and

∂F (Kt , Nt ) ∂(Nt f (kt ))

FK (Kt , Nt ) = =
Kt ∂Kt
1
= Nt f 0 (kt ) = f 0 (kt ) ,
Nt
∂F (Kt , Nt ) ∂(Nt f (kt ))
FN (Kt , Nt ) = =
Nt ∂Nt
K t
= f (kt ) − Nt f 0 (kt ) 0
2 = f (kt ) − kt f (kt ) .
(Nt )

Note that equilibrium profits will be zero.

Agent’s Problem
The only asset available is capital since agents are identical within each
generation and therefore equilibrium trading of any kind of bond will be zero
(convinced?).
st generation t’s saving.
Given (wt , rt+1 ) , an agent born at t ≥ 1 solves

max {u (c1t ) + βu (c2t+1 )}

{ct ,st }

subject to

c1t + st ≤ wt ,
c2t+1 ≤ (1 + rt+1 ) st ,
(c1t , c2t+1 ) ≥ 0.

The initial old generation owns k1 and then they consume c21 ≤ (1 + r1 ) k1 .
Also, note that generation t’s savings generate the stock of capital in t+1;
i.e., Nt st = Kt+1 . n o∞
Definition of a CE. A (Perfect Foresight) CE is an allocation b c21 , b
ct , b
kt ,
t=1
st }∞
savings{b t=1 and prices {w bt , rbt }∞
t=1 , such that
(1) Given {w bt , rbt+1 }, (b ct , sbt )solves
 the agent  t’s problem
 and
 bc21 = (1 + rb1 ) k1 .
0 b 0 b
(2) Given (w bt , rbt ) , rbt = f kt and w bt = f kt − kt f kt for all t.
b
(3) Market clearing conditions: For all t ≥ 1,
 
Nt f b kt = Nt−1b c2t + Ntb c1t + Kb t+1 − K
bt,
6.2. OLG ECONOMIES WITH PRODUCTION 143

K
b t+1 = Nt sbt .
Characterize the agent’s problem
Write the agent’s problem as
max u (wt − st ) + βu ((1 + rt+1 ) st ) .
0≤st ≤wt

Interior solution
st : u0 (wt − st ) = 1 + rt+1 βu0 ((1 + rt+1 ) st )
Apply the implicit function theorem (IFT) and write st = s(wt , rt+1 ).
Note that given (wt , rt+1 ), all generations have the same saving function
(i.e., it is time-invariant).
Let
1 + rt+1 0
G (st , wt , rt+1 ) = u0 (wt − st ) − u ((1 + rt+1 ) st )
1+θ
∂G
= −u00 (wt − st ) − (1 + rt+1 )2 βu00 ((1 + rt+1 ) st ) > 0,
∂st
∂G
= −β[u0 ((1 + rt+1 ) st )
∂rt+1
00
+ (1 + rt+1 ) st u ((1 + rt+1 )) R 0,
∂G
= u00 (wt − st ) < 0.
∂wt
IFT
∂s(wt , rt+1 )
sw =
∂wt
00
−u (wt − st )
=  ∈ (0, 1)
−u (wt − st ) − (1 + rt+1 )2 βu00 ((1 + rt+1 ) st )
00

∂s(wt , rt+1 )
sr = =
∂rt+1
βu0 (·) [1 + (1 + rt+1 ) st u00 (·) /u0 (·)]
R0
−u00 (·) − (1 + rt+1 )2 βu00 (·)
 

sw ∈ (0, 1): both consumption goods are normal; i.e., any change in wt
has just an income effect.
A change in rt+1 has both income and substitution effect and then its
impact depends upon the IES.
144 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

6.2.1 Dynamics and Steady State

Market clearing condition for capital

s (wt , rt+1 ) s (f (kt ) − kt f 0 (kt ) , f 0 (kt ))

kt+1 = = .
(1 + n) (1 + n)

Define
s (f (kt ) − kt f 0 (kt ) , f 0 (kt+1 ))
H (kt+1 , kt ) = kt+1 − ,
(1 + n)

and assume that

∂H (kt+1 , kt ) 1
=1− sr f 00 (kt+1 ) 6= 0
∂kt+1 1+n

to apply the IFT: kt+1 = g (kt ) where

dkt+1 sw f 00 (kt ) kt
=− R 0.
dkt (1 + n) − sr f 00 (kt )
dkt+1
dkt
> 0 if sr > 0.
General structure: we can say nothing about relation between kt+1 and
kt .
At steady state, kt+1 = kt = k ∗ for all t.
Some cases found interesting.
A steady state k ∗ is locally stable if

sw f 00 (k ∗ ) k ∗
− < 1.
(1 + n) − sr f 00 (k ∗ )

Why this? Suppose we take a linear expansion of g around k ∗ to get

kt+1 ≈ g (k ∗ ) + g 0 (k ∗ ) (kt − k ∗ )

where
sw f 00 (k ∗ ) k ∗
g 0 (k ∗ ) = −
(1 + n) − sr f 00 (k ∗ )
Thus, g 0 (k ∗ ) will be the characteristic root of the 1st. order linear differ-
ence equation.
6.2. OLG ECONOMIES WITH PRODUCTION 145

Figure 6.1: Stable Equilibrium OLG

SS is not only locally stable but also (almost) globally stable ,

0
k is locally stable and k ∗ is unstable.
∗

Figure 6.2: Multiple Equilibrium OLG

146 CHAPTER 6. OVERLAPPING GENERATION ECONOMIES

Figure 6.3: Cycles Equilibrium OLG

Endogenous business cycles. Note that here if the stock of capital

get either k ∗ or k ∗∗ , the economy will cycle for ever, even though it is not
subject to any type of exogenous shock. This is a two-period cycles. We
could have constructed cycles of any periodicity. In particular, periodicity 3
and then the economy could generate ”chaotic dynamic”.
Bibliography

[1] Aiyagari, S. Rao, 1994. “Uninsured Idiosyncratic Risk and Aggregate

Saving,” Quarterly Journal of Economics, Vol. 109(3), pp. 659–684.

