Macroeconomics I

Economic Growth: The Neoclassical Growth Model

Carlos Lizama
March 30, 2022

1
Introduction

• This lecture is important.

• We will cover tools ...

• Dynamic Programming (not an introduction!). We will discuss more general problems and
will study how to solve them.
• We are adding the “D” to our DSGE models. Main tool to study dynamic problems.
• Many (if not most) problems can be set up in this kind of framework.

• But we also cover a model...

• The Neoclassical growth model is
• A generalization of the Solow growth model.
• A workhorse model to study growth...
• but not only growth, also business cycles.

2
The Neoclassical Growth Model

3
The Neoclassical Growth Model: Introduction

• Also known as the Ramsey Growth Model or the Ramsey-Cass-Koopmans model.

• Similar to the Solow model

• Consumption/savings choice is microfounded i.e. endogenous savings rate.
• As a result, the savings rate is no longer constant in the transition to the staedy state.

• We will consider the planner’s problem through the lens of dynamic programming.
• The methods used to solve this type of dynamic, infinite horizon problems is an incredible
powerful tool used to solve a wide variety of macro models.

4
Centralized versus decentralized equilibrium

• Social planner’s problem focuses on allocations (no prices).

• We assume a benevolent social planner that wants to maximize the representative agent’s
welfare.

• An alternative approach is to solve the decentralized equilibrium.

• Each agents takes prices as given (households, firms), and solves their own maximization
problem determining allocations.

• Later, we will discuss under what conditions these two approaches yield the same
solution.

• Today we focus only on the social planner’s problem.

5
Neoclassical Growth Model: Setup
• Discrete time t = 0, 1, . . .

• The planner’s problem is to choose an infinite sequence for consumption and capital
{ct , kt+1 }∞
t=0 to maximize utility

∞
X
β t u(ct ) (1)
t=0

given k0 , where u(·) is an increase and concanve utility function, β is the discount
factor, ct is consumption, kt is the capital stock, subject to the resource constraint

ct + kt+1 ≤ (1 − δ)kt + f (kt ) ∀t (2)

and feasibility constraint ct ≥ 0 and kt+1 ≥ 0, ∀t.

• We call this problem the sequence problem.

6
Towards a Recursive formulation
• We pursue a new formulation of this problem called Dynamic Programming.

• We stated the original problem as one of choosing infinite sequence of consumption

and capital {ct , kt+1 }∞
t=0 , we can restate the problem as choosing today’s consumption
c0 and tomorrow’s capital k1 and nothing else. The rest can wait until the next period.

• What are the planner’s preferences over current consumption and next period’s
capital?

• Suppose we have solve the problem (1) subject to (2) for all possible values of k0 . Then
we could define a function v : R+ → R, where v is the maximized objective function.

• This means that v is such that, for a given k0 , if we solve (1) and obtain the optimal
path {c∗t , kt+1
∗
}∞
t=0 then v satisfy
∞
X
v(k0 ) = β t u(c∗t )
t=0 7
Towards a Recursive formulation
• In terms of the value function, the planner’s problem in period 0 could be written as

max u(c0 ) + βv(k1 ) (3)

c0 ,k1

• Subject to the resource constraint and the feasibility constraints

c0 + k1 ≤ γ(k0 ), c0 ≥ 0, k1 ≥ 0

and initial capital level k0 , where we have defined γ(k0 ) ≡ f (k0 ) + (1 − δ)k0

• If we knew the function v, we could define another function g : R+ → R+ as follows:

for each k0 > 0, define k1 = g(k0 ) and c0 = γ(k0 ) − g(k0 ) be the values that attain the
maximum in (3). We call g the policy function.

• With g defined, we completely describe the dynamics of capital accumulation from any
initial stock k0 .
8
The Recursive formulation: The Bellman equation
• Finally, we can convince ourselves that the function v must satisfy
v(k0 ) = max u(γ(k0 ) − k1 ) + βv(k1 )
0≤k1 ≤γ(k0 )

• Notice that we used the fact that the resource constraint will hold with equality (why?)

• Notice that when we write the problem in this recursive way, time subscripts are
unnecessary: the problem is the same every date. We can rewrite the planner’s
problem with current capital stock k as
v(k) = max u(γ(k) − k 0 ) + βv(k 0 ) (4)
0≤k0 ≤γ(k)

• (4) is the recursive formulation of problem (1). This is a functional equation and is
know as the Bellman equation.

