Classical approach Dynamic programming Brock-Mirman model

Macroeconomics IV
1. Intertemporal optimization in discrete
time

Hajime Tomura

Fall 2019/20

1 / 42
Classical approach Dynamic programming Brock-Mirman model

List of contents

1. Classical (variational) approach to intertemporal

optimization (the Euler equation and the transversality
condition)
2. Dynamic programming
3. Closed-form solution to a dynamic general equilibrium
model: Brock-Mirman model

2 / 42
Classical approach Dynamic programming Brock-Mirman model

Classical (variational) approach to

intertemporal optimization

3 / 42
Classical approach Dynamic programming Brock-Mirman model

Sequential problem
Standard representative-agent utility maximization problem:
∞
max
∞ ∑ β t u ( ct )
{ ct } t = 0 t = 0

s.t. kt +1 + ct = f (kt ) + (1 − δ)kt

ct ≥ 0, kt +1 ≥ 0

• kt : state variable (e.g., capital stock);

• ct : choice variable (e.g., consumption);
• β ∈ (0, 1): time discount factor;
• δ ∈ (0, 1): depreciation rate of capital.
• u, f : u 0 > 0, u 00 < 0; f 0 > 0, f 00 < 0.

How to solve this problem?

4 / 42
Classical approach Dynamic programming Brock-Mirman model

Classical approach to a sequential problem

1. Set up the Lagrangian, L:
∞
L= ∑ βt {u (ct ) + λt [f (kt ) + (1 − δ)kt − kt +1 − ct ]}
t =0
2. Take derivatives of L with respect to ct and kt +1 , and set
them to zero:
∂L
= βt [u 0 (ct ) − λt ] = 0;
∂ct
∂L
= βt {−λt + βλt +1 [f 0 (kt +1 ) + 1 − δ]} = 0;
∂kt +1
3. Eliminate the Lagrange multiplier, λt , from the two
equations to get the Euler equation:
u 0 (ct ) = βu 0 (ct +1 )[f 0 (kt +1 ) + 1 − δ]
5 / 42
Classical approach Dynamic programming Brock-Mirman model

Intuition behind the Euler equation

The benefit and cost of a marginal change in consumption and
saving are balanced at the optimum:

u 0 ( ct )
| {z }
marginal utility cost of increasing saving today
= β
|{z}
time discount factor
× u 0 ( ct + 1 )
| {z }
marginal utility of consumption tomorrow
× [ f 0 ( kt + 1 ) + 1 − δ ]
| {z }
marginal income gain tomorrow

6 / 42
Classical approach Dynamic programming Brock-Mirman model

Sufficient conditions for an optimum

Theorem (For a more general theorem, see Stokey
and Lucas (1989, Sec. 4.5 and 9.5) or Miao (2014,
Sec. 7.5).)
The Euler equation

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

and the transversality condition

lim βt u 0 (ct )kt +1 = 0

t →∞

are sufficient conditions for an optimum, if u is a strictly

increasing and concave, and continuously differentiable
function, and f is an increasing, concave, and continuously
differentiable function.
7 / 42
Classical approach Dynamic programming Brock-Mirman model

Preparation for the proof

Definition: The function f : Rn → R is concave if
∀a, b ∈ Rn , ∀t ∈ (0, 1), tf (a) + (1 − t )f (b) ≤ f (ta + (1 − t )b)
Theorem: If f : Rn → R is differentiable and concave, then for arbitrary
a ∈ Rn ,
∀ x ∈ Rn , f ( x ) − f ( a ) ≤ ∇ f ( a ) · ( x − a ) .
Proof for n = 1: By the definition of concavity, for all t ∈ (0, 1),
f (a + t (x − a)) − f (a)
f (x ) − f (a ) ≤ .
t
The proof is trivial if x = a. If x 6= a,
f (x ) − f (a ) f (a + t (x − a)) − f (a)
≤ lim = f 0 (a )
x −a t →0 t (x − a )
where the last equality is the definition of a derivative. 