[2] Arrow, Kenneth J., 1964. “The Role of Securities in the Optimal Alloca-
tion of Risk-Bearing,” Review of Economic Studies, Vol. 31, pp. 91–96.

[3] Arrow, K. and G. Debreu (1954). “Existence of an Equilibrium for a

Competitive Economy.” Econometrica 22, 265–290.

[4] Blanchard, Oliver J. and Stanley Fischer, 1989. Lectures on Macroeco-

nomics, MIT Press.

[5] Bewley, T. (1977). “The Permanent Income Hypothesis: A Theoretical

Formulation.” Journal of Economic Theory 16, 252–292.

[6] Brock, William A., 1982. “Asset Prices in a Production Economy.” In J.

McCall (ed.), The Economics of Information and Uncertainty. Chicago:
University of Chicago Press, pp. 1–43.

[7] Brock, William A., and Leonard Mirman, 1972. “Optimal Economic
Growth and Uncertainty: The Discounted Case,” Journal of Economic
Theory, Vol. 4(3), pp. 479–513.

[8] Cass, D. (1965). “Optimum Growth in an Aggregative Model of Capital

Accumulation.” Review of Economic Studies 32, 233–240.

[9] Cochrane, John (1991), ”A Simple Test of Consumption Insurance,”

Journal of Political Economy, Vol. 99, No. 5, pp. 957-976.

[10] Cooley, Thomas F., 1995. Frontiers of Business Cycle Research. Prince-
ton, NJ: Princeton University Press.

147
148 BIBLIOGRAPHY

[11] Debreu, Gerard, 1959. Theory of Value. New York: Wiley.

[12] Diamond, Peter A. (1965). “National Debt in a Neoclassical Growth

Model,” American Economic Review, Vol. 55, pp. 1126–1150.

[13] Gorman, T. (1953). “Community Preference Fields.” Econometrica 21,

63–80.

[14] Hansen, L. and R. Jagannathan (1991). “Implications of Security Market

Data for Models of Dynamic Economies.” Journal of Political Economy
99, 225–262.

[15] Huggett, M. (1993). “The Risk-free Rate in Heterogeneous, Incomplete-

Insurance Economies.” Journal of Economic Dynamics and Control 17,
953–969.

[16] Kehoe, Tim, (1989). ”Intertemporal General Equilibrium Models,” in F.

Hahn (ed.) The Economics of Missing Markets, Information and Games,
Claredon Press.

[17] Krueger, D. and F. Kubler (2002). “Intergenerational Risk Sharing via

Social Security when Markets are Incomplete.” American Economic Re-
view 92, 407–410.

[18] Kimball, Miles S. (1990). “Precautionary Saving in the Small and in the
Large,” Econometrica, 58, pp 53–73.

[19] Kydland, Finn E., and Edward C. Prescott (1977). “Rules Rather than
Discretion: The Inconsistency of Optimal Plans.” Journal of Political
Economy, Vol. 85(3), pp. 473–491.

[20] Lucas, R. E. Jr (1978). “Asset Prices in an Exchange Economy”, Econo-

metrica, 46, pp 1429-1445.

[21] Lucas, R. (1988). “On the Mechanics of Economic Development.” Jour-

nal of Monetary Economics 22, 3–42.

[22] Mace, Barbara (1991), ”Full Insurance in the Presence of Aggregate

Uncertainty,” Journal of Political Economy.

[23]
BIBLIOGRAPHY 149

[24] Mehra, R. and E. Prescott (1985). “The Equity Premium: A Puzzle.”

Journal of Monetary Economics 15, 145–161.

[25] Modigliani, F. and M. Miller (1958).“The Cost of Capital, Corporation

Finance, and the Theory of Investment.” American Economic Review
48, 261–297.

[26] Negishi, T., (1960). Welfare Economics and Existence of an Equilibrium

for a Competitive Economy,” Metroeconomica, 12, 92-97.

[27] Parker, Jonathan and B. Preston (2005), ”Precautionary Saving and

Consumption Fluctuations,” American Economic Review, Vol 95 No 4
(September), 1119-43.

[28] Ljungquist, Lars and Thomas Sargent (2004). Recursive Macroeconomic

Theory, 2nd Edition. MIT Press.

[29] Samuelson, P. (1958). “An Exact Consumption-Loan Model of Interest

With or Without the Social Contrivance of Money.” Journal of Political
Economy 66, 467–482.

[30] Shell, K. (1971). “Notes on the Economics of Infinity.” Journal of Polit-

ical Economy 79, 1002–1011.

[31] Weil, P. (1989). “The Equity Premium Puzzle and the Riskfree Rate
Puzzle.” Journal of Monetary Economics 24, 401–421.