• The study of dynamic optimization problems through the analysis of the functional
equation is called dynamic programming 9
Dynamic Programming

10
A general problem: The Sequence Problem (SP)
• The infinity horizon problems are as follows: choose an infinite sequence {xt+1 }∞
t=0 to
maximize the objective function
∞
max∞
X
β t F (xt , xt+1 ) (5)
{xt+1 }t=0
t=0

subject to
xt+1 ∈ Γ(xt ) ∀t, with x0 ∈ X given (6)
• Let X be the set of all possible values for x, most of the time is just a subset of the
Euclidean space. Let Γ be the correspondence describing feasibility constraints
Γ : X → X, i.e. for each x, Γ(x) is the set of feasible values for x0
• F (xt , xt+1 ) is the current period return and β is the discount factor.
• X, F, Γ, β are the “primitives” of the problem.
• Let’s call any sequence {xt+1 }∞ t=1 a plan. Given x0 ∈ X, define
Π(x0 ) = {{xt+1 }∞t=0 |x t+1 ∈ Γ(x t ), ∀t} be the set of all plans that are feasible from x0 .
11
A general problem: The Recursive Problem
• Corresponding to any such sequence problem, we have a recursive formulation.

• The functional equation (FE) associated with this problem is

v(x) = max F (x, y) + βv(y) ∀x ∈ X (7)

y∈Γ(x)

• The function v is called the value function, x is the beginning of the period state
variable and y is the control variable (or choice variable). F (x, y) is the current period
return given the state variable x and control variable y.

• In the neoclassical growth model

• the state x = k.
• control y = k0
• F (x, y) = u(γ(x) − y) or F (k, k0 ) = u(f (k) + (1 − δ)k − k0 ).
• Γ(x) = [0, γ(x)], i.e. y ∈ Γ(x), k0 ∈ [0, γ(k)] = [0, f (k) + (1 − δ)k]
12
The Principle of optimality: assumptions

• We want to study the relationship between these two problems and develop methods
for analyzing them.

Assumption 1
X is a convex subset of Rn . The correspondence Γ : X → X is non-empty, compact-valued,
and continuous. In this case, compact-valued just means Γ(x) ⊂ Rn is closed and bounded.

Assumption 2
The function F : X × X → R is bounded and continuous, and β ∈ (0, 1)

• These assumption can be relaxed (See Stokey, Lucas, Prescott, 1989).

13
The Principle of optimality: definitions

Definition
Supremum function Let’s define the sumpremum function v ∗ : X → R by
∞
sup
X
v ∗ (x0 ) = β t F (xt , xt+1 ) (8)
xt+1 ∈Π(x0 ) t=0

14
Principle of Optimality: Theorems

• The following theorem establishes the equivalence between the supremum function
(8) and the solution of the functional equation (7).

Theorem (1)
Let X, Γ, F and β satisfy Assumption 1 and 2. Then
1. Then the supremum function v ∗ satisfies the function equation (7),
2. If v is a solution of the functional equation and satisfies

lim v(xn ) = 0, ∀{x0 , x1 , . . .} ∈ Π(x0 ), x0 ∈ X

n→∞

then v = v ∗

15
Principle of Optimality: Theorems
• A feasible plan {x∗t+1 } ∈ Π(x0 ) is an optimal plan from x0 if it attains the supremum in
P∞
SP. That is, {x∗t+1 } is an optimal plan if t=0 β ∗ F (x∗t , x∗t+1 ) = v ∗ (x0 )

Theorem (2)
Let X, Γ, F and β satisfy Assumption 1 and 2.
1. Let {x∗t+1 } ∈ Γ(x0 ) be an optimal plan, i.e., it attains the supremum in SP for initial state x0 ,
Then

v ∗ (x∗t ) = F (x∗t , x∗t+1 ) + βv ∗ (xt+1 ), ∀t

2. Let {x∗t+1 } ∈ Π(x0 ) be a feasible plan from x0 that satisfies

v ∗ (x∗t ) = F (x∗t , x∗t+1 ) + βv ∗ (xt+1 ), ∀t

and limsupt→∞ β t v ∗ (x∗t ) ≤ 0. Then {x∗t+1 } attains the supremum in SP for initial state x0 .

16
Principle of Optimality: Theorems

• Theorem 1 gives the equivalence between the supremum function and the value
function.

• Theorem 2 gives the equivalente between an optimal plan and a plan that satisfies the
function equation with v = v ∗ .