For n ≥ 2, the theorem can be proved by considering the first-order Taylor

expansion such that
f (a + t (x − a)) − f (a) = t ∇f (a) · (x − a) + η (t (x − a))
η (t (x − a))
where η is a function satisfying lim = 0, given x and a.
t →0 t
8 / 42
Classical approach Dynamic programming Brock-Mirman model

Preparation for the proof

Theorem: The function f : Rn → R is concave, and the function g : Rm →

Rn is non-decreasing and concave. Then f ◦ g (i.e., g (f (·)) is also concave.
Proof:
g (f ((ta + (1 − t )b))) ≤ g ((tf (a) + (1 − t )f (b)))
≤ tg (f (a)) + (1 − t )g (f (b)) 

9 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the theorem (Stokey and Lucas

1989, Theorem 4.15; Miao 2014, Theorem
7.5.2.)
Suppose there exists {ct , kt +1 }t∞=0 satisfying ct > 0,
kt +1 > 0, the Euler equation, and the transversality condition,
and the budget constraint. Denote it by {ct∗ , kt∗+1 }t∞=0 .
Consider a positive integer T and any feasible plan (i.e.,
{ct , kt }t∞=0 satisfying all the constraints), given k0 . Compute
T T
DT ≡ ∑ βt u (ct∗ ) − ∑ βt u (ct )
t =0 t =0
T
= ∑ βt [u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 ) − u (f (kt ) + (1 − δ)kt − kt +1 )]
t =0
T
≥ ∑ βt {u 0 (ct∗ )[f 0 (kt∗ ) + (1 − δ)](kt∗ − kt ) + u 0 (ct∗ )(−1)(kt∗+1 − kt +1 )}
t =0
10 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the theorem (Stokey and Lucas

1989, Theorem 4.15; Miao 2014, Theorem
7.5.2.)
The last weak inequality in the previous slide hold because u
and f are increasing, differentiable, and concave functions (see
the preparation for the proof).

T
DT ≥ ∑ βt {u 0 (ct∗ )[f 0 (kt∗ ) + (1 − δ)](kt∗ − kt ) + u 0 (ct∗ )(−1)(kt∗+1 − kt +1 )}
t =0
= u 0 (c0∗ )[f 0 (k0 ) + (1 − δ)](k0 − k0 )
| {z }
because k0∗ =k0
T −1
+ ∑ βt { βu 0 (ct∗+1 )[f 0 (kt∗+1 ) + (1 − δ)] − u 0 (ct∗ )}(kt∗+1 − kt +1 )
t =0 | {z }
the Euler equation
T 0 ∗ ∗
+β u (cT )(−1)(kT +1 − kT +1 )
11 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the theorem (Stokey and Lucas

1989, Theorem 4.15; Miao 2014, Theorem
7.5.2.)
Thus, if the Euler equation is satisfied, then
T T
DT ≡ ∑ βt u (ct∗ ) − ∑ βt u (ct ) ≥ βT u 0 (cT∗ )(−1)(kT∗ +1 − kT +1 )
t =0 t =0
≥ βT u 0 (cT∗ )(−1)kT∗ +1
because u 0 > 0 and kT +1 ≥ 0 by assumption.
Take limit as T → ∞. If the transversality condition is
satisfied, then
∞ T
∑ βt [u (ct∗ ) − u (ct )] = lim ∑ βt [u (ct∗ ) − u (ct )]
T → ∞ t =0
t =0
≥ lim βT u 0 (cT∗ )(−1)kT∗ +1 =0 .
T →∞ |{z}
the transversality condition
12 / 42
Classical approach Dynamic programming Brock-Mirman model

Note on the theorem

The theorem does not guarantee that there is a sequence

{ct , kt +1 }t∞=0 that satisfies both the Euler equation and the
transversality condition.

This is why the theorem does not involve the condition for the
existence of an optimum, such as β ∈ (0, 1) and u is a
bounded function.

