Final Exam

Economics 720: Mathematics for Economists

Fall 1999

This exam has three questions on three pages; before you begin, please check to make
sure that your copy has all three questions and all three pages. Each question has five parts.
Each part of each question is worth 6 points, for a total of 6 × 5 × 3 = 90 points.

1 The Maximum Principle: Optimal Growth

This question asks you to apply the maximum principle to solve a social planner’s problem
in continuous time and with an infinite horizon. Consider an economy in which output is
produced with capital during each period t ∈ [0, ∞) according to the production function

F (k(t)) = k(t)α ,

with 1 > α > 0. Let c(t) denote the consumption of a representative consumer, and let δ
denote the depreciation rate of capital, where 1 > δ > 0. Then the capital stock evolves
according to
k(t)α − δk(t) − c(t) ≥ k̇(t)
for all t ∈ [0, ∞). The social planner takes the initial capital stock, k(0), as given and
chooses functions c(t) for t ∈ [0, ∞) and k(t) for t ∈ (0, ∞) to maximize the representative
consumer’s utility, given by Z ∞ ∙ ¸
1−σ
−ρt c(t)
e dt
0 1−σ
with ρ > 0, σ > 0, and σ 6= 1, subject to the constraint governing the evolution of k(t) for
all t ∈ [0, ∞).

a) To solve the social planner’s problem using the maximum principle, begin by setting up
the Hamiltonian.
b) Write down the first order condition for the static optimization problem that appears in
the definition of the Hamiltonian.
c) Next, write down the two differential equations that, according to the maximum principle,
must be satisfied by the problem’s solution.
d) Combine your results from parts (b) and (c) above to obtain a system of two differential
equations involving c(t) and k(t) that describe the problem’s solution.

1
 e) Your results from part (d) above should imply that starting from any initial k(0), the
economy will converge to a steady state in which k(t) remains constant at some value
k∗ and c(t) remains constant at some value c∗ . Use your results from part (d) to derive
solutions for the steady-state values k∗ and c∗ in terms of the model’s parameters: α,
δ, ρ, and σ.

2 Stochastic Dynamic Programming: Saving with a

Random Return
This problem asks you to use dynamic programming to solve a representative consumer’s
problem when savings earn a random rate of return. Let time be discrete and the horizon
be infinite, so that periods are indexed by t = 0, 1, 2, .... Let At denote the consumer’s assets
at the beginning of period t. During each period, the consumer divides these assets up into
an amount ct to be consumed and an amount st to be saved. The consumer’s savings earn
interest at the gross rate Rt+1 , where Rt+1 is random, possibly serially correlated, and does
not become known until the beginning of period t + 1. Thus, the consumer must choose st
before knowing the realized value of Rt+1 . The consumer takes the initial stock of assets A0
as given and chooses sequences {ct }∞ ∞
t=0 and {At }t=1 to maximize the expected utility function

X
∞ X
∞
t
E0 β u(ct ) = E0 β t u(At − st )
t=0 t=0

subject to the constraint

At+1 = Rt+1 st ,
which must hold for all t = 0, 1, 2, ... as well as for all possible realizations of Rt+1 .

a) Set up the Bellman equation for this problem, using At as the state variable, st as the
control variable, and recognizing that Rt+1 is random.
b) Now suppose that the consumer’s single-period utility function takes the form
(At − st )1−σ
u(ct ) = u(At − st ) = ,
1−σ
where σ > 0 and σ 6= 1. Suppose also that the random interest rate Rt+1 is indepen-
dently and identically distributed over time with
1−σ
Et Rt+1 =1
for all t = 0, 1, 2, .... Guess that under these conditions, the value function depends
only on At , where
KA1−σ
t
v(At ) =
1−σ
for some unknown constant K. Use this guess, together with the assumptions on u(ct )
and Rt+1 stated above, to rewrite your Bellman equation from part (a). Then derive
the first order and envelope conditions for the consumer’s problem.

2
 c) Use your results from part (b) above to derive expressions that show how the optimal
values of ct and st depend on the stock of assets At , the parameters β and σ, and the
unknown constant K. In doing this, it may be helpful to note that since st is known
at time t, Et st = st .

d) Use your results from parts (b) and (c) above to solve for the unknown K in terms of
the parameters β and σ.

e) Finally, combine your results from parts (c) and (d) above to derive a solution that shows
how the optimal ct depends on At and the parameters β and σ.

3 A System of Linear Differential Equations

Consider the system of linear differential equations consisting of

ẋ1 (t) = −x1 (t) + 3x2 (t)

and
ẋ2 (t) = 2x1 (t).
This system can be written more simply as

ẋ(t) = Ax(t),

where ∙ ¸
ẋ1 (t)
ẋ(t) = ,
ẋ2 (t)
∙ ¸
x1 (t)
x(t) = ,
x2 (t)
and ∙ ¸
−1 3
A= .
2 0

a) As a first step in solving this system, find the two eigenvalues of the matrix A.

b) Next, find eigenvectors corresponding to the two eigenvalues that you identified in part
(a) above.

c) Use your results from parts (a) and (b) above to construct a nonsingular matrix P and
a diagonal matrix D such that
P −1 AP = D.

d) Use your results from part (c) above to find the general solutions for x1 (t) and x2 (t).

e) Use your results from above to identify the stationary solution to the system. Is the
stationary solution asymptotically stable or unstable?

3
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 1999

1 The Maximum Principle: Optimal Growth

a) The Hamiltonian for the social planner’s problem is
∙ 1−σ
¸
−ρt c(t)
H(k(t), π(t); t) = max e + π(t)[k(t)α − δk(t) − c(t)]
c(t) 1−σ

b) The first order condition for the static optimization problem that appears in the definition
of H(k(t), π(t)) stated above is
e−ρt c(t)−σ − π(t) = 0.

c) The differential equations that, according to the maximum principle, must be satisfied
by the problem’s solution are
π̇(t) = −Hk (k(t), π(t); t) = −π(t)[αk(t)α−1 − δ]
and
k̇(t) = Hπ (k(t), π(t); t) = k(t)α − δk(t) − c(t).
d) Take the first order condition from part (b) and rewrite it as
e−ρt = c(t)σ π(t).
Differentiate both sides of this equation with respect to t to obtain
−ρe−ρt = σc(t)σ−1 ċ(t)π(t) + c(t)σ π̇(t).
Use the first order condition again, together with the differential equation for π̇(t) from
part (c), to rewrite this last result as
−ρc(t)σ π(t) = σc(t)σ−1 ċ(t)π(t) − c(t)σ π(t)[αk(t)α−1 − δ].
Now divide through by σπ(t)c(t)σ−1 and isolate ċ(t) on one side to obtain
ċ(t) = (1/σ)c(t)[αk(t)α−1 − δ − ρ],
which together with the differential equation for k̇(t) in part (c),
k̇(t) = k(t)α − δk(t) − c(t),
forms a system of two differential equations that determine the optimal c(t) and k(t).

1
e) The differential equation

ċ(t) = (1/σ)c(t)[αk(t)α−1 − δ − ρ]

from part (d) implies that in a steady-state where ċ(t) = 0, k(t) must satisfy
µ ¶1/(α−1)
∗ δ+ρ
k(t) = k = .
α
The differential equation
k̇(t) = k(t)α − δk(t) − c(t)
from part (d) then implies that when k(t) = k∗ and k̇(t) = 0,
µ ¶α/(α−1) µ ¶1/(α−1)
∗ δ+ρ δ+ρ
c(t) = c = −δ .
α α

2 Stochastic Dynamic Programming: Saving with a

Random Return
a) The Bellman equation for the consumer’s problem is

v(At , Rt ) = max u(At − st ) + βEt v(Rt+1 st , Rt+1 ).

st

b) Using the assumptions on u(ct ) and Rt+1 and the suggested guess for v(At ), the Bellman
equation becomes
KA1−σ
t (At − st )1−σ βKEt [(Rt+1 st )1−σ ]
= max + .
1−σ st 1−σ 1−σ
The first order condition is

−(At − st )−σ + βKEt (Rt+1

1−σ −σ
st ) = 0.

The envelope condition is

KA−σ
t = (At − st )−σ .
1−σ
c) Since st is known at time t and since Et Rt+1 = 1, the first order condition from part (b)
can be rewritten as
(At − st )−σ = βKs−σt .

Use this expression to solve for

∙ ¸
(βK)1/σ
st = At
1 + (βK)1/σ
and ∙ ¸
1
ct = At − st = At .
1 + (βK)1/σ

2
d) Substituting the results from part (c) back into the Bellman equation yields
µ ¶∙ ¸1−σ
KA1−σ
t 1 1
= A1−σ
t
1−σ 1−σ 1 + (βK)1/σ
µ ¶∙ ¸1−σ
βK (βK)1/σ
+ A1−σ
t
1−σ 1 + (βK)1/σ

or, more simply

∙ ¸1−σ ∙ ¸1−σ
1 (βK)1/σ
K = + βK
1 + (βK)1/σ 1 + (βK)1/σ
∙ ¸1−σ ∙ ¸1−σ
1 (1−σ)/σ 1
= + βK(βK)
1 + (βK)1/σ 1 + (βK)1/σ
∙ ¸1−σ ∙ ¸1−σ
1 1/σ 1
= + (βK)
1 + (βK)1/σ 1 + (βK)1/σ
∙ ¸1−σ
1
= [1 + (βK)1/σ ]
1 + (βK)1/σ
= [1 + (βK)1/σ ]σ .

Hence,
K 1/σ = 1 + (βK)1/σ ,
1
K 1/σ = ,
1 − β 1/σ
or
K = [1 − β 1/σ ]−σ .

e) The solution for K from part (d) implies

1/σ β 1/σ
(βK) = ,
1 − β 1/σ
1
1 + (βK)1/σ = ,
1 − β 1/σ
and
1
= 1 − β 1/σ .
1 + (βK)1/σ
Substitute this last result into the solution for ct from part (c) to obtain
∙ ¸
1
ct = At = (1 − β 1/σ )At ,
1 + (βK)1/σ

which shows how ct depends on At and the parameters β and σ.

3
3 A System of Linear Differential Equations
a) By definition, the eigenvalues of A are values of r that satisfy
∙ ¸
−1 − r 3
0 = det = r(1 + r) − 6 = r2 + r − 6 = (r + 3)(r − 2).
2 −r

Evidently, the eigenvalues are r1 = −3 and r2 = 2.

b) An eigenvector v1 corresponding to the eigenvalue r1 = −3 must satisfy
∙ ¸∙ ¸
2 3 v11
= 0.
2 3 v12

One such vector is ∙ ¸

3
v1 = .
−2
An eigenvector v2 corresponding to the eigenvalue r2 = 2 must satisfy
∙ ¸∙ ¸
−3 3 v21
= 0.
3 −2 v22

One such vector is ∙ ¸

1
v2 = .
1

c) Form the matrix P using v1 and v2 as its columns:

∙ ¸
3 1
P = .
−2 1

Form the matrix D using r1 and r2 as its diagonal elements:

∙ ¸
−3 0
D= .
0 2

It can be verified that P is nonsingular and that

P −1 AP = D.

d) Use the matrices P and D from above to rewrite the system as

ż(t) = Dz(t),

where ∙ ¸
ż1 (t)
ż(t) = = P −1 ẋ(t)
ż2 (t)
and ∙ ¸
z1 (t)
z(t) = = P −1 x(t).
z2 (t)

4
 Since D is diagonal, the transformed system consists of two self-contained equations

ż1 (t) = r1 z1 (t) = −3z1 (t)

and
ż2 (t) = r2 z2 (t) = 2z2 (t).
We know that these equations have general solutions

z1 (t) = k1 e−3t

and
z2 (t) = k2 e2t .
Now undo the transformation to solve for x(t):
∙ ¸ ∙ ¸∙ ¸
ẋ1 (t) 3 1 z1 (t)
= x(t) = P z(t) =
ẋ2 (t) −2 1 z2 (t)
or
x1 (t) = 3k1 e−3t + k2 e2t
and
x2 (t) = −2k1 e−3t + k2 e2t .

e) The stationary solution is obtained by setting k1 = k2 = 0, so that

x1 (t) = x2 (t) = 0

for all t. For any initial conditions x1 (0) and x2 (0) that imply k2 6= 0, however,

lim x1 (t) = lim (3k1 e−3t + k2 e2t ) = lim k2 e2t = ∞

t→∞ t→∞ t→∞

and
lim x2 (t) = lim (−2k1 e−3t + k2 e2t ) = lim k2 e2t = ∞,
t→∞ t→∞ t→∞

which show that the stationary solution is unstable.

5
 Final Exam
Economics 720: Mathematics for Economists
Fall 2001

This exam has four questions on four pages; before you begin, please check to make sure
that your copy has all four questions and all four pages. Each part of each question will be
given equal weight in determining your total exam score. Since this is a two-hour exam, I
suggest that you spend approximately (two hours)/(four questions) = 30 minutes on each
question.

1 The Kuhn-Tucker and Envelope Theorems: Cost Min-

imization
This problem asks you to apply the Kuhn-Tucker and envelope theorems to solve a static
cost-minimization problem faced by a firm. The firm produces output with capital k and
labor l according to the Cobb-Douglas production function

kα l1−α

with 0 < α < 1. Let r denote the rental rate for capital, and let w denote the wage rate for
labor. Then the firm’s total costs are

rk + wl.

The firm chooses k and l to minimize its total costs, subject to the requirement that it
produce at least y units of output. This problem can be stated formally as

min rk + wl subject to kα l1−α ≥ y.

k,l

When solving this problem, the firm takes the factor prices r and w and the output require-
ment y as given.

a) Set up the Lagrangian for the firm’s problem, and write down the first-order conditions
for the values k∗ and l∗ that, according to the Kuhn-Tucker theorem, must be satisfied
by the solution to the firm’s problem.

b) Together with the binding constraint

y = (k∗ )α (l∗ )1−α ,

1
 your two first-order conditions from part (a) form a system of three equations in three
unknowns: the values k∗ and l∗ that solve the firm’s problem and the corresponding
value of the multiplier that you introduced into the problem when setting up the
Lagrangian. Use these three equations to solve for the three unknowns in terms of the
model’s four parameters: α, r, w, and y.

c) Now define the minimum cost function as

C(α, r, w, y) = min rk + wl subject to kα l1−α ≥ y.

k,l

Use your results from part (b) to calculate the firm’s marginal cost, defined as the
derivative C4 (α, r, w, y) of the minimum cost function with respect to output, in terms
of the model’s parameters α, r, w, and y.

2 The Maximum Principle: Optimal Pricing with En-

try
This problem asks you to apply the maximum principle to solve the pricing problem faced
by a large firm in an industry where smaller firms gradually enter if the large firm’s price
becomes too high. Time is continuous and the horizon is infinite, so that periods are indexed
by t ∈ [0, ∞). The demand curve facing the industry as a whole during period t is

q(t) = a − bp(t),

where q(t) is quantity demanded, p(t) is the large firm’s price, and a and b are positive
constants.
New, smaller firms begin to enter when the large firm charges a price above some threshold
p̄. Hence, if x(t) denotes the output produced and sold by the smaller firms, this variable
evolves according to
ẋ(t) = k[p(t) − p̄]
for all t ∈ [0, ∞), where k is also a positive constant. The large firm’s sales are then given
by
q(t) − x(t) = a − bp(t) − x(t)
for all t = [0, ∞). If the large firm produces output at the constant marginal cost of c per
unit, then its profits during period t equal price minus cost times quantity sold:

[p(t) − c][a − bp(t) − x(t)].

The large firm’s problem can be stated formally as: choose p(t) for all t ∈ [0, ∞) and
x(t) for all t ∈ (0, ∞) to maximize the present discounted value of profits over the infinite
horizon, Z ∞
e−ρt [p(t) − c][a − bp(t) − x(t)]dt,
0

2
where ρ > 0 is the discount rate, subject to the constraints x(0) = x0 given and

ẋ(t) = k[p(t) − p̄]

for all t ∈ [0, ∞).

a) Set up the Hamiltonian for the large firm’s problem, treating p(t) as the flow variable
and x(t) as the stock variable. Then write down the first-order condition for the
choice of p(t) that solves the static optimization problem on the right-hand side of
the Hamiltonian and the two differential equations that, according to the maximum
principle, help characterize the solution to the firm’s problem.
b) Your results from part (a) form a system of three equations in three unknowns: the
flow variable p(t), the stock variable x(t), and the new variable that you introduced
when setting up the Hamiltonian. Reduce this three-equation system to a smaller,
two-equation system that involves only the flow variable p(t) and the stock variable
x(t). One of these two equations should be an equation for ṗ(t), and the other should
be an equation for ẋ(t).
c) Your two-equation system from part (b) should have a steady state, in which p(t) and
x(t) are constant, with p(t) = p∗ , x(t) = x∗ , ṗ(t) = 0, and ẋ(t) = 0. Use your results
from above to solve for the steady-state values of p∗ and x∗ in terms of the model’s
parameters: a, b, c, k, p̄, and ρ.

3 Dynamic Programming: Optimal Resource Deple-

tion
This question asks you to use dynamic programming to characterize the optimal strategy
for consuming an exhaustible, or nonrenewable, resource. Time is discrete and the horizon
is infinite, so that periods are indexed by t = 0, 1, 2, .... At the beginning of period t = 0,
the available stock of the resource is given by k0 . During each period t = 0, 1, 2, ..., the
representative agent consumes ct units of the resource; since the resource is nonrenewable,
this consumption simply subtracts from the available stock according to

kt+1 = kt − ct

for all t = 0, 1, 2, ..., where kt is the stock available at the beginning of period t.
Thus, the representative agent chooses {ct }∞ ∞
t=0 and {kt }t=1 to maximize the utility func-
tion
X
∞
β t ln(ct ),
t=0

with constant discount factor 0 < β < 1, subject to the constraints k0 given and

kt − ct ≥ kt+1

for all t = 0, 1, 2, ....

3
 a) Write down the Bellman equation for this problem, using kt as the state variable, using
ct as the control variable, and guessing that the value function takes the time-invariant
form
v(kt ) = E + F ln(kt ),
where E and F are constants to be determined. Then write down the first-order and
envelope conditions that characterize the solution to the problem.

b) Your first-order and envelope conditions from part (a) can be combined with the binding
constraint,
kt+1 = kt − ct ,
to form a system of three equations in three unknowns: the unknown variables kt and
ct that solve the original optimization problem and the unknown constant F . Use the
equations from this system to solve for the constant F in terms of the model’s single
parameter: the discount factor β. (Don’t worry about trying to solve for E, since this
constant does not enter into the expressions for the optimal kt and ct .)

c) Now combine your solution for F with the remaining optimality conditions to obtain a
system of two equations that can be used to construct the optimal sequences {ct }∞
t=0
and {kt }∞
t=1 , given the initial condition k0 and the discount factor β.

4 Differential Equations: Bank Accounts

Let y(t) keep track of a consumer’s bank account balances in a continuous-time, infinite-
horizon model where periods are indexed by t ∈ [0, ∞). At each instant in time, these
balances grow for two reasons. First, at every instant t, the balances y(t) earn interest at
the constant rate r. Second, at every instant t, the consumer deposits a constant amount b
of additional funds into the account.

a) Write down the differential equation that describes the evolution of y(t), the bank account
balances, over time.

b) Write down the general solution to this differential equation.

c) Now let y(0) = y0 denote the given amount that the consumer initially deposits in the
account. Write down the particular solution that satisfies this initial condition.

4
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2001

1 The Kuhn-Tucker and Envelope Theorems: Cost Min-

imization
a) With the Lagrangian defined as
L(k, l, λ) = rk + wl − λ(kα l1−α − y),
the first-order conditions for the optimal k∗ and l∗ are
r − αλ∗ (k∗ )α−1 (l∗ )1−α = 0 (1)
and
w − (1 − α)λ∗ (k ∗ )α (l∗ )−α = 0. (2)
b) Together with the binding constraint
y = (k∗ )α (l∗ )1−α , (3)
the first-order conditions (1) and (2) form a system of three equations in the three
unknowns k∗ , l∗ , and λ∗ . There are many ways to solve this system; here is one. Start
by using (3) to rewrite (1) and (2) as
k∗ = αλ∗ (y/r) (4)
and
l∗ = (1 − α)λ∗ (y/w). (5)
Now substitute these results into (3) to solve for λ∗ in terms of the parameters:
y = αα (λ∗ )α y α r−α (1 − α)1−α (λ∗ )1−α y 1−α w−(1−α)
1 = λ∗ αα r−α (1 − α)1−α w−(1−α)
λ∗ = α−α (1 − α)−(1−α) rα w1−α . (6)
And substitute (6) back into (4) and (5) to solve for k∗ and l∗ in terms of the parameters:
k∗ = α1−α (1 − α)−(1−α) r−(1−α) w1−α y (7)
and
l∗ = α−α (1 − α)α rα w−α y. (8)

1
c) The Kuhn-Tucker theorem implies that
C(α, r, w, y) = rk ∗ + wl∗ − λ∗ [(k∗ )α (l∗ )1−α − y],
and the envelope theorem implies that
C4 (α, r, w, y) = λ∗ .
Hence, marginal cost can be found simply by using (6):
C4 (α, r, w, y) = α−α (1 − α)−(1−α) rα w1−α . (9)

2 The Maximum Principle: Optimal Pricing with En-

try
a) The Hamiltonian is
H(x(t), π(t); t) = max e−ρt [p(t) − c][a − bp(t) − x(t)] + π(t)k[p(t) − p̄].
p(t)

The first-order condition for p(t) is

e−ρt [a − bp(t) − x(t)] − be−ρt [p(t) − c] + π(t)k = 0, (10)
and the two differential equations for π(t) and x(t) are
π̇(t) = −Hx (x(t), π(t); t) = e−ρt [p(t) − c] (11)
and
ẋ(t) = Hπ (x(t), π(t); t) = k[p(t) − p̄]. (12)
b) Rearrange the terms in (10) to obtain
π(t) = (e−ρt /k)[2bp(t) + x(t) − (a + bc)],
then differentiate both sides with respect to time to obtain
π̇(t) = −ρ(e−ρt /k)[2bp(t) + x(t) − (a + bc)] + (e−ρt /k)[2bṗ(t) + ẋ(t)]
or, using (11) and (12),
ṗ(t) = (1/2b)[2bρp(t) + ρx(t) + k(p̄ − c) − ρ(a + bc)]. (13)
Equations (12) and (13) now form a system of two equations in the two variables p(t)
and x(t).
c) In a steady-state with p(t) = p∗ , x(t) = x∗ , ṗ(t) = 0, and ẋ(t) = 0, (12) implies that
0 = k[p(t) − p̄]
or
p∗ = p̄. (14)
Equation (13) then implies that
0 = (1/2b)[2bρp(t) + ρx(t) + k(p̄ − c) − ρ(a + bc)]
or
x∗ = (a + bc) − (k/ρ)(p̄ − c) − 2bp̄. (15)

2
3 Dynamic Programming: Optimal Resource Deple-
tion
a) The Bellman equation is

E + F ln(kt ) = max ln(ct ) + βE + βF ln(kt − ct ).

ct

The first-order condition for ct is

1 βF
− = 0, (16)
ct kt − ct
and the envelope condition is
F βF
= . (17)
kt kt − ct
b) To solve for F , use (16) to obtain
µ ¶
1
ct = kt . (18)
1 + βF

Now substitute (18) into (17) to obtain

µ ¶
1
=1−β (19)
1 + βF
or
1
F = , (20)
1−β
which provides a solution for F in terms of β and verifies that F is a constant.

c) Combine (18) and (19) to obtain

ct = (1 − β)kt . (21)
Substitute (21) into the binding constraint

kt+1 = kt − ct

to obtain
kt+1 = βkt . (22)
Equations (21) and (22) can be used to construct the optimal sequences {ct }∞
t=0 and
∞
{kt }t=1 , given the initial condition k0 and the discount factor β.

3
4 Differential Equations: Bank Accounts
a) The differential equation is
ẏ(t) = ry(t) + b. (23)

b) The general solution to the differential equation (23) is

y(t) = −b/r + kert , (24)

since this solution implies that

ẏ(t) = rkert
equals
ry(t) + b = −b + rkert + b = rkert
as required.

c) Combining the initial condition y(0) = y0 given with the general solution (24) yields

y0 = y(0) = −b/r + k

or
k = y0 + b/r.
Hence, the particular solution is

y(t) = −b/r + (y0 + b/r)ert . (25)

4
 Midterm Exam
Economics 720: Mathematics for Economists
Fall 2002

This exam has four questions on five pages. Before you begin, please make sure that
your copy has all four questions and all five pages. Each question has four parts, and each
question will be given equal weight in determining your overall exam score. Since this is an
80-minute exam, I suggest that you spend approximately (80 minutes)/(4 questions) = 20
minutes on each question.

1 The Kuhn-Tucker Theorem and the Cobb-Douglas

Utility Function
A consumer likes two goods: good 1 and good 2. The consumer’s preferences are described
by the Cobb-Douglas utility function

U (c1 , c2 ) = ca1 c21−a ,

where c1 denotes consumption of good 1, c2 denotes consumption of good 2, and the para-
meter a lies between zero and one: 1 > a > 0. Let I denote the consumer’s income, let p1
denote the price of good 1, and let p2 denote the price of good 2. Then the consumer can
be viewed as choosing c1 and c2 to maximize utility, subject to the budget constraint

I ≥ p1 c1 + p2 c2 .

a) Define the Lagrangian for this constrained optimization problem.

b) Write down the full set of conditions that, according to the Kuhn-Tucker theorem, must
be satisfied by the consumer’s optimal choices c∗1 and c∗2 together with the associated
value of the Lagrange multiplier.
c) Use your results from above to solve for the optimal choices c∗1 and c∗2 in terms of the
model’s four parameters: a, I, p1 , and p2 . In doing this, it may help to notice in
advance that since the consumer’s utility function is strictly increasing, the budget
constraint will always bind at the optimum.
d) At the optimum, the consumer spends the fraction
p1 c∗1
I
1
 of his or her income on good 1 and the fraction
p2 c∗2
I
of his or her income on good 2. What do your results from above tell you about the
implications of the Cobb-Douglas utility function for these optimal expenditure shares?

2 The Maximum Principle in Discrete Time

Consider the same fairly general discrete-time dynamic optimization problem that we studied
in class. The horizon is finite, so that periods are indexed by t = 0, 1, ..., T . There is one
stock variable yt and one flow variable zt . Thus, sequences {zt }Tt=0 and {yt }Tt=1+1
must be
chosen to maximize the objective function

X
T
β t F (yt , zt ; t),
t=0

with 1 ≥ β > 0, subject to the constraints

yt + Q(yt , zt ; t) ≥ yt+1 for all t = 0, 1, ..., T,

c ≥ G(yt , zt ; t) for all t = 0, 1, ..., T,

y0 given,
and
yT +1 ≥ y ∗ .

a) Define the Hamiltonian for this problem.

b) Write down the first-order condition for the static optimization problem that appears in
the definition of the Hamiltonian.

c) Write down the pair of difference equations that, according to the Maximum Principle,
must be satisfied by the solution to the original, dynamic optimization problem.

d) Write down the pair of boundary conditions that, according to the Maximum Principle,
must also be satisfied by the solution to the original, dynamic optimization problem.

3 The Maximum Principle in Continuous Time: Cus-

tomer Flows
Consider the following model, where time is continuous and the horizon is infinite, so that
time periods can be indexed by t ∈ [0, ∞). At each time t, a representative firm has n(t)
customers. If the firm sets price p(t), each of its customers purchases p(t)−θ units of output,
where θ > 1 is a parameter that measures the absolute value of the price elasticity of demand.

2
Hence, the firm’s total sales during period t are n(t)p(t)−θ and its total revenues equal price
times quantity sold:
p(t)n(t)p(t)−θ = n(t)p(t)1−θ .
Assume that the representative firm faces increasing marginal costs. More specifically,
suppose that producing y units of output costs y α , where α > 1. It follows that the firm’s
total costs during period t are

[n(t)p(t)−θ ]α = n(t)α p(t)−θα .

The firm’s profits during period t equal revenues minus costs:

n(t)p(t)1−θ − n(t)α p(t)−θα .

Total discounted profits over the infinite horizon are therefore

Z ∞
e−ρt [n(t)p(t)1−θ − n(t)α p(t)−θα ],
0

where ρ > 0 is the discount rate.

Suppose, for simplicity, that all of the other firms in the industry charge the same price q
for their output, and that this price q remains constant over time. Suppose, therefore, that
the representative firm gains customers during periods when it charges a price p(t) that is
lower than q, and loses customers during periods when it charges a price p(t) that is higher
than q. This idea can be summarized by imposing a constraint governing the evolution of
n(t):
ṅ(t) = −φ[p(t) − q]n(t),
where φ > 0. Consistent with our assumptions, this constraint implies that ṅ(t) > 0, so
that the representative firm gains customers, if p(t) < q; conversely, ṅ(t) < 0, so that the
representative firm loses customers, if p(t) > q. We can generalize this constraint to allow
the representative firm to turn away customers; it would then read

−φ[p(t) − q]n(t) ≥ ṅ(t).

We can also assume, however, this constraint will always bind at the optimum.
The representative firm’s dynamic optimization problem may now be stated formally as:
choose continuously differentiable functions p(t) and n(t) for t ∈ [0, ∞) to maximize the
objective function Z ∞
e−ρt [n(t)p(t)1−θ − n(t)α p(t)−θα ],
0
subject to the constraints
n(0) given
and
−φ[p(t) − q]n(t) ≥ ṅ(t)
for all t ∈ [0, ∞).

3
 a) Define the Hamiltonian for the representative firm’s problem. In doing this, it may help
to view n(t) as the problem’s stock variable and p(t) as the problem’s flow variable.

b) Write down the first-order condition for the firm’s optimal choice of p(t), together with
the pair of differential equations that, according to the Maximum Principle, must be
satisfied by the solution to the firm’s dynamic optimization problem.

c) Now consider a symmetric, steady-state equilibrium for the industry as a whole. In such
an equilibrium, all firms, including the representative firm, charge the same constant
price q. And if, in addition, there are as many firms as total customers, each firm,
including the representative firm, will always serve a single customer. In terms of our
notation from above, these steady-state conditions can be written as

p(t) = q and ṗ(t) = 0 for all t ∈ [0, ∞)

and
n(t) = 1 and ṅ(t) = 0 for all t ∈ [0, ∞).
Substitute these steady-state conditions into the first-order condition and differential
equations that you derived above for the representative firm.

d) Finally, use your equations from above to derive a single equation involving only the
steady-state equilibrium price q and the model’s four parameters: θ > 1, α > 1, ρ > 0,
and φ > 0. IMPORTANT NOTE: It will take too much algebra to solve this equation
for q in terms of θ, α, ρ, and φ, so don’t bother to do this. Just derive the equation
itself, and don’t worry about its solution.

4 Dynamic Programming Under Certainty: Optimal

Monetary Policy
Consider the following macroeconomic model. Time is discrete and the horizon is infinite,
so that periods are indexed by t = 0, 1, 2., .... The output gap is denoted by yt ; the inflation
rate is denoted by π t ; the nominal interest rate is denoted by rt ; and the real interest rate is
calculated as rt − π t .
The central bank sets the nominal interest rate rt . The output gap is then determined
by the Keynesian IS curve,
yt = −σ(rt − πt ),
with σ > 0, which just says that output falls when the real interest rate rises; conversely,
output rises when the real interest rate falls. This equation implies that the central bank can
pick any value it wants for the output gap yt simply by choosing the appropriate value for
rt . So we can simplify things by just assuming from now on that the central bank chooses
yt ; then we can ignore the IS curve throughout the rest of our analysis.
Given the central bank’s choice of yt , inflation is determined by the Phillips curve

π t+1 = π t + ψyt ,

4
with ψ > 0, which just says that inflation rises between t and t + 1 when the output gap yt
is positive; conversely, inflation falls when the output gap is negative.
Now suppose that the central bank wants to stabilize the inflation rate πt and the output
gap yt around target values that both equal zero. These policy goals are reflected in the
central bank’s objective function for period t,
−π 2t − αyt2 ,
which penalizes departures of both variables from zero. In this single-period objective func-
tion, α > 0 measures the weight that the central bank places on its goals for the output gap,
relative to its goals for inflation.
The central bank’s dynamic optimization problem may now be stated formally as: choose
sequences for the output gap {yt }∞ ∞
t=0 and the inflation rate {π t }t=1 to maximize the additively
time-separable objective function
X
∞
β t [−π 2t − αyt2 ],
t=0

with discount factor β, 1 > β > 0, subject to the constraints

π 0 given
and
π t+1 = π t + ψyt
for all t = 0, 1, 2, ....
a) Write down the Bellman equation for the dynamic programming formulation of the
central bank’s problem. In doing so, it may help to view π t as the state variable and
yt as the control variable.
b) Now guess that the value function takes the time-invariant form
v(π t ) = −γπ 2t ,
where γ > 0 is an unknown constant. Using this guess, write down the first-order
condition for yt and the envelope condition for πt .
c) It will take too much algebra to solve for the unknown constant γ, so DON’T BOTHER
TO DO THIS. Instead, just use your first-order condition from above to derive an
equation that shows how the central bank’s optimal choice for yt depends on the infla-
tion rate π t , on the unknown constant γ, and on the model’s three parameters: β, α,
and ψ.
d) Finally, use your results from above to answer the following question: given that γ > 0,
1 > β > 0, α > 0, and ψ > 0, does the central bank find it optimal to ”lean against the
wind,” that is, to choose a negative value for the output gap when inflation is above
its target of zero, or does the central bank find it optimal to ”lean with the wind,”
that is, to choose a positive value for the output gap when inflation is above its target
of zero?

5
 Solutions to Midterm Exam
Economics 720: Mathematics for Economists
Fall 2002

1 The Kuhn-Tucker Theorem and the Cobb-Douglas

Utility Function
a) Define the Lagrangian as

L(c1 , c2 , λ) = ca1 c1−a

2 + λ(I − p1 c1 − p2 c2 ). (1.1)

b) According to the Kuhn-Tucker theorem, the values c∗1 and c∗2 that solve the consumer’s
problem and the associated value λ∗ of the Lagrange multiplier must satisfy the first-
order condition for c1 ,

L1 (c∗1 , c∗2 , λ∗ ) = ac∗a−1

1 c∗1−a
2 − λ∗ p1 = 0, (1.2)

the first-order condition for c∗2 ,

L2 (c∗1 , c∗2 , λ∗ ) = (1 − a)c∗a ∗−a

1 c2 − λ∗ p2 = 0, (1.3)

the constraint,
L3 (c∗1 , c∗2 , λ∗ ) = I − p1 c∗1 − p2 c∗2 ≥ 0, (1.4)
the nonnegativity condition,
λ∗ ≥ 0, (1.5)
and the complementary slackness condition,

λ∗ (I − p1 c∗1 − p2 c∗2 ) = 0. (1.6)

c) Begin by dividing (1.2) by (1.3) to eliminate λ∗ :

ac∗2 p1
∗
= .
(1 − a)c1 P2
Now use this result to obtain an expression for c∗2 in terms of c∗1 and the model’s
parameters: µ ¶µ ¶
∗ 1−a p1 ∗
c2 = c.
a p2 1

1
 Substitute this expression for c∗2 into the binding constraint to obtain
µ ¶ µ ¶
∗ 1−a ∗ 1
I = p1 c1 + p1 c1 = p1 c∗1 .
a a
This last result yields the solution for c∗1 :
aI
c∗1 = . (1.7)
p1
Substitute (1.7) back into the expression for c∗2 to obtain the solution for c∗2 :
(1 − a)I
c∗2 = . (1.8)
p2

d) Equation (1.7) reveals that the consumer spends the fraction

p1 c∗1
=a
I
of his or her income on good 1 and the fraction
p2 c∗2
=1−a
I
on good 2. A property of the Cobb-Douglas utility function is that it implies that
these expenditure shares are constant, and equal to the exponents or weights attached
to each good in the utility function itself.

2 The Maximum Principle in Discrete Time

a) Define the Hamiltonian as
H(yt , π t+1 ; t) = max β t F (yt , zt ; t) + π t+1 Q(yt , zt ; t) subject to c ≥ G(yt , zt ; t). (2.1)
zt

b) By the Kuhn-Tucker theorem,

H(yt , πt+1 ; t) = max β t F (yt , zt ; t) + π t+1 Q(yt , zt ; t) + λt [c − G(yt , zt ; t)],
zt

where zt satisfies the first-order condition

β t Fz (yt , zt ; t) + πt+1 Qz (yt , zt ; t) − λt Gz (yt , zt ; t) = 0. (2.2)

c) According to the Maximum Principle, the solution to the original dynamic optimization
problem must satisfy the pair of difference equations
π t+1 −π t = −Hy (yt , π t+1 ; t) = −[β t Fy (yt , zt ; t)+πt+1 Qy (yt , zt ; t)−λt Gy (yt , zt ; t)] (2.3)
and
yt+1 − yt = Hπ (yt , π t+1 ; t) = Q(yt , zt ; t). (2.4)

2
d) According to the Maximum Principle, the solution to the original dynamic optimization
problem must also satisfy two boundary conditions: the initial condition

y0 given (2.5)

and the terminal, or transversality, condition

πT +1 (yT +1 − y ∗ ) = 0. (2.6)

3 The Maximum Principle in Continuous Time: Cus-

tomer Flows
a) Define the Hamiltonian as

H(n(t), π(t); t) = max e−ρt [n(t)p(t)1−θ − n(t)α p(t)−θα ] − π(t)φ[p(t) − q]n(t). (3.1)
p(t)

b) According to the Maximum Principle, the solution to the firm’s dynamic optimization
problem must satisfy the first-order condition for p(t) on the right-hand side of (3.1),

e−ρt [(1 − θ)n(t)p(t)−θ + θαn(t)α p(t)−θα−1 ] − π(t)φn(t) = 0 (3.2)

and the pair of differential equations

π̇(t) = −Hn (n(t), π(t); t) = −e−ρt [p(t)1−θ − αn(t)α−1 p(t)−θα ] + π(t)φ[p(t) − q] (3.3)

and
ṅ(t) = Hπ (n(t), π(t); t) = −φ[p(t) − q]n(t). (3.4)

c) Substitute the steady-state conditions into (3.2)-(3.4) to obtain

e−ρt [(1 − θ)q−θ + θαq −θα−1 ] − π(t)φ = 0, (3.5)

π̇(t) = −e−ρt [q1−θ − αq−θα ], (3.6)

and
ṅ(t) = 0. (3.7)

d) Equation (3.7) just repeats one of the original steady-state conditions: ṅ(t) = 0. The
problem with (3.5) and (3.6) is that both make reference to the function π(t). Moreover,
(3.5) refers to π(t), while (3.6) refers to π̇(t). In order to eliminate all reference to π(t),
begin by rewriting (3.5) as

π(t) = (1/φ)e−ρt [(1 − θ)q−θ + θαq −θα−1 ].