17
Principle of Optimality: Theorems

• Suppose we have a solution v to the functional equation (7). Using v we can define a
policy function g by

g(x) ≡ {y ∈ Γ(x)|v(x) = F (x, y) + βv(y)}

• We say that a plan {xt+1 } ∈ Γ(x0 ) is generated by g if xt+1 = g(xt ),

• We define the optimal policy function g ∗ by

g ∗ (x) ≡ {y ∈ Γ(x)|v ∗ (x) = F (x, y) + βv ∗ (y)}

18
Principle of Optimality: Theorems
• We can rewrite theorem (2) as follows

Theorem (3)
Let X, Γ, F and β satisfy Assumption 1 and 2. Let

g ∗ (x) ≡ {y ∈ Γ(x)|v ∗ (x) = F (x, y) + βv ∗ (y)}

1. Let {x∗t+1 } be an optimal plan. Then {x∗t+1 } is generated by g ∗ , x∗t+1 = g ∗ (x∗t ) ∀t.
2. Any plan generated by g ∗ and satisfying lim supt→∞ β t v ∗ (x∗t ) ≤ 0 is an optimal plan.

• We have establish the tight connection between the solutions of the sequence
problem and the function equation (recursive formulation).

• Richard Bellman called this connection the Principle of Optimality.

19
Solution of the Bellman Equation

• Existence of the solution?

• Uniqueness?

• Algorithm to compute the solution?

• Before doing this, we need some mathematical background.

20
Mathematical Preliminaries
Definition (Metric Space)
A metric space is a set S, together with a metric (a distance function) ρ : S × S → R such
that for all x, y, z ∈ S
1. ρ(x, y) ≥ 0 and with equality if and only if x = y.
2. ρ(x, y) = ρ(y, x) (symmetry), and
3. ρ(x, z) ≤ ρ(x, y) + ρ(y, z) (triangle inequality).

Definition (Convergence)
A sequence {xn }∞ n=0 in S, converges to x ∈ S if for each ε > 0 there exists an Nε ∈ N such
that ρ(xn , x) < ε ∀n ≥ Nε . That is xn → x ∈ S if ρ(xn , x) → 0.

Definition (Cauchy Sequence)

A sequence {xn }∞ n=0 in S is a Cauchy sequence (satisfies the Cauchy criterion) if for each
ε > 0 there exists an Nε ∈ N such that ρ(xn , xm ) < ε ∀n, m ≥ Nε
21
Mathematical Preliminaries

Theorem
Every convergent sequence is Cauchy.

Definition (Complete metric space)

A metric space (S, ρ) is complete if every Cauchy sequence in S converges to an element in
S.

Proposition 1
S = Rn with the metric ρ(x, y) = ||x − y|| is a complete metric space.

22
Contraction Mapping

Definition (Contraction Mapping)

Let (S, ρ) be a metric space and let T : S → S be a function mapping S into itself. We say T
is a contraction mapping with modulus β if for some β ∈ (0, 1)

ρ(T x, T y) ≤ βρ(x, y), ∀x, y ∈ S

Definition (Fixed point)

Let T : S → S be a contraction mapping. We say that x ∈ S is a fixed point of the
contraction mapping T if x is such that T x = x.

Define the iterations of applying T by T 0 x = x and T n+1 = T (T n x), ∀n = 1, 2, . . .

23
The Contraction Mapping Theorem

Theorem (The Contraction Mapping Theorem)

If (S, ρ) is a complete metric space and T : S → S is a contraction mapping with modulus β,
then:
1. T has exactly one fixed point x ∈ S, T x = x and
2. for any x0 ∈ S, ρ(T n x0 , x) ≤ β n ρ(x0 , x) ∀n = 0, 1, 2, . . .

• Part 1 of the theorem states that a fixed point exists and is unique.

• Part 2 states that starting from any initial point in x0 ∈ S, repeated application of the
contraction T gets you closed and closer to the unique fixed point.

24
The Contraction Mapping Theorem
• We would like to apply this theorem on function spaces to prove existence and
uniqueness of a function. Consider X ⊂ Rn and let’s define C(X) to be the space of
bounded continuous functions f : X → R with the sup norm ||f || = supx∈X |f (x)|

Theorem
C(X) with the sup norm is a complete metric space.