13 / 42
Classical approach Dynamic programming Brock-Mirman model

Intuition for the transversality condition

Marginal utility of consumption, u 0 (ct ), is the marginal cost to
replace existing capital in terms of utility of consumption,
because a unit of new capital is produced from a unit of
goods. Thus, u 0 (ct ) is the internal marginal value (or “the
shadow value”) of capital for the representative household.
If
lim βt u 0 (ct )kt +1 > 0,
t →∞
then the household leaves unconsumed some positive present
discounted value of capital.
Such a saving plan is not optimal as the household can
increase consumption today by reducing saving without
sacrificing future consumption.

14 / 42
Classical approach Dynamic programming Brock-Mirman model

Tip: Easy way to find the transversality

condition
• In general, you can rewrite the maximization problem only
with state variables.
• Substitute each equality constraint into the objective
function:
∞
max∞ ∑ β t u ( f ( kt ) + ( 1 − δ ) kt − kt + 1 )
{ ks + 1 } s = 0 t = 0

• In a general form,
∞
max∞ ∑ β t F ( kt , kt + 1 )
{ ks + 1 } s = 0 t = 0

where kt is the state variable for period t.

15 / 42
Classical approach Dynamic programming Brock-Mirman model

Tip: Easy way to find the transversality

condition
• The transversality condition for the general form of the
maximization problem is
∂F (kt , kt +1 )
lim βt kt = 0
t →∞ ∂kt
• In the sequential problem we are considering,
lim βt [f 0 (kt ) + 1 − δ]u 0 (f (kt ) + (1 − δ)kt − kt +1 )kt = 0
t →∞ | {z }
ct
which is equivalent to
lim βt u 0 (ct )kt +1 = 0
t →∞
under the Euler equation, because the Euler equation is
u 0 ( ct ) = β [ f 0 ( kt + 1 ) + 1 − δ ] u 0 ( ct + 1 )
16 / 42
Classical approach Dynamic programming Brock-Mirman model

Necessity of the Euler equation

It is trivial to show the necessity of the Euler equation for an

optimum.

Consider the utility maximization problem over each pair of

two subsequent periods (e.g., t and t + 1) as a finite-horizon
problem given kt and {ks }s∞=t +2 . Then apply the
Kuhn-Tacker theorem.

17 / 42
Classical approach Dynamic programming Brock-Mirman model

Necessity of the transversality condition

Theorem (For a more general theorem, see
Kamihigashi 2002 or Miao 2014, Theorem 7.5.1.)
Suppose that β ∈ (0, 1), that u is a bounded, strictly
increasing and concave, and continuously differentiable
function, and that f is an increasing, concave, and
continuously differentiable function. Also, suppose that there
exists an optimum in which ct > 0 and kt +1 > 0 for all t.
Then, for all t, the optimum satisfies the Euler equation

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

and the transversality condition

lim βt u 0 (ct )kt +1 = 0

t →∞

18 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Lemma
Let g : [0, 1] → R ∪ {−∞} is a concave function with
g (1) > −∞. For any ξ ∈ [0, 1) and λ ∈ [ξ, 1),

g (1) − g ( λ ) g (1) − g ( ξ )
≤
1−λ 1−ξ

Proof.
Denote (1 − λ)/(1 − ξ ) by α. Because α ∈ (0, 1],
λ = αξ + 1 − α, and g is concave,

g (λ) ≥ αg (ξ ) + (1 − α)g (1)

. 19 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Denote the optimal sequence of ct and kt +1 by {ct∗ , kt∗+1 }t∞=0 .
Consider an alternative sequence {k0 , ξk1∗ , ξk2∗ , ...} for some
ξ ∈ [0, 1), which is feasible as ct simply increases by some
amount in each period. Given u is bounded, the value of the
utility function
∞
u (f (k0 ) + (1 − δ)k0 − ξk1∗ ) + ∑ βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )
t =1

is finite.
Consider a positive integer T and a sequence of kt such as
{k0 , k1∗ , ..., kT∗ , λkT∗ +1 , λkT∗ +2 , ...} for some λ ∈ [ξ, 1), which
is feasible (i.e., satisfies all the constraints). Denote by ĉt the
value of ct implied by this sequence of kt .
20 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Because {ct∗ , kt∗+1 }t∞=0 is optimal,
∞ ∞
0≥ ∑ βt u (ĉt ) − ∑ βt u (ct∗ )
t =0 t =0
T −1 T −1
= ∑ βt u (ct∗ ) − ∑ βt u (ct∗ )
t =0 t =0
| {z }
=0
+ βT u (f (kT∗ ) + (1 − δ)kT∗ − λkT∗ +1 ) − βT u (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 )
∞
+ ∑ βt u (f (λkt∗ ) + (1 − δ)λkt∗ − λkt∗+1 )
t =T +1
∞
− ∑ βt u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 )
t =T +1