Differentiate both sides of this expression by t to obtain

π̇(t) = −(ρ/φ)e−ρt [(1 − θ)q−θ + θαq −θα−1 ].

3
 Now substitute this expression for π̇(t) into (3.6) to obtain

−(ρ/φ)e−ρt [(1 − θ)q −θ + θαq−θα−1 ] = −e−ρt [q1−θ − αq−θα ]

or, more simply,

(ρ/φ)[(1 − θ)q −θ + θαq−θα−1 ] = [q 1−θ − αq −θα ]. (3.8)

Equation (3.8) is the equation involving only the steady-state equilibrium price q and
the model’s parameters. Although it would take some additional algebra, this equation
could be used to solve for q in terms of θ, α, ρ, and φ.

4 Dynamic Programming Under Certainty: Optimal

Monetary Policy
a) The Bellman equation is

v(π t ; t) = max −π 2t − αyt2 + βv(π t + ψyt ; t + 1). (4.1)

yt

b) Using the guess for v, the Bellman equation becomes

−γπ 2t = max −π2t − αyt2 − βγ(π t + ψyt )2 .

yt

The first-order condition for yt is

−2αyt − 2βγψ(π t + ψyt ) = 0 (4.2)

and the envelope condition for π t is

−2γπt = −2πt − 2βγ(π t + ψyt ). (4.3)

c) Use the first-order condition to solve for the optimal choice for yt in terms of the inflation
rate π t , the unknown constant γ, and the model’s parameters:

αyt = −βγψπt − βγψ 2 yt

(α + βγψ 2 )yt = −βγψπt

or µ ¶
βγψ
yt = − πt. (4.4)
α + βγψ 2
d) Given that γ, β, α, and ψ are all positive, (4.4) reveals that the optimal value for
the output gap yt is negative when inflation is positive; conversely, the optimal yt is
positive when inflation is negative. The central bank finds it optimal to ”lean against
the wind.”

4
 Final Exam
Economics 720: Mathematics for Economists
Fall 2002

This exam has six questions on seven pages. Before you begin, please check to make sure
that your copy has all six questions and all seven pages.
Questions 1, 2, 4, and 5 are longer questions, worth 10 points each. Questions 3 and 6
are shorter questions, worth 5 points each.
Since this is a 180-minute exam worth 50 points in total, I suggest that you spend
approximately (180 minutes)/(50 points total)×(10 points per long question) = 36 minutes
on each long question and (180 minutes)/(50 points total)×(5 points per short question) =
18 minutes on each short question.

1 Dynamic Optimization: Optimal Growth (10 points)

Consider the following version of the optimal growth model. Time is discrete and the horizon
is infinite, so that time periods can be indexed by t = 0, 1, 2, .... During each period t, output
yt is produced with capital kt according to the production function
yt = f (kt ),
where f is strictly increasing and strictly concave. Let δ, 1 > δ > 0, denote the depreciation
rate for capital, and let ct denote the representative household’s consumption. Then the
capital stock evolves over time according to the constraint
kt+1 = kt + f (kt ) − δkt − ct
for all t = 0, 1, 2, ..., which says that capital available during period t + 1 equals capital
available during period t, plus output produced during period t, minus the amount of capital
that depreciates away during period t, minus the amount of output that is consumed during
period t. This constraint can be rewritten as
f (kt ) − δkt − ct ≥ kt+1 − kt
for all t = 0, 1, 2, ..., to allow for the possibility that capital may be freely disposed of; at the
optimum, however, this constraint will always bind.
The representative household has utility over the infinite horizon that is measured by the
additively time-separable utility function
X
∞
β t u(ct ),
t=0

1
where the discount factor β satisfies 1 > β > 0 and where the single-period utility function
u is strictly increasing and strictly concave.
The household’s problem can now be stated formally as: choose sequences {ct }∞ t=0 and
{kt }∞
t=1 to maximize the utility function

X
∞
β t u(ct ),
t=0

subject to the constraints

k0 given
and
f (kt ) − δkt − ct ≥ kt+1 − kt
for all t = 0, 1, 2, ....

a) Define the Lagrangian for the household’s problem. Then write down the first-order
conditions and constraints that, according to the Kuhn-Tucker theorem, characterize
the solution to the household’s problem.

b) Define the Hamiltonian for the household’s problem. Then write down the first-order
condition and difference equations that, according to the maximum principle, charac-
terize the solution to the household’s problem.

c) Write down the Bellman equation for the household’s problem. Then write down the
first-order condition, envelope condition, and constraint that, according to the dynamic
programming approach, characterize the solution to the household’s problem.

2 Stochastic Dynamic Programming: Saving with a

Serially Correlated, Random Return (10 points)
Consider the following dynamic, stochastic optimization problem in discrete time. An
infinitely-lived representative consumer enters each period t = 0, 1, 2, ... with a stock of
financial wealth denoted by At . During period t, the consumer divides this wealth up into
an amount ct to be consumed and an amount st to be saved, subject to the constraint

At = ct + st ,

which must hold for all t = 0, 1, 2, ....

The gross return Rt+1 on savings between t and t + 1 is random and, more specifically,
follows the first-order autoregressive process

ln(Rt+1 ) = ρ ln(Rt ) + εt+1

where 1 > ρ > 0 and εt+1 is a shock that satisfies Et εt+1 = 0 for all t = 0, 1, 2, .... Thus,
last period’s return Rt is known when the consumer chooses st , but Rt+1 is still viewed as

2
random. Given the consumer’s choice of st and the realized value of Rt+1 , the consumer’s
financial wealth At+1 is determined by the constraint

Rt+1 st ≥ At+1 ,

which must hold for all t = 0, 1, 2, ... and for all possible realizations of Rt+1 .
The representative consumer seeks to maximize expected utility
X
∞
E0 β t ln(ct ),
t=0

where the discount factor satisfies 1 > β > 0 and where, as indicated, the single-period
utility function takes the natural logarithmic form. As a first step in solving the consumer’s
problem, it is convenient to substitute the constraint At = ct + st into the utility function to
obtain
X
∞ X
∞
t
E0 β ln(ct ) = E0 β t ln(At − st ).
t=0 t=0

The consumer’s problem can now be stated formally as: choose contingency plans for st ,
t = 0, 1, 2, ..., and At , t = 1, 2, 3, ..., to maximize the expected utility function
X
∞
E0 β t ln(At − st )
t=0

subject to the constraints

A0 given
and
Rt+1 st ≥ At+1
for all t = 0, 1, 2, ... and all possible realizations of Rt+1 .

a) Write down the Bellman equation for the household’s problem. In doing so, you might
find it helpful to view At as the state variable, st as the control variable, and Rt
as the random variable entering into the problem. You might also find it helpful to
recognize that since the consumer’s utility function is strictly increasing, the constraint
Rt+1 st ≥ At+1 will always bind at the optimum.

b) Now guess that the value function for this problem takes the form

v(At , Rt ) = F ln(At ) + G ln(Rt ) + H,

where F , G, and H are unknown constants. Substitute this guess into your Bellman
equation from above, and then write down the first-order condition for st and the
envelope condition for At . In deriving these optimality conditions, you might find it
helpful to note that since st is known at time t,

Et ln(st ) = ln(st ).

3
 You might also find it helpful to note that since

ln(Rt+1 ) = ρ ln(Rt ) + εt+1 ,

with Et εt+1 = 0, it follows that

Et ln(Rt+1 ) = ρ ln(Rt ).

c) Use your first-order condition from above to derive an expression linking the consumer’s
optimal choice st to the state variable At , the discount factor β, and the unknown
constant F .

d) Next, use your envelope condition from above to solve for the unknown constant F in
terms of the discount factor β.

e) Combine your results from above to obtain a set of two equations that can be used to
construct the optimal paths for st and At , given the initial value of A0 and the realized
path for Rt+1 . NOTE: It should not be necessary to solve for the unknown constants
G and H to derive these equations; since solving for G and H requires a lot of extra
algebra, DON’T BOTHER TO DO THIS.

3 A Scalar Differential Equation: General Solution and

Phase Diagrams (5 points)
Consider the scalar differential equation

ẏ(t) = ay(t),

where ẏ(t) is the derivative of the unknown function y(t), and a is a non-zero constant.

a) Write down the general solution to this differential equation.

b) Next, find the unique stationary solution to the differential equation.

c) Suppose now that a < 0, and draw the phase diagram for this differential equation.
What does the phase diagram tell you about the asymptotic stability or instability of
the unique stationary solution?

d) Suppose instead that a > 0, and redraw the phase diagram for this alternative case.
What does the phase diagram tell you about the asymptotic stability or instability of
the unique stationary solution?

4
4 A System of Linear Differential Equations (10 points)
Consider the system of linear differential equations consisting of

ẋ1 (t) = −x1 (t) + 3x2 (t)

and
ẋ2 (t) = −2x1 (t) + 4x2 (t).
This system can be written more simply as

ẋ(t) = Ax(t),

where ∙ ¸
ẋ1 (t)
ẋ(t) = ,
ẋ2 (t)
∙ ¸
x1 (t)
x(t) = ,
x2 (t)
and ∙ ¸
−1 3
A= .
−2 4

a) As a first step in solving this system, find the two eigenvalues of the matrix A.

b) Next, find eigenvectors corresponding to the two eigenvalues that you identified above.

c) Use your results from above to construct a nonsingular matrix P and a diagonal matrix
D such that
P −1 AP = D.

d) Use your results from above to find the general solutions for x1 (t) and x2 (t).

e) Identify the unique stationary solution to the system. Use your results from above to
answer the question: is the stationary solution asymptotically stable or unstable?

f) Finally, use your results from above to find the unique functions x1 (t) and x2 (t) that
solve the system of differential equations together with the initial conditions

x1 (0) = 2 and x2 (0) = 1.

5 Linear Difference Equations: Employment and Un-

employment Dynamics (10 points)
An economy consists of two types of workers: those that are employed (in state 1) and
those that are unemployed (in state 2). Let x1t denote the fraction of the population that is

5
employed during period t, let x2t denote the fraction of the population that is unemployed
during period t, and let xt denote the vector
∙ ¸
x1t
xt = .
x2t

Suppose that each worker who is employed during period t faces a 10 percent chance
of becoming unemployed during period t + 1, while each worker who is unemployed during
period t has a 40 percent chance of finding a job during period t + 1. Then if the population
is large, the evolution of xt is governed by the system of difference equations

xt+1 = Mxt ,

where ∙ ¸
0.90 0.40
M= .
0.10 0.60

a) As a first step in analyzing this model, find the eigenvalues of the matrix M.

b) Next, find the eigenvectors corresponding to the eigenvectors that you identified above.

c) Use your results from above to obtain the general solution to the system of difference
equations xt+1 = Mxt .

d) Now find the particular solution to the system under the assumption that the population
initially consists of equal numbers of employed and unemployed workers, so that
∙ ¸
0.50
x0 = .
0.50

e) Use your results from above to show that as t becomes arbitrarily large, the economy will
converge from x0 to a steady state in which the fractions of employed and unemployed
workers are constant.

f) In the steady state, what fraction of the population is employed? What fraction of the
population is unemployed?

6 Nonlinear Difference Equations: Capital Accumula-

tion and Economic Growth (5 points)
Consider an economy in which time is discrete and the horizon infinite, so that periods can
be indexed by t = 0, 1, 2, .... During each period t = 0, 1, 2, ..., output yt is produced with
capital kt according to the production function

yt = Aktα ,

where, by assumption, the parameters A and α satisfy A > 0 and 1 > α > 0.

6
 Now make two simplifying assumptions. First, assume that capital depreciates completely
between periods. Second, assume that in this economy, agents save and invest a constant
fraction s, 1 > s > 0, of output. Under these two additional assumptions, the capital stock
evolves according to the nonlinear difference equation

kt+1 = sAktα

for all t = 0, 1, 2, .... That is, the capital stock during period t + 1 equals the amount saved
during period t, which in turn just equals the constant savings rate s times the amount of
output Aktα .

a) Find the stationary solutions to this nonlinear difference equation.

b) Now draw a diagram that illustrates the asymptotic stability or instability of these
stationary solutions.

7
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2002

1 Dynamic Optimization: Optimal Growth

a) The Lagrangian is

X
∞ X
∞
t
L= β u(ct ) + λt+1 [kt + f (kt ) − δkt − ct − kt+1 ].
t=0 t=0

The first-order condition for ct , t = 0, 1, 2, ..., is

β t u0 (ct ) − λt+1 = 0.

The first-order condition for kt , t = 1, 2, 3, ..., is

λt+1 [1 + f 0 (kt ) − δ] − λt = 0.

The binding constraint, t = 0, 1, 2, ..., is

kt + f (kt ) − δkt − ct − kt+1 = 0.

b) The Hamiltonian is

H(kt , πt+1 ; t) = max β t u(ct ) + π t+1 [f (kt ) − δkt − ct ].

ct

The first-order condition for ct is

β t u0 (ct ) − π t+1 = 0.

The difference equations for π t and kt are

π t+1 − π t = −Hk (kt , π t+1 ; t) = −πt+1 [f 0 (kt ) − δ]

and
kt+1 − kt = Hπ (kt , π t+1 ; t) = f (kt ) − δkt − ct .

1
c) The Bellman equation is

v(kt ; t) = max u(ct ) + βv[kt + f (kt ) − δkt − ct ; t + 1].

ct

The first-order condition for ct is

u0 (ct ) − βv 0 (kt+1 ; t + 1) = 0.

The envelope condition for kt is

v 0 (kt ; t) = β[1 + f 0 (kt ) − δ]v0 (kt+1 ; t + 1).

The binding constraint is

kt+1 = kt + f (kt ) − δkt − ct .

Notice that the equations that you derived in parts (a), (b), and (c) of this question
coincide, as of course they must, with

λt+1 = π t+1 = β t+1 v0 (kt+1 ; t + 1).

2 Stochastic Dynamic Programming: Saving with a

Serially Correlated Random Return
a) The Bellman equation is

v(At , Rt ) = max ln(At − st ) + βEt v(Rt+1 st , Rt+1 ).

st

b) Using the guess for the form of the value function, and the facts about st and Rt+1 , the
Bellman equation can be written

F ln(At ) + G ln(Rt ) + H = max ln(At − st ) + βF ln(st ) + β(F + G)ρ ln(Rt ) + βH.

et

The first-order condition for st is

1 βF
− + = 0,
At − st st
while the envelope condition for At is
F 1
= .
At At − st

c) The first-order condition implies that

µ ¶
βF
st = At .
1 + βF

2
d) The envelope condition now implies that
µ ¶
βF
F At − F At = At
1 + βF
or ∙ µ ¶¸
βF
F 1− =1
1 + βF
or
F = 1 + βF
or
1
F = .
1−β
e) Return to µ ¶
βF
st = At
1 + βF
and substitute in the solution for F to obtain

st = βAt .

Next, substitute this solution for st into the binding constraint to obtain

At+1 = Rt+1 st = βRt+1 At .

Given the initial condition A0 and a realized path for Rt+1 , these last two equations
can be used to construct the optimal paths for st , t = 0, 1, 2, ..., and At , t = 1, 2, 3, ....

3 A Scalar Differential Equation: General Solution and

Phase Diagrams
a) The general solution is
y(t) = keat ,
where different values of the parameter k correspond, for instance, to different initial
conditions y(0) = y0 = k.
b) Stationary solutions take the form
y(t) = y
and
ẏ(t) = 0,
where the constant y must satisfy
0 = ay.
Since a is, by assumption, nonzero, it is clear from this last equation that the unique
stationary solution is
y(t) = 0.

3
c) In this case, with a < 0, the phase diagram reveals that the stationary solution is
asymptotically stable.

d) In this case, with a > 0, the phase diagram reveals that the stationary solution is
unstable.

4 A System of Linear Differential Equations

a) By definition, an eigenvalue r of A must solve the characteristic equation

0 = det(A − rI)
µ∙ ¸ ∙ ¸¶
−1 3 r 0
= det −
−2 4 0 r
∙ ¸
−1 − r 3
= det
−2 4−r
= (−1 − r)(4 − r) + 6
= −4 − 4r + r + r2 + 6
= r2 − 3r + 2
= (r − 1)(r − 2).

Evidently, the two eigenvalues are r1 = 1 and r2 = 2.

b) Again by definition, the eigenvector v1 corresponding to the eigenvalue r1 = 1 must

satisfy
(A − r1 I)v1 = 0
or ∙ ¸∙ ¸ ∙ ¸
−2 3 v11 0
=
−2 3 v12 0
or
−2v11 + 3v12 = 0.
Evidently, v1 must be of the form
∙ ¸
v11
v1 = .
(2/3)v11

A particularly simple choice for v1 is therefore

∙ ¸
3
v1 = .
2

Similarly, the eigenvector v2 corresponding to the eigenvalue r2 = 2 must satisfy

(A − r2 I)v2 = 0

4
Phase diagrams for y’(t) = ay(t):

0
y
When a < 0, y’(t) < 0 if y(t) > 0 and y’(t) > 0 if y(t) < 0.
The stationary solution y(t) = 0 is asymptotically stable.

When a > 0, y’(t) > 0 if y(t) > 0 and y’(t) < 0 if y(t) < 0.
The stationary solution y(t) = 0 is unstable.

0 y
 or ∙ ¸∙ ¸ ∙ ¸
−3 3 v21 0
=
−2 2 v22 0
or
−3v21 + 3v22 = 0
and
−2v21 + 2v22 = 0
Evidently, v2 must be of the form
∙ ¸
v21
v2 = .
v21

A particularly simple choice for v2 is therefore

∙ ¸
1
v2 = .
1

c) Form the matrix P using the eigenvectors v1 and v2 as its columns:

∙ ¸
3 1
P = .
2 1

Form the matrix D using the eigenvalues r1 and r2 as its diagonal elements:
∙ ¸
1 0
D= .
0 2

Since the eigenvectors v1 and v2 are linearly independent, P is nonsingular and

P −1 AP = D.

d) For this system of linear differential equations, the general solution takes the form

xt = k1 er1 t v1 + k2 er2 t v2 .

Plugging in the specific values of r1 , r2 , v1 , and v2 yields

∙ ¸ ∙ ¸
t 3 2t 1
xt = k1 e + k2 e .
2 1

e) A stationary solution to the system takes the form

∙ ¸ ∙ ¸
x1 (t) x1
x(t) = = =x
x2 (t) x2

and ∙ ¸ ∙ ¸
x1 (t) 0
ẋ(t) = = = 0,
x2 (t) 0

5
 where the constants x1 and x2 must satisfy
0 = Ax.
Since A is nonsingular, the unique stationary solution must have x = 0, that is,
x1 (t) = 0 and x2 (t) = 0.
Referring back to the general solution
∙ ¸ ∙ ¸
t 3 2t 1
x(t) = k1 e + k2 e ,
2 1
we can see that even for small values of k1 and k2 , corresponding to initial conditions
x1 (0) and x2 (0) close to the stationary solution,
lim x(t) = ∞,
t→∞

implying that the stationary solution is unstable. This last result follows from the fact
that the eigenvalues r1 and r2 are both positive.
f) We already know that the general solution takes the form
∙ ¸ ∙ ¸
t 3 2t 1
x(t) = k1 e + k2 e .
2 1
If we are also given the initial conditions x1 (0) = 2 and x2 (0) = 1, then we must have
∙ ¸ ∙ ¸ ∙ ¸ ∙ ¸ ∙ ¸
2 0 3 0 1 3 1
= x(0) = k1 e + k2 e = k1 + k2 .
1 2 1 2 1
Evidently, the parameters k1 and k2 must satisfy
2 = 3k1 + k2
and
1 = 2k1 + k2 .
Solve the first of these two equations for k2 :
k2 = 2 − 3k1 .
Now substitute this result into the second equation and solve for k1 :
1 = 2k1 + 2 − 3k1
or
k1 = 1.
Finally, return to the solution for k2 to obtain
k2 = 2 − 3k1 = −1.
Evidently, the unique functions x1 (t) and x2 (t) that satisfy the system of differential
equations and the initial conditions are
∙ ¸ ∙ ¸ ∙ ¸
x1 (t) t 3 2t 1
=e −e .
x2 (t) 2 1

6
5 Linear Difference Equations: Employment and Un-
employment Dynamics
a) By definition, r is an eigenvalue of M if it solves the characteristic equation

0 = det(M − rI)
∙ ¸
0.90 − r 0.40
= det
0.10 0.60 − r
= (0.90 − r)(0.60 − r) − (0.40)(0.10)
= 0.54 − 0.90r − 0.60r + r2 − 0.04
= r2 − 1.5r + 0.5
= (r − 1)(r − 0.5).

Evidently, the two eigenvalues are r1 = 1 and r2 = 0.5.

b) Again by definition, the eigenvector v1 corresponding to the eigenvalue r1 = 1 must

satisfy ∙ ¸∙ ¸ ∙ ¸
0.90 − 1 0.40 v11 0
= ,
0.10 0.60 − 1 v12 0
which requires that
−0.10v11 + 0.40v12 = 0
and
0.10v11 − 0.40v12 = 0
or, more simply,
v12 = (1/4)v11 .
Evidently, v1 must be of the form
∙ ¸
v11
v1 = .
(1/4)v12

A simple vector of this form is ∙ ¸

4
v1 = .
1
Similarly, the eigenvector v2 corresponding to the eigenvalue r2 = 0.5 must satisfy
∙ ¸∙ ¸ ∙ ¸
0.90 − 0.50 0.40 v21 0
= ,
0.10 0.60 − 0.50 v22 0

which requires that

0.40v21 + 0.40v22 = 0
and
0.10v21 + 0.10v22 = 0

7
 or, more simply,
v22 = −v21 .
Evidently, v2 must be of the form
∙ ¸
v21
v2 = .
−v22
A simple vector of this form is ∙ ¸
1
v2 = .
−1
c) The general solution to this system of difference equations takes the form
xt = k1 r1t v1 + k2 rt2 v2 ,
where r1 and r2 are the eigenvalues of M and v1 and v2 are the corresponding eigenvec-
tors. Since r1t = 1 for all t = 0, 1, 2, ..., the general solution to the difference equation
for xt is therefore ∙ ¸ ∙ ¸
4 t 1
xt = k1 + k2 (0.5) .
1 −1
d) Given the initial condition ¸
∙
0.50
x0 = ,
0.50
the general solution from above requires that
∙ ¸ ∙ ¸ ∙ ¸
0.50 4 1
= k1 + k2
0.50 1 −1
or
0.50 = 4k1 + k2
and
0.50 = k1 − k2 .
Use the second of these two equations to solve for k2 in terms of k1 :
k2 = k1 − 0.50.
Now substitute this result into the first equation, and solve for k1 :
0.50 = 4k1 + k2 = 4k1 + k1 − 0.50
1 = 5k1
k1 = 0.20.
Finally, substitute this solution for k1 back into the equation for k2 to obtain
k2 = k1 − 0.50 = 0.20 − 0.50 = −0.30.
Thus, starting from the initial condition given above, the evolution of xt is determined
by the particular solution
∙ ¸ ∙ ¸
4 t 1
xt = 0.20 − (0.30)(0.5) .
1 −1

8
e) Since
lim (0.5)t = 0,
t→∞

the particular solution from above implies that starting from x0 ,

∙ ¸ ∙ ¸
4 0.80
lim xt = 0.20 = .
t→∞ 1 0.20

Evidently, the economy converges to a steady state, in which the fractions of agents
employed and unemployed are constant.

f) The last result from above indicates that in the steady state, 80 percent of all workers
are employed and 20 percent of all workers are unemployed.

6 Nonlinear Difference Equations: Capital Accumula-

tion and Economic Growth
a) A stationary solution must have

kt = k and kt+1 = k,

where the constant k satisfies

k = sAk α .
Evidently, there are two stationary solutions, one with

kt = 0

for all t = 0, 1, 2, ..., and the other with

k = (sA)1/(1−α) > 0

for all t = 0, 1, 2, ....

b) A diagram with kt+1 on the y-axis and kt on the x-axis reveals that the stationary
solution kt = 0 is unstable, whereas the stationary solution kt = (sA)1/(1−α) > 0 is
asymptotically stable. These results follow mainly from the fact that the expression
on the right-hand side of the difference equation

kt+1 = sAktα

is strictly increasing and strictly concave when viewed as a function of kt , given the
assumptions that 1 > s > 0, A > 0, and 1 > α > 0.

9
kt+1 kt+1 = kt

kt+1 = sAktα

k2

k1

0 k0 k1 k2 (sA)1/(1-α) kt
 Final Exam
Economics 720: Mathematics for Economists
Fall 2003

This exam has five questions on seven pages; before you begin, please check to make sure
that your copy has all five questions and all seven pages. The five questions will receive equal
weight in determining your overall exam score. Since this is a two-hour exam, I suggest that
you spend approximately (120 minutes)/(5 questions) = 24 minutes on each question.

1 Applying the Kuhn-Tucker Theorem: The Linear

Expenditure System
This question asks you to derive a version of the linear expenditure system by applying the
Kuhn-Tucker theorem to a consumer’s static optimization problem.
The consumer likes two goods: c1 and c2 . The consumer’s utility function takes the
Stone-Geary form
U(c1 , c2 ) = (c1 − x1 )α (c2 − x2 )1−α ,
where x1 > 0 and x2 > 0 are positive parameters that measure the subsistence levels of
the two goods and where α is parameter that lies between zero and one: 0 < α < 1. The
consumer seeks to maximize this utility function subject to the budget constraint

I ≥ p1 c1 + p2 c2 ,

where I denotes the consumer’s income and p1 and p2 are the prices of the two goods.
To insure that this problem has a well-defined solution, it is necessary to assume that the
consumer’s income I is large enough to allow him or her to purchase and consume amounts
of each good that exceed the subsistence levels; this extra assumption can be written as

I > p1 x1 + p2 x2 .

The consumer’s problem can now be stated as

max(c1 − x1 )α (c2 − x2 )1−α subject to I ≥ p1 c1 + p2 c2 .

c1 ,c2

a) As a first step in applying the Kuhn-Tucker theorem, define the Lagrangian for this
problem.

1
 b) Now write down the complete set conditions that, according to the Kuhn-Tucker theorem,
must be satisfied by the values c∗1 and c∗2 that solve the consumer’s problem together
with the associated value of the Lagrange multiplier.

c) Use your results from part (b) above to obtain a mathematical expression that restates
the familiar graphical condition: the optimizing consumer acts so as to equate the
marginal rate of substitution between the two goods (the slope of the indifference
curve) to the ratio of their prices (the slope of the budget constraint).

d) Finally, use your results from parts (b) and (c) above to show how the consumer’s optimal
expenditures p1 c∗1 and p2 c∗2 on each good depend on the model’s parameters: x1 , x2 , α,
I, p1 , and p2 . HINT: There are a variety of ways to do this, but perhaps the easiest
is to start with your result from part (c) above, and use it to obtain an expression
linking optimal expenditures p2 c∗2 on good two to optimal expenditures p1 c∗1 on good
one. Then use this result, together with the budget constraint (which binds at the
optimum), to solve for p1 c∗1 in terms of the model’s parameters. Finally, substitute this
solution for p1 c∗1 back into your expression for p2 c∗2 to solve for p2 c∗2 in terms of the
model’s parameters.

e) Using your results from part (d) above, answer the following three questions.

i) Does the optimal p1 c∗1 rise, fall, or remain unchanged when income I rises?
ii) Does the optimal p1 c∗1 rise, fall, or remain unchanged when the price p1 of good
one rises?
iii) Does the optimal p1 c∗1 rise, fall, or remain unchanged when the price p2 of good
two rises?

2 The Maximum Principle in Discrete Time: Life Cy-

cle Saving and Borrowing Constraints
This problem asks you to use the maximum principle to study two versions of a model of
life cycle saving: one that allows the consumer to borrow and the other that imposes an
additional constraint that prevents the consumer from borrowing.
Consider, therefore, a consumer who is employed for periods t = 0, 1, ..., T . During each
period of employment, the consumer receives constant labor income w. Let kt denote the
consumer’s stock of assets at the beginning of period t, and assume that k0 = 0, so that the
consumer begins his or her career with no assets.
In the first version of this model, the consumer will be allowed to borrow by choosing
negative values for kt for any t = 1, 2, ..., T . However, the consumer must eventually save for
retirement: kT +1 must satisfy
kT +1 ≥ k∗ > 0,
where k ∗ denotes the amount of saving needed for retirement.

2
 If r denotes the constant interest rate, then the consumer’s stock of assets evolves ac-
cording to
w + rkt − ct ≥ kt+1 − kt
for all t = 0, 1, ..., T , where ct denotes consumption at each date t = 0, 1, ..., T . The con-
sumer’s utility function is
XT
β t ln(ct ),
t=0

where the discount factor lies between zero and one: 0 < β < 1.

a) In this first version of the model, where borrowing is permitted, the consumer chooses
sequences {ct }Tt=0 and {kt }Tt=1
+1
to maximize the utility function

X
T
β t ln(ct ),
t=0

subject to the constraints

w + rkt − ct ≥ kt+1 − kt
for all t = 0, 1, ..., T,
k0 = 0,
and
kT +1 ≥ k∗ .
As a first step in using the maximum principle, define the Hamiltonian corresponding
to this problem.

b) Now write down the first-order condition for ct and the pair of difference equations for
π t and kt that, according to the maximum principle, characterize the solution to the
consumer’s dynamic optimization problem.

c) Use your results from above to derive an expression for the optimal growth rate of
consumption ct /ct−1 in terms of the model’s parameters.

d) Next, consider a second version of this model, in which the consumer is prevented from
borrowing through the imposition of the additional constraints

kt+1 ≥ 0

for all t = 0, 1, ..., T , which require the stock of assets to always be nonnegative. Note
that since the constraint governing the evolution of kt will always bind at the optimum,
these borrowing constraints can conveniently be rewritten as

kt + w + rkt − ct ≥ 0

3
 for all t = 0, 1, ..., T . Thus, in this second version of the model, the consumer chooses
sequences {ct }Tt=0 and {kt }Tt=1+1
to maximize the utility function

X
T
β t ln(ct ),
t=0

subject to the constraints

w + rkt − ct ≥ kt+1 − kt
for all t = 0, 1, ..., T,
kt + w + rkt − ct ≥ 0
for all t = 0, 1, .., T ,
k0 = 0,
and
kT +1 ≥ k∗ .
As a first step in using the maximum principle to analyze this second version of the
model, define the Hamiltonian corresponding to the consumer’s problem.

e) Now write down the first-order condition for ct and the pair of difference equations for
π t and kt that, according to the maximum principle, characterize the solution to the
consumer’s dynamic optimization problem.

f) Finally, use your results from above to derive a lower bound on the optimal growth
rate of consumption ct /ct−1 that depends on the model’s parameters. What impact
does the imposition of the borrowing constraint have on the optimal growth rate of
consumption?

3 The Maximum Principle in Continuous Time: Opti-

mal Monetary Policy
This problem asks you to use the maximum principle to characterize optimal monetary policy
in an economy where there is Phillips-curve trade-off between inflation and unemployment.
More specifically, the Phillips curve in this economy is described by the equation

ṗ(t) = −k[u(t) − ū],

where p(t) denotes the inflation rate, u(t) denotes the unemployment rate, k > 0 measures
the slope of the Phillips curve, and ū > 0 is the “natural rate of unemployment.” Hence,
this equation indicates that inflation falls when the unemployment rate is above the natural
rate and inflation rises when the unemployment rate is below the natural rate.
The central bank in this economy attempts to maximize a social welfare function that
penalizes deviations of the unemployment rate u(t) from the natural rate ū and deviations

4
of the inflation rate p(t) from some target rate p̄ > 0. More specifically, over the infinite
horizon, the central bank attempts to maximize
Z ∞
−(1/2) e−ρt {[u(t) − ū]2 + [p(t) − p̄]2 }dt,
0

where ρ > 0 measures the discount rate. Note that the minus sign out in front of this ob-
jective function is essential to capture the idea that deviations of unemployment or inflation
from the central bank’s targets ought to be penalized instead of rewarded. Meanwhile, the
1/2 out in front of the objective function is there to make the algebra a little easier.
Thus, the central bank in this economy chooses continuously differentiable functions u(t)
and p(t) for all t ∈ [0, ∞) to maximize the objective function
Z ∞
−(1/2) e−ρt {[u(t) − ū]2 + [p(t) − p̄]2 }dt
0

subject to the constraints

p(0) given
and
ṗ(t) = −k[u(t) − ū]
for all t ∈ [0, ∞).

a) As a first step in applying the maximum principle, define the Hamiltonian for this prob-
lem. HINT: In this problem, where both positive and negative deviations of inflation
p(t) from the central bank’s target p̄ are penalized, it is necessary to view the con-
straint ṗ(t) = −k[u(t) − ū] as an equality constraint. Nevertheless, if you define the
Hamiltonian as if the constraint could be written as

−k[u(t) − ū] ≥ ṗ(t),

you’ll be led to the correct set of optimality conditions below.

b) Next, write down the first-order condition for u(t) and the pair of differential equations
for π(t) and p(t) that, according to the maximum principle, characterize the solution
to the central bank’s problem.
c) Your results from part (b) above form a system of three equations in the three unknowns
u(t), p(t), and π(t). The first two of these objects have straightforward economic in-
terpretations: u(t) measures the unemployment rate during period t and p(t) measures
the inflation rate during period t. But the third unknown, π(t), was introduced only
in defining the Hamiltonian and lacks a completely straightforward economic interpre-
tation. With these ideas in mind, use your results from above to reduce the system
of optimality conditions down to a pair of just two differential equations involving the
two economic variables u(t) and p(t).
d) Now use your results from part (c) above to draw a phase diagram, with inflation p on
the x-axis and unemployment u on the y-axis, that illustrates the qualitative properties
of the solution to the central bank’s dynamic optimization problem.

5
 e) Finally, use the phase diagram to answer the following questions:

i) What are the steady-state values of inflation and unemployment in this model?
ii) Starting from an initial condition p(0) for inflation that is above the steady state, is
it optimal for the central bank to reduce inflation down to its steady-state value?
If so, does the central bank accomplish this goal by raising unemployment above
the natural rate or by pushing unemployment below its natural rate?
iii) Starting from an initial condition p(0) for inflation that is below the steady state,
is it optimal for the central bank to increase inflation up to its steady-state value?
If so, does the central bank accomplish this goal by raising unemployment above
the natural rate or by pushing unemployment below its natural rate?

4 Dynamic Programming Under Certainty: Cake Eat-

ing
This question asks you to use dynamic programming to characterize the optimal strategy
for eating a cake that spoils, or depreciates away, over time.
Time is discrete and the horizon is infinite, so that periods are indexed by t = 0, 1, 2, ....
Let kt denote the fraction of the cake that remains uneaten at the beginning of each period
t = 0, 1, 2, ... and let ct denote the fraction of the cake that is eaten during period t. Assume,
too, that the fraction δ, 0 < δ < 1, of the remaining cake spoils or depreciates away during
each period t = 0, 1, 2, .... Then
k0 = 1
(the consumer starts off with one full cake) and

(1 − δ)kt − ct ≥ kt+1

for all t = 0, 1, 2, .... Subject to these constraints, the consumer chooses sequences {ct }∞
t=0
and {kt }∞t=1 to maximize the utility function

X
∞
β t ln(ct ),
t=0

with constant discount factor 0 < β < 1.

a) Write down the Bellman equation for this problem.

b) Now guess that the value function takes the time-invariant form

v(kt ) = E + F ln(kt ),

where E and F are unknown constants. Using this guess, derive the first-order condi-
tion for the control variable ct and the envelope condition for the state variable kt .

6
 c) Use your results from part (b) above to show how the optimal choice of how much cake
to eat during period t, ct , depends on the amount of cake remaining at the beginning
of period t, kt , and on the parameters δ and β.
d) Finally, use your results from above to show how kt+1 , the amount of cake left at the
beginning of period t + 1, depends on kt , the amount of cake left at the beginning of
period t, as well as the parameters δ and β.

5 Stochastic Dynamic Programming: Optimal Growth

This problem asks you to use dynamic programming to solve a stochastic version of the
optimal growth model, in which an explicit solution for the value function can be found.
Suppose that the social planner or representative consumer maximizes the expected utility
function
X∞
E0 β t ln(ct )
t=0
subject to the constraints k0 given and

zt ktα ≥ ct + kt+1

for all t = 0, 1, 2, ..., where ct denotes consumption during period t, kt denotes the capital
stock at the beginning of period t, and zt represents a shock to the productivity of capital.
The value of zt is known when ct and kt+1 are chosen during period t, but the value of zt+1
is random and satisfies Et ln(zt+1 ) = 0. The model’s parameters α and β both lie between
zero and one: 0 < α < 1 and 0 < β < 1.

a) Write down the Bellman equation for this problem.

b) Now guess that the value function takes the form

v(kt , zt ) = E + F ln(kt ) + G ln(zt ),

where E, F , and G are unknown constants. Using this guess, derive the first order
condition for the control variable ct and the envelope condition for the state variable kt .
When you do this, you may find it helpful to recall that, by assumption, Et ln(zt+1 ) = 0.
You might also find it helpful to notice that since kt+1 is known during period t,

Et ln(kt+1 ) = Et ln(zt ktα − ct ) = ln(zt ktα − ct ) = ln(kt+1 ).

c) Use your results from part (b) above to show how the optimal choice of ct depends on the
capital stock kt and the productivity shock zt at t as well as the model’s parameters α
and β.
d) Finally, use your results from above to show how the capital stock kt+1 at t+1 depends on
the capital stock kt and the productivity shock zt at t as well as the model’s parameters
α and β.