Theorem (The Contraction Mapping Theorem, restated)

Let X ⊆ Rn and let C(X) be the space of bounded continuous functions f : X → R with the
sup norm. If T : C(X) → C(X) is a contraction mapping with modulus β, that is
||T f − T h|| ≤ β||f − h||, ∀f, h ∈ C(X) then
1. T has exactly one fixed point f ∈ C(X), T f = f and
2. For any f0 ∈ C(X), ||T n f0 − f || ≤ β n ||f0 − f || ∀n = 0, 1, 2, . . .

25
Blackwell’s sufficient conditions

Theorem (Blackwell’s sufficient conditions)

Let X ⊆ Rn and let B(X) be a space of bounded functions f : X → R with the sup norm. Let
T : B(X) → B(X) be an operator satisfying the following two properties:
1. (monotonicity) f, g ∈ B(X) and f (x) ≤ g(x) ∀x ∈ X implies T f (x) ≤ T g(x), ∀x ∈ X.
2. (discounting) there exists some β ∈ (0, 1) such that

[T (f + a)](x) ≤ T f (x) + βa, ∀f ∈ B(X), a ∈ R+ , x ∈ X

where (f + a)(x) is the function defined by (f + a)(x) = f (x) + a

Then T is a contraction with modulus β.

26
The Theorem of the Maximum
Theorem (Theorem of the Maximum)
Let X ⊆ Rn and let Y ⊆ Rm . Let φ : X × Y → R be a continuous (single-valued) function, and
let Γ : X → Y be a nonempty, compact-valued, and continuous correspondence.
Define the function h : X → R by

h(x) = max φ(x, y)

y∈Γ(x)

and the correspondence g : X → Y by

g(x) = {y ∈ Γ(x)|φ(x, y) = h(x)}

Then h is continuous and the correspondence g is nonempty, compact-valued, and

upper-hemi-continuous.
• For each x ∈ X, the set Γ(x) is nonempty and compact and the function φ(x, ·) is
continuous in y. Hence the maximum is attained and the set g(x) of maximizers is
nonempty.
27
The Theorem of the Maximum

• The Theorem of The Maximum tells us that:

• (i) h is continuous. (The Value Function is continuous)
• (ii) g is nonempty, compact-valued, and upper-hemi-continuous. (The policy function is well
behaved).

• Furthermore, it can be shown that if φ(x, y) is strictly concave, the solution of the
correspondence is single-valued.
• The policy function is in fact a function, not a correspondence.
• Since it is single-valued and upper-hemi-continuous, it is continuous.

28
Existence and uniqueness of the value function

• The general formulation of the functional equation we are trying to solve is

v(x) = max F (x, y) + βv(y) ∀x ∈ X (9)

y∈Γ(x)

• We call the function v : X → R the value function.

• Assuming we know v, we can define the correspondence g : X → R by

g(x) ≡ {y ∈ Γ(x)|v(x) = F (x, y) + βv(y)} (10)

• in other words, given x, g(x) is the (set of) value(s) of y that attains the maximum.

29
Existence and uniqueness of the value function

• We want to prove that the solution v exists and is unique.

• Define the operator T on the space C(X)

T f (x) = max F (x, y) + βf (y) (11)

y∈Γ(x)

for each f ∈ C(X). Therefore, (10) is simply

v = Tv

• We have defined the operator T so that the value function is a fixed point of this
operator.

30
Existence and uniqueness of the value function

Theorem
Let X, Γ, F and β satisfy assumptions 1-2. Then the operator defined in (11) maps C(X) into
itself
T : C(X) → C(X)
and has a unique fixed point v ∈ C(X),
Tv = v
and for all v0 ∈ C(X),
||T n v0 − v|| ≤ β n ||v0 − v||
, ∀n.
Moreover, given the fixed point v, the policy correspondence g defined in (10) is non-empty,
compact-valued and upper-hemi-continuous.

31
Existence and uniqueness of the value function: proof

• We need to show that the operator T defined in (11) is a contraction on the function
space C(X).

• First we need to show that T : C(X) → C(X).

• Since F and f are continuous and bounded and Γ is non-empty, compact-valued, and
continuous. Therefore, the problem is to maximize a continuous function over a compact
set, hence the maximum is attained.
• Since F and f are bouded, T f must be bounded.
• By the theorem of the maximum, T f is also continuous. Therefore, T : C(X) → C(X).

• Now we have to show that T is a contraction. We can use Blackwell’s sufficient

conditions, so we need to show
• Monotonicity.
• Discounting.