21 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Multiplying both sides by 1/(1 − λ) yields
βT u (f (kT∗ ) + (1 − δ)kT∗ − λkT∗ +1 ) − βT u (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 )
1−λ
∞
1
1 − λ t =∑
≤ [ βt u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 )
T +1
− βt u (f (λkt∗ ) + (1 − δ)λkt∗ − λkt∗+1 )]
∞
1
≤
1−ξ ∑ [ βt u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 )
t =T +1
− βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )]

The last inequality holds by the lemma derived at the

beginning of the proof, given the concavity of u.
22 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Because kt∗+1 > 0, the partial derivative of u with respect to
kt +1 at kt +1 = kt∗+1 exists. Thus, taking λ → 1 yields

βT u (f (kT∗ ) + (1 − δ)kT∗ − λkT∗ +1 ) − βT u (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 )

lim
λ →1 1−λ
∂ n T o
=− β u (f (kT∗ ) + (1 − δ)kT∗ − λkT∗ +1 )
∂λ
= − βT u 0 (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 ))(−1)kT∗ +1
= βT u 0 (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 ))kT∗ +1
∞
1
≤
1−ξ ∑ [ βt u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 )
t =T +1
− βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )]

23 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition
Therefore,
lim βT u 0 (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 ))kT∗ +1
T →∞
∞
1
≤ lim
T →∞ 1 − ξ
∑ [ βt u (f (kt∗ ) + (1 − δ)kt∗ − kt∗+1 )
t =T +1
− βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )]
=0

The last equality holds as

u (f (k0 ) + (1 − δ)k0 − ξk1∗ ) + ∑t∞=1 βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )
and ∑t∞=0 βt u (ct∗ ) are both finite.

24 / 42
Classical approach Dynamic programming Brock-Mirman model

Proof for the necessity of the transversality

condition

Also,

lim βT u 0 (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 ))kT∗ +1 ≥ 0

T →∞

given u 0 > 0 and kT∗ +1 ≥ 0 by assumption. Thus,

lim βT u 0 (f (kT∗ ) + (1 − δ)kT∗ − kT∗ +1 ))kT∗ +1 = 0 

T →∞

25 / 42
Classical approach Dynamic programming Brock-Mirman model

Note on the boundedness condition on u

The conditions that β ∈ (0, 1) and that u is a bounded

function guarantee that ∑t∞=0 βt u (ct ) is finite for any feasible
sequence of ct .

Even without the boundedness condition, the theorem holds as

long as
u (f (k0 ) + (1 − δ)k0 − ξk1∗ ) + ∑t∞=1 βt u (f (ξkt∗ ) + (1 − δ)ξkt∗ − ξkt∗+1 )
and ∑t∞=0 βt u (ct∗ ) are both finite.

26 / 42
Classical approach Dynamic programming Brock-Mirman model

Dynamic programming

27 / 42
Classical approach Dynamic programming Brock-Mirman model

Dynamic programming: Bellman equation

To use the classical approach for a problem with an inequality
constraint, you need to know beforehand in which period
(more generally, in which state) a constraint binds.

Dynamic programming is an alternative solution method free

of this restriction.

First step: rewrite the original utility maximization problem (a

sequential problem [SP]) into a Bellman equation (a functional
equation [FE]).