7
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2003

1 Applying the Kuhn-Tucker Theorem: The Linear

Expenditure System
a) The Lagragian for the consumer’s problem is defined as

L(c1 , c2 , λ) = (c1 − x1 )α (c2 − x2 )1−α + λ(I − p1 c1 − p2 c2 ).

b) According to the Kuhn-Tucker theorem, if c∗1 and c∗2 are the values of c1 and c2 that
solve the consumer’s problem, then there exists a value λ∗ of the Lagrange multiplier
λ such that the following conditions are satisfied: the first-order conditions

L1 (c∗1 , c∗2 , λ∗ ) = α(c∗1 − x1 )α−1 (c∗2 − x2 )1−α − λ∗ p1 = 0

for c∗1 and

L2 (c∗1 , c∗2 , λ∗ ) = (1 − α)(c∗1 − x1 )α (c∗2 − x2 )−α − λ∗ p2 = 0
for c∗2 , the constraint

L3 (c∗1 , c∗2 , λ∗ ) = I − p1 c∗1 − p2 c∗2 ≥ 0,

the nonnegativity condition

λ∗ ≥ 0,
and the complementary slackness condition

λ∗ (I − p1 c∗1 − p2 c∗2 ) = 0.

c) Divide the first-order condition for c∗1 by the first-order condition for c∗2 to obtain

α(c∗1 − x1 )α−1 (c∗2 − x2 )1−α p1

∗ α ∗ −α
= ,
(1 − α)(c1 − x1 ) (c2 − x2 ) p2

which restates the familiar graphical condition that says that the optimizing consumer
acts so as to equate the marginal rate of substitution between the two goods to the
ratio of their prices.

1
d) To solve for expenditures on each good, start with the optimality condition from part
(c) above, which can be simplified to read
α(c∗2 − x2 ) p1
∗
= ,
(1 − α)(c1 − x1 ) P2
and use to it obtain the expression
µ ¶ µ ¶
∗ 1−α ∗ 1−α
p2 c2 = p1 c1 − x1 p1 + x2 p2 .
α α
Next, substitute this expression into the budget constraint, which is binding at the
optimum, to solve for optimal expenditures on good one:

p1 c∗1 = αI + (1 − α)x1 p1 − αx2 p2 .

And finally, substitute this solution for p1 c∗1 back into the previous expression for p2 c∗2
to solve for optimal expenditures on good two:

p2 c∗2 = (1 − α)I − (1 − α)x1 p1 + αx2 p2 .

e) The solutions for the optimal expenditures reveal that:

i) Expenditures p1 c∗1 on good one rise when income I rises.

ii) Expenditures p1 c∗1 on good one rise when the price p1 of good one rises.
iii) Expenditures p1 c∗1 on good one fall when the price p2 of good two rises.

2 The Maximum Principle in Discrete Time: Life Cy-

cle Saving and Borrowing Constraints
a) The Hamiltonian for the first problem, where borrowing is permitted, is given by

H(kt , π t+1 ; t) = max β t ln(ct ) + π t+1 (w + rkt − ct ).

ct

b) According to the maximum principle, the solution to the consumer’s problem must satisfy
the first-order condition
βt
− π t+1 = 0
ct
for all t = 0, 1, ..., T and the pair of difference equations

π t+1 − π t = −Hk (kt , π t+1 ; t) = −π t+1 r

for all t = 1, 2, ..., T and

kt+1 − kt = Hπ (kt , π t+1 ; t) = w + rkt − ct

for all t = 0, 1, ..., T .

2
c) Rewrite the difference equation for πt as

(1 + r)π t+1 = πt .

Next, consider the first-order condition for ct , which implies that

βt
π t+1 =
ct
and
β t−1
πt =
.
ct−1
Combining these three results yields the expression for the optimal growth rate of
consumption:
ct /ct−1 = β(1 + r).

d) The Hamiltonian for the second problem, where borrowing is not permitted, is given by

H(kt , π t+1 ; t) = max β t ln(ct ) + π t+1 (w + rkt − ct ) subject to kt + w + rkt − ct ≥ 0.

ct

e) By the Kuhn-Tucker theorem,

H(kt , π t+1 ; t) = max β t ln(ct ) + π t+1 (w + rkt − ct ) + λt (kt + w + rkt − ct ),

ct

where λt denotes the Lagrange multiplier on the borrowing constraint for period t.
Thus, according to the maximum principle, the solution to the consumer’s problem
must satisfy the first-order condition

βt
− π t+1 − λt = 0
ct
for all t = 0, 1, ...T and the pair of difference equations

π t+1 − π t = −Hk (kt , πt+1 ; t) = −π t+1 r − λt (1 + r)

for all t = 1, 2, ..., T and

kt+1 − kt = Hπ (kt , π t+1 ; t) = w + rkt − ct

for all t = 0, 1, ..., T .

f) Rewrite the difference equation for π t as

(1 + r)(π t+1 + λt ) = π t .

Next, consider the first-order condition for ct , which implies that

βt
= π t+1 + λt
ct

3
 and
β t−1
= πt + λt−1 ≥ π t ,
ct−1
where the last inequality in the expression for β t−1 /ct−1 follows from the Kuhn-Tucker
condition the requires λt−1 to be nonnegative. Combining these results yields
µ t¶
β β t−1
(1 + r) ≤ ,
ct ct−1
which places a lower bound on the optimal growth rate for consumption:

ct /ct−1 ≥ β(1 + r).

Evidently, the imposition of the borrowing constraint tends to increase the optimal
growth rate of consumption. This makes sense: since borrowing allows the consumer
to choose a higher value of ct−1 in exchange for a lower value of ct , borrowing reduces the
growth rate of consumption. Equivalently, the imposition of the borrowing constraint
tends to increase the growth rate of consumption.

3 The Maximum Principle in Continuous Time: Opti-

mal Monetary Policy
a) The Hamiltonian for the central bank’s problem is given by

H(p(t), π(t); t) = max −(1/2)e−ρt {[u(t) − ū]2 + [p(t) − p̄]2 } − π(t)k[u(t) − ū].
u(t)

b) According to the maximum principle, the solution to the central bank’s problem must
satisfy the first-order condition

−e−ρt [u(t) − ū] − π(t)k = 0

and the pair of differential equations

π̇(t) = −Hp (p(t), π(t); t) = e−ρt [p(t) − p̄]

and
ṗ(t) = Hπ (p(t), π(t); t) = −k[u(t) − ū]
for all t ∈ [0, ∞).

c) Rewrite the first-order condition as

π(t) = −(1/k)e−ρt [u(t) − ū]

and differentiate both sides with respect to t to obtain

π̇(t) = ρ(1/k)e−ρt [u(t) − ū] − (1/k)e−ρt u̇(t).

4
 Next, substitute this expression into the differential equation for π(t) to obtain

ρ(1/k)e−ρt [u(t) − ū] − (1/k)e−ρt u̇(t) = e−ρt [p(t) − p̄]

or, more simply,

u̇(t) = ρ[u(t) − ū] − k[p(t) − p̄].
Combine this last equation with

ṗ(t) = −k[u(t) − ū]

to obtain a system of two differential equations involving u(t) and p(t).

d) First, consider the differential equation for p(t), which implies that:

ṗ(t) = 0 if u(t) = ū
ṗ(t) < 0 if u(t) > ū
ṗ(t) > 0 if u(t) < ū

Next, consider the differential equation for u(t), which implies that:

u̇(t) = 0 if u(t) = ū + (k/ρ)[p(t) − p̄]

u̇(t) > 0 if u(t) > ū + (k/ρ)[p(t) − p̄]
u̇(t) < 0 if u(t) < ū + (k/ρ)[p(t) − p̄]

These conditions can be illustrated on a phase diagram.

e) The phase diagram reveals that:

i) The steady-state rate of inflation is p̄ and the steady-state rate of unemployment

is ū.
ii) Starting from an initial inflation rate p(0) that is above the steady steady state,
it is optimal for the central bank to lower inflation to p̄ by choosing values for
unemployment that are above the natural rate.
iii) Starting from an initial inflation rate p(0) that is below the steady state, it is
optimal for the central bank to raise inflation to p̄ by choosing values for unem-
ployment that are below the natural rate.

4 Dynamic Programming Under Certainty: Cake Eat-

ing
a) The Bellman equation for the consumer’s problem is

v(kt ; t) = max ln(ct ) + βv[(1 − δ)kt − ct ; t + 1].

ct

5
 u

u’ = 0 locus
u = ū + (k/ρ)(p − p-bar)

steady state at (p-bar,ū)

ū
p’ = 0 locus
(u = ū)

stable manifold
ū − (k/ρ)(p-bar)

p-bar p
b) Using the conjectured form for the value function, the Bellman equation becomes

E + F ln(kt ) = max ln(ct ) + βE + βF ln[(1 − δ)kt − ct ].

ct

The first-order condition for ct is

1 βF
− = 0,
ct (1 − δ)kt − ct
and the envelope condition for kt is
F β(1 − δ)F
= .
kt (1 − δ)kt − ct

c) The first-order condition for ct implies that

(1 − δ)kt − ct = βF ct

or µ ¶
1−δ
ct = kt .
1 + βF
Substitute this result into the envelope condition for kt to obtain
µ ¶
1−δ
F (1 − δ)kt − F kt = β(1 − δ)F kt
1 + βF
or, more simply,
1
= 1 − β.
1 + βF
And substitute this result back into the previous expression for ct to obtain

ct = (1 − δ)(1 − β)kt ,

which shows how the optimal choice of ct depends on kt , δ, and β.

d) Substitute the solution for ct into the binding constraint

kt+1 = (1 − δ)kt − ct

to obtain
kt+1 = (1 − δ)βkt ,
which shows how the optimal choice of kt+1 depends on kt , δ, and β. In fact, given the
initial condition k0 = 1, the optimal sequences {ct }∞ ∞
t=0 and {kt }t=1 can be constructed
using
ct = (1 − δ)(1 − β)kt
and
kt+1 = (1 − δ)βkt
for all t = 0, 1, 2, ....

6
5 Stochastic Dynamic Programming: Optimal Growth
a) The Bellman equation for the social planner’s problem is

v(kt , zt ) = max ln(ct ) + βEt v(zt ktα − ct , zt+1 ).

ct

b) Using the conjectured form for the value function and the facts that Et ln(kt+1 ) = ln(kt+1 )
and Et ln(zt+1 ) = 0, the Bellman equation becomes

E + F ln(kt ) + G ln(zt ) = max ln(ct ) + βE + βF ln(zt ktα − ct ).

ct

The first-order condition for ct is

1 βF
− = 0,
ct zt ktα − ct
and the envelope condition for kt is

F αβF zt ktα−1
= .
kt zt ktα − ct

c) The first-order condition implies that

µ ¶
1
ct = zt ktα .
1 + βF

Substitute this expression into the envelope condition to obtain

1
= 1 − αβ.
1 + βF
And substitute this result back into the previous expression for ct to obtain

ct = (1 − αβ)zt ktα ,

which reveals that it is optimal for consumption ct to be a fixed fraction 1 − αβ of

output zt ktα .

d) Substitute the solution for ct into the binding constraint to obtain

kt+1 = αβzt ktα ,

which shows how the optimal choice for kt+1 depends on kt , zt , α, and β.

7
 Final Exam
Economics 720: Mathematics for Economists
Fall 2005

This exam has six questions on seven pages; before you begin, please check to make sure
that your copy has all six questions and all seven pages. The six questions will receive equal
weight in determining your overall exam score. Since this is a two-hour exam, I suggest
that you spend approximately (120 minutes)/(6 questions) = 20 minutes on each question.

1 Quasi-Linear Preferences
This question asks you to apply the Kuhn-Tucker theorem to characterize the solution to a
consumer’s maximization problem where the utility function takes the special quasi-linear
form. Suppose, in particular, that there are two goods, x and y, which sell at prices p and
q. The consumer seeks to maximize his or her utility, given by

U(x, y) = ln(x) + y

subject to the budget constraint

I ≥ px + qy,
where I denotes the consumer’s income. Also, the consumer’s choices of x and y must be
nonnegative, as dictated by the nonnegativity constraints

x≥0

and
y ≥ 0.

a) Define the Lagrangian for this consumer’s problem: choose x and y to maximize utility,
subject to the budget and nonnegativity constraints. As we discussed in class, there are
a number of different ways to incorporate nonnegativity constraints into the definition
of the Lagrangian, so you can choose whichever way is easiest for you: all will eventually
lead you to the same results provided you write the associated optimality conditions
in the appropriate way in part (b) below.
b) Let x∗ and y ∗ denote the values of x and y that solve this problem. Write down the full
set of first-order conditions, constraints, nonnegativity conditions, and complementary
slackness conditions that, according to the Kuhn-Tucker theorem, must be satisfied by
x∗ , y ∗ , and the associated values of the Lagrange multipliers.

1
 c) In order to more sharply characterize the solution to the consumer’s problem, it is helpful
to make the following observations. First, since the utility function is strictly increasing
in both x and y, the budget constraint I ≥ px + qy will always bind at the optimum.
Second, the logarithmic form for utility from good x implies that the marginal utility
of consuming this good becomes infinite as the quantity consumed approaches zero.
Hence, the consumer will always choose a positive amount of good x and the constraint
x ≥ 0 will not bind at the optimum. Third, the linear form for utility from good y
implies that the marginal utility of consuming this good is constant and finite. It is
possible, therefore, for the constraint y ≥ 0 to bind at the optimum at least for certain
values of the parameters I, p, and q. In light of these observations, use your results
from part (b) above to answer: under what conditions on the parameters I, p, and q
will the consumer find it optimal to choose y ∗ = 0? What is the optimal value for x∗ ,
expressed in terms of the parameters I, p, and q, in this first case?
d) Similarly, use your results from part (b) to answer: under what conditions on the para-
meters I, p, and q will the consumer find it optimal to choose y ∗ > 0? What are the
optimal values for x∗ and y ∗ , expressed again in terms of the parameters I, p, and q,
in this second case?
e) Finally, use your results from parts (c) and (d) above to answer: which of the two goods,
x or y, might more appropriately be called a “luxury good,” in the sense that it is
purchased only by higher-income consumers?

2 Investment with Adjustment Costs

This question asks you to use the maximum principle to characterize the optimal investment
strategy of a firm that faces a quadratic cost of adjusting its capital stock. Suppose, in
particular, that time is continuous and the horizon infinite, so that periods can be indexed
by t ∈ [0, ∞). During each period t, the firm produces output Y (t) with capital K(t)
according to the production function
Y (t) = K(t)α ,
where 1 > α > 0. By investing I(t) units of output at t, the firm increases its capital stock
according to
I(t) ≥ K̇(t),
but also incurs a quadratic adjustment cost given by
(φ/2)[I(t)]2 ,
where φ > 0. In this problem, the firm’s choice of I(t) can be positive, in which case the
firm is installing new capital, or negative, in which case the firm is selling off existing capital;
but either way, the formulation implies that it will incur adjustment costs in making these
changes. The firm sells off whatever output remains after its investment choice is made; its
profits during period t are therefore
K(t)α − I(t) − (φ/2)[I(t)]2 .

2
Finally, let r > 0 denote the constant discount rate. The firm’s problem can now be stated
as: choose continuously differentiable functions I(t) and K(t) for t ∈ [0, ∞) to maximize the
discounted value of profits,
Z ∞
e−rt {K(t)α − I(t) − (φ/2)[I(t)]2 },
0

subject to the capital accumulation constraint I(t) ≥ K̇(t), which must hold for all t ∈ [0, ∞),
taking the initial capital stock K(0) = K0 > 0 as given.

a) Define the Hamiltonian for the firm’s problem. Again as we discussed in class, there
are a number of ways to do this, but so long as you write down the corresponding
optimality conditions in the appropriate way, you’ll be led to the same solution below.

b) Write down the first order condition for the firm’s optimal choice of I(t) in each period t ∈
[0, ∞), together with the pair of differential equations that according to the maximum
principle help characterize the solution to the firm’s problem.

c) Your answers from part (b) above form a system of three equations in three unknowns.
Use one of these equations to eliminate one of the variables–the additional variable
that you introduced into the problem when defining the Hamiltonian–to rewrite the
system as one involving two equations in the two unknowns: investment I(t) and the
capital stock K(t).

d) Use your results from part (c) above to calculate the steady-state values of investment
and capital, that is, the constant values of I(t) = I ∗ and K(t) = K ∗ that prevail when
˙ = 0 and K̇(t) = 0.
I(t)

e) Finally, use your results from above to draw a phase diagram that illustrates the following
property of the solution to the firm’s problem: starting from any value of the initial
capital stock K0 , there is a unique value of I(0) such that starting from K(0) = K0 and
I(0), the firm’s investment I(t) and capital stock K(t) converge to their steady-state
values. In drawing this phase diagram, it may be helpful to note that while investment
I(t) may take on either positive or negative values, depending on whether the firm is
accumulating new capital or selling off existing capital, the capital stock K(t) must
always remain positive.

3 Optimal Resource Depletion

This problem asks you to use the maximum principle to derive a system of differential
equations that characterize the optimal consumption of an exhaustible resource and then to
use the guess-and-verify method to solve that system of differential equations. Once again,
let time be continuous and the horizon infinite, so that periods can be indexed by t ∈ [0, ∞).
Let x(t) denote the stock of an exhaustible resource (like oil or coal) that remains available
during period t, and let c(t) denote the amount of this resource that is consumed during

3
period t. Since no new units of the resource are ever created, the amount consumed simply
subtracts from the available stock according to
−c(t) ≥ ẋ(t)
for all t ∈ [0, ∞). The optimization problem then involves choosing continuously differen-
tiable functions c(t) and x(t) for t ∈ [0, ∞) to maximize utility from consuming the resource
over the infinite horizon, given by
Z ∞
e−ρt ln(c(t))dt,
0

where ρ > 0 is the discount rate, subject to the constraint −c(t) ≥ ẋ(t) for all t ∈ [0, ∞),
taking as given the level of the initial resource stock x(0) = x0 > 0.
a) Define the Hamiltonian for this problem, using π(t) to denote the multiplier correspond-
ing to the constraint −c(t) ≥ ẋ(t) for period t.
b) Next, write down the first-order condition for c(t) and the pair of differential equations for
π̇(t) and ẋ(t) that, according to the maximum principle, help characterize the solution
to this problem.
c) Probably the easiest way to solve the pair of differential equations that you derived in
part (b) above is to use the guess-and-verify method. So guess that these differential
equations have general solutions of the form
π(t) = π
and µ
¶
1
x(t) = e−ρt + k,
πρ
where π and k are two constants that remain to be determined, then verify that these
two functions do in fact satisfy the differential equations that you derived in part (b)
above.
d) The maximum principle also gives us two boundary conditions that must be satisfied by
the functions π(t) and x(t): the initial condition
x(0) = x0 given
and the terminal, or transversality, condition
lim π(T )x(T ) = 0.
T →∞

Use these boundary conditions to find specific values for the two unknown constants
π and k that enter into the general solutions for π(t) and x(t) that you found in part
(c) above.
e) Finally, combine your specific solutions to the differential equations and boundary con-
ditions with the first-order condition for c(t) to obtain an expression that shows how
the optimal choice of c(t) depends on the initial resource stock x0 and the discount
rate ρ.

4
4 Saving Under Certainty
This problem asks you to use dynamic programming to characterize the optimizing behavior
of an infinitely-lived consumer who acts under conditions of perfect foresight or complete
certainty. Let time be discrete; since the horizon is infinite, time periods are indexed by
t = 0, 1, 2, .... Let At denote this consumer’s bank account balance at the beginning of each
period t = 0, 1, 2, .... The initial condition A0 is given, but for t = 0, 1, 2, ... the consumer
can choose negative values for At , that is, borrowing is permitted. During each period the
consumer receives a constant level of labor income w. Let ct denote consumption and st
denote savings; these variables are linked via the constraint
At + w ≥ ct + st
for all t = 0, 1, 2, .... Savings earn interest at the constant rate r, so that
(1 + r)st ≥ At+1
must also hold for all t = 0, 1, 2, .... The consumer’s utility function is
X
∞ X
∞
t
β u(ct ) = β t u(At + w − st ),
t=0 t=0

where the second expression for utility uses the fact that so long as u is strictly increasing,
the constraint linking ct and st will always bind (as will the constraint (1 + r)st ≥ At+1 ).
The consumer’s problem can now be stated as: choose sequences {st }∞ ∞
t=0 and {At }t=1 to
maximize
X∞
β t u(At + w − st )
t=0
subject to the constraints A0 given and (1 + r)st ≥ At+1 for all t = 0, 1, 2, ....
a) Write down the Bellman equation for this problem.
b) Next, write down the first-order condition for the control variable and the envelope
condition for the state variable.
c) A problem with your answers from part (b) above is that both make reference to an
unknown function: the derivative of the value function that you introduced into the
problem when setting up the Bellman equation. Combine your two equations from
part (b) above to obtain a single optimality condition that only involves these objects
with direct economic interpretations: consumption ct and ct+1 during periods t and
t + 1 and the parameters β and r. In doing this, you may find it helpful to also use
the binding constraints At + w = ct + st and (1 + r)st = At+1 .
d) Suppose that the consumer’s single-period utility function takes the logarithmic form,
u(ct ) = ln(ct ).
Under this additional assumption, use your result from part (c) above to obtain an ex-
pression for the optimal growth rate of consumption, ct+1 /ct , in terms of the parameters
β and r.

5
 e) Finally, use your result from part (d) above to answer the following question: will a more
patient consumer choose a faster or slower growth rate for consumption?

5 Optimal Stochastic Growth

This problem asks you to use dynamic programming to solve a stochastic version of the
optimal growth model, for which an explicit solution for the value function can be found.
In the model, time is discrete and the horizon is infinite, so that periods are indexed by
t = 0, 1, 2, .... Suppose that the representative consumer chooses contingency plans for
consumption ct for t = 0, 1, 2, ... and the capital stock kt for t = 1, 2, 3, ... to maximize the
expected utility function
X∞
E0 β t ln(ct )
t=0
subject to the constraints k0 given and
zt ktα ≥ ct + kt+1
for all t = 0, 1, 2, ..., where zt ktα measures output produced during period t, so that zt
represents a shock to the productivity of capital. The value of zt is known when ct and kt+1
are chosen during period t, but the value of zt+1 is random and satisfies Et ln(zt+1 ) = 0. The
model’s parameters α and β both lie between zero and one: 0 < α < 1 and 0 < β < 1.
a) Write down the Bellman equation for this problem.
b) Now guess that the value function takes the form
v(kt , zt ) = E + F ln(kt ) + G ln(zt ),
where E, F , and G are unknown constants. Using this guess, derive the first order
condition for the control variable ct and the envelope condition for the state variable kt .
When you do this, you may find it helpful to recall that, by assumption, Et ln(zt+1 ) = 0.
You might also find it helpful to notice that since kt+1 is known during period t,
Et ln(kt+1 ) = Et ln(zt ktα − ct ) = ln(zt ktα − ct ) = ln(kt+1 ).

c) Use your results from part (b) above to show how the optimal choice of ct depends on
the capital stock kt and the productivity shock zt as well as the model’s parameters α
and β.
d) Use your results from above to show how the capital stock kt+1 at t + 1 depends on the
capital stock kt and the productivity shock zt as well as the model’s parameters α and
β.
e) Suppose that, just by chance, the productivity shocks z0 and z1 during periods t = 0
and t = 1 both turn out to equal one: z0 = 1 and z1 = 1. Under these additional
assumptions, use your results from parts (c) and (d) above to derive expressions that
show how the optimal consumption choices for periods t = 0 and t = 1, c0 and c1 ,
depend on the initial capital stock k0 and the model’s parameters α and β.

6
6 Scalar Differential Equations
a) Write down the general solution to the scalar differential equation

ẏ(t) = −y(t).

b) Write down the unique solution to the initial value problem consisting of the same
differential equation
ẏ(t) = −y(t)
from part (a) above and the initial condition

y(0) = y0 given.

c) Find the stationary solutions to the scalar differential equation

ẏ(t) = y(t)[1 − y(t)2 ].

d) Now draw a phase diagram to illustrate the properties of all of the solutions to the same
differential equation
ẏ(t) = y(t)[1 − y(t)2 ]
from part (c) above.

e) Which stationary solutions to the differential equation

ẏ(t) = y(t)[1 − y(t)2 ]

from parts (c) and (d) above are asymptotically stable? Which are unstable?

7
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2005

1 Quasi-Linear Preferences
The consumer’s problem is to choose quantities of the two goods x and y to maximize

U(x, y) = ln(x) + y

subject to the budget constraint

I ≥ px + qy
and the nonnegativity constraints
x≥0
and
y ≥ 0.

a) There are a number of ways to define the Lagrangian for this problem, depending on
how one chooses to treat the nonnegativity constraints; here, I will work with the
formulation that explicitly incorporates them into the Lagrangian, defined as

L(x, y, λ, μ, φ) = ln(x) + y + λ(I − px − qy) + μx + φy.

b) Given the way the Lagrangian has been defined above, the Kuhn-Tucker theorem im-
plies that the values x∗ and y ∗ that solve the consumer’s problem, together with the
associated values λ∗ , μ∗ , and φ∗ of the three multipliers, must satisfy the first-order
conditions
L1 (x∗ , y ∗ , λ∗ , μ∗ , φ∗ ) = 1/x∗ − λ∗ p + μ∗ = 0 (1)
and
L2 (x∗ , y ∗ , λ∗ , μ∗ , φ∗ ) = 1 − λ∗ q + φ∗ = 0, (2)
the constraints
L3 (x∗ , y ∗ , λ∗ , μ∗ , φ∗ ) = I − px∗ − qy ∗ ≥ 0, (3)
L4 (x∗ , y ∗ , λ∗ , μ∗ , φ∗ ) = x∗ ≥ 0, (4)
and
L5 (x∗ , y ∗ , λ∗ , μ∗ , φ∗ ) = y ∗ ≥ 0, (5)

1
 the nonnegativity conditions
λ∗ ≥ 0, (6)
μ∗ ≥ 0, (7)
and
φ∗ ≥ 0, (8)
and the complementary slackness conditions

λ∗ (I − px∗ − qy ∗ ) = 0, (9)

μ∗ x∗ = 0, (10)
and
φ∗ y ∗ = 0. (11)

c) The form of the utility function implies that the budget constraint will always bind. The
form of the utility function also implies that x∗ > 0 and hence, by (10), that μ∗ = 0 as
well. Suppose first that y ∗ = 0. In this case, (1)-(3) and (8) require that

1/x∗ − λ∗ p = 0,

1 − λ∗ q + φ∗ = 0,
I − px∗ = 0,
and
φ∗ ≥ 0
must hold. The binding budget constraint implies that the consumer spends all of his
or her income on good x:
x∗ = I/p.
The first-order condition for x∗ then implies that

λ∗ = p/x∗ = 1/I.

The first-order condition for y ∗ and the nonnegativity condition for φ∗ then require
that
0 ≤ φ∗ = λ∗ q − 1 = q/I − 1
or more simply that
q ≥ I.
Thus, whenever q ≥ I it is optimal for the consumer to choose x∗ = I/p and y ∗ = 0.

d) Suppose now that y ∗ > 0. In this case, (11) implies that φ∗ = 0, so that (1)-(3) require
that
1/x∗ − λ∗ p = 0,
1 − λ∗ q = 0,

2
 and
I − px∗ − qy ∗ = 0.
Note that in this case, the first-order condition for y ∗ immediately implies that
λ∗ = 1/q.
Hence, the first-order condition for x∗ implies that
x∗ = q/p
and the budget constraint implies that
y ∗ = (I − q)/q,
which is strictly positive as required if
I > q.
Thus, whenever I > q it is optimal for the consumer to choose x∗ = q/p and y ∗ =
(I − p)/q.
e) The results from parts (c) and (d) above imply that only consumers with income I larger
than q will purchase good y. In this sense, y is best regarded as the “luxury” good in
this example.

2 Investment with Adjustment Costs

The firm’s problem is to choose continuously differentiable functions I(t) and K(t) for t ∈
[0, ∞) to maximize the discounted value of profits,
Z ∞
e−rt {K(t)α − I(t) − (φ/2)[I(t)]2 },
0

subject to the capital accumulation constraint I(t) ≥ K̇(t), which must hold for all t ∈ [0, ∞),
taking the initial capital stock K(0) = K0 > 0 as given.
a) The Hamiltonian for this problem can be written in present value form as
H(k(t), π(t); t) = max e−rt {K(t)α − I(t) − (φ/2)[I(t)]2 } + π(t)I(t).
I(t)

b) According to the maximum principle, the firm’s optimal choices must satisfy the first-
order condition
−e−rt [1 + φI(t)] + π(t) = 0 (12)
as well as the pair of differential equations
π̇(t) = −Hk (k(t), π(t); t) = −αe−rt K(t)α−1 (13)
and
K̇(t) = Hπ (k(t), π(t); t) = I(t) (14)
for all t ∈ [0, ∞).

3
c) Rewrite (12) as
π(t) = e−rt [1 + φI(t)]
and differentiate both sides with respect to t to obtain
˙
π̇(t) = −re−rt [1 + φI(t)] + φe−rt I(t).

Substitute this last result into (13) to obtain

˙ = re−rt [1 + φI(t)] − αe−rt K(t)α−1
φe−rt I(t)

or more simply
˙ = (1/φ){r[1 + φI(t)] − αK(t)α−1 }.
I(t) (15)
Together, (14) and (15) form a system of two differential equations in the two unknowns
I(t) and K(t).
˙
d) Equations (14) and (15) implies that when I(t) = 0 and K̇(t) = 0, I(t) = I ∗ and
K(t) = K ∗ where
I∗ = 0
and
K ∗ = (r/α)1/(α−1) .

e) Equations (14) and (15) also imply that

K̇(t) = 0 whenever I(t) = 0,

K̇(t) > 0 whenever I(t) > 0,

K̇(t) < 0 whenever I(t) < 0,
µ ¶
˙ α 1
I(t) = 0 whenever I(t) = K(t)α−1 − ,
φr φ
µ ¶
˙ α 1
I(t) = 0 whenever I(t) > K(t)α−1 − ,
φr φ
and µ ¶
˙ < 0 whenever I(t) < α 1
I(t) K(t)α−1 − .
φr φ
Using these observations, it is possible to draw a phase diagram that illustrates that
starting from any value of the initial capital stock K0 , there is a unique value of I(0)
such that starting from K(0) = K0 and I(0), the firm’s investment I(t) and capital
stock K(t) converge to their steady-state values.

4
 For each value of K0, there is a unique value
of I0 that leads the system to the steady state.
I

stable manifold
(saddle path)

steady state
K*=(r/α)1/(α−1)
K’=0 isocline
I*=0
I=0

0
K

−1/Φ

I’=0 isocline
I=(1/Φ)[(α/r)Kα-1−1]
3 Optimal Resource Depletion
The problem is to choose continuously differentiable functions c(t) and x(t) for t ∈ [0, ∞) to
maximize Z ∞
e−ρt ln(c(t))dt
0

subject to the constraint −c(t) ≥ ẋ(t) for all t ∈ [0, ∞), taking as given the level of the
initial resource stock x(0) = x0 > 0.

a) The Hamiltonian for this problem can be written as

H(x(t), π(t); t) = max e−ρt ln(c(t)) − π(t)c(t).

c(t)

b) According to the maximum principle, the optimal choices must satisfy the first-order
condition
e−ρt
= π(t) (16)
c(t)
as well as the pair of differential equations

π̇(t) = −Hx (x(t), π(t); t) = 0 (17)

and
e−ρt
ẋ(t) = Hπ (x(t), π(t); t) = −c(t) = − (18)
π(t)
for all t ∈ [0, ∞).

c) Probably the easiest way to solve the pair of differential equations (17) and (18) is by
the guess-and-verify method. Differentiating both sides of

π(t) = π

with respect to t yields

π̇(t) = 0,
verifying that (17) is satisfied by this choice. Differentiating both sides of
µ ¶
1
x(t) = e−ρt + k
πρ

with respect to t yields µ ¶

1 −ρt
ẋ(t) = − e ,
π
verifying that (18) is satisfied as well.

5
d) The maximum principle also states that the solution to the problem must satisfy the
initial condition
x(0) = x0 given
and the terminal, or transversality, condition

lim π(T )x(T ) = 0.

T →∞

These boundary conditions pin down values for the unknown constants π and k that
appear in the general solutions for π(t) and x(t). In particular, the initial condition
requires that µ ¶
1 1
x0 = x(0) = e−ρ0 + k = + k,
πρ πρ
While the terminal condition requires that
∙µ ¶ ¸ µ ¶
1 −ρt 1 −ρt
0 = lim π(T )x(T ) = lim π e + k = lim e + πk = πk.
T →∞ T →∞ πρ T →∞ ρ

Rewrite the first of these two equations as

1
k = x0 −
πρ
and substitute it into the second to obtain
1
0 = πx0 − .
ρ
Hence
1
π=
ρx0
and
k = 0.

e) The results from parts (c) and (d) from above imply that
1
π(t) = π =
ρx0

for all t ∈ [0, ∞). Substitute this result into the first-order condition (16) to obtain
the solution
e−ρt
c(t) = = ρx0 e−ρt . (19)
π(t)
Intuitively, (19) implies that consumption during period t is higher when the initial
resource stock x0 is larger and when the discount rate ρ is larger, that is, when the
planner cares less about the future. In addition, consumption declines over time and
converges gradually to zero as the resource stock is depleted over the infinite horizon.

6
4 Saving Under Certainty
The problem is to choose sequences {st }∞ ∞
t=0 and {At }t=1 to maximize

X
∞
β t u(At + w − st )
t=0

subject to the constraints A0 given and (1 + r)st ≥ At+1 for all t = 0, 1, 2, ....

a) The Bellman equation for this problem is

v(At ; t) = max u(At + w − st ) + βv[(1 + r)st ; t + 1].

st

b) The first order condition for the control variable st is

−u(At + w − st ) + β(1 + r)v 0 [(1 + r)st ; t + 1] = 0. (20)

The envelope condition for the state variable At is

v0 (At ; t) = u0 (At + w − st ). (21)

c) Use the constraints to rewrite (20) and (21) more simply as

u0 (ct ) = β(1 + r)v0 (At+1 ; t + 1) (22)

and
v0 (At ; t) = u0 (ct ). (23)
Since (23) must hold for all t = 0, 1, 2, ..., it also implies that

v 0 (At+1 ; t + 1) = u0 (ct+1 ).

Substitute this last condition into (22) to obtain

u0 (ct ) = β(1 + r)u0 (ct+1 ), (24)

a restatement of the optimality conditions (20) and (21) that makes reference only to
objects with direct economic interpretations.

d) The additional assumption that utility is logarithmic implies that (24) can be specialized
to
ct+1 /ct = β(1 + r). (25)

e) A more patient consumer puts a higher weight β on future consumption relative to

current consumption. Hence, according to (25), more patient consumers will choose
faster growth rates for consumption.

7
5 Optimal Stochastic Growth
The problem is to choose contingency plans for ct , t = 0, 1, 2, ..., and kt , t = 1, 2, 3, ..., to
maximize
X∞
E0 β t ln(ct )
t=0
subject to the constraints k0 given and zt ktα ≥ ct + kt+1 for all t = 0, 1, 2, ....
a) The Bellman equation for this problem is
v(kt , zt ) = max ln(ct ) + βEt v(zt ktα − ct , zt+1 ).
ct

b) With the suggested guess, the Bellman equation becomes

E + F ln(kt ) + G ln(zt ) = max ln(ct ) + βE + βF ln(zt ktα − ct ),
ct

after also using the results that, by assumption, Et ln(zt+1 ) = 0 and Et ln(zt ktα − ct ) =
ln(zt ktα − ct ). The first-order condition for the control variable ct is then
1 βF
− = 0, (26)
ct zt ktα − ct
and the envelope condition for the state variable kt is
F αβF zt ktα−1
= . (27)
kt zt ktα − ct
c) The first-order condition (26) implies
µ ¶
1
ct = zt ktα (28)
1 + βF
and, in light of this result, the envelope condition (27) implies
1
= 1 − αβ. (29)
1 + βF
Substituting (29) back into (28) yields the desired solution for ct :
ct = (1 − αβ)zt ktα . (30)

d) Finally, in light of (30), the binding constraint yields the desired solution for kt+1 :
kt+1 = zt ktα − ct = αβzt ktα (31)

e) Given k0 , (30) implies that if z0 = 1, then

c0 = (1 − αβ)k0α .
Likewise, (31) implies that
k1 = αβk0α .
Substituting this last result into (30) then implies that if z1 = 1 as well, then
c1 = (1 − αβ)(αβk0α )α .

8
6 Scalar Differential Equations
a) The general solution to the scalar differential equation

ẏ(t) = −y(t)

can be found by the guess-and-verify method. Guess that the general solution is

y(t) = ke−t ,

where k ∈ R. To verify that this guess works, simply note that it implies

ẏ(t) = −ke−t = −y(t)

as required.

b) To find the unique solution to the initial value problem consisting of the same differential
equation
ẏ(t) = −y(t)
from part (a) above and the initial condition

y(0) = y0 given,

take the general solution y(t) = ke−t from above and impose the initial condition to
pin down k:
y0 = y(0) = ke−0 = k.
Evidently, the unique solution is

y(t) = y0 e−t .

c) The stationary solutions to the scalar differential equation

ẏ(t) = y(t)[1 − y(t)2 ]

take the form

y(t) = c
for all t and hence
ẏ(t) = 0.
Hence, the stationary solutions can be found be solving for the values of c that satisfy

0 = c(1 − c2 ),

namely, c = −1, c = 0, and c = 1. Evidently, there are three stationary solutions

y(t) = −1, y(t) = 0, and y(t) = 1.