32
Existence and uniqueness of the value function: proof

Monotonicity: We want to show that if f (x) ≤ h(x), then T f (x) ≤ T h(y)

Denote g(x; f ) ≡ {y ∈ Γ(x)|T f x) = F (x, y) + βf (y)}

Note that

F (x, g(x; f )) + βf (g(x; f )) ≤ F (x, g(x; f )) + βh(g(x; f )) ≤ max F (x, y) + βh(y)

| {z } y∈Γ(x)
=T f (x) | {z }
=T h(x)

Therefore, monotonicity is satisfied.

33
Existence and uniqueness of the value function: proof

Discounting.

T (f + a)(x) = max {F (x, y) + β(f (y) + a)} = max {F (x, y) + βf (y)} + βa = T f (x) + βa
y∈Γ(x) y∈Γ(x)

Since monotonicity and disconuting are satisfied, T satisfies the Blackwell’s conditions and
is a contraction mapping. Contracting mapping theorem proves that T has a unique fixed
point.

34
Properties of the value function and the policy function

Assumption 3
(i) For every y, F (·, y) is strictly increasing in each of its first n arguments.

(ii) Γ is monotone in the sense that x ≤ x0 implies Γ(x) ⊆ Γ(k 0 )

Theorem (Value function is strictly increasing)

Let X, Γ, F and β satisfy Assumptions 1-3, and let v be the unique solution to (9). Then v is
strictly increasing

35
Properties of the value function and the policy function

Assumption 4
(i) F is concave. That is, for all θ ∈ (0, 1)

F (θ(x, y) + (1 − θ)(x0 , y 0 )) ≥ θF (x, y) + (1 − θ)F (x0 , y 0 )

and with equality if and only if x = x0 and y = y 0 .

(ii) Γ is convex in the sense that for any θ ∈ [0, 1], and x, x0 ∈ X, y ∈ Γ(x) and y 0 ∈ Γ(x0 )
implies θy + (1 − θ)y 0 ∈ Γ(θx + (1 − θ)x0 )

Theorem (Value function is strictly concave, policy function is single-valued)

Let X, Γ, F and β satisfy Assumptions 1-4, and let v be the unique solution to (9) and g satisfy
(10). Then v is strictly concave and g is continuous, single-valued function.

36
Properties of the value function and the policy function

Theorem (Envelope theorem)

Let X, Γ, F and β satisfy Assumptions 1-4, assume F is continuously differentiable and let v be
the unique solution to (9) and g satisfy (10). If x0 ∈ intX and g(x0 ) ∈ intΓ(x0 ), then v is
continuously differentiable at x0 with derivative given by

vi (x0 ) = Fi (x0 , g(x0 )), ∀i = 1, 2, . . . , n. (12)

37
The Neoclassical Growth Model (again)

38
Neoclassical Growth model and Dynamic Programming

• So far, many mathematical results...

• Bottom line... everything is well-behaved

• Value function exists and is unique, we can find it by iterating on the value function (value
function iteration).
• Value function is continuous, strictly increasing, strictly concave, twice differentiable.
• Policy function is a function (not a correspondence) and continuous.
• We can use standard optimization techniques to solve the maximization problem (v is twice
differentiable, we can solve for g using the F.O.C. and the envelope theorem).

39
Neoclassical Growth
• The problem is
∞
max∞
X
β t u(f (kt ) + (1 − δ)kt − kt+1 )
{kt+1 }t=0
t=0

• We want to find the optimal sequence that maximizes lifetime utility.

• We know that the problem is equivalent to

V (k) = max
0
{u(f (k) + (1 − δ)k − k 0 ) + V (k 0 )}
k

• We know V exists and has nice properties.

• We can solve this problem using standard maximization techniques.

40
Solving the Neoclassical Growth Model

V (k) = max
0
{u(f (k) + (1 − δ)k − k 0 ) + V (k 0 )}
k

• The F.O.C. is

−uc (c) + βV 0 (k 0 ) = 0

• Envelope theorem

V 0 (k) = uc (c)(fk (k) + (1 − δ))

41
Envelope Theorem

Consider k 0 = g(k). Take the derivative of V (k) with respect to k:

dV (k)
V 0 (k) = = uc (c)(fk (k) + (1 − δ) − g 0 (k)) + βV 0 (g(k))g 0 (k)
dk
= uc (c)(fk (k) + 1 − δ) + g 0 (k) (−uc (c) + βV 0 (k 0 ))
| {z }
=0 F OC
0
V (k) = uc (c)(fk (k) + 1 − δ)

42
The Euler Equation
• Using the F.O.C. and the Envelope Condition we obtain

uc (c) = βuc (c0 )(fk (k 0 ) + 1 − δ) (13)

• This equation is an optimality condition.