V (kt ) = max u (ct ) + βV (kt +1 )

kt + 1
s.t. kt +1 + ct = f (kt ) + (1 − δ)kt
ct ≥ 0, kt +1 ≥ 0

28 / 42
Classical approach Dynamic programming Brock-Mirman model

Dynamic programming: Bellman equation

Make sure that the solution to the FE is equivalent to that to
the SP. (See Stokey and Lucas, 1989, Theorems 4.2 and 4.3.)
• Necessity of FE: Suppose that the constraints in SP are
defined for all possible values of state variables and that
limT →∞ ∑T t
t =0 β u (ct ) exists for any feasible sequence of
ct , {ct }t∞=0 . Then, the solution for SP is that for FE.
• Sufficiency of FE: Suppose that the constraints in SP are
defined for all possible values of state variables and that
limT →∞ ∑T t
t =0 β u (ct ) exists for any feasible sequence of
ct , {ct }t∞=0 . If

lim βT V (kT ) = 0
T →∞

for any feasible sequence of kt +1 , i.e., {kt +1 }t∞=0 , given

k0 , then the solution for FE is that for SP.
29 / 42
Classical approach Dynamic programming Brock-Mirman model

Dynamic programming: Euler equation

Second step: Solve the Bellman equation:

V ( kt ) = max u (ct ) + βV (kt +1 )

{ct ≥0, kt +1 ≥0}
s.t. kt +1 + ct = f (kt ) + (1 − δ)kt

Suppose V is differentiable (see Stokey and Lucas, 1989, Ch.

4 and 9 for sufficient conditions; you can guess and verify the
differentiability in an actual model.)

Use the Lagrangian method to obtain the first-order condition

with respect to kt +1 :

−u 0 (ct ) + βV 0 (kt +1 ) = 0

How to obtain V 0 (kt +1 )?

30 / 42
Classical approach Dynamic programming Brock-Mirman model

Dynamic programming: Euler equation

Third step: Applying the chain rule,
df (g (x ))/dx = f 0 (g (x ))g 0 (x ), to V (kt ) yields:

V 0 (kt ) = u 0 (f (kt ) + (1 − δ)kt − kt +1 )[f 0 (kt ) + (1 − δ)]

dkt +1
+ [−u 0 (f (kt ) + (1 − δ)kt − kt +1 ) + βV 0 (kt +1 )]
| {z } dkt
=0 by the Euler equation
Under the guess that the Euler equation holds with ct > 0 in
each period (which must be verified), the envelop theorem
holds for t + 1:
V 0 (kt +1 ) = u 0 (ct +1 )[f 0 (kt +1 ) + (1 − δ)]
In this case, the Euler equation in the classical approach is
derived:
u 0 (ct ) = βu 0 (ct +1 )[f 0 (kt +1 ) + (1 − δ)]
31 / 42
Classical approach Dynamic programming Brock-Mirman model

Closed-form solution to a dynamic

general equilibrium model:
Brock-Mirman model

32 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model: A

deterministic case
• Sequential problem:
∞
max
{ct ≥0,kt +1 ≥0}t∞=0
∑ βt ln ct
t =0
s.t. kt +1 + ct = A1t −α ktα
where: At is an exogenous technological level; α ∈ (0, 1);
β ∈ (0, 1).
• Bellman equation:
V (At , kt ) = max ln(A1t −α ktα − kt +1 ) + βV (At +1 , kt +1 )
kt +1
• Euler equation:
1 A1−α αk α−1
= β t +1 t +1
ct ct + 1
33 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model:

Closed-form solution to the policy function
Definition of a policy function: it returns the optimal value of
each choice variable given the values of state variables.
Guess and verify ct = γA1t −α ktα with an undetermined
coefficient γ (thus the Euler equation holds with ct > 0 in
each period).
1 αA1−α k α−1
= β t +1 t +1
ct ct + 1
1− α α −1
1 αA k
∴ = β t1+−1α tα+1
ct γAt +1 kt +1
1 α
∴ 1− α α
=β
γAt kt γ(1 − γ)A1t −α ktα
∴ γ = 1 − αβ
34 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model: Verify

the guess

The policy function was derived under the conjecture that the
value function, V (kt ), is differentiable.

To complete the proof for the solution, it is necessary to

confirm this conjecture by deriving the value function.

Also, it is not yet confirmed that the solution for FE is

equivalent to that for SP.