9
d) The analysis from part (c) above has already told us that the differential equation

ẏ(t) = y(t)[1 − y(t)2 ]

implies that
ẏ(t) = 0 if y(t) = −1, y(t) = 0, or y(t) = 1.
Note that the equation also implies that

ẏ(t) > 0 if y(t) > 0 and 1 − y(t)2 > 0, that is, if 1 > y(t) > 0,

ẏ(t) > 0 if y(t) < 0 and 1 − y(t)2 < 0, that is, if − 1 > y(t),
ẏ(t) < 0 if y(t) > 0 and 1 − y(t)2 < 0, that is, if y(t) > 1,
and
ẏ(t) < 0 if y(t) < 0 and 1 − y(t)2 > 0, that is, if − 1 > y(t) > 0.
A phase diagram that illustrates these properties of the differential equation can be
used to highlight the features of solutions starting from any initial condition y(0) = y0 .

e) In addition, the phase diagram reveals that the stationary solutions

y(t) = −1 and y(t) = 1

are asymptotically stable, while the stationary solution

y(t) = 0

is unstable.

10
 -1 0 1

y’(t) = y(t)[1-y(t)2]

The phase diagram reveals that:

If y0 < -1, y converges to -1.
If y0 = -1, y remains at -1.
If -1 < y0 < 0, y converges to -1.
If y0 = 0, y remains at 0.
If 0 < y0 < 1, y converges to 1.
If y0 = 1, y remains at 1.
If 1 < y0, y converges to 1.
 Final Exam
Economics 720: Mathematics for Economists
Fall 2006

This exam has six questions on eight pages; before you begin, please make sure that your
copy has all six questions and all eight pages. The six questions will receive equal weight in
determining your overall exam score. Since this is a 110-minute exam, I suggest that you
spend approximate (110 minutes)/(6 questions) ≈ 18 minutes on each question.

1 Expenditure Minimization
A consumer has preferences over two goods, as described by the Cobb-Douglas utility func-
tion
U (c1 , c2 ) = ca1 c21−a ,
where c1 denotes consumption of good 1, c2 denotes consumption of good 2, and a is a
parameter that lies between zero and one: 0 < a < 1. Assume that the consumer can
purchase as many or as few units of each good as he or she chooses in perfectly competitive
markets at the strictly positive prices p1 > 0 for good 1 and p2 > 0 for good 2.
Now suppose that this consumer chooses c1 ≥ 0 and c2 ≥ 0 in an attempt to minimize
the total expenditures p1 c1 + p2 c2 required to attain a level of utility that meets or exceeds
some prespecified level Ū. That is, suppose that the consumer solves
min p1 c1 + p2 c2 subject to ca1 c1−a
2 ≥ Ū.
c1 ,c2

a) Define (set up) the Lagrangian for this expenditure minimization problem. In doing this,
it may be helpful for you to note that the form of the utility function implies that the
nonnegativity constraints c1 ≥ 0 and c2 ≥ 0 will not bind at the optimum; hence, it is
safe to ignore these constraints in setting up the Lagrangian and in working through
the rest of the analysis below.
b) Let c∗1 and c∗2 denote the values of c1 and c2 that solve this problem. Write down the full
set of first-order conditions, constraints, nonnegativity conditions, and complementary
slackness conditions that, according to the Kuhn-Tucker theorem, must be satisfied by
c∗1 , c∗2 , and the associated value of the Lagrange multiplier.
c) Next, use these Kuhn-Tucker conditions to find solutions for c∗1 , c∗2 , and the associated
value of the Lagrange multiplier in terms of the model’s four parameters: p1 , p2 , Ū,
and a. [Note: these solutions for c∗1 and c∗2 define the Hicksian or compensated demand
functions for the two goods.]

1
 d) Now define the minimum expenditure function E(p1 , p2 , Ū) as the minimized value of
the objective function, taken subject to the constraint; that is, define

E(p1 , p2 , Ū ) = min p1 c1 + p2 c2 subject to ca1 c1−a

2 ≥ Ū .
c1 ,c2

Use your results from above to obtain expressions for the derivatives of this minimum
expenditure function, E1 (p1 , p2 , Ū), E2 (p1 , p2 , Ū), and E3 (p1 , p2 , Ū), in terms of the
parameters p1 , p2 , Ū, and a.

e) What do your results from above tell you about the relationship between the Hicksian
demand functions for the two consumption goods and the derivatives of the minimum
expenditure function? What economic interpretation does your results provide of the
Lagrange multiplier in this constrained minimization problem?

2 Utility Maximization
Consider the same consumer as in question 1 from above, with preferences over the two
goods described by the Cobb-Douglas utility function

U (c1 , c2 ) = ca1 c21−a ,

where c1 denotes consumption of good 1, c2 denotes consumption of good 2, and a is a

parameter that lies between zero and one: 0 < a < 1. Assume again that the consumer can
purchase as many or as few units of each good as he or she chooses in perfectly competitive
markets at the strictly positive prices p1 > 0 for good 1 and p2 > 0 for good 2.
Now, however, suppose that this consumer chooses c1 ≥ 0 and c2 ≥ 0 in an attempt to
maximize his or her utility, subject to the budget constraint

I ≥ p1 c1 + p2 c2 ,

where I denotes the consumer’s fixed level of income. That is, suppose that the consumer
solves
max ca1 c1−a
2 subject to I ≥ p1 c1 + p2 c2 .
c1 ,c2

a) Define (set up) the Lagrangian for this utility maximization problem. In doing this, it
may be helpful for you to note that the form of the utility function implies that the
nonnegativity constraints c1 ≥ 0 and c2 ≥ 0 will not bind at the optimum; hence, it is
safe to ignore these constraints in setting up the Lagrangian and in working through
the rest of the analysis below.

b) Let c∗1 and c∗2 denote the values of c1 and c2 that solve this problem. Write down the full
set of first-order conditions, constraints, nonnegativity conditions, and complementary
slackness conditions that, according to the Kuhn-Tucker theorem, must be satisfied by
c∗1 , c∗2 , and the associated value of the Lagrange multiplier.

2
 c) Next, use these Kuhn-Tucker conditions to find solutions for c∗1 , c∗2 , and the associated
value of the Lagrange multiplier in terms of the model’s four parameters: p1 , p2 , I, and
a. [Note: these solutions for c∗1 and c∗2 define the Marshallian or Walrasian demand
functions for the two goods.]

d) Now define the indirect utility function V (p1 , p2 , I) as the maximized value of the objec-
tive function, taken subject to the constraint; that is, define

V (p1 , p2 , I) = max ca1 c1−a

2 subject to I ≥ p1 c1 + p2 c2 .
c1 ,c2

Use your results from above to obtain expressions for the derivatives of this indirect
utility function, V1 (p1 , p2 , I), V2 (p1 , p2 , I), and V3 (p1 , p2 , I), in terms of the parameters
p1 , p2 , I, and a.

e) What do your results from above tell you about the relationship between the Marshallian
demand functions for the two consumption goods and the derivatives of the indirect
utility function? What economic interpretation does your results provide of the La-
grange multiplier in this constrained maximization problem?

3 Fishing
Let y(t) denote the number of fish in a lake and assume that in the absence of fishing, the
fish population evolves according to the so-called logistic model

ẏ(t) = ay(t)[b − y(t)]

for all t ∈ [0, ∞), where a > 0 and b > 0 are both positive parameters. Catching c(t) fish
yields a consumer utility Z ∞
e−ρt ln(c(t))dt
0
over the infinite horizon, where ρ > 0 is a positive discount rate, but subtracts from the fish
population so that
ẏ(t) = ay(t)[b − y(t)] − c(t)
for all t ∈ [0, ∞). In working through the analysis below, it may be helpful for you to assume
that the model’s parameters also satisfy ab > ρ.
Hence, the consumer must choose continuously differentiable functions c(t) for t ∈ [0, ∞)
and y(t) for all t ∈ (0, ∞) to maximize
Z ∞
e−ρt ln(c(t))dt
0

subject to the constraints

ay(t)[b − y(t)] − c(t) ≥ ẏ(t)
for all t ∈ [0, ∞), taking the initial stock of fish y(0) > 0 as given.

3
 a) As a first step in using the maximum principle to characterize the solution to the con-
sumer’s problem, define (set up) the Hamiltonian for this problem.

b) Now write down the first-order condition and the pair of differential equations that,
according to the maximum principle, are necessary conditions for the values of c(t)
and y(t) that solve the consumer’s problem.

c) Your answer to part (b) from above should take the form of a system of three equations in
three unknowns. Use one of these equations to eliminate one of the variables–the addi-
tional variable that you introduced into the problem when defining the Hamiltonian–to
rewrite the system as one involving two equations in two unknowns: consumption c(t)
and the fish population y(t).

d) Use your results from part (c) above to find solutions for the optimal steady-state
values c∗ and y ∗ of consumption and the fish population size in terms of the model’s
parameters: a, b, and ρ.

e) Finally, use your results from above to draw a phase diagram that illustrates the following
property of the solution to the consumer’s problem: starting from any value y(0) > 0
for the initial fish population size, there is a unique value of c(0) such that starting
from y(0) and c(0), optimally-chosen consumption c(t) and the population size y(t)
converge to their steady-state values c∗ and y ∗ . In drawing this phase diagram, it
may be helpful to note that although we did not specifically impose the nonnegativity
constraints c(t) ≥ 0 and y(t) ≥ 0 for all t ∈ [0, ∞) in setting up the consumer’s
dynamic optimization problem, the form of the utility function (which implies that the
marginal utility of consumption becomes arbitrarily large as consumption approaches
zero) and the law of motion for the fish population (which implies that no new fish are
born when the fish population is fully depleted) imply that these constraints will never
bind when the variables are chosen optimally.

4 Investment with Adjustment Costs

During each period t ∈ [0, ∞), a firm produces output Y (t) with capital K(t) according to
the production function
Y (t) = K(t)α ,
where 0 < α < 1. Capital depreciates at the constant rate δ, where 0 < δ < 1; hence, by
investing I(t) units of output at t, the firm augments its capital stock according to

I(t) − δK(t) ≥ K̇(t).

but at the same time also incurs a quadratic adjustment cost given by

(φ/2)[I(t)]2 ,

where φ > 0. In this problem, the firm’s choice of I(t) can be positive, in which case the
firm is installing new capital, or negative, in which case the firm is selling off existing capital;

4
but either way, the formulation implies that it will incur adjustment costs in making these
changes. The firm sells off whatever output remains after its investment choice is made; its
profits during period t are therefore

K(t)α − I(t) − (φ/2)[I(t)]2 .

Finally, let r > 0 denote the constant discount rate.

The firm’s problem can now be stated as: choose continuously differentiable functions
I(t) for t ∈ [0, ∞) and K(t) for t ∈ (0, ∞) to maximize the discounted value of profits,
Z ∞
e−rt {K(t)α − I(t) − (φ/2)[I(t)]2 }dt,
0

subject to the capital accumulation constraints

I(t) − δK(t) ≥ K̇(t)

for all t ∈ [0, ∞), taking the initial capital stock K(0) > 0 as given.

a) As a first step in using the maximum principle to characterize the solution to the firm’s
problem, define (set up) the Hamiltonian for this problem.

b) Now write down the first-order condition and the pair of differential equations that,
according to the maximum principle, are necessary conditions for the values of I(t)
and K(t) that solve the consumer’s problem.

c) Your answer to part (b) from above should take the form of a system of three equations in
three unknowns. Use one of these equations to eliminate one of the variables–the addi-
tional variable that you introduced into the problem when defining the Hamiltonian–to
rewrite the system as one involving two equations in two unknowns: investment I(t)
and the capital stock K(t).

d) Using your results from part (c) above, write down a set of two equations that determine
the optimal steady-state values I ∗ and K ∗ . NOTE: In this case, it may be too difficult
to actually solve these two equations for the steady-state values of I ∗ and K ∗ , so DON’T
BOTHER TO DO THIS. Instead, just write down the two equations that determine
I ∗ and K ∗ .

e) Finally, use your results from above to draw a phase diagram that illustrates the following
property of the solution to the firm’s problem: starting from any value of the initial
capital stock K(0) > 0, there is a unique value of I(0) such that starting from K(0)
and I(0), optimally-chosen investment I(t) and capital stock K(t) converge to their
steady-state values K ∗ and I ∗ . In drawing this phase diagram, it may be helpful to note
that while investment I(t) may take on either positive or negative values, depending on
whether the firm is accumulating new capital or selling off existing capital, the capital
stock K(t) must always remain positive when the variables are chosen optimally.

5
5 Human Capital and Growth
For t = 0, 1, 2, ..., let ht denote the level of skill, or stock of human capital, attained by
a representative consumer who splits his or her time between two activities: studying to
increase his or her stock of human capital or working to produce output. More specifically,
by allocating the fraction ut to his or her time to studying, the consumer adds to his or her
stock of human capital according to

ht+1 = (1 + γut )ht ,

where γ > 0 is a positive parameter. Note that this specification implies that when ut = 0,
so that no time is spent studying, ht+1 = ht , so that human capital remains unchanged.
When ut = 1, so that all available time is spent studying, ht+1 = (1 + γ)ht , so that human
capital grows at the fastest possible rate γ. For values of ut between zero and one, human
capital grows at a rate that lies between zero and γ. By allocating the remaining fraction
1 − ut of his or her time to working, the consumer produces yt units of output according to
the technology described by the production function

yt = [(1 − ut )ht ]α ,

where α lies between zero and one: 0 < α < 1. This specification implies that more output
can be obtained when either: (i) the consumer spends a larger fraction 1 − ut of his or her
time working or (ii) the consumer is more skilled, that is, has a larger stock of human capital
ht .
Let the consumer have preferences over consumption ct at each date t = 0, 1, 2, ..., as
described by the utility function
X
∞
β t ln(ct ),
t=0

where the discount factor β also lies between zero and one: 0 < β < 1. If output cannot
be stored, the optimizing consumer will always choose ct = yt = [(1 − ut )ht ]α or, taking
logs, ln(ct ) = α ln(1 − ut ) + α ln(ht ). In working through the analysis that follows, it will be
helpful to assume that the parameters γ and β are such that β(1 + γ) > 1; the results that
you will derive below imply that when this condition is satisfied, it will always be optimal
for the consumer to choose a value of ut that satisfies 0 < ut < 1, so that in formulating
the consumer’s problem it will not be necessary to impose nonnegativity constraints on the
times spent studying or working.
Hence, the consumer’s problem can now be formulated as one of choosing sequences
{ut }∞ ∞
t=0 and {ht }t=1 to maximize

X
∞
β t [α ln(1 − ut ) + α ln(ht )]
t=0

subject to the constraints

(1 + γut )ht ≥ ht+1
for all t = 0, 1, 2, ..., taking the initial stock of human capital h0 as given.

6
 a) Write down the Bellman equation for the consumer’s problem, using ht as the state
variable and ut as the control variable.
b) Next, guess that the value function takes the time-invariant form

v(ht ) = E + F ln(ht ),

where E and F are constants to be determined. Using this guess, derive the first-order
and envelope conditions that characterize the optimal choices of ut and ht .
c) Use your results from part (b) above to obtain an expression that shows how the optimal
choice of ut depends on the model’s parameters: γ, α, and β.
d) Use your results from above to obtain an expression that shows how the optimal growth
rate of human capital ht+1 /ht depends on the model’s parameters: γ, α, and β.
e) Finally, use your results from above to obtain an expression that shows how the optimal
growth rate of output and consumption yt+1 /yt = ct+1 /ct depends on the model’s
parameters: γ, α, and β.

6 Monopolistic Price Adjustment

A monopolist faces the downward-sloping demand curve

dt = p−ε
t

in each period t = 0, 1, 2, ..., where dt denotes demand for the firm’s output, pt denotes
the price that the firm sets for each unit of output, and the parameter ε > 1 measures the
absolute value of the constant price elasticity of demand. Suppose that the firm can produce
each unit of output at the constant marginal cost c > 0. Suppose also that it faces the
quadratic cost µ ¶2
φ pt
−1
2 pt−1
of adjusting its price between periods; this cost of adjustment can either be interpreted
literally as reflecting the administrative costs within the firm of changing its price or more
broadly as reflecting the negative impact on the firm resulting from consumers’ resistance
to and dissatisfaction with rapid price changes. The firm’s total profits during each period
t then equal revenues from goods sold (price times quantity demanded dt pt = p1−ε t ) minus
−ε
costs of goods sold (marginal cost times quantity demanded cdt = cpt ) minus costs of
adjustment; if the firm discounts future profits at the constant rate r > 0, then the present
value of profits over the infinite horizon is measured by
X∞ µ ¶t " µ ¶2 #
1 φ p t
p1−ε
t − cp−ε
t − −1 .
t=0
1 + r 2 pt−1

This model lacks what might clearly be identified as “stock” and “flow” variables. How-
ever, the firm’s problem can be solved using dynamic programming methods by employing

7
the following notational trick. For all t = 0, 1, 2, ..., let yt denote “last period’s price during
period t.” Then yt = pt−1 , yt+1 = pt , and so on. Now the firm can be viewed as solving the
problem: maximize by choice of sequences {pt }∞ ∞
t=0 and {yt }t=1 the objective function

X ∞ µ ¶t " µ ¶2 #
1 φ pt
p1−ε
t − cp−ε
t − −1
t=0
1 + r 2 yt

subject to the “binding constraints”

yt+1 = pt
for all t = 0, 1, 2, ..., taking the initial condition y0 = p−1 as given.

a) Write down the Bellman equation for the firm’s problem, using yt as the state variable
and pt as the control variable. In doing this, you may find it helpful to substitute
the “binding constraint” yt+1 = pt into the expression on the right-hand side of the
Bellman equation, so that the optimization problem on the right-hand side of the
Bellman equation becomes an unconstrained maximization problem.

b) Next, write down the first-order and envelope conditions that characterize the optimal
choices of pt and yt .

c) Together with the “binding constraint” yt+1 = pt , your answer from part (b) above
should take the form of a system of three equations in three unknowns. Use one of
these equations to eliminate one of the unknowns–the unknown value function that
you introduced into the problem when setting up the Bellman equation–to rewrite
the system as one involving two equations in two unknowns: pt and yt .

d) Now use the “binding constraints” yt+1 = pt and yt = pt−1 to rewrite your answer to
part (c) from above in the form of a single equation in a single unknown: the firm’s
optimal price pt .

e) Finally, let p∗ denote the firm’s optimal price in a steady state where pt+1 = pt = pt−1 = p∗
for all t = 0, 1, 2, .... Use your answer from part (d) above to solve for p∗ in terms of the
model’s parameters: ε, c, φ, and r. How does the monopolist’s steady-state “markup”
of price over marginal cost, as measured by the ratio p∗ /c, change as the absolute value
of the elasticity of demand ε rises?

8
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2006

1 Expenditure Minimization
The consumer solves
min 1 1 + 2 2 subject to 1 1−
2 ≥ ̄
1 2

a) There are a number of ways to define the Lagrangian for this expenditure minimization
problem, but one way is to subtract the product of the multiplier  and the constraint
from the objective function to obtain

(1  2  ) = 1 1 + 2 2 − (1 1−

2 − ̄ )

b) With the Lagrangian defined as above, the Kuhn-Tucker theorem states that if ∗1 and
∗2 solve the consumer’s problem, then there exists a value ∗ of  such that, together,
∗1 , ∗2 , and ∗ satisfy the first-order conditions

1 (∗1  ∗2  ∗ ) = 1 − ∗ (∗1 )−1 (∗2 )1− = 0 (1)

and
2 (∗1  ∗2  ∗ ) = 2 − (1 − )∗ (∗1 ) (∗2 )− = 0 (2)
the constraint
(∗1 ) (∗2 )1− − ̄ ≥ 0 (3)
the nonnegativity condition
∗ ≥ 0 (4)
and the complementary slackness condition

∗ [(∗1 ) (∗2 )1− − ̄ ] = 0 (5)

c) Again, there are many ways of using these Kuhn-Tucker conditions to find solutions
for ∗1 , ∗2 , and ∗ in terms of the parameters 1 , 2 , ̄ , and , but one way starts by
rearranging the first-order condition (1) to read
1
∗ = ∗ −1 ∗ 1−
 (6)
(1 ) (2 )

1
 Since the objects in both the numerator and the denominator on the right-hand side
of (6) are all strictly positive, the multiplier ∗ must also be strictly positive; the
complementary slackness condition (5) then implies that the constraint (3) must bind
at the optimum. Now use (6) to substitute out for ∗ in (2) and rearrange to obtain
µ ¶µ ¶
∗ 1− 1 ∗
2 =  (7)
 2 1

Substitute this last result into the binding constraint to obtain the desired solution for
∗1 :
µ ¶1− µ ¶1−
∗  2
1 = ̄ (8)
1− 1
Substitute this expression back into (7) to obtain the desired solution for ∗2 :
µ ¶ µ ¶
∗ 1− 1
2 = ̄ (9)
 2

Finally, substitute (8) and (9) back into (6) to obtain the desired solution for ∗ :

∗ = − (1 − )−(1−) 1 1−

2  (10)

d) Now define the minimum expenditure function (1  2  ̄) as

(1  2  ̄ ) = min 1 1 + 2 2 subject to 1 1−

2 ≥ ̄ 
1 2

Related, let (8)-(10) define the functions ∗1 (1  2  ̄), ∗2 (1  2  ̄), and ∗ (1  2  ̄),
describing how optimal consumptions and the associated value of the Lagrange multi-
pliers change as the parameters 1 , 2 , and ̄ vary. Then, in light of the complementary
slackness condition (5), the minimum expenditure function can be evaluated as

(1  2  ̄) = 1 ∗1 (1  2  ̄) + 2 ∗2 (1  2  ̄ )

© ª
−∗ (1  2  ̄) [∗1 (1  2  ̄)] [∗2 (1  2  ̄)]1− − ̄ 

The envelope theorem implies that in differentiating both sides of this last expression,
it is safe to ignore the dependence of ∗1 , ∗2 , and ∗ on 1 , 2 , and ̄ and write
µ ¶1− µ ¶1−
 2
1 (1  2  ̄) = ∗1 (1  2  ̄ )
= ̄  (11)
1− 1
µ ¶ µ ¶
∗ 1− 1
2 (1  2  ̄) = 2 (1  2  ̄) = ̄ (12)
 2
and
3 (1  2  ̄) = ∗ (1  2  ̄) = − (1 − )−(1−) 1 1−
2  (13)

2
 e) Equations (11) and (12) indicate that the Hicksian demand function for each good
coincides with the derivative of the minimum expenditure function with respect to the
price of that good; these results resemble those given to us by Sheppard’s lemma, which
says that for the cost minimizing firm, the conditional factor demand functions coincide
with the derivatives of the minimum cost with respect to the relevant factor prices. And
since 3 (1  2  ̄ ) measures the additional expenditure that the optimizing consumer
would require to attain a slightly higher level of utility, (13) provides an economic
interpretation of the Lagrange multiplier as the “marginal cost” of utility.

2 Utility Maximization
Now the consumer solves

max 1 1−

2 subject to  ≥ 1 1 + 2 2 
1 2

a) Once again there are a number of ways to define the Lagrangian for this utility maximiza-
tion problem, but one way is to add the product of the multiplier  and the constraint
to the objective function to obtain

(1  2  ) = 1 1−

2 + ( − 1 1 − 2 2 )

b) With the Lagrangian defined as above, the Kuhn-Tucker theorem states that if ∗1 and
∗2 solve the consumer’s problem, then there exists a value ∗ of  such that, together,
∗1 , ∗2 , and ∗ satisfy the first-order conditions

1 (∗1  ∗2  ∗ ) = (∗1 )−1 (∗2 )1− − ∗ 1 = 0 (14)

and
2 (∗1  ∗2  ∗ ) = (1 − )(∗1 ) (∗2 )− − ∗ 2 = 0 (15)
the constraint
 − 1 ∗1 − 2 ∗2 ≥ 0 (16)
the nonnegativity condition
∗ ≥ 0 (17)
and the complementary slackness condition

∗ ( − 1 ∗1 − 2 ∗2 ) = 0 (18)

c) Again, there are many ways of using these Kuhn-Tucker conditions to find solutions
for ∗1 , ∗2 , and ∗ in terms of the parameters 1 , 2 , , and , but one way starts by
rearranging the first-order condition (14) to read

(∗1 )−1 (∗2 )1−

∗ =  (19)
1

3
 Since the objects in both the numerator and the denominator on the right-hand side
of (19) are all strictly positive, the multiplier ∗ must also be strictly positive; the
complementary slackness condition (18) then implies that the constraint (16) must
bind at the optimum. Now use (19) to substitute out for ∗ in (15) and rearrange to
obtain µ ¶µ ¶
∗ 1− 1 ∗
2 =  (20)
 2 1
Substitute this last result into the binding constraint to obtain the desired solution for
∗1 :

∗1 =  (21)
1
Substitute this expression back into (20) to obtain the desired solution for ∗2 :
(1 − )
∗2 =  (22)
2
Finally, substitute (21) and (22) back into (6) to obtain the desired solution for ∗ :
−(1−)
∗ =  (1 − )1− −
1 2  (23)

d) Now define the indirect utility function  (1  2  ) as

 (1  2  ) = max 1 1−

2 subject to  ≥ 1 1 + 2 2 
1 2

Related, let (21)-(23) define the functions ∗1 (1  2  ), ∗2 (1  2  ), and ∗ (1  2  ),
describing how optimal consumptions and the associated value of the Lagrange multi-
pliers change as the parameters 1 , 2 , and  vary. Then, in light of the complementary
slackness condition (18), the indirect utility function can be evaluated as

 (1  2  ) = [∗1 (1  2  )] [∗2 (1  2  )]1−

+∗ (1  2  )[ − 1 ∗1 (1  2  ) − 2 ∗2 (1  2  )]

The envelope theorem implies that in differentiating both sides of this last expression,
it is safe to ignore the dependence of ∗1 , ∗2 , and ∗ on 1 , 2 , and  and write

1 (1  2  ) = −∗ (1  2  )∗1 (1  2  ) (24)

−(1−)
= − (1 − )1− −
1 2 (1 )
−(1+) −(1−)
= −1+ (1 − )1− 1 2 

2 (1  2  ) = −∗ (1  2  )∗2 (1  2  ) (25)

−(1−)
= − (1 − )1− −
1 2 [(1 − )2 ]
−(2−)
= − (1 − )2− −
1 2 

and
−(1−)
3 (1  2  ) = ∗ (1  2  ) =  (1 − )1− −
1 2  (26)

4
 e) Equations (24)-(26) indicate that the Marshallian demand functions and the derivatives
of the indirect utility function are related via

1 (1  2  )
∗1 (1  2  ) = −
3 (1  2  )

and
2 (1  2  )
∗2 (1  2  ) = − 
3 (1  2  )
These last two expressions are statements of Roy’s identity, which like Shephard’s
lemma turns out to be a special case or application of the envelope theorem. And
since 3 (1  2  ) measures the additional utility the optimizing consumer can attain
if provided with a slightly higher level of income, (26) also provides an economic inter-
pretation of the Lagrange multiplier as the “marginal utility” of income.

3 Fishing
The consumer must choose continuously differentiable functions () for  ∈ [0 ∞) and ()
for all  ∈ (0 ∞) to maximize Z ∞
− ln(())
0
subject to the constraints
()[ − ()] − () ≥ ̇()
for all  ∈ [0 ∞), taking the initial stock of fish (0)  0 as given.

a) The Hamiltonian for this problem can be defined in present value terms as

(() (); ) = max − ln(()) + (){()[ − ()] − ()}

()

b) According to the maximum principle, the values of () and () that solve the consumer’s
problem and the associated value of () must satisfy the first-order condition

−
− () = 0 (27)
()

and the pair of differential equations

̇() = − (() (); ) = −()[ − 2()] (28)

and
̇() =  (() (); ) = ()[ − ()] − () (29)
for all  ∈ [0 ∞).

5
c) Rewrite (27) as
− = ()()
and differentiate both sides with respect to  to obtain
−− = ̇()() + ()̇()
Now use (27) and (28) to rewrite this last result as
−()() = −()[ − 2()]() + ()̇()
or, after dividing through by () and isolating ̇() on the left-hand side,
̇() = ()[ − 2() − ] (30)
Together, (29) and (30) form a system of two equations in two unknowns: consumption
() and the fish population ().
d) In a steady state where () = ∗ and () =  ∗ , ̇() = 0 and ̇() = 0. Substituting
these steady-state conditions into (29) and (30) yields
0 =  ∗ ( −  ∗ ) − ∗
and
0 = ∗ ( − 2 ∗ − )
Use the second of these two conditions to solve for
 − 
∗ = 
2
Then substitute this solution into the first condition to find
µ ¶µ ¶
∗  −   −  2 2 − 2
 = − = 
2 2 4
e) Consider first (30), which implies that
̇() = 0 when () =  ∗ 
̇()  0 when ()   ∗ 
and
̇()  0 when ()   ∗ 
Consider next (29), which implies that
̇() = 0 when () = () − ()2 
̇()  0 when ()  () − ()2 
and
̇()  0 when ()  () − ()2 
The phase diagram that reflects these implications of (29) and (30) also reveals that
starting from any value (0)  0 for the initial fish population size, there is
a unique value of (0) such that starting from (0) and (0), optimally-chosen
consumption () and the population size () converge to their steady-state values
∗ and  ∗ .

6
c c’ = 0 isocline or locus
(y = yy*))

steady state
(y*,c*)
c* y’ = 0 isocline or locus
(c = aby – ay2)

saddle
ddl path
th or
stable manifold

y* y
4 Investment with Adjustment Costs
The firm chooses continuously differentiable functions () for  ∈ [0 ∞) and () for  ∈
(0 ∞) to maximize the discounted value of profits,
Z ∞
− {() − () − (2)[()]2 }
0

subject to the capital accumulation constraints

() − () ≥ ̇()

for all  ∈ [0 ∞), taking the initial capital stock (0)  0 as given.

a) The Hamiltonian for this problem can be defined in present value terms as

(() (); ) = max − {() − () − (2)[()]2 } + ()[() − ()]

()

b) According to the maximum principle, the values of () and () that solve the firm’s
problem and the associated value of () must satisfy the first-order condition

−− [1 + ()] + () = 0 (31)

and the pair of differential equations

̇() = − (() (); ) = −− ()−1 + () (32)

and
̇() =  (() (); ) = () − () (33)
for all  ∈ [0 ∞).

c) Rewrite (31) as
() = − [1 + ()]
and differentiate both sides with respect to  to obtain
˙
̇() = −− [1 + ()] + − ()

Now substitute this last result together with (31) into (32) to obtain
˙ = −− ()−1 + − [1 + ()]
−− [1 + ()] + − ()
˙ on the left-hand side,
or, after dividing through by − and isolating ()
˙ = (1){( + )[1 + ()] − ()−1 }
() (34)

Together, (33) and (34) form a system of two equations in two unknowns: investment
() and the capital stock ().

7
 ˙ = 0 and ̇() = 0. Substituting
d) In a steady state where () =  ∗ and () =  ∗ , ()
these steady-state conditions into (33) and (34) yields

 ∗ =  ∗

and
( + )(1 +  ∗ ) = ( ∗ )−1 
These last two equations determine the optimal steady-state values  ∗ and  ∗ .
e) Consider first (33), which implies that

̇() = 0 when () = ()

̇()  0 when ()  ()

and
̇()  0 when ()  ()
Consider next (34), which implies that
µ ¶
˙ = 0 when () = 1
()
 1
()−1 − 
 + 
µ ¶
˙  0 when ()  1
()
 1
()−1 − 
 + 
and µ ¶
˙  0 when ()  1
()
 1
()−1 − 
 + 
A phase diagram that reflects these implications of (33) and (34) also reveals that
starting from any value of the initial capital stock (0)  0, there is a unique value
of (0) such that starting from (0) and (0), optimally-chosen investment () and
capital stock () converge to their steady-state values  ∗ and  ∗ .

5 Human Capital and Growth

The consumer chooses sequences { }∞ ∞
=0 and { }=1 to maximize

X
∞
  [ ln(1 −  ) +  ln( )]
=0

subject to the constraints

(1 +  ) ≥ +1
for all  = 0 1 2 , taking the initial stock of human capital 0 as given.

a) The Bellman equation for the consumer’s problem is

( ; ) = max  ln(1 −  ) +  ln( ) + [(1 +  ) ;  + 1]



8
 K’ = 0 isocline or locus
I (I = δK)

saddle path or
stable manifold

steady state (K*,I*)

I*
I’ = 0 isocline or locus
(I = (1/Φ){[α/(δ+r)]K
Φ 1-1})
α-1

0
K* K

-1/Φ
b) Using the conjectured form of the value function, the Bellman equation becomes

 +  ln( ) = max  ln(1 −  ) +  ln( ) +  +  ln(1 +  ) +  ln( )



The first-order condition for  is then

  
− + =0 (35)
1 −  1 + 
while the envelope condition for  is
  
= +  (36)
  

c) For this problem, the envelope condition (36) yields an immediate solution for  :

 = 
1−

Substitute this solution into (35) to obtain the desired solution for  :

 +  − 1
 = (37)

for all  = 0 1 2 . Evidently, it is optimal to choose a constant value of  . This
constant value of  is strictly less than one since   0 and 1  , and this constant
value of  is strictly positive since (1 + )  1.

d) Using the binding constraint and the solution (37) for  ,

+1  = 1 +  = (1 + ) (38)

for all  = 0 1 2 . Evidently, it is optimal to have human capital grow at a constant
rate, which is positive since (1 + )  1 and that is larger when  is larger, so that
the consumer is more patient, and when  increases, so time spent studying yields a
larger payoff in terms of human capital growth.

e) Since
 =  = [(1 −  ) ] 
(37) and (38) imply that

+1  = (+1  ) = [(1 + )] (39)

for all  = 0 1 2 . Since time spent producing 1 −  is constant and since human
capital growth +1  is constant, consumption and output growth +1  are constant
as well.

9
6 Monopolistic Price Adjustment
The firm can be viewed as choosing sequences { }∞ ∞
=0 and { }=1 to maximize the objective
function ¶ " ¶2 #
X∞ µ µ
1   
1−
 − −
 − −1
=0
1 +  2 

subject to the “binding constraints”

+1 = 
for all  = 0 1 2 , taking the initial condition 0 = −1 as given.
a) The Bellman equation for the firm’s problem is
µ ¶2 µ ¶
1− −   1
( ; ) = max  −  − −1 + ( ;  + 1)
 2  1+

b) The first-order condition for  is

µ ¶ µ ¶
 1 1
(1 − )−
 + −−1
 − −1 +  0 ( ;  + 1) = 0 (40)
  1+
and the envelope condition for  is
µ ¶
0  
 ( ; ) =  − 1 2 (41)
 

c) Substitute the “binding constraint” +1 =  into the last term on the left-hand side of
(40) to obtain
µ ¶ µ ¶
− −−1  1 1
(1 − ) +  − −1 +  0 (+1 ;  + 1) = 0
  1+
Then consider (41), which since it must hold for all  = 0 1 2  implies that
µ ¶
0 +1 +1
 (+1 ;  + 1) =  −1 2

+1 +1
Substitute the second of these last two equations into the first to obtain
µ ¶ µ ¶ µ ¶
− −−1  1 1 +1 +1
(1 − ) +  − −1 +  −1 2
= 0 (42)
  1+ +1 +1
Now (42) and the binding constraint form a system of two equations in two unknowns,
 and  , that makes no reference to the unknown value function .
d) Substitute +1 =  and  = −1 into (42) to obtain
µ ¶ µ ¶ µ ¶
− −−1  1 1 +1 +1
(1 − ) +  − −1 +  −1 = 0 (43)
−1 −1 1+  2
which is now a single equation in a single unknown: the firm’s optimal price  .

10
e) In a steady state where +1 =  = −1 = ∗ for all  = 0 1 2 , (43) implies that

(1 − )(∗ )− + (∗ )−−1 = 0

Use this last equation to find the desired solution for ∗ :

µ ¶
∗ 
 =  (44)
−1

Equation (44) restates a familiar result from microeconomics, which says that an op-
timizing monopolist should set its steady-state markup ∗  equal to ( − 1), where
 measures the absolute value of the elasticity of demand. According to (44) and con-
sistent with intuition, the markup falls as  rises, that is, as demand becomes more
elastic with respect to price changes.

11
 Midterm Exam
Economics 720: Mathematics for Economists
Fall 2007

This exam has five questions on five pages; before you begin, please make sure that your
copy has all five questions and all five pages. The five questions will receive equal weight
in determining your overall exam score. Since this is a 75-minute exam, I suggest that you
spend approximately (75 minutes/5 questions) = 15 minutes on each question.

1 Cost Minimization
A firm produces output y with capital k and labor l according to the Cobb-Douglas specifi-
cation
ka lb ≥ y,
where the parameters a and b both lie, individually, between zero and one and sum to a
number less than one, so that production exhibits decreasing returns to scale: 0 < a < 1,
0 < b < 1, and 0 < a + b < 1.
Suppose that the firm chooses capital and labor inputs in order to minimize the total
cost rk + wl of producing at least ȳ units of output, where r > 0 is the rental rate for capital
and w > 0 is the wage rate for labor. That is, suppose that the firm solves

min rk + wl subject to ka lb ≥ ȳ.

k,l

a) Define (set up) the Lagrangian for this cost minimization problem. In doing this, you
can ignore the nonnegativity constraints k ≥ 0 and l ≥ 0 for the capital and labor
inputs, as the form of the production function implies that these constraints will not
bind at the optimum.

b) Let k∗ and l∗ denote the values of k and l that solve this problem. Write down the full set
of first-order conditions, constraints, conditions on the sign of the Lagrange multiplier,
and complementary slackness conditions that, according to the Kuhn-Tucker theorem,
must be satisfied by k∗ , l∗ , and the associated value of the Lagrange multiplier.

c) Next, use these Kuhn-Tucker conditions to find solutions for k∗ , l∗ , and the associated
value of the Lagrange multiplier in terms of the model’s five parameters: r, w, ȳ, a,
and b. [Note: these solutions for k∗ and l∗ define the firm’s conditional factor demand
curves for capital and labor.]