• It tells us that the optimal allocation of consumption/savings is such that the marginal
utility from consumption today must be equal to the discounted marginal utility of
consumption tomorrow, adjusted by the marginal gross rate of capital and the discount
factor.

• Standard utility function (CRRA preferences) u(c) = c1−σ

1−σ if σ 6= 1 and u(c) = log(c) if
σ = 1. In this case
c0
= (β(fk (k 0 ) + 1 − δ)1/σ
c
43
Transversality Condition

• F.O.C. are necessary but not sufficient.

• Consider a finite horizon economy

∞
max
X
β t u(f (kt ) + (1 − δ)kt − kt+1 )
{kt+1 }∞
t=0 t=0
kt+1 ≥ 0

• This is a finite-dimensional optimization problem ⇒ we can solve it using Kuhn-Tucker

conditions.

44
Transversality Condition
• Constrained optimization problem
T
X T
X T
X
L= β t u(ct ) + λt [f (kt ) + (1 − δ)kt − ct − kt+1 ] + µt kt+1
t=0 t=0 t=0

• First order necessary conditions

ct : 0 = β t u0 (ct ) − λt
kt+1 : 0 = −λt + λt+1 [f 0 (kt+1 ) + (1 − δ)] + µt
kT +1 : 0 = −λT + µT

• Complementary slackness conditions

0 = λt [f (kt ) + (1 − δ)kt − ct − kt+1 ]

0= µt kt+1

45
Transversality condition
• Optimal cT
λT = β T u0 (cT )

• Optimal kT +1
µT = λT = β T u0 (cT )

• Complementary slackness
β T u0 (cT )kT +1 = 0

• The present discount utility value of “left over” resources must be equal to zero.

• We can generalize this to the infinite horizon case

lim β T u0 (cT )kT +1 = 0
T →∞

46
Characterizing the solution

Finally, we can say that optimal path of {ct , kt+1 }∞

t=0 is such that
• The Euler equation is satisfied

uc (ct ) = βuc (ct+1 )(fk (kt+1 ) + 1 − δ)

• The resource constraint is satisfy

ct + kt+1 = f (kt ) + (1 − δ)kt

• The transversality condition is satisfied

lim β T u0 (cT )kT +1 = 0

T →∞

47
Steady State and Balanced Growth Path

• In the steady state, all variables are constant.

• In the balanced growth path, all variables grow at a constant rate (not necessarily the same
rate).

• If c∗t = c∗t+1 = c∗ and If kt∗ = kt+1

∗
= k∗
 
1
1 = β(fk (k ∗ ) + (1 − δ)) ⇒ k ∗ = fk−1 −1+δ
β
c(k ∗ ) = f (k ∗ ) − δk ∗

48
 Practical Dynamic Programming:
Solving the Neoclassical Growth Model

49
Solving dynamic programming problems

• There are many methods to solve dynamic programming problems.

• Guess and verify. Literally, you guess and verify that the proposed solution is in fact a
solution of the problem. In practice, not very useful.

• Value function iteration. This relies on iterating on the value function, we know that
the operator T is a contraction, so we know that iterating on this operator converges
to the true solution.

• Policy function iteration. Based on similar theory, but instead of solving the value
function, we iterate on the policy function.

• Other numerical approaches... (not covered in this course).

50
Guess and verify

• Consider the followig setup:

• Log-utility u(c) = log(c)
• Production function f (k) = kα
• Full depreciation δ = 1

• Guess the following solution

V (k) = A + B log(k) (14)

A and B are coefficients to be determined

51
Guess and verify

• Let’s solve the problem

max log(k α − k 0 ) + β(A + B log(k 0 ))

0≤k0 ≤kα

• The F.O.C.
1 1
− 0
+ βB 0
kα −k k

• Hence,

βB
k0 = kα
1 + βB

52
Guess and verify
• Evaluate the solution:
    
βB βB
V (k) = log k −α
k α
+ β A + B log k α
1 + βB 1 + βB
kα
   
βB
V (k) = log + βA + βB log kα
1 + βB 1 + βB
 
βB
= − log(1 + βB) + βA + βB log + (1 + βB)α log(k)
1 + βB | {z }
| {z } ≡B
≡A

• Therefore,
α
B=
1 − αβ
  
1 αβ
A= log(1 − αβ) + log(αβ)
1−β 1 − αβ

• The policy function is k 0 = αβk α 53

Value function iteration: numerical implementation

• Suppose we approximate the continuous state space K with a finite grid of possible
capital stocks kmin < . . . < ki < . . . < kmax , i = 1, . . . , n, i.e. a vector of size n.