35 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model:

Detrending the model
If ln At +1 = ln(1 + g ) + ln At , At has a deterministic trend.
For a general variable xt , denote xt /At by x̂t . The model can
be detrended by At , so that the value function is bounded:
∞
max
{ct ≥0,kt +1 ≥0}t∞=0
∑ βt ln ct s.t. kt +1 + ct = A1t −α ktα
t =0
is equivalent to
∞
max
{ĉt ≥0,k̂t +1 ≥0}t∞=0
∑ βt (ln ĉt + ln At ) s.t. (1 + g )k̂t +1 + ĉt = k̂tα
t =0
whose solution coincides with the solution to
∞
max
{ĉt ≥0,k̂t +1 ≥0}t∞=0
∑ βt ln ĉt s.t. (1 + g )k̂t +1 + ĉt = k̂tα
t =0
36 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model:

Detrending the model
Let’s confirm limT →∞ βT V (k̂T ) = 0 by deriving the value
function for the detrended case.
αβk̂tα
Policy function: ĉt = (1 − αβ)k̂tα , k̂t +1 = 1+g

α
Value function: V (k̂t ) = constant + ln k̂t
1 − αβ

The differentiability of V (k̂t ) for k̂t > 0 is verified.

Also, because k̂t +1 can be k̂tα /(1 + g ) at most given ĉt ≥ 0,

limT →∞ βT V (k̂T ) = 0 holds for any feasible sequence of k̂t ,
given α ∈ (0, 1) and β ∈ (0, 1). Thus, the solution for FE is
equivalent to that for SP.
37 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model:

Detrending the model
Caveat: The instantaneous utility function, ln ct , is not
well-defined for ct = 0.

A Remedy:
• Assume
(
ln At c if ct < At c,
u (ct ) =
ln ct if ct ≥ At c,

where c is a positive constant arbitrarily close to zero, so

that V (k̂t ) ≥ u/(1 − β) for any feasible path of {k̂t }t∞=1 .

This assumption does not affect the solution for SP.

38 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model: Phase

diagram
Given ln At +1 = ln(1 + g ) + ln At for all t, draw the phase
diagram.
αβ α
Equilibrium law of motion for capital: k̂t +1 = 1+g k̂t
k̂t +1

αβ α
k̂t +1 = 1+g k̂t

45o
k̂t
O

39 / 42
Classical approach Dynamic programming Brock-Mirman model

Brock and Mirman (1972) model: A

stochastic case
Try if you are interested (not mandatory): Consider the
sequential problem in page 31. Assume that
ln At +1 = ρ ln At + et +1
ρ ∈ (0, 1)
for all t, where et +1 is an independent white noise. Guess and
verify that the value function takes the following form:
V (At , kt ) = F + G ln At + H ln kt
for some constants F , G , and H. Also show
h i
lim E0 βT V (AT , kT ) = 0
T →∞
for any feasible sequence of kt , where E0 is the expectation
operator given A0 and k0 in period 0.
40 / 42
Classical approach Dynamic programming Brock-Mirman model

Further readings

• Miao (2014), Ch. 7.

• Stokey and Lucas (1989), Ch. 3, 4 and 9.
• Ljungqvist and Sargent (2012), Ch. 3 and 4.

41 / 42
Classical approach Dynamic programming Brock-Mirman model

References
• Brock, William A., and Leonard Mirman, 1972, “Optimal
Economic Growth and Uncertainty: The Discounted
Case,” Journal of Economic Theory, 4(3): 479-513.
• Kamihigashi, Takashi, 2002, “A Simple Proof of the
Necessity of the Transversality Condition,” Economic
Theory, 20(2): 427-433.
• Ljungqvist, Lars, and Thomas J. Sargent, 2012, Recursive
Macroeconomic Theory, third edition, MIT Press.
• Miao, Jianjun, 2014, Economic Dynamics in Discrete
Time, MIT Press.
• Stokey, Nancy L., and Robert E. Lucas, Jr., 1989,
Recursive Methods in Economic Dynamics, Harvard
University Press.
42 / 42