1
 d) Now define the minimum cost function C(r, w, ȳ) as the minimized value of the objective
function, taken subject to the constraint:

C(r, w, ȳ) = min rk + wl subject to k a lb ≥ ȳ.

k,l

Use your results from above to obtain expressions for the derivatives of the minimum
cost function, C1 (r, w, ȳ), C2 (r, w, ȳ), and C3 (r, w, ȳ), in terms of the parameters r, w,
ȳ, a, and b.

e) What do your results from above tell you about the relationship between the condi-
tional factor demand curves and the derivatives of the minimum cost function? What
economic interpretation does your results provide for the Lagrange multiplier in this
constrained minimization problem?

2 Profit Maximization
Consider the same firm as in question 1 from above, producing output y with capital k and
labor l according to the Cobb-Douglas specification

ka lb ≥ y,

where the parameters a and b both lie, individually, between zero and one and sum to a
number less than one, so that production exhibits decreasing returns to scale: 0 < a < 1,
0 < b < 1, and 0 < a + b < 1.
Assume, as above, that the firm faces the rental rate r > 0 for capital and the wage
rate w > 0 for labor. Now, however, suppose that the firm maximizes profits py − rk − wl,
where p > 0 denotes its output price, subject to the constraint imposed by its production
possibilities. That is, suppose that the firm solves

max py − rk − wl subject to ka lb ≥ y.
y,k,l

a) Define (set up) the Lagrangian for this profit maximization problem. In doing this, you
can ignore the nonnegativity constraints y ≥ 0, k ≥ 0, and l ≥ 0 for output and the
capital and labor inputs, as the set-up of the problem implies these constraints will
not bind at the optimum.

b) Let y ∗ , k∗ , and l∗ denote the values of y, k, and l that solve this problem. Write
down the full set of first-order conditions, constraints, conditions on the sign of the
Lagrange multiplier, and complementary slackness conditions that, according to the
Kuhn-Tucker theorem, must be satisfied by y ∗ , k∗ , l∗ , and the associated value of the
Lagrange multiplier.

c) Next, use these Kuhn-Tucker conditions to find solutions for y ∗ , k∗ , l∗ , and the associated
value of the Lagrange multiplier in terms of the model’s five parameters: p, r, w, a,
and b. [Note: this solution for y ∗ defines the firm’s supply function, and these solutions
for k∗ and l∗ define the firm’s factor demand curves for capital and labor.]

2
 d) Now define the profit function Π(p, r, w) as the maximized value of the objective function,
taken subject to the constraint:
Π(p, r, w) = max py − rk − wl subject to ka lb ≥ y.
y,k,l

Use your results from above to obtain expressions for the derivatives of the profit
function, Π1 (p, r, w), Π2 (p, r, w), and Π3 (p, r, w), in terms of the parameters p, r, w, a,
and b.
e) What do your results from above tell you about the relationship between the supply and
factor demand curves and the derivatives of the profit function?

3 A Dynamic Optimization Problem in Discrete Time

Consider a discrete-time dynamic optimization problem in which the terminal date is T = 2,
so that the time periods are t = 0, t = 1, and t = 2. Let zt , t = 0, 1, 2, denote the value
of a flow variable during each period t, and let yt , t = 0, 1, 2, 3, denote the value of a stock
variable at the beginning of each period t, so that in particular y0 is the initial value of the
stock and y3 is the terminal value of the stock.
Suppose that in this case, the objective function and the constraints cannot be written
in an additively time-separable way, so that the objective function
F (y0 , z0 , y1 , z1 , y2 , z2 )
depends simultaneously on variables from all periods t = 0, 1, 2 and the constraint
c ≥ G(y0 , z0 , y1 , z1 , y2 , z2 , y3 )
depends simultaneously on variables from all periods t = 0, 1, 2 as well as the terminal value
of the stock y3 .
Hence, the problem can be stated formally as: choose values z0 , z1 , and z2 for the flow
variable and values y1 , y2 , and y3 for the stock variable to maximize the objective function
F (y0 , z0 , y1 , z1 , y2 , z2 )
subject to the constraints
c ≥ G(y0 , z0 , y1 , z1 , y2 , z2 , y3 ),
y0 given,
and
y3 ≥ y ∗ ,
where c in the first constraint is a constant parameter and y ∗ represents a lower bound that
is imposed on the terminal value of the stock.
a) Since this problem lacks the additively time-separable structure that is required for an
application of the maximum principle, the best way to solve it is to use the Kuhn-
Tucker theorem. Accordingly, begin by defining (setting up) the Lagrangian for this
problem.

3
 b) Next, write down the set of first-order conditions that must be satisfied by the values
of z0 , z1 , z2 , y1 , y2 , and y3 that solve the problem, together with the associated values
of the multipliers that you introduced into the problem above, when writing down the
Lagrangian.
c) Write down the constraints that must be satisfied by the values of z0 , z1 , z2 , y1 , y2 , and
y3 that solve the problem.
d) Write down the nonnegativity or nonpositivity constraints that must be satisfied by the
values of the Lagrange multipliers.
e) Write down the complementary slackness conditions associated with each constraint and
its Lagrange multiplier.

4 The Maximum Principle in Discrete Time

Consider the same fairly general discrete-time dynamic optimization problem that we studied
in class. The horizon is finite, so that periods are indexed by t = 0, 1, ..., T . There is one
stock variable yt and one flow variable zt . Thus, sequences {zt }Tt=0 and {yt }Tt=1+1
must be
chosen to maximize the objective function
X
T
β t F (yt , zt ; t),
t=0

with 1 ≥ β > 0, subject to the constraints

yt + Q(yt , zt ; t) ≥ yt+1 for all t = 0, 1, ..., T,
c ≥ G(yt , zt ; t) for all t = 0, 1, ..., T,
y0 given,
and
yT +1 ≥ y ∗ .
a) Define the Hamiltonian for this problem.
b) Write down the first-order condition for the static optimization problem that appears in
the definition of the Hamiltonian.
c) Write down the pair of difference equations that, according to the Maximum Principle,
must be satisfied by the solution to the original, dynamic optimization problem (here,
as in class, you can assume that the constraint governing the evolution of the stock
variable always binds, as it will in virtually all economic applications).
d) Write down a fourth equation that, together with the three equations you derived to in
parts (b) and (c) above, can be used to help find the sequences {zt }Tt=0 and {yt }Tt=1
+1

that solve the original dynamic optimization problem.

e) Write down the pair of boundary conditions that, according to the Maximum Principle,
must also be satisfied by the solution to the original, dynamic optimization problem.

4
5 Optimal Growth in Discrete Time
Consider an infinite-horizon economy in which output is produced with capital during each
period t = 0, 1, 2, ... according to the production function
yt = F (kt ) = ktα ,
where the parameter α lies between zero and one: 0 < α < 1. Let a representative house-
hold’s consumption during each period t = 0, 1, 2, ... be denoted by ct , and let the parameter
δ, with 0 < δ ≤ 1, denote the rate at which physical capital depreciates between periods.
Then, the capital stock kt+1 at the beginning of period t + 1 equals the capital stock at the
beginning of period t, plus the amount of new output ktα produced during period t, less the
amount of capital δkt that depreciates away during period t, less the amount of output ct
consumed during period t; or, allowing for the free disposal of capital:
kt + ktα − δkt − ct ≥ kt+1
for all t = 0, 1, 2, ....
In this economy, a benevolent social planner takes the initial capital stock k0 as given,
and chooses sequences {ct }∞ ∞
t=0 and {kt }t=1 to maximize the representative household’s utility,
given by the additively-time separable and logarithmic specification
X∞
β t ln(ct ),
t=0

subject to the constraints

kt + ktα − δkt − ct ≥ kt+1
for all t = 0, 1, 2, ... and
k0 given.
a) As a first step in solving this problem using the maximum principle, begin by defining
(setting up) the Hamiltonian.
b) Next, write down the first-order condition for the static optimization problem that
appears in the definition of the Hamiltonian.
c) Write down the pair of difference equations that, according to the Maximum Principle,
must be satisfied by the solution to the social planner’s original, dynamic optimization
problem.
d) Now assume that δ = 1, so that the capital stock depreciates fully between periods, and
use this assumption to simplify the three equations that you derived in parts (b) and
(c) above.
e) Your three equations from part (d) above should involve three unknowns: the values
of ct and kt that solve the social planner’s problem and the value of the additional
variable that you introduced into the problem in part (a) from above when setting up
the Hamiltonian. Use one of the equations as a solution for the additional variable,
and rewrite the system as one involving just two equations in the two variables with
direct economic interpretations: ct and kt .

5
 Solutions to Midterm Exam
Economics 720: Mathematics for Economists
Fall 2007

1 Cost Minimization
The firm solves
min rk + wl subject to ka lb ≥ ȳ.
k,l

a) There are a number of ways to define the Lagrangian for this cost minimization problem,
but one way is to subtract the product of the multiplier λ and the constraint from the
objective function to obtain

L(k, l, λ) = rk + wl − λ(ka lb − ȳ).

b) With the Lagrangian defined as above, the Kuhn-Tucker theorem states that if k ∗ and
l∗ solve the firm’s problem, then there exists a value λ∗ of λ such that, together, k∗ ,
l∗ , and λ∗ satisfy the first-order conditions

L1 (k ∗ , l∗ , λ∗ ) = r − aλ∗ (k ∗ )a−1 (l∗ )b = 0 (1)

and
L2 (k ∗ , l∗ , λ∗ ) = w − bλ∗ (k ∗ )a (l∗ )b−1 = 0, (2)
the constraint
(k∗ )a (l∗ )b ≥ ȳ, (3)
the nonnegativity condition
λ∗ ≥ 0, (4)
and the complementary slackness condition

λ∗ [(k∗ )a (l∗ )b − ȳ] = 0. (5)

c) Again, there are many ways of using these Kuhn-Tucker conditions to find solutions for
k∗ , l∗ , and λ∗ in terms of the parameters r, w, ȳ, a, and b, but one way starts by
rearranging the first-order condition (1) to read
r
λ∗ = . (6)
a(k∗ )a−1 (l∗ )b

1
 Since the objects in both the numerator and the denominator on the right-hand side
of (6) are all strictly positive, the multiplier λ∗ must also be strictly positive; the
complementary slackness condition (5) then implies that the constraint (3) must bind
at the optimum. Now use (6) to substitute out for λ∗ in (2) and rearrange to obtain
µ ¶
∗ br
l = k∗ . (7)
aw
Substitute this last result into the binding constraint to obtain the desired solution for
k∗ : ³ aw ´b/(a+b)
k∗ = (ȳ)1/(a+b) . (8)
br
Substitute this expression back into (7) to obtain the desired solution for l∗ :
µ ¶a/(a+b)
∗ br
l = (ȳ)1/(a+b) . (9)
aw
Finally, substitute (8) and (9) back into (6) to obtain the desired solution for λ∗ :
³ r ´a/(a+b) ³ w ´b/(a+b)
λ∗ = (ȳ)(1−a−b)/(a+b) . (10)
a b
d) Now define the minimum cost function C(r, w, ȳ) as
C(r, w, ȳ) = min rk + wl subject to k a lb ≥ ȳ.
k,l

Related, let (8)-(10) define the functions k∗ (r, w, ȳ), l∗ (r, w, ȳ), and λ∗ (r, w, ȳ) describ-
ing how the optimal capital and labor inputs and the associated value of the Lagrange
multiplier depend on the parameters r, w, and ȳ. Then, in light of the complementary
slackness condition (5), the minimum cost function can be evaluated as
C(r, w, ȳ) = rk ∗ (r, w, ȳ) + wl∗ (r, w, ȳ) − λ∗ (r, w, ȳ){[k ∗ (r, w, ȳ)]a [l∗ (r, w, ȳ)]b − ȳ}.
The envelope theorem implies that in differentiating both sides of this last expression,
it is safe to ignore the dependence of k∗ , l∗ , and λ∗ on r, w, and ȳ and write
³ aw ´b/(a+b)
C1 (r, w, ȳ) = k∗ (r, w, ȳ) = (ȳ)1/(a+b) , (11)
br
µ ¶a/(a+b)
∗ br
C2 (r, w, ȳ) = l (r, w, ȳ) = (ȳ)1/(a+b) , (12)
aw
and ³ r ´a/(a+b) ³ w ´b/(a+b)
∗
C3 (r, w, ȳ) = λ (r, w, ȳ) = (ȳ)(1−a−b)/(a+b) . (13)
a b
e) Equations (11) and (12) reveal that the derivatives of the minimum cost function with
respect to the rental rate for capital and the wage rate for labor coincide with the
conditional factor demand curves for capital and labor: they provide a statement of
Shephard’s lemma. Equation (13) provides an interpretation of the Lagrange multiplier
as a measure of marginal cost at the optimum.

2
2 Profit Maximization
Now the firm solves
max py − rk − wl subject to ka lb ≥ y.
y,k,l

a) Once again there are a number of ways to define the Lagrangian for this profit maximiza-
tion problem, but one way is to add the product of the multiplier λ and the constraint
to the objective function to obtain

L(y, k, l, λ) = py − rk − wl + λ(ka lb − y).

b) With the Lagrangian defined as above, the Kuhn-Tucker theorem states that if y ∗ , k∗ ,
and l∗ solve the firm’s problem, then there exists a value λ∗ of λ such that, together,
y ∗ , k∗ , l∗ and λ∗ satisfy the first-order conditions

L1 (y ∗ , k∗ , l∗ , λ∗ ) = p − λ∗ = 0. (14)

L2 (y ∗ , k∗ , l∗ , λ∗ ) = −r + aλ∗ (k∗ )a−1 (l∗ )b = 0 (15)

and
L3 (y ∗ , k∗ , l∗ , λ∗ ) = −w + bλ∗ (k∗ )a (l∗ )b−1 = 0 (16)
the constraint
(k∗ )a (l∗ )b ≥ y ∗ , (17)
the nonnegativity condition
λ∗ ≥ 0, (18)
and the complementary slackness condition

λ∗ [(k∗ )a (l∗ )b − y ∗ ] = 0. (19)

c) Again, there are many ways of using these Kuhn-Tucker conditions to find solutions for
y ∗ , k∗ , l∗ , and λ∗ in terms of the model’s five parameters: p, r, w, a, and b, but one
way starts by observing from (14) that

λ∗ = p, (20)

which provides the desired solution for λ∗ . Since the output price p is strictly positive,
(20) implies that λ∗ is also strictly positive; the complementary slackness condition
(19) then implies that the constraint (17) must bind at the optimum. Use (20) to
substitute out for λ∗ in (15), and rearrange to obtain
µ ¶1/b
∗ r
l = (k∗ )(1−a)/b . (21)
ap
Substitute this last result into (16) and use (14) again to obtain
³ ´(1−b)/(1−a−b) µ b ¶b/(1−a−b)
∗ 1/(1−a−b) a
k =p , (22)
r w

3
 which provides the desired solution for k ∗ . Substitute (22) back into (21) to obtain the
desired solution for l∗ :
³ ´a/(1−a−b) µ b ¶(1−a)/(1−a−b)
∗ 1/(1−a−b) a
l =p . (23)
r w

Finally, substitute (22) and (23) into the binding constraint to obtain the desired
solution for y ∗ :
³ ´a/(1−a−b) µ b ¶b/(1−a−b)
∗ (a+b)/(1−a−b) a
y =p . (24)
r w

d) Now define the profit function Π(p, r, w) as

Π(p, r, w) = max py − rk − wl subject to ka lb ≥ y.

y,k,l

Related, let (20) and (22)-(24) define the functions y ∗ (p, r, w), k∗ (p, r, w), l∗ (p, r, w),
and λ∗ (p, r, w) describing how the optimal level of output, the optimal capital and labor
inputs, and the associated value of the Lagrange multiplier depend on the parameters
p, r, and w. Then, in light of the complementary slackness condition (19), the profit
function can be evaluated as

Π(p, r, w) = py ∗ (p, r, w) − rk∗ (p, r, w) − wl∗ (p, r, w)

+λ∗ (p, r, w){[k ∗ (p, r, w)]a [l∗ (p, r, w)]b − y ∗ (p, r, w)}.

The envelope theorem implies that in differentiating both sides of this last expression,
it is safe to ignore the dependence of y ∗ , k ∗ , l∗ , and λ∗ on p, r, and w and write
³ a ´a/(1−a−b) µ b ¶b/(1−a−b)
∗ (a+b)/(1−a−b)
Π1 (p, r, w) = y (p, r, w) = p , (25)
r w
³ a ´(1−b)/(1−a−b) µ b ¶b/(1−a−b)
∗ 1/(1−a−b)
Π2 (p, r, w) = −k (p, r, w) = −p , (26)
r w
and
³ a ´a/(1−a−b) µ b ¶(1−a)/(1−a−b)
∗ 1/(1−a−b)
Π3 (p, r, w) = −l (p, r, w) = −p . (27)
r w

e) Equations (25)-(27) provide a statement of Hotelling’s lemma. According to (25), the

derivative of the profit function with respect to the output price p coincides with the
supply function. According to (26) and (27), the derivative of the profit function with
respect to the two factor prices r and w coincide with the factor demand curves for
capital and labor, multiplied by -1, since an increase in these factor prices leads to a
decrease in profits.

4
3 A Dynamic Optimization Problem in Discrete Time
The problem can be stated as
max F (y0 , z0 , y1 , z1 , y2 , z2 )
z0 ,z1 ,z2 ,y1 ,y2 ,y3
subject to c ≥ G(y0 , z0 , y1 , z1 , y2 , z2 , y3 ), y0 given, and y3 ≥ y ∗ .
a) There are a number of ways to define the Lagrangian for this constrained maximization
problem, but one way is as
L(z0 , z1 , z2 , y1 , y2 , y3 , λ, φ) = F (y0 , z0 , y1 , z1 , y2 , z2 )
+λ[c − G(y0 , z0 , y1 , z1 , y2 , z2 , y3 )] + φ(y3 − y ∗ ).

b) Since there are six choice variables, there are six first-order conditions, each obtained by
differentiating the Lagrangian through by one of the choice variables and equating to
zero:
L1 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = F2 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ )−λ∗ G2 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) = 0,
L2 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = F4 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ )−λ∗ G4 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) = 0,
L3 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = F6 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ )−λ∗ G6 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) = 0,
L4 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = F3 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ )−λ∗ G3 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) = 0,
L5 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = F5 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ )−λ∗ G5 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) = 0,
and
L6 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = −λ∗ G7 (y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) + φ∗ = 0

c) The optimal choices must also satisfy the constraints

L7 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = c − G(y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ ) ≥ 0,
L8 (z0∗ , z1∗ , z2∗ , y1∗ , y2∗ , y3∗ , λ∗ , φ∗ ) = y3∗ − y ∗ ≥ 0,
and
y0 given.
d) And with the Lagrangian defined as above, the Kuhn-Tucker theorem implies that the
multipliers must satisfy the nonnegativity constraints
λ∗ ≥ 0
and
φ∗ ≥ 0.
e) Finally, the Kuhn-Tucker theorem implies that the complementary slackness conditions
λ∗ [c − G(y0 , z0∗ , y1∗ , z1∗ , y2∗ , z2∗ , y3∗ )] = 0
and
φ∗ (y3∗ − y ∗ ) = 0
hold at the optimum.

5
4 The Maximum Principle in Discrete Time
a) Define the Hamiltonian as

H(yt , πt+1 ; t) = max β t F (yt , zt ; t) + π t+1 Q(yt , zt ; t) subject to c ≥ G(yt , zt ; t).

zt

b) By the Kuhn-Tucker theorem,

H(yt , πt+1 ; t) = max β t F (yt , zt ; t) + π t+1 Q(yt , zt ; t) + λt [c − G(yt , zt ; t)],

zt

where zt satisfies the first-order condition

β t Fz (yt , zt ; t) + πt+1 Qz (yt , zt ; t) − λt Gz (yt , zt ; t) = 0. (28)

c) According to the Maximum Principle, the solution to the original dynamic optimization
problem must satisfy the pair of difference equations

π t+1 − π t = −Hy (yt , π t+1 ; t) = −[β t Fy (yt , zt ; t) + π t+1 Qy (yt , zt ; t) − λt Gy (yt , zt ; t)] (29)

and
yt+1 − yt = Hπ (yt , π t+1 ; t) = Q(yt , zt ; t). (30)

d) According to the Kuhn-Tucker theorem, the solution to the original dynamic optimization
problem must also satisfy the complementary slackness condition

λt [c − G(yt , zt ; t)] = 0. (31)

Equations (28)-(31) form a system of four equations in four unknowns, that helps to
identify the values of yt and zt that solve the original dynamic optimization problem
as well as the associated values of the multipliers πt+1 and λt .

e) According to the Maximum Principle, the solution to the original dynamic optimization
problem must also satisfy two boundary conditions: the initial condition

y0 given

and the terminal, or transversality, condition

πT +1 (yT +1 − y ∗ ) = 0.

5 Optimal Growth in Discrete Time

a) Define the Hamiltonian as

H(kt , π t+1 ; t) = max β t ln(ct ) + π t+1 (ktα − δkt − ct ).

ct

6
b) By the Kuhn-Tucker theorem, ct must satisfy the first-order condition
βt
− π t+1 = 0. (32)
ct

c) According to the Maximum Principle, the solution to the social planner’s original, dy-
namic optimization problem must satisfy the pair of difference equations
πt+1 − πt = −Hk (kt , π t+1 ; t) = −π t+1 (αktα−1 − δ) (33)
and
kt+1 − kt = Hπ (kt , π t+1 ; t) = ktα − δkt − ct . (34)
d) When δ = 1, (32) remains as before, but (33) and (34) simplify to
π t = απ t+1 ktα−1 (35)
and
kt+1 = ktα − ct . (36)
e) A potential problem with the three-equation system formed by (32), (35), and (36) is that
it involves the variable π t+1 , which lacks an immediate economic interpretation. To
eliminate this variable from the system, use the fact that, according to the maximum
principle, (32) must hold for all t = 0, 1, 2, ... to eliminate
βt
π t+1 =
ct
and
β t−1
πt =
ct−1
from (35), so that this equation becomes
ct
= αβktα−1 . (37)
ct−1
Equations (36) and (37) now form a system of two equations in two unknowns: the
values of ct and kt that solve the social planner’s problem. In terms of the economics,
(36) just says that when capital depreciates completely between periods, the capital
stock kt+1 at the beginning of period t + 1 consists entirely of output ktα − ct that is
not consumed during period t. Equation (37) can be rearranged slightly to read
ct
= αktα−1 .
βct−1
The left-hand side of this last expression measures the intertemporal marginal rate of
substitution, that is, the marginal rate of substitution between consumption during
periods t and t − 1. The right-hand side of this last expression measures the marginal
product of capital. Hence, (37) expresses the intertemporal efficiency condition that
the intertemporal marginal rate of substitution ought to equal the marginal product
of capital.

7
 Final Exam
Economics 720: Mathematics for Economists
Fall 2007

This exam has five questions on six pages; before you begin, please make sure that your
copy has all five questions and all six pages. The five questions will receive equal weight in
determining your overall exam score. Since this is a 120-minute exam, I suggest that you
spend approximately (120 minutes/5 questions) = 24 minutes on each question.

1 The Kuhn-Tucker Theorem

A consumer likes two goods: apples and bananas. Let a denote his or her consumption of
apples; let b denote his or her consumption of bananas; and let the function U(a, b) measure
the utility that the consumer obtains from eating apples and bananas. Assume that U(a, b)
is strictly increasing in both its arguments, so that Ua (a, b) > 0 and Ub (a, b) > 0 for all values
of a ≥ 0 and b ≥ 0, where Ua and Ub denote derivatives of U with respect to its first and
second arguments. Suppose also that the consumer can purchase as many or as few apples
and bananas as he or she likes in perfectly competitive markets at the prices pa for apples
and pb for bananas. Finally, let I denote the consumer’s income. Now the consumer’s static
utility maximization problem can be stated formally as

max U(a, b) subject to I ≥ pa a + pb b, a ≥ 0, and b ≥ 0,

a,b

where the last two constraints specifically allow for the possibility that the optimizing con-
sumer can be at a corner solution, consuming either no apples or no bananas.

a) Define (write down) the Lagrangian for this problem, allowing for the possibility that
the nonnegativity constraint on a or b might bind at the optimum.

b) Write down the first-order conditions for the consumer’s problem, again allowing for
the possibility that the nonnegativity constraint on a or b might bind at the opti-
mum. NOTE: Your first-order conditions may depend on precisely how you defined
the Lagrangian for the problem in part (a), above.

c) Write down the full set of constraints that must be satisfied by the optimal choices of
a and b, as well as any additional conditions on the sign (positive or negative) of the
Lagrange multiplier or multipliers that you introduced into the problem in part (a)
above.

1
 d) Write down the complementary slackness conditions that must be satisfied at the opti-
mum.

e) Use your results from above to answer the following question: under what circumstances
will the the optimizing consumer’s marginal rate of substitution Ua (a∗ , b∗ )/Ub (a∗ , b∗ )
between apples and bananas NOT equal the relative price pa /pb of apples and bananas?

2 The Maximum Principle in Continuous Time

Consider the continuous-time, dynamic optimization problem of choosing continuously dif-
ferentiable functions z(t) for t ∈ [0, T ] to describe the evolution of a flow variable and y(t)
for t ∈ (0, T ] to describe the evolution of a stock variable to maximize the objective function
Z T
e−ρt F (y(t), z(t); t)dt,
0

with ρ ≥ 0, subject to the constraints

Q(y(t), z(t); t) ≥ ẏ(t) for all t ∈ [0, T ],

c ≥ G(y(t), z(t); t) for all t ∈ [0, T ],

y(0) given
and
y(T ) ≥ y ∗ .

a) Define the Hamiltonian for this problem.

b) Write down the first-order condition for the static optimization problem that appears in
the definition of the Hamiltonian.

c) Write down the pair of differential equations that, according to the maximum principle,
must be satisfied by the solution to the original, dynamic optimization problem (here,
as in class, you can assume that the constraint governing the evolution of the stock
variable always binds, as it will in virtually all economic applications).

d) Write down a fourth equation that, together with the three equations you derived to in
parts (b) and (c) above, can be used to help find the values of y(t) and z(t) that solve
the original dynamic optimization problem.

e) Write down the pair of boundary conditions that, according to the maximum principle,
must also be satisfied by the solution to the original, dynamic optimization problem.

2
3 Optimal Growth in Continuous Time
Consider an infinite-horizon economy in which output is produced with capital according to
the production function
F (k(t)) = k(t)α ,
where 0 < α < 1. Let c(t) denote consumption and let δ ≥ 0 denote the depreciation rate
for capital. Then the capital stock evolves according to

k(t)α − δk(t) − c(t) ≥ k̇(t)

for all t ∈ [0, ∞). Then, subject to this constraint and taking the initial value k(0) for
the capital stock as given, a benevolent social planner chooses the continuously differential
functions c(t) for all t ∈ [0, ∞) and k(t) for all t ∈ (0, ∞) to maximize a representative
consumer’s utility, given by the additively time-separable and logarithmic specification
Z ∞
e−ρt ln(c(t))dt,
0

with ρ ≥ 0.

a) As a first step in solving this problem using the maximum principle, begin by defining
(setting up) the Hamiltonian.
b) Next, write down the first-order condition for the static optimization problem that
appears in the definition of the Hamiltonian.

c) Write down the pair of difference equations that, according to the maximum principle,
must be satisfied by the solution to the social planner’s original, dynamic optimization
problem.
d) Your three equations from parts (b) and (c) above should involve three unknowns: the
values of c(t) and k(t) that solve the social planner’s problem and the value of the
additional variable that you introduced into the problem in part (a) from above when
setting up the Hamiltonian. Rewrite this system as one involving just two equations
in the two variables with direct economic interpretations: c(t) and k(t).
e) Your equations from part (d) above imply that the economy has a unique, nontrivial
steady state in which c(t) = c∗ > 0 and k(t) = k∗ > 0 for all t ∈ [0, ∞). Use those
equations to solve for the steady-state values c∗ and k ∗ of consumption and capital in
terms of the model’s parameters α, δ, and ρ.

4 Optimal Resource Depletion

Consider a discrete-time, infinite horizon model that characterizes the optimal consumption
of an exhaustible resource (like oil or coal). Let time periods be indexed by t = 0, 1, 2, ...;
and let xt , t = 0, 1, 2, ..., denote the stock of the exhaustible resource that remains at the
beginning of period t. Let ct , t = 0, 1, 2, ..., denote the amount of this resource that is

3
consumed during period t. Since no new units of the resource are ever created, the amount
consumed simply subtracts from the available stock according to
xt − ct ≥ xt+1
for all t = 0, 1, 2, ..., where the inequality constraint (which you can assume will always
bind at the optimum) simply recognizes that the resource can be freely disposed of. The
optimization problem then involves choosing sequences {ct }∞ ∞
t=0 and {xt }t=1 to maximize
utility from consuming the resource over the infinite horizon, given by
X
∞
β t ln(ct ),
t=0

where the discount factor lies between zero and one, 0 < β < 1, subject to the constraint
xt − ct ≥ xt+1 for all t = 0, 1, 2, ..., taking as given the level of the initial resource stock
x0 > 0.
a) Define (write down) the Lagrangian for this dynamic optimization problem. Then, write
down the first-order conditions and constraints that, according to the Kuhn-Tucker
theorem, characterize the solution to the dynamic optimization problem.
b) Define (write down) the Hamiltonian for this dynamic optimization problem. Then, write
down the first-order condition and the pair of difference equations that, according to
the maximum principle, characterize the solution to the dynamic optimization problem.
c) Write down the Bellman equation for this dynamic optimization problem. Then, write
down the first-order condition, the envelope condition, and the constraint that, accord-
ing to the dynamic programming approach, characterize the solution to the dynamic
optimization problem.
d) In each of parts (a), (b), and (c) from above, you should have derived a system of three
equations in three unknowns: the values of ct and xt that solve the optimal resource
depletion problem and the value of a third unknown variable or function that you
introduced into the problem when setting up the Lagrangian, Hamiltonian, or Bellman
equation. Rewrite this three-equation system as one involving just two equations in
the two variables with direct economic interpretations: ct and xt .
e) Use your answers from part (d) above to answer the following question: is the optimal
consumption ct of the exhaustible resource rising, falling, or staying constant over time?

5 Saving with a Random Return

Consider the following dynamic, stochastic optimization problem in discrete time. An
infinitely-lived representative consumer enters each period t = 0, 1, 2, ... with a stock of
financial wealth denoted by At . During period t, the consumer divides this wealth up into
an amount ct to be consumed and an amount st to be saved, subject to the constraint
At = ct + st ,

4
which must hold for all t = 0, 1, 2, ....
The gross return Rt+1 on savings between t and t + 1 is random and, more specifically,
follows the first-order autoregressive process

ln(Rt+1 ) = ρ ln(Rt ) + εt+1

where 0 < ρ < 1 and εt+1 is an independently and identically distributed shock that satisfies
Et εt+1 = 0 for all t = 0, 1, 2, .... Thus, last period’s return Rt is known when the consumer
chooses st , but Rt+1 is still viewed as random. Given the consumer’s choice of st and the
realized value of Rt+1 , the consumer’s financial wealth At+1 is determined by the constraint

Rt+1 st ≥ At+1 ,

which must hold for all t = 0, 1, 2, ... and for all possible realizations of Rt+1 .
The representative consumer seeks to maximize expected utility
X
∞
E0 β t ln(ct ),
t=0

where the discount factor satisfies 0 < β < 1 and where, as indicated, the single-period
utility function takes the natural logarithmic form. As a first step in solving the consumer’s
problem, it is convenient to substitute the constraint At = ct + st into the utility function to
obtain
X
∞ X
∞
E0 β t ln(ct ) = E0 β t ln(At − st ).
t=0 t=0

The consumer’s problem can now be stated formally as: choose contingency plans for st ,
t = 0, 1, 2, ..., and At , t = 1, 2, 3, ..., to maximize the expected utility function
X
∞
E0 β t ln(At − st )
t=0

subject to the constraints

A0 given
and
Rt+1 st ≥ At+1
for all t = 0, 1, 2, ... and all possible realizations of Rt+1 .

a) Write down the Bellman equation for the household’s problem. In doing so, you might
find it helpful to view At as the state variable, st as the control variable, and Rt
as the random variable entering into the problem. You might also find it helpful to
recognize that since the consumer’s utility function is strictly increasing, the constraint
Rt+1 st ≥ At+1 will always bind at the optimum.

5
b) Now guess that the value function for this problem takes the form

v(At , Rt ) = F ln(At ) + G ln(Rt ) + H,

where F , G, and H are unknown constants. Substitute this guess into your Bellman
equation from above, and then write down the first-order condition for st and the
envelope condition for At . In deriving these optimality conditions, you might find it
helpful to note that since st is known at time t,

Et ln(st ) = ln(st ).

You might also find it helpful to note that since

ln(Rt+1 ) = ρ ln(Rt ) + εt+1 ,

with Et εt+1 = 0, it follows that

Et ln(Rt+1 ) = ρ ln(Rt ).

c) Use your first-order condition from above to derive an expression linking the consumer’s
optimal choice st to the state variable At , the discount factor β, and the unknown
constant F .

d) Next, use your envelope condition from above to solve for the unknown constant F in
terms of the discount factor β.

e) Combine your results from above to obtain a set of two equations that can be used to
construct the optimal paths for st and At , given the initial value of A0 and the realized
path for Rt+1 . NOTE: It should not be necessary to solve for the unknown constants
G and H to derive these equations and, since solving for G and H requires some extra
algebra, DON’T BOTHER TO DO THIS.

6
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2007

1 The Kuhn-Tucker Theorem

The consumer solves

max U(a, b) subject to I ≥ pa a + pb b, a ≥ 0, and b ≥ 0.

a,b

a) There are a number of ways to define the Lagrangian for this problem, but one way is to
treat the nonnegativity constraints symmetrically with the budget constraint and let

L(a, b, λ, μ, φ) = U(a, b) + λ(I − pa a − pb b) + μa + φb.

b) With the Lagragian defined as above, the first order conditions are

L1 (a∗ , b∗ , λ∗ , μ∗ , φ∗ ) = Ua (a∗ , b∗ ) − λ∗ pa + μ∗ = 0 (1)

and
L2 (a∗ , b∗ , λ∗ , μ∗ , φ∗ ) = Ub (a∗ , b∗ ) − λ∗ pb + φ∗ = 0 (2)

c) The Kuhn-Tucker theorem implies that in addition to the two first-order conditions stated
above, the values a∗ and b∗ that solve the consumer’s problem and the associated values
λ∗ , μ∗ , and φ∗ of the multipliers must satisfy the constraints

I ≥ pa a∗ + pb b∗ ,

a∗ ≥ 0,
and
b∗ ≥ 0
and the nonnegativity conditions
λ∗ ≥ 0,
μ∗ ≥ 0,
and
φ∗ ≥ 0.

1
 d) The Kuhn-Tucker theorem also implies that the a∗ , b∗ , λ∗ , μ∗ , and φ∗ must satisfy the
complementary slackness conditions

λ∗ (I − pa a∗ − pb b∗ ) = 0,

μ∗ a∗ = 0, (3)
and
φ∗ b∗ = 0. (4)

e) The first-order conditions (1) and (2) imply that

Ua (a∗ , b∗ ) λ∗ pa − μ∗
= .
Ub (a∗ , b∗ ) λ∗ pb − φ∗

The right-hand side of this expression fails to equal pa /pb if either μ∗ > 0 or φ∗ > 0;
the complementary slackness conditions (3) and (4) imply that one of these multipliers
will be strictly positive if either a∗ = 0 or b∗ = 0. Hence, if the consumer is at a
corner solution, consuming either no apples or no bananas, then the marginal rate
of substitution Ua (a∗ , b∗ )/Ub (a∗ , b∗ ) between apples and bananas will not equal the
relative price pa /pb of apples and bananas.

2 The Maximum Principle in Continuous Time

The problem involves choosing continuously differentiable functions z(t) for t ∈ [0, T ] and
y(t) for t ∈ (0, T ] to maximize
Z T
e−ρt F (y(t), z(t); t)dt
0

subject to the constraints

Q(y(t), z(t); t) ≥ ẏ(t) for all t ∈ [0, T ],

c ≥ G(y(t), z(t); t) for all t ∈ [0, T ],

y(0) given
and
y(T ) ≥ y ∗ .

a) Again, there are a number of different ways to set up the Hamiltonian for this problem,
but one definition is

H(y(t), π(t); t) = max e−ρt F (y(t), z(t); t)+π(t)Q(y(t), z(t); t) subject to c ≥ G(y(t), z(t); t).
z(t)

2
 b) Letting λ(t) denote the multiplier on the constraint that appears in the static optimiza-
tion problem that appears in the definition of the Hamiltonian, the first-order condition
for the value of z(t) that solves this problem can be written as

e−ρt Fz (y(t), z(t); t) + π(t)Qz (y(t), z(t); t) − λ(t)Gz (y(t), z(t); t) = 0. (5)

The maximum principle implies that this same first-order condition must also be sat-
isfied by the value of z(t) that solves the original, dynamic, optimization problem.
c) The maximum principle also implies that the values of z(t) and y(t) that solve the
dynamic optimization problem must satisfy the pair of differential equations

π̇(t) = −Hy (y(t), π(t); t) (6)

= −[e−ρt Fy (y(t), z(t); t) + π(t)Qy (y(t), z(t); t) − λ(t)Gy (y(t), z(t); t)]

and
ẏ(t) = Hπ (y(t), π(t); t) = Q(y(t), z(t); t), (7)
where the second equality in each of these expressions follows from an application of
the envelope theorem to the Hamiltonian.
d) Together with the complementary slackness condition

λ(t)[c − G(y(t), z(t); t)] = 0,

(5)-(7) form a system that can be used to solve for the values of z(t) and y(t) that
solve the dynamic optimization problem and the associated values of π(t) and λ(t).
e) According to the maximum principle, the solution to the dynamic optimization problem
must also satisfy two boundary conditions: the initial condition

y(0) given

and the terminal, or transversality, condition

π(T )[y(T ) − y ∗ ] = 0.