• On this grid of points, the value function is also a finte vector

v(kmin ), . . . , v(ki ), . . . , v(kmax ), i = 1, . . . , n.

• Consider ki a typical element of the grid of capital and vi a typical element of the value
function.

• Let cij be the consumption if current capital is k = ki and capital chosen for next
period is k = kj , cij = f (ki ) − kj . We need to be careful to respect the feasibility
constraints 0 ≤ kj ≤ f (ki ).

• Let uij denote the flow utility associated with cij , so it is an n × n matrix.

54
Value function iteration: numerical implementation

• The Bellman equation can be written

vi = max{uij + βvj }
j

• The associated policy function is

gi = arg max{uij + βvj }

j

such that gi = g(ki ) attains the maximum given k = ki .

55
Value function iteration: numerical implementation

• Value function iteration:

1. Set n = 0. Start with an initial guess v n = v 0 .
2. Calculate
vin+1 = T v n = max{uij + βvjn }, ∀i
j

Compute the error ε = ||v n+1 − v n || = maxi |T vin − vin |

3. If ε < tol, some tolerance, stop. Otherwise n ← n + 1 and go back to 2.

• Since T is a contraction, this algorithm will converge.

56
Value function iteration: numerical implementation

• We know this algorithm converges, T is a contraction.

• It is a bit of “brute force”.

• The larger the state space, the slowest it becomes. This algorithm suffers severely from
curse of dimensionality.

• Instead of fully discretize space and decisions, we could interpolate between nodes.
This is much more precise, but also slower.

• In general, there is a trade-off between speed and accuracy.

57
Policy function iteration
• We have just solved a functional equation. The unknown is a function.

• Instead of solving the functional equation associated to the value function, we could
solve the functional equation associated to the policy function.

uc (fk (kt ) + (1 − δ)kt − g(kt )) = β(fk (kk+1 ) + (1 − δ))uc (fk (kt+1 ) + (1 − δ)kt+1 − g(kt+1 ))

• This is called policy function iteration.

• The same idea, start with a guess for g n and update the guess by finding g n+1 that
satisfy this equation.

• Note that we have not proved that this algorithm converges!! In general, it does, even
though we have not shown a proof.

• This algorithm is usually faster than value function iteration.

58
More facts, balanced growth path, and extensions

59
Kaldor’s Facts

• In the balanced growth path, our models have to be consistent with some basics facts
about growth, called Kaldor’s facts.

• Kaldor’s facts:
1. Labor productivity has grown at a sustained rate.
2. Capital per worker has also grown at a sustained rate.
3. The real interest rate, or the return on capital, has been stable.
4. The ratio of capital to output has also been stable.
5. Capital and labor share have capture stable shares of national income.
6. Among the fast growing countries of the world, there is an appreciable variation in the rate
of growth, of the order of 2-5 percent.

60
Kaldor’s facts
• Our models should be able to account for these facts.
1. Labor productivity has grown at a sustained rate.
Y
⇒ L
grows at a sustained rate.
2. Capital per worker has also grown at a sustained rate.
K
⇒ L
grows at a sustained rate
3. The real interest rate, or the return on capital, has been stable.
⇒ r = f 0 (k) − δ is constant.
4. The ratio of capital to output has also been stable.
K
Y
has been constant.
5. Capital and labor share have capture stable shares of national income.
For instance, it can be shown that when F (Kt , Lt ) = Ktα L1−α
t , capital share = α and labor
share = (1 − α)
6. Among the fast growing countries of the world, there is an appreciable variation in the
rate of growth, of the order of 2-5 percent.

61
More on BGP

• So far, we have considered a production function like this Yt = F (Kt , At Lt ) with F

homogeneous of degree one (Constant returns to scale), increasing and concave in
both arguments.

• Consider Lt+1 = (1 + n)Lt and At+1 = (1 + g)At

• Why do we consider a labor augmenting technology?

• Answer: Uzawa’s theorem.

• Uzawa’s theorem tells us what conditions a production function must satisfy to be

consistent with a balanced growth path.

62
Usawa’s Theorem
Theorem (Usawa’s Theorem)
Consider a growth model with aggregate production function Yt = F̃ (Ãt , Kt , Lt ) with CRS in Kt
and Lt . Population grows at a constant rate Lt+1 = (1 + n)Lt . The aggregate resource
constraint is

Ct + Kt+1 − (1 − δ)Kt = Yt

There exists T ≤ ∞ such that ∀t > T , Yt+1 = (1 + gY )Yt , Ct+1 = (1 + gC )Ct ,

Kt+1 = (1 + gK )Kt , then
1. gY = gC = gK
2. for any t > T , there exists F homogeneous of degree 1 such that the aggregate production
function can be represented as

Yt = F (Kt , At Lt )
1+gY
with At+1 = (1 + gA )At with (1 + gA ) = 1+n
63
Uzawa´s Theorem: Proof, part 1.

1. gY = gC = gK

For all t > T

Kt+1 Yt − Ct
=1−δ+
Kt Kt
Yt Ct
gK + δ = −
Kt Kt
Yt Ct
The LHS constant. Therefore, the RHS must be constant i.e., Kt and Kt must be
constant which implies gK = gY = gC

64
Uzawa’s Theorem: Proof, part 2.

2.
   
Yt Yt Yt Yt
Yt = YT · = F̃ ÃT , KT , LT = F̃ ÃT , Kt , LT
YT YT YT YT
Yt LT
Define At ≡ L t YT and F (Kt , At Lt ) ≡ F̃ (ÃT , Kt , At Lt ). By construction, F is

homogeneous of degree 1. Note that

At+1 Yt+1 /Yt (1 + gY )

= = = (1 + gA )
At Lt+1 /Lt (1 + n)

65
Neoclassical growth model and endogenous labor supply
• Households care not only about consumption, but also about leisure.

• Households preferences are increasing in consumption and leisure. Households

choose between work and leisure. They are endowed with one unit of time.

• Households solve
∞
max
X
β t u(ct , 1 − ht )
ct ,ht ,kt+1
t=0
s.t. ct + kt+1 = f (kt , ht ) + (1 − δ)kt
0 ≤ ht ≤ 1
ct ≥ 0, kt+1 ≥ 0

• Is ht a state or a control?

66
NGM with labor supply

• Recursive formulation

V (k) = max u(f (k, h) + (1 − δ)k − k 0 , 1 − h) + βV (k 0 )

k0 >0,h∈[0,1]

• F.O.C.
∂V
0= = −u1 (c, 1 − h) + βV1 (k 0 )
∂k 0
∂V
0= = u1 (c, 1 − h)f2 (k, h) − u2 (c, 1 − h)
∂h

• Envelope theorem

Vk (k) = u1 (c)(f1 (k, h) + 1 − δ)

67
NGM with labor supply
• Using the F.O.C. w.r.t k 0 and the envelope condition we obtain
u1 (c, 1 − h) = β(1 + f1 (k 0 , h0 ) − δ)u1 (c0 , 1 − h0 )

• This is the standard Euler equation or the intertemporal condition.

• Using the F.O.C. w.r.t h we obtain

u2 (c, 1 − h)
w = f2 (k, h) = (15)
u1 (c, 1 − h)

• It can be shown that in a decentralized equilibrium, w is the wage and it equals the
marginal productivity of labor f2 (k, h).

• Equation (15) is the labor supply equation or the intratemportal condition.

• These conditions and the transversality condition determine the equilibrium path of
this economy. 68
NGM with labor supply: condition on preferences
• We showed that a technical change must be labor augmenting to be consistent with a
balanced growth path.

• Are there any restrictions on preferences?

• In an economy where technology is labor improving, we need to impose that hours worked
do not change.
• This is equivalent to say that income and substitution effects of productivity growth must
exactly offsetting effects on labor supply.

• King, Plosser, and Rebelo (1988) show that preferences must be of the following form

C 1−σ

1−σ v(L) if σ > 0, σ 6= 1
u(C, L) =
log(C) + v(L)

where C is consumption and L is leisure, with v(L) and increasing and concave
function.
69
Endogenous growth

70
Endogenous growth

• In Solow growth model and the neoclassical growth model, growth is driven by
technology At . At evolves exogenously.

• An important strand of literature, study what is called “endogenous growth models”.

• In this models, endogenous decisions drive economic growth.

• Other issues in growth models:

• AK model (think of α → 1 in a Cobb Douglas technology).
• Human capital accumulation.
• Spillovers.
• R&D models
• Creative destruction (Aghion and Howitt).

71