3 Optimal Growth in Continuous Time

The problem involves choosing continuously differential functions c(t) for all t ∈ [0, ∞) and
k(t) for all t ∈ (0, ∞) to maximize
Z ∞
e−ρt ln(c(t))dt
0

subject to the constraints k(0) given and

k(t)α − δk(t) − c(t) ≥ k̇(t)

for all t ∈ [0, ∞).

3
a) The Hamiltonian for this problem can be defined as
H(k(t), π(t); t) = max e−ρt ln(c(t)) + π(t)[k(t)α − δk(t) − c(t)].
c(t)

b) The first-order condition for the value of c(t) that solves the static optimization problem
that appears in the definition of the Hamiltonian can be written as
e−ρt
− π(t) = 0. (8)
c(t)
The maximum principle implies that this same first-order condition must also be sat-
isfied by the value of c(t) that solves the original, dynamic, optimization problem.
c) The maximum principle also implies that the values of c(t) and k(t) that solve the
dynamic optimization problem must satisfy the pair of differential equations
π̇(t) = −Hk (k(t), π(t); t) = −π(t)[αk(t)α−1 − δ] (9)
and
k̇(t) = Hπ (k(t), π(t); t) = k(t)α − δk(t) − c(t), (10)
where the second equality in each of these expressions follows from an application of
the envelope theorem to the Hamiltonian.
d) A problem with the three-equation system (8)-(10) is that it involves not just the variables
c(t) and k(t) from the original social planner’s problem but also the new variable
π(t) that lacks a straightforward economic interpretation. To eliminate π(t) from the
system, rewrite (8) as
e−ρt = π(t)c(t)
and differentiate both sides to obtain
−ρe−ρt = π(t)ċ(t) + π̇(t)c(t).
Now use (8) and (9) again to rewrite this last expression as
−ρπ(t)c(t) = π(t)ċ(t) − π(t)[αk(t)α−1 − δ]c(t).
Divide both sides of this expression by π(t) and rearrange to obtain
ċ(t) = [αk(t)α−1 − δ − ρ]c(t). (11)
Together, (10) and (11) now form a two-equation system involving only c(t) and k(t).
e) In a steady-state, ċ(t) = 0. Hence, (11) requires that
µ ¶1/(α−1)
∗ δ+ρ
k(t) = k = .
α
Since k̇(t) = 0 also holds in steady state, (10) implies that
µ ¶α/(α−1) µ ¶1/(α−1)
∗ ∗ α ∗ δ+ρ δ+ρ
c(t) = c = (k ) − δk = −δ .
α α

4
4 Optimal Resource Depletion
The problem involves choosing sequences {ct }∞ ∞
t=0 and {xt }t=1 to maximize

X
∞
β t ln(ct )
t=0

subject to the constraints x0 given and xt − ct ≥ xt+1 for all t = 0, 1, 2, ....

a) With the Lagrangian defined as

X
∞ X
∞
t
L= β ln(ct ) + π t+1 (xt − ct − xt+1 ),
t=0 t=0

the first-order condition for ct is

βt
− π t+1 = 0 (12)
ct
and the first-order condition for xt is

π t+1 − π t = 0. (13)

Together with the binding constraint

xt+1 = xt − ct , (14)

(12) and (13) form a system of three equations in the three unknowns: the values of
ct and xt that solve the dynamic optimization problem and the associated value of π t .

b) With the Hamiltonian defined as

H(xt , π t+1 ; t) = max β t ln(ct ) − π t+1 ct ,

ct

the first-order condition for the static problem,

βt
− π t+1 = 0, (15)
ct
and the pair of difference equations

π t+1 − π t = −Hx (xt , π t+1 ; t) = 0 (16)

and
xt+1 − xt = Hπ (xt , π t+1 ; t) = −ct (17)
form a system of three equations in ct , xt , and πt that coincides, of course, with (12)-
(14).

5
c) The Bellman equation for the problem is

v(xt ; t) = max ln(ct ) + βv(xt − ct ; t + 1).

ct

The first-order condition for the value of ct that solves the static and unconstrained
optimization problem on the right-hand side of the Bellman equation is
1
− βv 0 (xt − ct ; t + 1) = 0. (18)
ct
The envelope condition for xt is

v0 (xt ; t) = βv 0 (xt − ct ; t + 1). (19)

And the binding constraint is

xt+1 = xt − ct . (20)

d) We know from our general analysis that (12)-(14), (15)-(17), and (18)-(20) provide
identical information about the solution to the original dynamic optimization problem,
so that we can work with whichever system is most convenient. Consider (18)-(20).
Using (20), (18) can be written more simply as
1
= βv 0 (xt+1 ; t + 1).
ct
Since this condition must hold for all t = 0, 1, 2, ..., it also implies that
1
= βv 0 (xt ; t).
ct−1

Substituting (20) into (19) yields

v 0 (xt ; t) = βv0 (xt+1 ; t + 1)

and hence
1 1
=
βct−1 ct
or more simply
ct = βct−1 . (21)
Equations (20) and (21) form a system of two equations the two unknowns: the values
of ct and xt that solve the dynamic optimization problem.

e) Since the discount factor satisfies 0 < β < 1, (21) implies that the optimal consumption
ct of the exhaustible resource is falling over time.

6
5 Saving with a Random Return
The problem involves choosing contingency plans for st , t = 0, 1, 2, ..., and At , t = 1, 2, 3, ...,
to maximize
X∞
E0 β t ln(At − st )
t=0

subject to the constraints

A0 given
and
Rt+1 st ≥ At+1
for all t = 0, 1, 2, ... and all possible realizations of Rt+1 .

a) The Bellman equation for this problem is

v(At , Rt ) = max ln(At − st ) + βEt v(Rt+1 st , Rt+1 ).

st

b) Using the conjectured form of the value function, the Bellman equation becomes

F ln(At ) + G ln(Rt ) + H = max ln(At − st ) + βF Et ln(Rt+1 st ) + βGEt ln(Rt+1 ) + βH

st

or, more simply,

F ln(At ) + G ln(Rt ) + H = max ln(At − st ) + βF ln(st ) + βρ(F + G) ln(Rt ) + βH.

st

The first-order condition for st is therefore

1 βF
− + =0
At − st st
and the envelope condition for At is
F 1
= .
At At − st

c) The first-order condition for st implies that

µ ¶
βF
st = At .
1 + βF

d) Substituting this last expression into the envelope condition yields a solution for F in
terms of β: µ ¶
βF
F At − F At = At
1 + βF
∙ µ ¶¸
βF
F 1− =1
1 + βF

7
 µ ¶
1
F =1
1 + βF
F = 1 + βF
1
F = . (22)
1−β
e) Now substitute the solution (22) for F back into the first-order condition for st to obtain

st = βAt . (23)

Finally, substitute (23) into the binding constraint to obtain

At+1 = βRt+1 At . (24)

Equations (23) and (24) form a system that can be used to construct the optimal paths
for st and At , given the initial value of A0 and the realized path for Rt+1 . Although
the solution does not require these last steps, it is also possible to substitute (22) -(24)
back into the Bellman equation to find
βρ
G=
(1 − βρ)(1 − β)

and µ ¶∙ µ ¶ ¸
1 β
H= ln(1 − β) + ln(β)
1−β 1−β
and verify that the conjectured form of the value function does satisfy the Bellman
equation.

8
 Midterm Exam
Economics 720: Mathematics for Economists
Fall 2008

This exam has three questions on three pages; before you begin, please make sure that
your copy has all three questions and all three pages. The three questions will receive equal
weight in determining your overall exam score. Since this is a 75-minute exam, I suggest
that you spend approximately (75 minutes/3 questions) = 25 minutes on each question.

1) A consumer derives utility U(c1 , c2 ) from consuming two goods, where c1 is his or her
consumption of good 1 and c2 is his or her consumption of good 2. Let p1 > 0 denote the
price of good 1, let p2 > 0 denote the price of good 2, and assume that the consumer can
purchase as many or as few units of each good as he or she likes at these competitively
determined prices. Assume that the utility function is strictly increasing in both of
its arguments, so that U1 (c1 , c2 ) > 0 and U2 (c1 , c2 ) > 0 for all values of c1 > 0 and
c2 > 0, where Ui denotes the derivative of U with respect to its ith argument. Finally,
assume that the utility function satisfies limc1 →0 U1 (c1 , c2 ) = ∞ for all values of c2 and
limc2 →0 U1 (c1 , c2 ) = ∞ for all values of c1 ; these last assumptions will allow you to
avoid having to explicitly impose nonnegativity constraints on the consumer’s choices
of c1 and c2 in all of the analysis that follows, since they imply that the consumer will
always wish to purchase at least a small, positive amount of each good.

a) Suppose first that the consumer takes his or her income I as given, and seeks to
maximize utility subject to a budget constraint that says that the consumer’s
total expenditures cannot exceed income I. Write down (define) the Lagrangian
for this utility maximization problem; then, write down the set of conditions
that, according to the Kuhn-Tucker theorem, must be satisfied by the values c∗1
and c∗2 that solve the consumer’s problem together with the associated value of
the Lagrange multiplier.
b) Explain how you would use your results from part (a) to derive demand curves for
the two consumption goods.
c) Suppose next that the consumer seeks to minimize the total cost of achieving a level
of utility that is greater than or equal to Ū. Write down (define) the Lagrangian
for this cost or expenditure minimization problem; then, write down the set of
conditions that, according to the Kuhn-Tucker theorem, must be satisfied by the
values c∗∗ ∗∗
1 and c2 that solve the consumer’s problem together with the associated
value of the Lagrange multiplier.

1
 d) Explain how you would use your results from part (c) to derive demand curves for
the two consumption goods.
e) What is the difference between the demand curves that you derived in part (b) and
the demand curves that you derived in part (d)? Under what conditions, if any,
will the solutions c∗1 and c∗2 that you described in parts (a) and (b) coincide with
the solutions c∗∗ ∗∗
1 and c2 that you described in parts (c) and (d)?

2) An econometrician collects data on a firm’s output y > 0, the competitively-determined

wage rate w > 0 and the rental rate r > 0 at which it hired labor and capital to
produce that output, as well as the firm’s total costs C > 0. Then, by fitting the
log-linear equation
ln(C) = α ln(w) + β ln(r) + γ ln(y)
where ln denotes the natural logarithm of each variable, to those data, the econome-
trician obtains the parameter estimates α = 0.5, β = 0.5, and γ = 1. Using these
estimates, assuming that the firm has always acted to minimize the costs of producing
output, and assuming that the firm produces output y with capital k and l according to
the production function y = f(k, l), what conclusions can you draw about the specific
form of this production function f ?
3) A consumer works for periods t = 0, 1, ..., T , receiving constant labor income w during
each of those periods. Let kt denote the consumer’s stock of wealth, that is, his or her
bank account balance at the beginning of period t. Suppose that the consumer starts
period t = 0 with nothing in the bank: k0 = 0. Assume, also, that the bank does not
allow the consumer to borrow, so that kt ≥ 0 must hold for all t = 1, 2, ..., T +1 as well.
Let r denote the constant interest rate, and let ct denote the consumer’s consumption
during period t. Finally, suppose that the consumer’s utility function is additively-time
separable and discounted, and that the single-period utility function is logarithmic in
form. Putting all of these building blocks together, the consumer’s problem can be
stated as: choose sequences {ct }Tt=0 and {kt }Tt=1
+1
to maximize utility
X
T
β t ln(ct ),
t=0

where β, with 0 < β < 1, denotes the discount factor, subject to the constraints

w + rkt − ct ≥ kt+1 − kt

governing the evolution of the bank account balance for all t = 0, 1, ..., T , the con-
straints
kt ≥ 0
ruling out borrowing for all t = 1, 2, ..., T , the initial condition

k0 = 0 given,

and the constraint

kT +1 ≥ 0

2
on the terminal value of the bank account balance (assuming here for simplicity that
the consumer receives public pension income when retired and does not need to draw
on funds from his or her bank account to consume after he or she has retired).

a) Write down the Hamiltonian for the consumer’s problem. Hint: the easiest way to
do this, and to be ready to derive the additional results below, is to note that
since the constraints governing the evolution of the bank account balance will
always bind, the no-borrowing constraints from the consumer’s dynamic problem
can be expressed equivalently as

w + (1 + r)kt − ct ≥ 0

for all t = 0, 1, ..., T . Writing the no-borrowing constraint in this way is conve-
nient, since it clarifies how that constraint restricts the feasible choice of ct during
each period t = 0, 1, ...T given the beginning-of-period bank account balance kt .
b) Now write down the first-order condition, the pair of difference equations, and
the initial and transversality conditions that, according to the maximum prin-
ciple, must be satisfied by the sequences for {ct }Tt=0 and {kt }Tt=1
+1
that solve the
consumer’s dynamic optimization problem.
c) Suppose now that the discount factor and the interest rate satisfy β(1 + r) = 1
and guess (correctly, it turns out) that with this additional restriction on the
parameters, the no-borrowing constraint will never bind, in the sense that the
Lagrange multiplier attached to the additional constraint

w + (1 + r)kt − ct ≥ 0

will equal zero for all t = 0, 1, ..., T . Use your results from part (b) to answer the
follow question: under these conditions, will the consumer’s optimal consumption
be rising, falling, or staying constant over time?
d) Use your results from parts (b) and (c) to find the sequences {ct }Tt=0 and {kt }Tt=1
+1

that solve the consumer’s dynamic optimization problem.

3
 Solutions to Midterm Exam
Economics 720: Mathematics for Economists
Fall 2008

1)

a) With the Lagrangian for the utility maximization problem defined as

Lu (c1 , c2 , λ) = U(c1 , c2 ) + λ(I − p1 c1 − p2 c2 ),

the Kuhn-Tucker theorem implies that the optimal values c∗1 and c∗2 and the asso-
ciated value of the Lagrange multiplier λ∗ must satisfy the first-order conditions

Lu1 (c∗1 , c∗2 , λ∗ ) = U1 (c∗1 , c∗2 ) − λ∗ p1 = 0

and
Lu2 (c∗1 , c∗2 , λ∗ ) = U2 (c∗1 , c∗2 ) − λ∗ p2 = 0,
the constraint
I − p1 c∗1 − p2 c∗2 ≥ 0,
the nonnegativity condition for the multiplier

λ∗ ≥ 0,

and the complementary slackness condition

λ∗ (I − p1 c∗1 − p2 c∗2 ) = 0.

b) Since U is strictly increasing and p1 and p2 are strictly positive, the first-order
conditions imply that λ∗ is strictly positive as well. It then follows from the com-
plementary slackness condition that the budget constraint binds at the optimum.
Use the two first-order conditions, together with the binding constraint, as a sys-
tem of three equations to solve for the three unknowns c∗1 , c∗2 , and λ∗ in terms of
the parameters p1 , p2 , and I. The solutions c∗1 (p1 , p2 , I) and c∗2 (p1 , p2 , I) can be
interpreted as the demand curves for the two consumption goods.
c) With the Lagrangian for the expenditure minimization problem defined as

Le (c1 , c2 , μ) = p1 c1 + p2 c2 − μ[U (c1 , c2 ) − Ū],

1
 the Kuhn-Tucker theorem implies that the optimal values c∗∗ ∗∗
1 and c2 and the as-
sociated value of the Lagrange multiplier μ∗ must satisfy the first-order conditions

Le1 (c∗∗ ∗∗ ∗ ∗ ∗∗ ∗∗
1 , c2 , μ ) = p1 − μ U1 (c1 , c2 ) = 0

and
Le2 (c∗∗ ∗∗ ∗ ∗ ∗∗ ∗∗
1 , c2 , μ ) = p2 − μ U2 (c1 , c2 ) = 0,

the constraint
U(c∗∗ ∗∗
1 , c2 ) − Ū ≥ 0,

the nonnegativity condition for the multiplier

μ∗ ≥ 0,

and the complementary slackness condition

μ∗ [U (c∗∗ ∗∗
1 , c2 ) − Ū ] = 0.

d) Since U is strictly increasing and p1 and p2 are strictly positive, the first-order
conditions imply that μ∗ is strictly positive as well. It then follows from the
complementary slackness condition that the constraint binds at the optimum. Use
the two first-order conditions, together with the binding constraint, as a system
of three equations to solve for the three unknowns c∗∗ ∗∗ ∗
1 , c2 , and μ in terms of the
parameters p1 , p2 , and Ū. The solutions c∗∗ ∗∗
1 (p1 , p2 , Ū) and c2 (p1 , p2 , Ū ) can be
interpreted as the demand curves for the two consumption goods.
e) The demand curves c∗1 (p1 , p2 , I) and c∗2 (p1 , p2 , I) describe how the consumer’s choices
change as prices and income vary; these are Marshallian demand curves. The
demand curves c∗∗ ∗∗
1 (p1 , p2 , Ū ) and c2 (p1 , p2 , Ū ) describe how the consumer’s choices
change as prices and utility vary; these are Hicksian demand curves. Define the
indirect utility function associated with the utility maximization problem as

V (p1 , p2 , I) = max U(c1 , c2 ) subject to I ≥ p1 c1 + p2 c2

c1 ,c2

and define the expenditure function associated with the expenditure minimization
problem as as

E(p1 , p2 , Ū ) = min p1 c1 + p2 c2 subject to U(c1 , c2 ) ≥ Ū .

c1 ,c2

The solutions to the two problems will coincide when I is such that

I = E(p1 , p2 , Ū)

or, equivalently, when Ū is such that

Ū = V (p1 , p2 , I).

2
2) The econometrician’s estimates, together with the assumptions made about the firm’s
behavior, imply that the firm’s minimum cost function, defined by

C(w, r, y) = min wl + rk subject to f (k, l) ≥ y

w,l

takes the specific form

C(w, r, y) = w1/2 r1/2 y.
Shephard’s lemma, which is just a special case of the envelope theorem applied to
the firm’s minimum cost function, implies that the firm’s conditional factor demand
curves can be found by differentiating the minimum cost function with respect to the
two factor prices. Hence, in particular,

l∗ (w, r, y) = C1 (w, r, y) = (1/2)w−1/2 r1/2 y

and
k∗ (w, r, y) = C2 (w, r, y) = (1/2)w1/2 r−1/2 y.
There are a number of ways to use these last two expressions to determine the specific
form of the firm’s production function, but the easiest is to just multiply the first by
the second to obtain
l∗ (w, r, y)k∗ (w, r, y) = (1/4)y 2 .
Multiply both sides of this last equation by 4 and take the square root of both sides
to obtain
y = 2[k ∗ (w, r, y)]1/2 [l∗ (w, r, y)]1/2 ,
which, when compared to the binding constraint y = f(k, l), reveals that

f(k, l) = 2k1/2 l1/2 .

Evidently, the firm’s production function takes the Cobb-Douglas form, with equal
shares for capital and labor.
3)

a) The Hamiltonian for the consumer’s problem is

H(kt , π t+1 ; t) = max β t ln(ct ) + π t+1 (w + rkt − ct ) subject to w + (1 + r)kt − ct ≥ 0.

ct

b) The maximum principle implies that if λt denotes the Lagrange multiplier on the
constraint from the static optimization problem on the right-hand-side of the
equation defining the Hamiltonian, the sequences {ct }Tt=0 and {kt }Tt=1
+1
that solve
the original, dynamic optimization problem must satisfy the first-order condition
βt
− π t+1 − λt = 0,
ct
the pair of difference equations

π t+1 − π t = −Hk (kt , π t+1 ; t) = −π t+1 r − λt (1 + r)

3
 and
kt+1 − kt = Hπ (kt , π t+1 ; t) = w + rkt − ct ,
the initial condition
k0 = 0,
and the transversality condition

π T +1 kT +1 = 0.

c) After guessing that λt = 0 for all t = 0, 1, ..., T , the first-order condition simplifies
to
βt
− π t+1 = 0
ct
and the first of the two difference equations simplifies to

(1 + r)π t+1 = π t .

Using the first-order condition to substitute out for π t+1 and π t and imposing the
restriction that β(1 + r) = 1, this last expression becomes

ct = ct−1 ,

revealing that optimal consumption is constant over time, with

ct = c

for some constant c for all t = 0, 1, ...T .

d) To pin down the constant level of consumption c and to determine the optimal time
path for the bank account balance kt , consider the second of the two difference
equations, which just corresponds to the binding constraint

kt+1 = w + (1 + r)kt − ct .

Using the result that ct = c for all t = 0, 1, ..., T , this equation implies that for
t = 0,
k1 = (1 + r)k0 + w − c.
For t = 1,

k2 = (1 + r)k1 + w − c = (1 + r)2 k0 + (1 + r)(w − c) + w − c.

For t = 2,

k3 = (1 + r)k2 + w − c = (1 + r)3 k0 + (1 + r)2 (w − c) + (1 + r)(w − c) + w − c.

And hence for t = T ,

X
T
kT +1 = (1 + r)T +1 k0 + (1 + r)t (w − c).
t=0

4
The initial condition k0 = 0 has been given to us by the specification of the
original problem. Meanwhile, since the first-order condition from part (b) implies
that
βT
π T +1 = > 0,
cT
the transversality condition requires that kT +1 = 0 as well. So it must be true
that
XT
0= (1 + r)t (w − c),
t=0

which can only hold if c = w. Putting all these results together, it follows that
the optimal path for consumption has

ct = w

for all t = 0, 1, ..., T and the optimal path for the bank account balance has

kt = 0

for all t = 0, 1, ..., T + 1.

5
 Final Exam
Economics 720: Mathematics for Economists
Fall 2008

This exam has three questions on five pages; before you begin, please make sure that
your copy has all three questions and all five pages. The three questions will receive equal
weight in determining your overall exam score. Since this is a 120-minute exam, I suggest
that you spend approximately (120 minutes/3 questions) = 40 minutes on each question.

1 A Taste for Variety

Consider a static model in which a continuum of goods indexed by  ∈ [0 ] can be produced
and consumed at a single date. Hence, in this model,  is a parameter that measures the
number of distinct goods that are available. Let () denote the number of units of each
good  ∈ [0 ] that a representative household consumes, let
∙Z  ¸1

= () 
0

define an index of total consumption enjoyed by the household, and suppose that the house-
hold’s preferences are described by the utility function

ln()

defined in terms of the index . With this specification, the parameter  lies between zero
and one, 0    1, and is related to the degree of substitutability between the various
goods; in particular, this specification implies that the elasticity of substitution between any
two goods is 1(1 − ), so that a value of  closer to zero means that the household is less
willing to substitute between goods as their relative price changes and a value of  closer to
one means that the household is more willing to substitute between goods as their relative
price changes.
Suppose that each unit of each good can be produced with one unit of labor and that the
household has  units of labor available in total to produce all of the various goods. Hence,
this household chooses () for all  ∈ [0 ] to maximize its utility
(∙Z ¸ )
 1

ln () 
0

1
subject to the production possibilities constraint
Z 
≥ ()
0

a) Set up (define) the Lagrangian for this problem.

b) Next, write down the set of conditions that, according to the Kuhn-Tucker theorem,
must be satisfied by the values ∗ (),  ∈ [0 ], that solve the household’s problem
together with the associated value of the Lagrange multiplier.
c) Guess (correctly, it turns out) that it is optimal for the household to produce and consume
equal amounts of all goods, so that ∗ () = ∗ for all  ∈ [0 ]. Use this conjecture,
together with your results from above, to solve for the constant ∗ in terms of the
model’s parameters: , , and .
d) Finally, define the household’s indirect utility function as
(∙Z ¸1 ) Z
 

 (  ) = max ln ()  subject to  ≥ ()
() ∈[0] 0 0

Using this definition, together with your results from above, to derive an expression
that shows how the household’s maximized utility depends on the parameters , ,
and .
e) Does the household’s maximized utility increase, decrease, or stay the same when the
number  of available goods increases?

2 Expanding Product Variety

The previous static problem suggests that in a dynamic model, households with preferences as
described above might be willing to allocate scare labor resources not just to the production
of consumption goods but also to a process of research and development that leads to the
invention of new products over time. To explore this possibility, consider a dynamic model
cast in continuous time with an infinite horizon, so that  ∈ [0 ∞), and let the representative
household have preferences over this infinite horizon as described by the utility function
Z ∞
−  (())
0

where   0 is the discount rate and  is a single-period utility function defined in terms of
the index of total consumption () at each date , given as before by
"Z #1
()
() = (; )  
0

where (; ) denotes consumption of each good  ∈ [0 ()] at time  and now the number
of product varieties () is also allowed to depend on time.

2
 In this dynamic model, the household allocates its time across two activities. First, as
before, it allocates total time () during each period  to producing the various consumption
goods in existence using technologies that yield one unit of output for each unit of labor input.
Hence, production possibilities in this “manufacturing sector” are described by
Z ()
() ≥ (; )
0

for all  ∈ [0 ∞). Second, the household allocates total time () to the process of research
and development and, by doing so, comes up with new inventions that expand the range of
goods available for production and consumption in future periods according to

() ≥ ̇()

for all  ∈ [0 ∞), where  is a parameter that measures the productivity of labor in inventive
activity: a higher value of  means that more inventions are found using a given amount of
time allocated to research and development.
To complete the specification of the model, let the household be endowed with a constant
amount of time  to be allocated within each period to either manufacturing or research and
development, as summarized by the constraint

 ≥ () + ()

for all  ∈ [0 ∞). Finally, assume that the form of the household’s utility function implies
that the household will always find it optimal to produce at least some consumption goods by
devoting at least some time to manufacturing, so that the nonnegativity constraint () ≥ 0,
if imposed, will never bind. On the other hand, it is possible that the household will choose
not to “invest” any time in research and development, so that the nonnegativity constraint
() ≥ 0 might bind at the optimum. It turns out that it is most convenient to express this
nonnegativity constraint for () by requiring that

 − () ≥ 0

hold for all  ∈ [0 ∞).

The household’s optimization problem is easiest to solve in two stages. First, during any
period , let’s ask how the household will use the time () that it allocates to manufacturing
to produce an optimal mixture of the () goods that are available during that period. This
static problem involves choosing (; ) for all  ∈ [0 ()] to maximize
⎧" #1 ⎫
⎨ Z () ⎬
 (; ) 
⎩ 0 ⎭

subject to the production possibilities constraint for manufacturing

Z ()
() ≥ (; )
0

3
This problem is very similar to the static problem that you solved above, except that now
() and () may vary from period to period, so rather than work through all of the details
of this sort of problem all over again, just assume for the sake of convenience that the single-
period utility function  is such that the indirect utility function for each period , defined
in a manner that is similar to before as
⎧" #1 ⎫
⎨ Z () ⎬ Z ()

 [() () ] = max  (; )  subject to () ≥ (; )
(;) ∈[0] ⎩ 0 ⎭ 0

takes the double-log form

 [() () ] = ln[()] +  ln[()]

where   0 is a positive parameter that depends in a nonlinear way on the elasticity of

substitution 1(1 − ).
Now the second, dynamic, part of the household’s problem can be stated as: choose
continuously differentiable functions () for  ∈ [0 ∞) and () for  ∈ (0 ∞) to maximize
utility over the infinite horizon,
Z ∞
− {ln[()] +  ln[()]}
0

subject to the constraint

[ − ()] ≥ ̇()
linking the invention of new goods ̇() to the amount of labor () =  − () allocated to
research and development in each period  ∈ [0 ∞), and the constraint

 − () ≥ 0

that requires the amount of labor () =  − () allocated to research and development to
be nonnegative in each period  ∈ [0 ∞), taking the initial condition (0) = 0 as given.

a) Set up (define) the Hamiltonian for the household’s dynamic problem as just described.

b) Now write down the first-order condition, the pair of differential equations, and the
complementary slackness condition that, according to the maximum principle, must
be satisfied by the functions () and () that solve the household’s problem.

c) The most interesting solutions to this problem are such that Lagrange multiplier on the
nonnegativity constraint  − () ≥ 0 equals zero for all  ∈ [0 ∞); solutions of this
form exist, it turns out, so long as the initial number of product varieties (0) = 0
is not too large. So suppose that the multipliers on the constraints  − () ≥ 0 are
zero for all  ∈ [0 ∞) and assume, as well, that the economy has reached a steady
state in which () =  and () =  and hence ̇() = ̇() = 0 as well. Then,
using your results from above, solve for the steady-state values  and  in terms of
the parameters , , , and .

4
 d) Recalling that  measures the number of distinct goods that are available in the steady
state that you just described, answer the following questions. Does the number 
of distinct goods in the steady state increase, decrease, or stay the same when the
household gets more patient, that is, when  falls? Does  increase, decrease, or stay
the same when the “weight”  on product variety in the household’s indirect utility
function rises? Does  increase, decrease, or stay the same when the total amount of
productive time  rises? Does  increase, decrease, or stay the same when productivity
in research and development  rises?

3 Stochastic Growth
Consider a discrete-time version of the optimal growth (Ramsey) model in which the horizon
is infinite, so that time periods are indexed by  = 0 1 2 , the single-period utility function
is logarithmic and the aggregate production function is Cobb-Douglas, capital depreciates
fully in use during each period, and there are random, but serially uncorrelated, shocks to
productivity. In this model, the representative consumer maximizes the expected utility
function
X
∞
0   ln( )
=0

subject to the constraints 0 given and

  ≥  + +1

for all  = 0 1 2 , where  represents the shock to the productivity of capital. The value
of  is known when  and +1 are chosen during period , but the value of +1 is random
and satisfies  ln(+1 ) = 0 for all  = 0 1 2 .

a) Write down the Bellman equation for this problem.

b) Now guess that the value function takes the form

(   ) =  +  ln( ) +  ln( )

where ,  , and  are unknown constants. Using this guess, derive (write down) the
first order condition for the control variable  and the envelope condition for the state
variable  .

c) Use your results from above to derive two equations that can be used to recursively
construct the optimal values for  and +1 for each period  = 0 1 2 , given values
for the parameters  and , the initial condition for 0 , and the productivity shocks 
that are realized at each date  = 0 1 2 .

d) For the sake of completeness, write down solutions that show how the unknown constants
,  , and  depend on the parameters  and .

5
 Solutions to Final Exam
Economics 720: Mathematics for Economists
Fall 2008

1 A Taste for Variety

In the static model, the representative household chooses () for all  ∈ [0 ] to maximize
its utility (∙Z
 ¸1 )

ln () 
0

subject to the production possibilities constraint

Z 
≥ ()
0

a) The Lagrangian for this problem can be defined as

(∙Z ¸1 ) ∙ Z ¸
 

[() ] = ln ()  +  − () 
0 0

b) According to the Kuhn-Tucker theorem, if the values ∗ (),  ∈ [0 ], solve the house-
hold’s problem, then there exists a value ∗ of the Lagrange multiplier such that,
together, these values satisfy the first-order conditions
∗ ()−1
R
∗ 
− ∗ = 0
0
 () 

for all  ∈ [0 ], the constraint

Z 
≥ ∗ ()
0

the nonnegativity condition

∗ ≥ 0
and the complementary slackness condition
∙ Z  ¸
∗ ∗
 −  () = 0
0

1
c) Rearrange the first-order conditions so that they read
∙Z  ¸
∗ −1 ∗ ∗ 
 () =  ()  
0

for all  ∈ [0 ]. Since the right-hand side of this expression does not depend on , the
condition implies that it is optimal for the household to produce and consume equal
amounts of all goods, so that ∗ () = ∗ for all  ∈ [0 ], where ∗ and ∗ must satisfy
∙Z  ¸
∗
∗ −1
( ) = ( )  = ∗ (∗ ) 
∗ 
0

or, more simply,

1 = ∗ (∗ ) 
Since ∗ and  are both strictly positive, this last expression implies that ∗ must be
strictly positive as well. The complementary slackness condition then implies that the
budget constraint must bind at the optimum, so that ∗ must also satisfy
Z 
= ∗  = ∗ 
0

Rearranging this last result yields the desired solution for ∗ :

∗ = 

d) The indirect utility function, defined as

(∙Z ¸1 ) Z
 

 (  ) = max ln ()  subject to  ≥ ()
() ∈[0] 0 0

is a maximum value function. Hence, to evaluate this function, we need to use a two-
step procedure. But we’ve already finished step one: given settings for the parameters
, , and , we know that the optimal choices for consumptions are
∗ () = 
for all  ∈ [0 ]. Step two now involves substituting this solution back into the objective
function to obtain
(∙Z ¸1 )

 (  ) = ln ∗ () 
0
(∙Z ¸1 )

= ln () 
0

= ln[() ]1
µ ¶
1−
= ln() + ln()


e) Since 0    1, the last expression from part (d) above implies that the household’s
maximized utility increases when the number  of available goods increases.

2
2 Expanding Product Variety
In the dynamic model, the representative household chooses continuously differentiable func-
tions () for  ∈ [0 ∞) and () for  ∈ (0 ∞) to maximize utility over the infinite horizon,
Z ∞
− {ln[()] +  ln[()]}
0

subject to the constraint

[ − ()] ≥ ̇()
linking the invention of new goods ̇() to the amount of labor () =  − () allocated to
research and development in each period  ∈ [0 ∞), and the constraint

 − () ≥ 0

that requires the amount of labor () =  − () allocated to research and development to
be nonnegative in each period  ∈ [0 ∞), taking the initial condition (0) = 0 as given.
Note that the form of the indirect utility function from part (d) of question 1 implies that
the new parameter  is related to the elasticity parameter  according to  = (1 − ).

a) The Hamiltonian for this problem can be defined as

[() (); ] = max − {ln[()]+ ln[()]}+()[−()] subject to −() ≥ 0.

()

b) According to the maximum principle, the functions () and () that solve the house-
hold’s problem must satisfy the first-order condition for the static problem that is
described on the right-hand side of the Hamiltonian. Letting () denote the Lagrange
multiplier on the constraint in that problem, this first-order condition can be written
as
−
− () − () = 0
()
According to the maximum principle, the solution to the household’s problem must
also satisfy the pair of differential equations

− 
̇() = − [() (); ] = −
()

and
̇() =  [() (); ] = [ − ()]
and the complementary slackness condition corresponding to the constraint from the
static problem:
()[ − ()] = 0

3
 c) In a steady state with () =  and () =  and hence ̇() = ̇() = 0, the second
differential equation from part (b), above, implies that the steady-state value of  is
given by
 = 
This part of the solution simply confirms that in a steady state where no new products
are being invented, all of the household’s time gets allocated to manufacturing. When
() = 0 as well, the first-order condition from above implies that
− −
() = =
() 
so that
−
̇() = − 

Substituting this last result into the first differential equation yields
− − 
− =− 
 ()
which can be rearranged to obtain the desired solution for the steady-state value of :

= 


d) The solution described in part (c), above, implies that the steady-state number of product
varieties  increases when  falls. This makes sense: a more patient household, with a
smaller discount rate , will be more willing to forgo consumption in order to “invest”
in research and development. The solution also implies that the number of product
varieties will increase when  increases, so that the household attaches a higher weight
to product variety in preferences, when  increases, so that the household has more
productive time to allocate to research and development as well as manufacturing, and
when  increases, so that productivity in research in development is higher.

3 Stochastic Growth
In this stochastic version of the Ramsey model, the representative consumer maximizes the
expected utility function
X
∞
0   ln( )
=0

subject to the constraints 0 given and

  ≥  + +1

for all  = 0 1 2 , where  represents the shock to the productivity of capital. The value
of  is known when  and +1 are chosen during period , but the value of +1 is random
and satisfies  ln(+1 ) = 0 for all  = 0 1 2 .

4
a) The Bellman equation for this problem can be written as

(   ) = max ln( ) +  (  −   +1 )



b) Using the guess

(   ) =  +  ln( ) +  ln( )
the Bellman equation becomes

 +  ln( ) +  ln( ) = max ln( ) +  +   ln(  −  ) +  ln(+1 )


= max ln( ) +  +  ln(  −  )


The first-order condition for  is therefore

1 
− =0
   − 
and the envelope condition for  is

   −1
= 
   − 

c) There are many ways to derive the desired set of equations, but one way starts by
rearranging the first-order condition to obtain
µ ¶
1
 =   
1 + 

then substituting this expression into the envelope condition and rearranging to obtain
1
= 1 − 
1 + 
Substitute the second of these last two results into the first to obtain

 = (1 − )  

then substitute this result into the binding constraint to obtain

+1 =   

These two equations confirm that even in the presence of uncertainty, it is optimal for
the household to consume the fixed fraction 1− of output and to save and invest the
rest. Given the values of the parameters  and , the initial condition for 0 , and the
productivity shocks  that are realized at each date  = 0 1 2 , these two equations
can also be used to construct the optimal values of  and +1 for all  = 0 1 2 .

5
d) The condition
1
= 1 − 
1 + 
derived above can be rearranged to obtain the desired solution for  :

 = 
1 − 
Substituting this solution for  , together with the solutions for  and +1 derived
above, back into the Bellman equation yields
µ ¶

+ ln( ) +  ln( ) = max ln( ) +  +  ln(  −  )
1 −  

= ln(1 − ) + ln( ) +  ln( )

µ ¶

+ + ln()
1 − 
µ ¶ µ ¶
 
+ ln( ) +  ln( )
1 −  1 − 

Since µ ¶
 
+ = 
1 −  1 − 
the terms involving ln( ) cancel from both sides, leaving
µ ¶ µ ¶
 
 +  ln( ) = ln(1 − ) + ln( ) +  + ln() + ln( )
1 −  1 − 

And since this last expression must hold for all possible values of  , it requires that
 1
=1+ =
1 −  1 − 
and µ¶

 = ln(1 − ) +  + ln()
1 − 
or µ ¶∙ µ ¶ ¸
1 
= ln(1 − ) + ln() 
1− 1 − 

6
 Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2009

Sunday, December 20, 2009

This exam has two questions on four pages; before you begin, please check to make sure
your copy has all two questions and all four pages. The first question is worth 40 points
and the second question is worth 20 points; as indicated at the beginning of the semester,
this 60-point final will account for 60 percent of your course grade and the problem sets will
account for the remaining 40 percent of your course grade.

1. The Maximum Principle in Continuous Time

This question asks you to use the maximum principle to solve a continuous-time, infinite-
horizon model in which economic growth is driven by the accumulation of both physical and
human capital.

During each period t ∈ [0, ∞), a representative consumer divides up a total stock of physical
capital k(t) into a fraction u(t) used to produce goods for consumption and investment in
physical capital and a fraction 1 − u(t) used to produce additional human capital. Likewise,
the consumer “divides up” a total stock of human capital h(t) by spending a fraction v(t) of
his or her time producing goods for consumption and investment in physical capital and a
fraction 1 − v(t) of his or her time accumulating additional human capital. Hence, u(t)k(t)
and v(t)h(t) measure the total amounts of physical and human capital used to produce goods
and [1 − u(t)]k(t) and [1 − v(t)]h(t) measure the total amounts of physical and human capital
used to accumulate more human capital.

Suppose that goods and new human capital both get produced according to the same Cobb-
Douglas production function in which the parameter α, satisfying 0 < α < 1, measures the
share of physical versus human capital in production. Suppose also that both capital stocks
depreciate at the same rate, measured by the parameter δ satisfying 0 < δ < 1.

Then in this economy, physical capital gets accumulated according to the constraint

[u(t)k(t)]α [v(t)h(t)]1−α − δk(t) − c(t) ≥ k̇(t), (1)

for all t ∈ [0, ∞). In (1), c(t) denotes the amount of goods consumed during period t. Thus,
when this constraint holds as an equality, as it will when the consumer chooses quantities
optimally, it says that the total amount [u(t)k(t)]α [v(t)h(t)]1−α of goods produced during
each period t gets used for three purposes: (1) to replace physical capital that depreciates
away, (2) for consumption, and (3) to add, on net, to the stock of physical capital.

Meanwhile, human capital gets accumulated according to the constraint

{[1 − u(t)]k(t)}α {[1 − v(t)]h(t)}1−α − δh(t) ≥ ḣ(t), (2)

1
for all t ∈ [0, ∞). This constraint will also hold as an equality when the consumer chooses
quantities optimally. Hence, it says that the total amount {[1−u(t)]k(t)}α {[1−v(t)]h(t)}1−α
of human capital produced during each period t gets used for two purposes: (1) to replace
human capital that depreciates away and (2) to add, on net, to the stock of human capital.

Comparing (1) and (2) also reveals that even under the simplifying assumptions made above
that the production functions and depreciation rates are the same for physical and human
capital, there is a key economic distinction between goods and human capital, because goods
can be eaten but human capital cannot.

Suppose, finally, that the representative consumer’s utility from consuming c(t) units of the
good at each date t ∈ [0, ∞) is given by
Z ∞
e−ρt ln(c(t)) dt, (3)
0

where the single-period utility function takes the natural log form and where the discount
rate ρ satisfies ρ > 0.

Thus, the consumer’s problem can be stated as one of choosing continuously differentiable
functions c(t), u(t), v(t), k(t), and h(t) for all t ∈ [0, ∞) to maximize the utility function
in (3) subject to the constraints in (1) and (2) for all t ∈ [0, ∞), taking as given the initial
stocks k(0) and h(0) of physical and human capital.

a. (10 points) Write down the Hamiltonian for the consumer’s problem. In doing so, you
will need to account for the fact that the underlying dynamic optimization problem has
three flow variables, c(t), u(t), and v(t), so that the static optimization problem on the
right-hand side of your definition of the Hamiltonian should have three choice variables
as well. You will also need to account for the fact that the underlying problem has two
stock variables, k(t) and h(t), so that you will have to introduce two new variables into
the problem, corresponding to multipliers on the capital accumulation constraints (1)
and (2), and then add two extra terms to the period-t objective function when setting
up the Hamiltonian. To keep things simple, however, you can assume at this point
that nonnegativity constraints on c(t), u(t), 1 − u(t), v(t), 1 − v(t), k(t), and h(t) will
never bind at the optimum, so that these constraints can be left out when defining the
Hamiltonian. And you can assume, as indicated above, that the constraints (1) and
(2) will always hold with equality at the optimum.

b. (20 points) Now write down three first-order conditions and four differential equations
that, according to the maximum principle, must be satisfied by the values of c(t), u(t),
v(t), k(t), and h(t) that solve the consumer’s problem together with the associated
values of the two new variables that you introduced into the problem when defining
the Hamiltonian.

c. (10 points) The equations that you derived in part (b) above imply that this economy
has a “balanced growth path,” along which the fractions u(t) and v(t) are constant,

2
 with u(t) = u and v(t) = v for all t ∈ [0, ∞) and along which consumption c(t) and
the two capital stocks k(t) and h(t) all grow at the same constant rate γ, so that
ċ(t) k̇(t) ḣ(t)
= = =γ
c(t) k(t) h(t)
for all t ∈ [0, ∞). Use your results from above to obtain solutions for these three
constants, u, v, and γ, in terms of the model’s parameters: α, δ, and ρ.

2. Stochastic Dynamic Programming

This question asks you to use dynamic programming to characterize the solution to a stochas-
tic version of the Ramsey model in discrete time where labor supply as well as consumption
gets chosen optimally during each period t = 0, 1, 2, ... in an infinite horizon.

To start by fixing notation, let Kt denote the stock of physical capital at the beginning of
each period t = 0, 1, 2, .... Let Ct denote consumption and let Ht denote hours worked by a
representative consumer during each period t = 0, 1, 2, .... Suppose that during each period
t, output gets produced according to the Cobb-Douglas production function
Zt Ktα Ht1−α , (4)
where the parameter α, measuring the share of capital versus labor in production, satisfies
0 < α < 1 and where Zt is a stochastic (random) shock to productivity. As in our general
analysis from class, assume that Zt gets realized at the very beginning of period t. Hence,
when the consumer chooses Ct and Ht during period t, he or she knows the true value of the
period-t shock Zt but still views the value of the period-t + 1 shock Zt+1 as random. Assume
that Zt is serially correlated but follows a Markov process; that is, that the probability
distribution of Zt+1 , though it can depend on Zt , cannot depend also on Zt−1 , Zt−2 , and so
on backwards in time.

During each period t, the representative consumer divides up the output measured by (4)
into amounts to be consumed and invested. Thus, if physical capital depreciates at a rate
measured by the parameter δ satisfying 0 < δ < 1, then the resource constraint
Zt Ktα Ht1−α ≥ Ct + Kt+1 − (1 − δ)Kt (5)
must hold for all periods t = 0, 1, 2, ... and all possible realizations of Zt . Note that (5)
is consistent with the fact that the capital stock Kt+1 that is available at the beginning of
period t + 1 equals the amount of capital Kt that is available at the beginning of period t,
plus the new output Zt Ktα Ht1−α that gets produced during period t, minus the amount of
output Ct consumed during period t, and minus the amount of capital δKt that depreciates
away during period t.

Finally, suppose that the representative consumer’s expected utility over the infinite horizon
is given by
∞
X
E0 β t [ln(Ct ) + θ ln(1 − Ht )], (6)
t=0

3
where the discount factor β satisfies 0 < β < 1, the total amount of time available to the
consumer during each period is normalized to equal one, so that 1 − Ht denotes leisure
enjoyed by the consumer during each period t = 0, 1, 2, ..., and the positive parameter θ > 0
measure the weight on leisure versus consumption in a single-period utility function where
both consumption and leisure enter in the natural log form.

Although this problem can be solved in a variety of ways, it turns out to be easiest to start
by defining a new variable
St = Zt Ktα Ht1−α + (1 − δ)Kt − Ct (7)
as a measure of saving during each period t and to use this definition to rewrite the expected
utility function in (6) as
∞
X
E β t {ln[Zt Ktα Ht1−α + (1 − δ)Kt − St ] + θ ln(1 − Ht )} (8)
t=0

and then to rewrite the capital accumulation constraint in (5) more simply as
St ≥ Kt+1 (9)
for all t = 0, 1, 2, ....

Now the representative consumer’s problem can be stated as one of choosing contingency
plans for St and Ht for all t = 0, 1, 2, ... and Kt for all t = 1, 2, 3, ... to maximize the expected
utility function in (8) subject to the capital accumulation constraint in (9), which again must
hold for for all periods t = 0, 1, 2, ..., taking the initial conditions K0 and Z0 as given.

a. (10 points) Write down the Bellman equation for this stochastic problem. Then use the
Bellman equation to derive the first-order and envelope conditions that characterize
the values of St , Ht , and Kt that solve the consumer’s problem.
b. (10 points) A difficulty that arises in trying to interpret the optimality conditions that
you derived in part (a) above is that they make reference not just to variables like Ct ,
Ht , Kt , Zt , and St and parameters like α, δ, β, and θ that have straightforward economic
interpretations, but also to the derivative of the unknown value function that you
introduced into this problem when setting up the Bellman equation. Accordingly, use
one of the optimality conditions to eliminate the derivative of the value function from
the system. Then, by invoking the constraints (5), (7), and (9) as needed, assuming
correctly that all will hold as equalities at the optimum, rewrite your results from part
(a) above as a system of three equations involving only the unknown variables Ct , Ht ,
and Kt that solve the consumer’s problem, the model’s parameters α, δ, β, and θ, and
the exogenous shock Zt . Note: You will not be able to solve this system of equations
explicitly for optimal choices of Ct , Ht , and Kt , but you should be able to derive the
system of three equations that, in principle at least, could be used to solve for these
unknown values for given values of the parameters α, δ, β, and θ and the exogenous
shock Zt .

4
 Solutions to Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2009

Sunday, December 20, 2009

1. The Maximum Principle in Continuous Time

The consumer’s problem can be stated as one of choosing continuously differentiable func-
tions c(t), u(t), v(t), k(t), and h(t) for all t ∈ [0, ∞) to maximize the utility function
Z ∞
e−ρt ln(c(t)) dt, (3)
0

subject to the constraints

[u(t)k(t)]α [v(t)h(t)]1−α − δk(t) − c(t) ≥ k̇(t) (1)

and
{[1 − u(t)]k(t)}α {[1 − v(t)]h(t)}1−α − δh(t) ≥ ḣ(t), (2)
each of which must hold for all t ∈ [0, ∞), taking k(0) and h(0) as given.

a. The Hamiltonian for this problem can be written in present-value terms as

H(k(t), h(t), π(t), θ(t)) = max e−ρt ln(c(t))

c(t),u(t),v(t)

+π(t){[u(t)k(t)]α [v(t)h(t)]1−α − δk(t) − c(t)}

+θ(t){[(1 − u(t))k(t)]α [(1 − v(t))h(t)]1−α − δh(t)}.

b. According to the maximum principle, the values of c(t), u(t), v(t), k(t), and h(t) that
solve the consumer’s problem together with the associated values of π(t) and θ(t), must
satisfy the first-order conditions

e−ρt
− π(t) = 0, (10)
c(t)

απ(t)u(t)α−1 k(t)α v(t)1−α h(t)1−α

(11)
− αθ(t)(1 − u(t))α−1 k(t)α (1 − v(t))1−α h(t)1−α = 0,
and
(1 − α)π(t)u(t)α k(t)α v(t)−α h(t)1−α
(12)
− (1 − α)θ(t)(1 − u(t))α k(t)α (1 − v(t))−α h(t)1−α = 0,

1
 and the differential equations

π̇(t) = − Hk (k(t), h(t), π(t), θ(t))

= − π(t)[αu(t)α k(t)α−1 v(t)1−α h(t)1−α − δ] (13)
− θ(t)[α(1 − u(t))α k(t)α−1 (1 − v(t))1−α h(t)1−α ],

θ̇(t) = − Hh (k(t), h(t), π(t), θ(t))

= − π(t)[(1 − α)u(t)α k(t)α v(t)1−α h(t)−α ] (14)
− θ(t)[(1 − α)(1 − u(t))α k(t)α (1 − v(t))1−α h(t)−α − δ],
k̇(t) = Hπ (k(t), h(t), π(t), θ(t)) = [u(t)k(t)]α [v(t)h(t)]1−α − δk(t) − c(t), (15)
and

ḣ(t) = Hθ (k(t), h(t), π(t), θ(t)) = [(1 − u(t))k(t)]α [(1 − v(t))h(t)]1−α − δh(t) (16)

c. Along the balanced growth path, u(t) = u, v(t) = v, and

ċ(t) k̇(t) ḣ(t)

= = =γ
c(t) k(t) h(t)

for all t ∈ [0, ∞). Hence, (11) and (12) imply that along this balanced growth path

π(t)uα−1 v 1−α = θ(t)(1 − u)α−1 (1 − v)1−α

and
π(t)uα v −α = θ(t)(1 − u)α (1 − v)−α .
Dividing the first of these equations by the second yields
v 1−v
= ,
u 1−u
which requires that
u = v. (17)
Hence, (11) and (12) also require that

π(t) = θ(t) (18)

and hence
π̇(t) = θ̇(t) (19)
for all t ∈ [0, ∞) along the balanced growth path as well. In light of (17)-(19), equations
(13) and (14) require that

π̇(t) = −π(t)[αk(t)α−1 h(t)1−α − δ]

and
π̇(t) = −π(t)[(1 − α)k(t)α h(t)−α − δ].

2
These last two equations imply that the ratio of physical to human capital must be
constant, with
k(t) α
= (20)
h(t) 1−α
for all t ∈ [0, ∞), along the balanced growth path. Substituting (20) back into either
(13) or (14) then reveals that

π̇(t)
= δ − αα (1 − α)1−α (21)
π(t)

for all t ∈ [0, ∞) as well. Rewriting (10) as

e−ρt = c(t)π(t)

and differentiating both sides with respect to t yields

−ρe−ρt = ċ(t)π(t) + c(t)π̇(t)

or, using (10) again,

−ρc(t)π(t) = ċ(t)π(t) + c(t)π̇(t).
Dividing this last expression through by π(t) and using (21) yields

ċ(t)
= αα (1 − α)1−α − δ − ρ
c(t)

providing the solution for γ:

γ = αα (1 − α)1−α − δ − ρ. (22)

To determine u = v, return to the binding constraint (16) for human capital accumu-
lation, which along the balanced growth path requires

ḣ(t) = (1 − u)k(t)α h(t)1−α − δh(t)

or
ḣ(t)
= (1 − u)k(t)α h(t)−α − δ.
h(t)
In light of (20) and (22), this last expression requires that
 α
α 1−α α
α (1 − α) − δ − ρ = (1 − u) −δ
1−α

or more simply  α
1−α
u=v =α+ρ . (23)
α

3
2. Stochastic Dynamic Programming

The consumer’s problem can be stated as one of choosing contingency plans for St and Ht
for all t = 0, 1, 2, ... and Kt for all t = 1, 2, 3, ... to maximize the expected utility function
∞
X
E β t {ln[Zt Ktα Ht1−α + (1 − δ)Kt − St ] + θ ln(1 − Ht )} (8)
t=0

subject to the constraints

St ≥ Kt+1 (9)
for all t = 0, 1, 2, ..., taking K0 and Z0 as given.

a. With the Bellman equation for this stochastic problem written as

v(Kt , Zt ) = max ln[Zt Ktα Ht1−α + (1 − δ)Kt − St ] + θ ln(1 − Ht ) + βEt v(St , Zt+1 ),
St ,Ht

the first-order conditions characterizing the optimal choices of St and Ht can be written
as
1
− α 1−α
+ βEt v1 (St , Zt+1 ) = 0
Zt Kt Ht + (1 − δ)Kt − St
and
(1 − α)Zt Ktα Ht−α θ
1−α − =0
Zt Ktα Ht + (1 − δ)Kt − St 1 − Ht
and the envelope condition characterizing the optimal choice of Kt can be written as

αZt Ktα−1 Ht1−α + 1 − δ

v1 (Kt , Zt ) = .
Zt Ktα Ht1−α + (1 − δ)Kt − St

b. To simplify the optimality conditions to facilitate their economic interpretation, start by

using the definition of St and the binding constraint St = Kt+1 to rewrite the first-order
and envelope conditions from above as
1
= βEt v1 (Kt+1 , Zt+1 ),
Ct

(1 − α)Zt Ktα Ht−α θ

= ,
Ct 1 − Ht
and
αZt Ktα−1 Ht1−α + 1 − δ
v1 (Kt , Zt ) = .
Ct
Next, note that in characterizing the solution to the original dynamic problem, the
envelope condition must hold for period t + 1 as well as for period t. Hence, it implies
that
α−1 1−α
αZt+1 Kt+1 Ht+1 + 1 − δ
v1 (Kt+1 , Zt+1 ) = .
Ct+1

4
Use this last result to rewrite the first-order condition for Ct as
α−1 1−α
Ht+1 + 1 − δ
 
1 αZt+1 Kt+1
= βEt . (24)
Ct Ct+1

In terms of the economics, (24) indicates that the optimizing consumer acts so that
the intertemporal marginal rate of substitution
βCt
mt+1 =
Ct+1
and the stochastic return on capital
K α−1 1−α
Rt+1 = αZt+1 Kt+1 Ht+1 + 1 − δ

satisfy the familiar capital asset pricing relationship

K
1 = Et (mt+1 Rt+1 ).

The first-order condition for Ht can be rearranged slightly to obtain

θCt
= (1 − α)Zt Ktα Ht−α , (25)
1 − Ht
which states that the optimizing consumer acts so as to equate the marginal rate of
substitution between consumption and leisure to the marginal product of labor. Hence
(24) and (25) are just generalizations of basic results from consumer and producer
theory to this dynamic, stochastic economic environment. Together with the binding
resource constraint (5):

Zt Ktα Ht1−α = Ct + Kt+1 − (1 − δ)Kt , (26)

equations (24) and (25) form a system of three equations that could, in principle,
be used to characterize the optimal choice of Ct , Ht , and Kt for given values of the
parameters α, δ, β, and θ and the exogenous shock Zt .

5
 Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2010

Due Tuesday, December 14, 2010 at 11:00am

This exam has three questions on five pages; before you begin, please check to make sure
your copy has all three questions and all five pages. Each of the three questions is worth 20
points; as indicated at the beginning of the semester, this 60-point final will account for 60
percent of your course grade and the problem sets will account for the remaining 40 percent
of your course grade.

This is an open-book exam, meaning that it is fine for you to consult your notes, textbooks,
and other references when working on your answers to the questions. I expect you to work
independently on the exam, however, so that the answers you submit are yours and yours
alone.

1. Growth and Pollution, Part I

This question asks you to use the Kuhn-Tucker theorem to solve a static constrained opti-
mization problem that can be used to think about patterns of pollution and environmental
quality seen when looking across a sample of countries at different stages of economic devel-
opment.

Suppose, in particular, that a country has k > 0 units of physical capital that can be used to
produce output of a single consumption good using any one from a continuum of technologies
indexed by z ∈ [0, 1]. Lower values of z represent “cleaner” technologies that generate less
pollution and higher values of z represent “dirtier” technologies that generate more pollution.
People dislike pollution, but they also like consumption; and because dirtier technologies can
be used to produce output at a lower cost, a trade-off arises between economic growth and
environmental quality.

To make these ideas more specific, suppose that the country’s k units of capital, when used
with technology z, produce c = Akz units of output, where A > 0 is a parameter governing
the overall level of productivity across all available technologies. Hence, according to this
specification, more capital used with dirtier technologies yields more output. Suppose also,
however, that the same k units of capital, when used with the same technology z, generate
p = Akz β units of pollution, where β > 1 is a parameter governing how the use of dirtier
technologies causes environmental quality to deteriorate.

Suppose, finally, that a representative consumer in this country gets utility

c1−σ − 1
u(c) =
1−σ

1
from consuming c units of output but suffers disutility
 
B
v(p) = pγ
γ
when the level of pollution is given by p, where the preference parameters satisfy σ > 0,
B > 0, and γ > 1.

A social planner for this economy takes the stock of physical capital k as given, and chooses
the levels of consumption c and pollution p to maximize the representative consumer’s total
utility subject to the constraints imposed by the description of the technologies from above.
Although there are a number of ways to formalize the social planner’s problem, perhaps
the easiest is to start by observing that because preferences over consumption are such that
limc→0 u0 (c) = ∞ and because the cleanest technology z = 0 yields no output, the constraints
c ≥ 0 and z ≥ 0 will never bind at the optimum and can therefore be dropped from further
consideration. On the other hand, for certain parameter configurations at least, it may be
optimal to chose the dirtiest technology z = 1; hence the constraint 1 ≥ z must still be
accounted for.

Substituting the technological relations c = Akz and p = Akz β into the utility and disutility
functions u(c) and v(p) then allows the social planner’s problem to be written as one with a
single choice variable and a single constraint:
(Akz)1−σ − 1
 
B
max − (Akz β )γ subject to 1 ≥ z.
z 1−σ γ

a. Define (write down) the Langragian for this problem.

b. Next, write down the full set of first-order conditions, constraints, nonnegativity con-
ditions, and complementary slackness conditions that, according to the Kuhn-Tucker
theorem, characterize the value z ∗ that solves the social planner’s problem together
with the associated value of the Lagrange multiplier.
c. Now use your results from part (b), above, to solve for the optimal choice of z ∗ as a
function of the model’s parameters: k, A, β, σ, B, and γ.
d. Next, use your result from part (c), above, to obtain an expression that shows how total
pollution depends on the model’s parameters k, A, β, σ, B, and γ when z ∗ is chosen
optimally.
e. Finally, in order to obtain a sharper economic characterization of your result from part
(d), above, suppose that the preference parameter σ is larger than one: σ > 1. Under
this extra assumption, what does your result imply for the relationship between the
economy’s level of economic development, as measured by its stock of physical capital
k, and its optimal choice of pollution? Looking at a cross-section of countries, under
what circumstances will more developed countries have better environmental quality?
Under what circumstances will more developed countries have worse environmental
quality?

2
2. Growth and Pollution, Part II

This question asks you to use the maximum principle to characterize the solution to a
dynamic version of the problem that you just considered in a static context. The dynamics
are added to the model by allowing the social planner to increase the capital stock k over
time by allocating output to investment instead of consumption.

Suppose that time is continuous and the horizon is infinite, so that periods can be indexed by
t ∈ [0, ∞). Let k(t) denote the economy’s stock of physical capital at time t and, extending
the specification from above, suppose that by using a technology indexed by z(t) during time
t, the economy produces Ak(t)z(t) units of output but also generates p(t) = Ak(t)z(t)β units
of pollution where, as before, A > 0 and β > 1 are parameters.

Now, however, suppose that in addition to choosing an optimal technology z(t) at each date
t ∈ [0, ∞), the social planner must also choose how to optimally divide up output Ak(t)z(t)
into an amount c(t) to be consumed and an amount Ak(t)z(t) − c(t) to be invested. If, in
addition, the parameter δ, with 0 < δ < 1, measures the depreciation rate of physical capital,
then the stock of capital evolves according to

Ak(t)z(t) − δk(t) − c(t) ≥ k̇(t)

for all t ∈ [0, ∞).

With the additional dynamics introduced through physical capital accumulation complicat-
ing the problem, assume now that the constraints 1 > z(t) and z(t) > 0 can be ignored,
either because there are ever “dirtier” technologies than the dirtiest z = 1 considered before
or because the configuration of parameters insures that these constraints will not bind at
the optimum. Then, with the specification for the utility over consumption

c(t)1−σ − 1
u[c(t)] =
1−σ
with σ > 0 and the disutility from pollution
 
B
v[p(t)] = p(t)γ
γ

with B > 0 and γ > 1 the same as before, the social planner’s problem can be stated as
follows: given the initial capital stock k(0), choose continuously differentiable functions c(t),
z(t), and k(t) for all t ∈ [0, ∞) to maximize
Z ∞
c(t)1−σ − 1
   
−ρt B β γ
e − [Ak(t)z(t) ] dt
0 1−σ γ

subject to
Ak(t)z(t) − δk(t) − c(t) ≥ k̇(t)
for all t ∈ [0, ∞), where ρ > 0 denotes the discount rate.

3
 a. Write down the Hamiltonian for the social planner’s problem.

b. Next, write down the first-order conditions and the pair of differential equations that,
according to the maximum principle, characterize the solution to the social planner’s
problem.

c. The conditions you derived for part (b), above, should form a system of equations
involving the not only optimal choices of c(t), z(t), and k(t) but also the associated
values for a new variable, corresponding to the Lagrange multiplier on the capital
accumulation constraint that you would have introduced into the problem if you had
decide to solve it using the Kuhn-Tucker theorem instead. Use one of your optimality
conditions to eliminate this new variable from the system, so as to obtain a smaller set
of three equations involving only the model’s parameters and the three variables c(t),
z(t), and k(t) that enter into the original statement of the planner’s problem.

d. The smaller system you derived for part (c), above, implies that starting from any
initial value of k(0), there are unique values of c(0) and z(0) that place the economy
on a saddle path that converges to a steady state, in which c(t) = c∗ , z(t) = z ∗ , and
k(t) = k ∗ for all t. Use your answers from part (c), above, to derive a system of three
equations that can be used to solve for the steady-state values c∗ , z ∗ , and k ∗ in terms
of the model’s parameters: A β, δ, σ, B, and γ. Note: you don’t have to actually solve
for the steady-state values; just write down the three-equation system that would, in
principle, allow you to find these values.

3. Stochastic Growth

This problem asks you to use the guess-and-verify method to characterize the solution to a
version of the stochastic growth model.

Suppose that time is discrete and the horizon is infinite, so that periods can be indexed by
t = 0, 1, 2, .... Let output produced during each period t = 0, 1, 2, ... be given by yt = At kt ,
where kt is the beginning-of-period capital stock and At is a random shock to productivity.

By consuming ct units of output at each date t = 0, 1, 2, ..., the representative consumer gets
expected utility
∞
X
E0 β t ln(ct ),
t=0

where the discount factor satisfies 0 < β < 1. Letting δ, with 0 < δ < 1, denote the
depreciation rate for physical capital, the evolution of the capital stock is described by

At kt + (1 − δ)kt − ct ≥ kt+1 ,

a constraint that must hold for all t = 0, 1, 2, ... and all possible realizations of the produc-
tivity shock At at each date t = 0, 1, 2, ....

4
Assume, finally, that the productivity shock At gets realized at the very beginning of pe-
riod t, before ct and kt+1 are chosen, and that the total return on capital, accounting for
depreciation,
RtK = At + 1 − δ
follows the first-order autoregressive process
K
ln(Rt+1 ) = (1 − ρ) ln(R) + ρ ln(RtK ) + εt+1 ,

where the parameter ρ, satisfying 0 < ρ < 1, measures the degree of serial correlation in the
productivity shock, the parameter R, satisfying R > 0, determines the average value of RtK ,
and where εt+1 is a serially uncorrelated shock, satisfying Et εt+1 = 0 for all t = 0, 1, 2, ...,
that gets realized at the beginning of period t + 1. Hence, when ct and kt+1 are chosen, RtK
K
is known but Rt+1 is still viewed by the representative consumer as random.

The representative consumer’s problem can now be stated as: given the initial capital stock
k0 , choose contingency plans for ct , t = 0, 1, 2, ..., and kt , t = 1, 2, 3, ..., to maximize
∞
X
E0 β t ln(ct ),
t=0

subject to the constraints (which are more conveniently rewritten in terms of RtK )

RtK kt − ct ≥ kt+1

for all t = 0, 1, 2, ... and for all possible realizations of RtK at each date t = 0, 1, 2, ....

a. Write down the Bellman equation for the representative consumer’s problem.

b. Next, guess that the value function takes the time-invariant form

v(kt , RtK ) = E + F ln(kt ) + G ln(RtK )

where E, F , and G are unknown constants to be determined, and, using this guess,
write down the first-order and envelope conditions that help characterize the con-
sumer’s optimal decisions.

c. Use your results from part (b), above, to solve for the constant F in terms of the model’s
parameters β, R, and ρ.

d. Use your results from parts (b) and (c) to derive a pair of equations that can be used
to construct the optimal choices for ct , t = 0, 1, 2, ..., and kt , t = 1, 2, 3, ..., for a given
value of the initial capital stock k0 , a given set of realizations for RtK at each date
t = 0, 1, 2, ..., and for given values of the model’s parameters β, R, and ρ.

5
 Solutions to Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2010

Due Tuesday, December 14, 2010 at 11:00am

1. Growth and Pollution, Part I

The social planner’s problem is

(Akz)1−σ − 1
 
B
max − (Akz β )γ subject to 1 ≥ z.
z 1−σ γ

a. The Lagrangian for the social planner’s problem can be written as

(Akz)1−σ − 1
 
B
L(z, λ) = − (Akz β )γ + λ(1 − z).
1−σ γ

b. According to the Kuhn-Tucker theorem, if z ∗ is the value of z that solves the social
planner’s problem, then there exists an associated value λ∗ of λ such that, together,
z ∗ and λ∗ satisfy the first-order condition

L1 (z ∗ , λ∗ ) = (Ak)1−σ (z ∗ )−σ − βB(Ak)γ (z ∗ )βγ−1 − λ∗ = 0,

the constraint
L2 (z ∗ , λ∗ ) = 1 − z ∗ ≥ 0,
the nonnegativity condition
λ∗ ≥ 0,
and the complementary slackness condition

λ∗ (1 − z ∗ ) = 0.

c. Let’s look first for a solution with a nonbinding constraint, that is, with 1 > z ∗ . By
the complementary slackness condition, this solution must have λ∗ = 0. Hence, the
first-order condition implies that
 1
 βγ+σ−1  γ+σ−1
 βγ+σ−1
∗ 1 1
z =
βB Ak

For this solution to apply, it must be that z ∗ < 1 or, in light of the expression just
derived, that
  1
1 γ+σ−1
Ak > .
βB

1
 If, on the other hand, the constraint is binding, then z ∗ = 1 and the first-order condition
implies that
λ∗ = (Ak)1−σ − βB(Ak)γ .
For this solution to apply, it must be that λ∗ ≥ 0 or, in light of the expression just
derived, that
  1
1 γ+σ−1
≥ Ak.
βB
Putting these results together, the optimal choice for z ∗ is given by
1
   γ+σ−1
1
 1 if βB ≥ Ak

z∗ = 1
  βγ+σ−1 γ+σ−1   1 .
γ+σ−1
 1 1 βγ+σ−1 1


if Ak >

βB Ak βB

Evidently, if we take k as an index of the level of economic development, then the

optimal technology becomes progressively cleaner as the level of economic development
increases.

d. Since total pollution is linked to the choice of technology via p = Akz β , the solution for
z ∗ implies that
1
   γ+σ−1
1
 Ak if βB ≥ Ak

p∗ =   β
(β−1)(σ−1)   1 .
 1 βγ+σ−1 1
βγ+σ−1

if Ak > 1 γ+σ−1
βB Ak βB

e. The solution for p∗ just derived implies that in general, total pollution increases with the
index of economic development for low levels of k. This is because at low levels of de-
velopment, the dirtiest technology is always chosen, so that pollution always increases
with output. The same solution implies, however, that at higher levels of development
the relationship between development and environmental quality is ambiguous. This
is because as economies develop, they choose cleaner technologies, reducing pollution,
but also produce more, increasing pollution; the total effect is ambiguous. However,
with the additional condition σ > 1 imposed, the effects of cleaner technologies out-
weigh the effects of additional production, so that environmental quality improves with
economic development. Hence, when σ > 1, the model predicts that in a cross-section
of countries, environmental quality should at first first deteriorate but then improve
with economic development. In fact, patterns of exactly this sort have been found in
numerous studies: see, for example, Gene M. Grossman and Alan B. Krueger, “Eco-
nomic Growth and the Environment,” Quarterly Journal of Economics, vol.110 (May
1995), pp.353-377.

2
2. Growth and Pollution, Part II

The social planner’s problem is to choose continuously differentiable functions c(t), z(t), and
k(t) for all t ∈ [0, ∞) to maximize
Z ∞
c(t)1−σ − 1
   
−ρt B β γ
e − [Ak(t)z(t) ] dt
0 1−σ γ
subject to
Ak(t)z(t) − δk(t) − c(t) ≥ k̇(t)
for all t ∈ [0, ∞), taking the initial capital stock k(0) as given.

a. The Hamiltonian for the social planner’s problem can be written as

c(t)1−σ − 1
   
−ρt B β γ
H(k(t), π(t); t) = max e − [Ak(t)z(t) ]
c(t),z(t) 1−σ γ
+π(t)[Ak(t)z(t) − δk(t) − c(t)].

b. According to the maximum principle, the values of c(t), z(t), and k(t) that solve the
social planner’s problem must satisfy the first-order conditions
e−ρt c(t)−σ − π(t) = 0
and
−e−ρt βB[Ak(t)]γ z(t)βγ−1 + π(t)Ak(t) = 0
and the pair of differential equations
π̇(t) = −Hk (k(t), π(t); t) = e−ρt BAγ k(t)γ−1 z(t)βγ − π(t)[Az(t) − δ]
and
k̇(t) = Hπ (k(t), π(t); t) = Ak(t)z(t) − δk(t) − c(t).
c. The first-order condition for c(t) implies that
π(t) = e−ρt c(t)−σ
and hence
π̇(t) = −ρe−ρt c(t)−σ − σe−ρt c(t)−σ−1 ċ(t).
Using these expressions to eliminate all references to π(t) and π̇(t), the first-order
condition for z(t) can be rewritten as
βB[Ak(t)]γ z(t)βγ−1 = c(t)−σ Ak(t)
and the two differential equations can be rewritten as
ċ(t) = (1/σ)[Az(t) − δ − ρ]c(t) − (1/σ)BAγ k(t)γ−1 z(t)βγ c(t)σ+1
and
k̇(t) = Ak(t)z(t) − δk(t) − c(t).
This three-equation system involves only the three variables, c(t), z(t), and k(t), from
the original economic problem.

3
 d. In the steady state, c(t) = c∗ , z(t) = z ∗ , k(t) = k ∗ , ċ(t) = 0, and k̇(t) = 0, where the
three-equation system from above implies that

βB(Ak ∗ )γ (z ∗ )βγ−1 = (c∗ )−σ Ak ∗ ,

Az ∗ − δ − ρ = BAγ (k ∗ )γ−1 (z ∗ )βγ (c∗ )σ ,

and
c∗ = Ak ∗ z ∗ − δk ∗ .
The existence of a steady state with constant consumption, capital, and output is an
interesting implication of this model. The simpler “Ak” model of economic growth,
in which output y(t) is produced with capital k(t) according to the constant-returns-
to-scale specification y(t) = Ak(t), features sustained economic growth in which con-
sumption, capital, and output all grow forever at the same constant, positive rate.
Here, with pollution added to the specification, the added environmental effect makes
it optimal to set “limits on growth.” For a detailed analysis of this model and some
further extensions, see Nancy L. Stokey. “Are There Limits to Growth?” International
Economic Review, vol.39 (February 1998), pp.1-31.

3. Stochastic Growth

The representative consumer takes the initial capital stock k0 as given and chooses contin-
gency plans for ct , t = 0, 1, 2, ..., and kt , t = 1, 2, 3, ..., to maximize
∞
X
E0 β t ln(ct ),
t=0

subject to the constraints

RtK kt − ct ≥ kt+1
for all t = 0, 1, 2, ... and for all possible realizations of RtK at each date t = 0, 1, 2, ..., where
the total return on capital,
RtK = At + 1 − δ,
follows the first-order autoregressive process
K
ln(Rt+1 ) = (1 − ρ) ln(R) + ρ ln(RtK ) + εt+1 .

a. The Bellman equation for the representative consumer’s problem can be written as

v(kt , RtK ) = max ln(ct ) + βEt v(RtK kt − ct , Rt+1

K
).
ct

b. Using the guess that the value function takes the time-invariant form

v(kt , RtK ) = E + F ln(kt ) + G ln(RtK )

4
 where E, F , and G are unknown constants to be determined, as well as the assumptions
K
made about the stochastic behavior of Rt+1 , the Bellman equation becomes

E + F ln(kt ) + G ln(RtK )
= max ln(ct ) + βE + βF ln(RtK kt − ct ) + β(1 − ρ)G ln(R) + βρG ln(RtK ).
ct

The first-order condition for ct is then

1 βF
− K =0
ct Rt kt − ct

and the envelope condition for kt is

F βF RK
= K t .
kt Rt kt − ct

c. Rewrite the first-order condition as

 
1
ct = RtK kt ,
1 + βF

and substitute this expression into the envelope condition to solve for
1
F = .
1−β

d. Substituting the solution for F back into the first-order condition yields

ct = (1 − β)RtK kt .

This equation, together with the constraint

kt+1 = RtK kt − ct = βRtK kt

can be used to construct the optimal choices for ct , t = 0, 1, 2, ..., and kt , t = 1, 2, 3, ...
for any given value of k0 and any given set of realizations for RtK at each date t =
0, 1, 2, ....

5
 Midterm Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2012

Thursday, October 25, 1:30 - 2:45pm

This exam has four questions on four pages; before you begin, please check to make sure
that your copy has all four questions and all four pages. Each part of each question is worth
five points. Hence, question 1 is worth 15 points, question 2 is worth 5 points, question 3 is
worth 15 points, and question 4 is worth 10 points, for a total of 45 points overall.

1. Utility Maximization (15 points)

Consider a consumer who uses his or her income I to purchase c1 units of good one at the
price of p1 per unit and c2 units of good 2 at the price of p2 per unit, subject to the budget
constraint
I ≥ p 1 c1 + p 2 c2 . (1.1)
In addition, the consumer’s choices of c1 and c2 must satisfy the nonnegativity constraints

c1 ≥ 0 (1.2)

and
c2 ≥ 0. (1.3)
Finally, suppose that the consumer has preferences over the two goods described by the
utility function
U (c1 , c2 ) = ln(c1 ) + ln(c2 + α), (1.4)
where ln denotes the natural logarithm and α > 0 is a positive parameter.

a. Set up the Lagrangian for the consumer’s problem: choose c1 and c2 to maximize the
utility function (1.4) subject to the constraints (1.1)-(1.3). Recall from our conversa-
tions in class that this can be done in at least two ways: feel free to use whichever
formulation of the Lagrangian you find most convenient. Then write down the first-
order conditions, constraints, nonnegativity conditions, and complementary slackness
conditions that, according to the Kuhn-Tucker theorem, must be satisfied by the values
of c∗1 and c∗2 that solve the consumer’s problem, together with the associated values
of the Lagrange multiplier (or multipliers, if you use more than one). Again, as we
discussed in class, there are a number of ways to do this, but make sure that your
statement of the Kuhn-Tucker conditions is consistent with your chosen definition of
the Lagrangian.

b. As a first step in using your answers from part (a), above, to find the solution to
the consumer’s problem, observe that since the utility function (1.4) implies that the
marginal utility of consuming good 1 goes to infinity as c1 approaches zero, it will

1
 always be optimal for the consumer to choose c∗1 > 0. With this in mind, use your
results from above to find the condition on the parameters I, p1 , p2 , and α under which
the consumer will find it optimal to consume only good 1, so that the constraint c∗2 ≥ 0
binds at the optimum. In this case where c∗2 = 0, find the solution for c∗1 in terms of
the parameters I, p1 , p2 , and α.

c. Still keeping in mind that c∗1 > 0 will always hold, find the condition on the parameters
I, p1 , p2 , and α under which the consumer will find it optimal to consume both goods,
so that c∗2 > 0 holds as well. In this case, find the solutions for c∗1 and c∗2 in terms of
the parameters I, p1 , p2 , and α.

2. Cost Minimization (5 points)

Consider a firm that uses k units of capital and l units of labor to produce

(k α l1−α )1/β

units of output, where α and β, satisfying 0 < α < 1 and β > 1, are parameters. If the firm
obtains these inputs at the competitive rental rate r for capital and the competitive wage
rate w for labor, and seeks to minimize the cost of producing at least y units of output, it
solves the problem
min rk + wl subject to (k α l1−α )1/β ≥ y, (2.1)
k,l

where, in this cost-minimization problem, the output requirement y is another parameter.

It turns out that, for this firm, the minimum cost function given by
 
1
C(r, w, y) = rα w1−α y β
αα (1 − α)1−α

measures the minimized value of costs in (2.1), when k and l are chosen optimally subject
to the constraint.

Recall from our discussions in class that the conditional factor demand curves

k ∗ = k ∗ (r, w, y)

for capital and

l∗ = l∗ (r, w, y)
for labor express the values of k and l that solve the problem in (2.1) as functions of the
three parameters r, w, and y.

Given this information, please find the explicit expressions for the two conditional factor
demand curves k ∗ (r, w, y) and l∗ (r, w, y).

2
3. Cigarette Smoking (15 points)

Consider a dynamic model with two periods, t = 0 and t = 1. During each period, a
consumer divides his or her income up into an amount to be spent on cigarettes and an
amount to be spent on all other goods. Let c0 and c1 denote his or her consumption of
cigarettes during periods t = 0 and t = 1, and let x0 and x1 denote his or her consumption
of all other goods during the same two periods. The consumer’s preferences are described
by the utility function

U (c0 , x0 , c1 , x1 ) = ln(c0 ) + x0 + β[ln(c1 − γc0 ) + x1 ], (3.1)

where ln denotes the natural logarithm and the parameters β and γ both lie between zero
and one: 0 < β < 1, and 0 < γ < 1. Therefore, the parameter β is the consumer’s discount
factor – a higher value of β implies that the consumer is more patient – and the parameter
γ measures the addictiveness of cigarettes – the more the consumer smokes during period
t = 0, the higher will be his or her marginal utility of smoking during period t = 1.

For simplicity, suppose that the price of cigarettes p and the consumer’s income I both
remain constant across the two periods. Also for simplicity, suppose that the price of all
other goods is constant and equal to one in both periods. Then the consumer will choose c0 ,
x0 , c1 , and x1 to maximize the utility function in (3.1) subject to the budget constraints

I ≥ pc0 + x0 (3.2)

during period t = 0 and

I ≥ pc1 + x1 (3.3)
during period t = 1.

a. Set up the Lagrangian for the consumer’s problem: choose c0 , x0 , c1 , and x1 to maximize
the utility function in (3.1) subject to the constraints in (3.2) and (3.3). In doing this,
you can assume that the consumer’s constant income level I is large enough that he or
she will always consume positive amounts of cigarettes and all other goods in both pe-
riods, so that there is no need to explicitly impose nonnegativity constraints on c0 , x0 ,
c1 , and x1 . Then write down the first-order conditions, constraints, nonnegativity con-
ditions, and complementary slackness conditions that, according to the Kuhn-Tucker
theorem, must be satisfied by the values of c∗0 , x∗0 , c∗1 , and x∗1 , that solve the consumer’s
problem, together with the associated values of the Lagrange multipliers.

b. Use your results from part (a), above, to solve for the optimal choices of c∗0 and c∗1 as
functions of the parameters β, γ, I, and p.

c. Use your results from part (b), above, to answer the following questions:

(i) Does the optimal choice of c∗0 increase or decrease when the parameter β rises?
(ii) Does the optimal choice of c∗1 increase or decrease when the parameter β rises?
(iii) Who smokes more: patient or impatient consumers?

3
4. Life Cycle Saving (10 points)

Consider a consumer who is employed for T + 1 periods: t = 0, 1, . . . , T . During each

period of employment, the consumer receives labor income w; since w is not indexed by t,
labor income remains constant over time. Let kt denote the consumer’s stock of financial
assets at the beginning of period t, and assume that k0 = 0, so that the consumer begins
his or her career with no assets. For all t = 1, 2, . . . , T , kt can be negative; that is, the
consumer is allowed to borrow. However, the consumer must eventually save for retirement,
a requirement that is captured by imposing the constraint

kT +1 ≥ k ∗ > 0 (4.1)

on the terminal stock of wealth.

Let r denote the interest rate earned on savings, or paid on debt, during each period t =
0, 1, . . . , T ; since r is not indexed by t, this interest rate remains constant over time. Then
the consumer’s stock of assets evolves over time according to

kt+1 = kt + w + rkt − p1t c1t − p2t c2t ,

where c1t and c2t denote the amounts of two goods purchased by the consumer during each
period t = 0, 1, . . . , T and p1t and p2t denote the prices of these two goods. Allowing for
the possibility of free disposal of wealth, which will never be optimal under the preference
specification to be introduced next, these constraints can be written as

w + rkt − p1t c1t − p2t c2t ≥ kt+1 − kt , (4.2)

for all t = 0, 1, . . . , T .

Finally, assume that the consumer’s preferences over consumption of the two goods during
periods t = 0, 1, . . . , T are described by the utility function
T
X
β t [α ln(c1t ) + (1 − α) ln(c2t )], (4.3)
t=0

where ln denotes the natural logarithm and the discount factor β and the parameter α both
lie between zero and one: 0 < β < 1 and 0 < α < 1. The consumer’s problem can now be
stated as: choose sequences {c1t }Tt=0 , {c2t }Tt=0 , and {kt }Tt=1
+1
to maximize the utility function
in (4.3) subject to the constraints k0 = 0 given, (4.1), and (4.2) for for all t = 0, 1, . . . , T .

a. To derive the optimality conditions describing the solution to the consumer’s problem
using the maximum principle, begin by setting up the Hamilton.

b. Then write down the first-order conditions, the difference equations, and the initial
and terminal (or transversality) conditions that, according to the maximum principle,
characterize the solution to the consumer’s dynamic optimization problem.

4
 Solutions to Midterm Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2012

Thursday, October 25, 1:30 - 2:45pm

1. Utility Maximization (15 points)

The consumer chooses c1 and c2 to maximize the utility function

U (c1 , c2 ) = ln(c1 ) + ln(c2 + α), (1.4)

subject to the budget constrant

I ≥ p 1 c1 + p 2 c2 (1.1)
and the nonnegativity constraints
c1 ≥ 0 (1.2)
and
c2 ≥ 0. (1.3)

a. There are a number of ways in which the Lagrangian for this problem can be defined, but
one simply treats the nonnegativity constraints symmetrically to the budget constraint
and assigns a Lagrange multiplier to each. Taking this approach, and letting λ denote
the multiplier of the budget constraint and µ1 and µ2 denote the multipliers on the
nonnegativity constraints, the Lagrangian becomes

L(c1 , c2 , λ, µ1 , µ2 ) = ln(c1 ) + ln(c2 + α) + λ(I − p1 c1 − p2 c2 ) + µ1 c1 + µ2 c2 .

With this definition of the Lagrangian, Kuhn-Tucker theorem implies that the values
c∗1 and c∗2 that solve the consumer’s problem and the associated values λ∗ , µ∗1 , and µ∗2
of the multipliers must satisfy the first-order conditions
1
L1 (c∗1 , c∗2 , λ∗ , µ∗1 , µ∗2 ) = ∗
− λ∗ p1 + µ∗1 = 0
c1

and
1
L2 (c∗1 , c∗2 , λ∗ , µ∗1 , µ∗2 ) = − λ∗ p2 + µ∗2 = 0,
c∗2 +α
the constraints
L3 (c∗1 , c∗2 , λ∗ , µ∗1 , µ∗2 ) = I − p1 c∗1 − p∗2 ≥ 0,
L4 (c∗1 , c∗2 , λ∗ , µ∗1 , µ∗2 ) = c∗1 ≥ 0,
and
L5 (c∗1 , c∗2 , λ∗ , µ∗1 , µ∗2 ) = c∗2 ≥ 0,

1
 the nonnegativity conditions
λ∗ ≥ 0,
µ∗1 ≥ 0,
and
µ∗2 ≥ 0,
and the complementary slackness conditions

λ∗ (I − p1 c∗1 − p∗2 ) = 0,

µ∗1 c∗1 = 0,
and
µ∗2 c∗2 = 0.

b. Since the marginal utility of consuming good 1 goes to infinity as c1 approaches zero, the
nonnegativity constraint c1 ≥ 0 will never bind; the associated complementary slack-
ness condition then implies that µ∗1 = 0 will always hold. Suppose, on the other hand,
that the constraint c2 ≥ 0 does bind at the optimum. With c∗2 = 0, the complementary
slackness condition for that second nonnegativity constraint is satisfied for any value
of µ∗2 ≥ 0. Under these circumstances, the first-order condition for c1 implies that

p1 c∗1 = λ∗ .

Substituting this condition and c∗2 = 0 into the budget constraint, which will always
bind at the optimum, yields
1
I = p1 c∗1 + p2 c∗2 =
λ∗
and hence
1
λ∗ = .
I
∗
Using this solution for λ , together with the first-order condition for c1 again, provides
the solution
I
c∗1 = .
p1
This makes sense: if it is optimal to consume only good 1, then it follows that the
consumer will spend all of his or her income on that good. It only remains to determine
the circumstances under which c∗2 = 0 is, in fact, the optimal choice. To do this,
substitute c∗2 = 0 and λ∗ = 1/I into the first-order condition for c2 to obtain
1 p2
− + µ∗2 = 0
α I
or
p2 1
µ∗2 = − .
I α

2
 The nonnegativity condition µ∗2 ≥ 0 therefore requires that the parameters satisfy

p2 α ≥ I.

Evidently, when α is large, income is low, or the price p2 is high, it will be optimal for
the consumer to only purchase a positive amount of good 1.

c. In this case where it is optimal for the consumer to choose positive amounts of both goods,
the complementary slackness conditions for both nonnegativity constraints imply that
µ∗1 = µ∗2 = 0. Substituting these results into the two first-order conditions yields
1
p1 c∗1 =
λ∗
and
1
p2 c∗2 =− p2 α,
λ∗
which can then be substituted into the binding budget constraint to obtain
1 1
I= ∗
+ ∗ − p2 α
λ λ
or
2
λ∗ = .
I + p2 α
Substituting this solution for λ∗ back into the first-order conditions then yields the
solutions
I + p2 α
c∗1 = .
2p1
and
I + p2 α
c∗2 = − α.
2p2
The last expression reveals that c∗2 will only be positive, as conjectured, if the param-
eters satisfy
I > p2 α.
Evidently, when α is small, income is high, or the price p2 is low, it will be optimal for
the consumer to purchase positive amounts of both goods.

2. Cost Minimization (5 points)

The firm solves solves

min rk + wl subject to (k α l1−α )1/β ≥ y, (2.1)

k,l

which leads to the minimum cost function

 
1
C(r, w, y) = rα w1−α y β ,
αα (1 − α)1−α

3
measuring the minimized value of costs in (2.1), when k and l are chosen optimally subject
to the constraint. Applying Shephard’s lemma or, equivalently, the envelope theorem to this
problem leads directly to solutions for the two conditional factor demand curves:
 1−α  
∗ α w 1−α β
k (r, w, y) = Cr (r, w, y) = y
1−α r

and  α  
∗ 1−α r α β
l (r, w, y) = Cw (r, w, y) = y
α w

3. Cigarette Smoking (15 points)

The consumer chooses c0 , x0 , c1 , and x1 to maximize the utility function

U (c0 , x0 , c1 , x1 ) = ln(c0 ) + x0 + β[ln(c1 − γc0 ) + x1 ], (3.1)

subject to the budget constraints

I ≥ pc0 + x0 (3.2)
during period t = 0 and
I ≥ pc1 + x1 (3.3)
during period t = 1.

a. Let λ0 and λ1 denote the multipliers on the two budget constraints, and define the
Lagrangian as

L(c0 , x0 , c1 , x1 , λ0 , λ1 ) = ln(c0 )+x0 +β[ln(c1 −γc0 )+x1 ]+λ0 (I−pc0 −x0 )+λ1 (I−pc1 −x1 ).

The Kuhn-Tucker theorem then implies that the values c∗0 , x∗0 , c∗1 , and x∗1 that solve
the consumer’s problem, together with the associated values λ∗0 and λ∗1 of the Lagrange
multipliers, must satisfy the first-order conditions
1 βγ
L1 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = ∗
− ∗ ∗
− λ∗0 p = 0,
c0 c1 − γc0

L2 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = 1 − λ∗0 = 0,

β
L3 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = − λ∗1 p = 0,
c∗1 ∗
− γc0
and
L4 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = β − λ∗1 = 0,
the constraints
L5 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = I − pc∗0 − x∗0 ≥ 0
and
L6 (c∗0 , x∗0 , c∗1 , x∗1 , λ∗0 , λ∗1 ) = I − pc∗1 − x∗1 ≥ 0,

4
 nonnegativity conditions
λ∗0 ≥ 0
and
λ∗1 ≥ 0,
and complementary slackness conditions
λ∗0 (I − pc∗0 − x∗0 ) = 0
and
λ∗1 (I − pc∗1 − x∗1 ) = 0.

b. The first-order conditions for x0 and x1 lead immediately to the solutions

λ∗0 = 1
and
λ∗1 = β,
allowing the first-order conditions for c0 and c1 to be written more simply as
1 βγ
− ∗ =p
c0 c1 − γc∗0
∗

and
1
= p.
− γc∗0
c∗1
Substituting the second in this last pair of relationships into the first yields
1
− βγp = p,
c∗0
which leads to the solution
1
c∗0 = .
(1 + βγ)p
Finally, substituting this solution back into the first-order condition for c1 yields
γ
1 = pc∗1 − ,
1 + βγ
which leads to the solution
1 + βγ + γ
c∗1 = .
(1 + βγ)p
c. Differentiating the solutions found above reveals that
∂c∗0 γ
=− <0
∂β (1 + βγ)2 p
and
∂c∗1 γ2
=− <0
∂β (1 + βγ)2 p
It follows immediately from these expressions that

5
 (i) The optimal choice of c∗0 decreases when the parameter β rises.
(ii) The optimal choice of c∗1 decreases when the parameter β rises.
(iii) Impatient consumers smoke more than patient consumers.

4. Life Cycle Saving (10 points)

The consumer chooses sequences {c1t }Tt=0 , {c2t }Tt=0 , and {kt }Tt=1
+1
to maximize the utility
function
XT
β t [α ln(c1t ) + (1 − α) ln(c2t )], (4.3)
t=0

subject to the constraints

w + rkt − p1t c1t − p2t c2t ≥ kt+1 − kt , (4.2)

for all t = 0, 1, . . . , T ,
k0 = 0,
and
kT +1 ≥ k ∗ > 0 (4.1)

a. The Hamiltonian for the consumer’s problem can be defined as

H(kt , πt+1 ; t) = max β t [α ln(c1t ) + (1 − α) ln(c2t )] + πt+1 (w + rkt − p1t c1t − p2t c2t ).
c1t ,c2t

b. According to the maximum principle, the solution to the consumer’s problem is charac-
terized by the first-order conditions
β tα
− πt+1 p1t = 0
c1t
and
β t (1 − α)
− πt+1 p2t = 0,
c2t
the pair of difference equations

πt+1 − πt = −Hk (kt , πt+1 ; t) = −πt+1 r

and
kt+1 − kt = Hπ (kt , πt+1 ; t) = w + rkt − p1t c1t − p2t c2t ,
the initial condition
k0 = 0,
and the terminal, or transversality, condition

πT +1 (kt+1 − k ∗ ) = 0.

6
 Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2012

Due Saturday, December 15, 2012 at 12 noon

This exam has three questions on seven pages; before you begin, please check to make sure
that your copy has all three questions and all seven pages. Each question has three parts,
and each part of each question is worth five points. Hence, each question is worth 15 points,
for a total of 45 points overall.

This is an open-book exam, meaning that it is fine for you to consult your notes, textbooks,
and other written references when working on your answers to the questions. I expect you
to work independently on the exam, however, without discussing the questions or answers
with anyone else, inside or outside of the class: the answers you submit must be yours and
yours alone.

1. Optimal Growth and the Relative Income Hypothesis

In his 1949 doctoral thesis, titled “Income, Saving and the Theory of Consumer Behavior,”
James Duesenberry developed his “relative income hypothesis,” which builds on the assump-
tion that people care less about the absolute level of their own consumption than about their
own consumption level relative to everyone else’s consumption in the economy as a whole.
This problem asks you to apply the maximum principle to solve a version of the neoclassical
(Ramsey) model of optimal growth in continuous time in which the representative consumer
has a utility function that reflects Duesenberry’s ideas.

In this version of the model, as in the standard case, output y(t) gets produced with capital
k(t) during each period t ∈ [0, ∞) according to the production function y(t) = k(t)α , where
the parameter α lies between zero and one: 0 < α < 1. Also as in the standard case, capital
depreciates at the constant rate δ > 0, and c(t) denotes the representative consumer’s
consumption during each period t. Hence, as usual, the capital stock evolves according to
k(t)α − δk(t) − c(t) ≥ k̇(t), (1.1)
for all t ∈ [0, ∞), where k̇(t) = dk(t)/dt. While this constraint as written allows for the free
disposal of capital, the preference specification to be introduced next will imply that (1.1)
will always hold with equality when the functions c(t) and k(t) are chosen optimally.

As noted above, what sets this version of the growth model apart from the standard case
is the specification of the representative consumer’s preferences. Suppose, in particular,
that while c(t) denotes the representative consumer’s own consumption during each period
t, x(t) denotes the economy-wide average level of per-capita consumption, and that the
representative consumer’s utility at each date t depends on the level of c(t) relative to x(t),
according to ln(c(t)−γx(t)), where the parameter γ also lies between zero and one: 0 < γ < 1.
Since utility at t depends on the natural logarithm of c(t) − γx(t), the larger the value of

1
γ the more the representative consumer cares about his or her own consumption relative to
the average. Over the infinite horizon, the consumer’s utility is then
Z ∞
e−ρt ln(c(t) − γx(t)) dt (1.2)
0

where ρ > 0 is the discount rate.

Since the representative consumer is imagined to be just one of many identical consumers in
the economy as a whole, he or she takes the behavior of average consumption x(t) as given,
and chooses continuously differentiable functions c(t) for t ∈ [0, ∞) and k(t) for t ∈ (0, ∞)
to maximize the utility function in (1.2) subject to the constraint in (1.1) for all t ∈ [0, ∞),
also taking the initial capital stock k(0) as given. Thus, for the representative consumer,
x(t) appears in (1.2) as a variable that changes the form of the single-period utility function
over time, not as a variable that can be chosen.

a. Define (write down) the Hamiltonian for this problem. Then write down the first-order
conditions and differential equations that, according to the maximum principle, char-
acterize the solution to the representative consumer’s problem. Note: as we discussed
in class, there is more than one way of defining the Hamiltonian for a problem like
this one; you can use whichever definition you want, but make sure your optimality
conditions are consistent with your chosen form of the Hamiltonian. In addition, to
derive these optimality conditions, you don’t need to impose nonnegativity constraints
on c(t) and k(t), as the specification of the utility and production functions implies
that these constraints will never bind.

b. Recalling that the representative consumer is just one of many identical consumers in the
economy as a whole, note that c(t) = x(t) will hold for all t ∈ [0, ∞) in the economy’s
equilibrium: while each individual does not take this into account when solving his or
her optimization problem, in the end each individual’s consumption must also equal
the economy-wide average. If you use this equilibrium condition to substitute out
for x(t), your results from part (a) above should take the form of three equations
in three unknowns: the values of c(t) and k(t) that solve the consumer’s problem,
plus the value of the additional variable that you introduced into the problem when
setting up the Hamiltonian. Since that additional variable lacks an immediate economy
interpretation, rewrite those three equations as a system of two equations that make
reference only to the two functions c(t) and k(t) and the model’s parameters α, δ, γ,
and ρ.

c. If this economy starts in its nontrivial steady state, c(t) = c∗ > 0 and k(t) = k ∗ > 0
for all t ∈ [0, ∞), where c∗ and k ∗ are positive constants. Thus, ċ(t) = k̇(t) = 0 for all
t ∈ [0, ∞) in this steady state as well. Use these steady-state conditions, together with
the two equations you derived in part (b) above, to solve for the steady-state values c∗
and k ∗ of consumption and capital in terms of the model’s parameters α, δ, γ, and ρ.

2
2. Optimal Growth with Habit Formation

Duesenberry’s model of preferences provides one explanation as to why people don’t seem
too much happier today that they were generations ago, even though their incomes and
consumptions are much higher. Another model that explains the same phenomenon adapts
to a growth framework the preference specifications used in microeconomics by Gary Becker
and co-authors George Stigler (“De Gustibus Non Est Disputandum,” American Economic
Review 1977) and Kevin Murphy (“A Theory of Rational Addiction,” Journal of Political
Economy 1988) to explain the optimal consumption of addictive goods. In this alterna-
tive model of “habit formation,” people care most about how their own consumption today
compares to their own consumption in the recent past, and therefore feel badly when con-
sumption grows slowly or not at all, even if the level of their own consumption remains
quite high. This question asks you to use dynamic programming to solve a version of the
neoclassical growth model in discrete time in which the representative consumer has a utility
function that features habit formation.

In this version of the model, output yt gets produced with capital kt during each period
t = 0, 1, 2, . . . according to the production function yt = ktα , with 0 < α < 1. Capital
depreciates at the constant rate δ, and ct denotes the representative consumer’s consumption
during period t. hence the capital stock evolves according to
ktα + (1 − δ)kt − ct ≥ kt+1 (2.1)
for all t = 0, 1, 2, . . . . Once more, this constraint as written allows for the free disposal of
capital, but this will never be optimal with the preferences to be described next; (2.1) will
always hold with equality when the sequences for ct and kt are chosen optimally.

Now what sets this version of the model apart from the standard case is that the represen-
tative consumer’s utility at each date t depends on his or her consumption during period
t relative to his or her own consumption during period t − 1 according to ln(ct − γct−1 )
where ln denotes the natural logarithm and the parameter γ, satisfying 0 < γ < 1, governs
the strength of “habits,” or addictive effects, from consumption during the previous period.
Over the infinite horizon, the consumer’s utility is then
∞
X
β t ln(ct − γct−1 ), (2.2)
t=0

where the discount factor satisfies 0 < β < 1. In this version of the model, therefore, the
representative consumer takes the initial capital stock k0 and the initial level of the habit in
consumption c−1 as given, and chooses sequences {ct }∞ ∞
t=0 and {kt }t=1 to maximize the utility
function in (2.2) subject to the constraint in (2.1) for all t = 0, 1, 2, . . ..

The utility function in (2.2) is not additively time-separable, since marginal utility at t
depends not just on consumption during period t but also on consumption during period
t − 1. Still, the problem can be converted into one that can be solved using dynamic
programming methods by defining a second state variable xt that is equal to the previous
period’s consumption level ct−1 and therefore keeps track of the “habit stock” according to
xt = ct−1 (2.3)

3
for all t = 0, 1, 2, . . .. In terms of this new state variable, the utility function in (2.2) can be
rewritten equivalently as
X∞
β t ln(ct − γxt ). (2.4)
t=0

The consumer’s problem can now be restated as one of choosing a sequence {ct }∞ t=0 for the
control variable and sequences {xt }∞ t=1 and {k }∞
t t=1 for the state variables to maximize the
utility function in (2.4) subject to the constraints in (2.1) and (2.3) for all t = 0, 1, 2, . . .,
taking the initial state variables x0 = c−1 and k0 as given. The extra constraint (2.3)
then summarizes the key difference between this model and the previous model: here, the
representative consumer explicitly recognizes that his or her choice of ct at t will affect
the level of the habit stock xt+1 at t + 1; whereas in Duesenberry’s model, each individual
consumer ignores the effect that the choice of his or her own consumption has on the economy-
wide average. Before moving on to solve this problem, note once again that the form of the
utility and production functions imply that you don’t need to explicitly impose nonnegativity
constraints on the choices of ct , xt , and kt , since these will never bind at the optimum.

a. In this model of habit formation, the stock variables xt and kt completely summarize the
state of the economy as it appears to the representative consumer at the beginning of
each period t. Hence, the value function that enters into the Bellman equation for the
consumer’s problem will depend on xt and kt , but not separately on t: that is, it can
be written as v(xt , kt ). Using this insight, write down the Bellman equation for the
consumer’s problem. Then, write down the first-order condition describing the optimal
choice of consumption ct for period t and the two envelope conditions describing the
optimal choices of xt and kt for period t.

b. Together with the two constraints, (2.1) with equality and (2.3), the first-order and
envelope conditions you derived above form a system of five equations in five unknowns:
the three unknown variables ct , xt , and kt and the two partial derivatives v1 (xt , kt ) and
v2 (xt , kt ) of the unknown value function that enters into the Bellman equation. Use the
constraint (2.3) to replace xt in this system with the lagged value of consumption ct−1 .
Then use two of your three optimality conditions to eliminate reference to v1 (xt , kt )
and v2 (xt , kt ) in the two equations that remain: the result should be a system of two
equations that make reference only to the variables ct and kt for different values of t
(that is, two difference equations in the sequences {ct }∞ ∞
t=0 and {kt }t=0 ) and the model’s
parameters α, δ, γ, and β.

c. If this economy starts in its nontrivial steady state, ct = c∗ > 0 and kt = k ∗ > 0 for all
t = 0, 1, 2, . . ., where c∗ and k ∗ are positive constants. Use the two equations involving
ct and kt that you derived in part (b) above to solve for the steady-state values c∗ and
k ∗ of consumption and capital in terms of the model’s parameters α, δ, γ, and β.

4
3. Real and Nominal Interest Rates

This problem asks you to use dynamic programming to solve a stochastic model that draws
links between nominal interest rates, real interest rates, and the rate of inflation in a way
that generalizes the relationships between the same three variables described, most famously,
by Irving Fisher in his 1896 monograph, “Appreciation and Interest.”

Let At denote the dollar value of a representative consumer’s financial assets at the beginning
of each period t = 0, 1, 2, . . .. During each period, the consumer divides these assets up into
an amount to be spent on consumption ct and amounts to be invested in two types of bonds.

The first type of bond is called a “real bond” because its price at time t and its payoff at time
t + 1 are both expressed in “real terms,” that is, in units of consumption. More specifically,
each real bond costs the same as qt units of consumption at t and pays off the cost of one
unit of consumption at t + 1. The second type of bond is called a “nominal bond” because
its price at t and its payoff at t + 1 are both expressed in “nominal terms.” that is, in units
of dollars. More specifically, each nominal bond costs Qt dollars at time t and pays off one
dollar at t + 1.

Let Pt denote the dollar price of consumption at time t, let bt denote the number of real
bonds purchased by the consumer during period t, and let Bt denote the number of nominal
bonds purchased by the consumer during period t. Then, during each period t, the consumer
faces the constraint
At ≥ Pt ct + Pt qt bt + Qt Bt , (3.1)
where, on the right-hand side, the number of real bonds purchased gets multiplied by both
Pt and qt because each real bond costs the same as qt units of consumption which, in turn,
cost Pt qt dollars at time t.

Given that he or she purchases bt real bonds and Bt nominal bonds during period t, the
consumer then has assets of dollar value

At+1 = Pt+1 bt + Bt (3.2)

at the beginning of period t + 1 where, on the right-hand side, the number of real bonds gets
multiplied by Pt+1 since each real bond pays off the cost one unit of consumption, which is
Pt+1 dollars at time t + 1.

Suppose that in this economy, the prices of goods and bonds fluctuate stochastically, so
that the vector εt = (Pt , qt , Qt ) appears as a random shock in the consumer’s optimization
problem. Suppose, as we did in class, that the shocks follow a Markov process, so that the
distribution of εt+1 depends on εt but not on previous values of εt−1 , εt−2 , . . .. Consistent
with timing assumptions we made in solving similar stochastic problems in class, assume that
today’s prices in εt are known, but next period’s prices in εt+1 are still viewed as random,
when the consumer chooses ct , bt , and Bt during period t. Then, in this model, the consumer
takes his or her initial value of assets A0 and the initial set of prices in ε0 as given and chooses
contingency plans for ct , bt , and Bt for t = 0, 1, 2, . . . and At+1 for t = 1, 2, 3, . . . to maximize

5
the expected utility function
∞
X
E0 β t u(ct ),
t=0
where the discount factor satisfies 0 < β < 1, the single-period utility function u is strictly
increasing, strictly concave, and satisfies limc→0 u0 (c) = ∞, and E0 denotes the consumer’s
expectation at t = 0, subject to the constraints (3.1) and (3.2), each of which must hold for all
periods t = 0, 1, 2, . . . and for all possible realizations of the shocks in εt+1 . The assumption
on u0 (c) as c approaches zero implies that a nonnegativity constraint on the choice of ct does
not have to be imposed, as it will never bind at the optimum. Nonnegativity constraints on
bt , Bt , and At will not be imposed either; hence, in effect, the consumer is allowed to borrow
as well as save by choosing negative values for bt and/or Bt .

In order to solve this problem using dynamic programming, it is helpful to recognize, first,
that because the single-period utility function is strictly increasing, the constraint in (3.1)
will bind for all periods t = 0, 1, 2, . . .. Substituting this binding constraint into the expected
utility function yields
∞  
X
t At Qt Bt
E0 βu − q t bt − . (3.3)
t=0
P t P t

The consumer’s problem now simplifies to one of choosing contingency plans for bt and Bt
for t = 0, 1, 2, . . . and At+1 for t = 1, 2, 3, . . . to maximize (3.3) subject to the constraint in
(3.2) for all t = 0, 1, 2, . . . and all possible realizations of εt+1 , taking A0 and ε0 as given. In
this problem, the bond purchases bt and Bt are the control variables, the value of assets At
is the state variable, and the value function v(At , εt ) will depend on both the state variable
At and the vector of shocks εt .

a. Write down the Bellman equation for the consumer’s problem. Then write down the
first-order conditions for bt and Bt and the envelope condition for At that help describe
the solution of this problem.
b. Your three equations from part (a) above should make reference to the derivative
v1 (At , εt ) of the unknown value function from the Bellman equation, which lacks an
immediate economic interpretation. Use one of these optimality conditions, together
with the constraints (3.1) with equality and (3.2), to rewrite the remaining two equa-
tions in a form that makes reference only to consumption ct and the prices Pt , qt , and
Qt at different periods t and the discount factor β.
c. Since in this model, the representative consumer can invest in real bonds to exchange qt
units of consumption at t for one unit of consumption at t + 1, the real interest rate rt
can be defined by
1
rt = .
qt
Similarly, since the representative consumer can invest in nominal bonds to exchange
Qt dollars at t for one dollar at t + 1, the nominal interest rate Rt can be defined by
1
Rt = .
Qt

6
The gross rate of inflation πt+1 between t and t + 1 is measured by
Pt+1
πt+1 = ,
Pt
and the consumer’s intertemporal marginal rate of substitution mt+1 between t and
t + 1 is
βu0 (ct+1 )
mt+1 = .
u0 (ct )
Use these definitions to rewrite your two equations from part (b) above as two equa-
tions that link the real interest rate rt and t and the nominal interest rate Rt at t
to the inflation rate πt+1 between t and t + 1 and the intertemporal marginal rate of
substitution mt+1 between t and t + 1.

7
 Solutions to Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2012

Due Saturday, December 15, 2012 at 12 noon

1. Optimal Growth and the Relative Income Hypothesis

The representative consumer takes the initial capital stock k(0) and the time path for the
economy-wide average level of per-capita consumption x(t) as given, and chooses continu-
ously differentiable functions c(t) for t ∈ [0, ∞) and k(t) for t ∈ (0, ∞) to maximize the
utility function Z ∞
e−ρt ln(c(t) − γx(t)) dt (1.2)
0
subject to the constraint
k(t)α − δk(t) − c(t) ≥ k̇(t), (1.1)
for all t ∈ [0, ∞).

a. The Hamiltonian for this problem can be defined in present-value form as

H(k(t), π(t); t) = max e−ρt ln(c(t) − γx(t)) + π(t)[k(t)α − δk(t) − c(t)].

c(t)

The maximum principle then implies that the values of c(t) and k(t) that solve the
representative consumer’s problem must satisfy the first-order condition

e−ρt
− π(t) = 0
c(t) − γx(t)

for the value of c(t) that solves the static, unconstrained problem on the right-hand
side of the definition of the Hamilton, and the pair of differential equations

π̇(t) = −Hk (k(t), π(t); t) = −π(t)[αk(t)α−1 − δ]

and
k̇(t) = Hπ (k(t), π(t); t) = k(t)α − δk(t) − c(t).

b. Use the equilibrium condition c(t) = x(t) to rewrite the first-order condition for c(t) as

e−ρt = (1 − γ)c(t)π(t),

then differentiate both sides of this expression to obtain

−ρe−ρt = (1 − γ)ċ(t)π(t) + (1 − γ)c(t)π̇(t).

1
 Use the first-order condition again, together with the differential equation for π̇(t), to
rewrite this last equation as

−ρ(1 − γ)c(t)π(t) = (1 − γ)ċ(t)π(t) − (1 − γ)c(t)π(t)[αk(t)α−1 − δ].

Dividing both sides of this expression by (1 − γ)π(t) and isolating ċ(t) on the left-hand
side, the differential equation can be rewritten as

ċ(t) = [αk(t)α−1 − δ − ρ]c(t)

and combined with

k̇(t) = k(t)α − δk(t) − c(t)
to form a system of two equations that make reference only to the two functions c(t)
and k(t) and the model’s parameters α, δ, γ, and ρ.

c. If this economy starts in its nontrivial steady state, c(t) = c∗ > 0 and k(t) = k ∗ > 0
for all t ∈ [0, ∞), where c∗ and k ∗ are positive constants. Since ċ(t) = k̇(t) = 0 for
all t ∈ [0, ∞) in this steady state as well, two differential equations from above require
the steady-state values c∗ and k ∗ to satisfy

0 = [α(k ∗ )α−1 − δ − ρ]c∗

and
0 = (k ∗ )α − δk ∗ − c∗ .
The first of these two equations can be used to solve for
 1/(α−1)
∗ δ+ρ
k = ,
α

and the second can be used to solve for

 α/(α−1)  1/(α−1)
∗ δ+ρ δ+ρ
c = −δ .
α α

Interestingly, the differential equations for c(t) and k(t) are the same in this version of
the model as they are in the standard neoclassical growth model. The representative
consumer in this model may not feel is happy as a consumer in the standard model,
but the dynamic behavior of consumption and capital is unchanged.

2. Optimal Growth with Habit Formation

The representative consumer takes the initial capital stock k0 and the initial habit stock
x0 = c−1 as given, and chooses sequences {ct }∞ ∞ ∞
t=0 , {xt }t=1 and {kt }t=1 to maximize the
utility function
X∞
β t ln(ct − γxt ), (2.4)
t=0

2
subject to the constraints
ktα + (1 − δ)kt − ct ≥ kt+1 (2.1)
and
xt = ct−1 (2.3)
for all t = 0, 1, 2, . . ..

a. The Bellman equation for this problem is

v(xt , kt ) = max ln(ct − γxt ) + βv[ct , ktα + (1 − δ)kt − ct ].

ct

The first-order condition for ct is therefore

1
+ βv1 [ct , ktα + (1 − δ)kt − ct ] − βv2 [ct , ktα + (1 − δ)kt − ct ] = 0,
ct − γxt
and the envelope conditions for xt and kt are
γ
v1 (xt , kt ) = −
ct − γxt
and
v2 (xt , kt ) = β(αktα−1 + 1 − δ)v2 [ct , ktα + (1 − δ)kt − ct ].

b. Use the constraints (2.1) and (2.3) to rewrite the first-order condition for ct as
1
+ βv1 (xt+1 , kt+1 ) − βv2 (xt+1 , kt+1 ) = 0
ct − γct−1
and the envelope conditions for xt and kt as
γ
v1 (xt , kt ) = −
ct − γct−1
and
v2 (xt , kt ) = β(αktα−1 + 1 − δ)v2 (xt+1 , kt+1 ).
Since the envelope condition for xt must hold for all t = 0, 1, 2, . . ., it implies that
γ
v1 (xt+1 , kt+1 ) = − .
ct+1 − γct
Substitute this expression into the first-order condition for ct to obtain
1 βγ
βv2 (xt+1 , kt+1 ) = − .
ct − γct−1 ct+1 − γct
Since this condition must also hold for all t = 0, 1, 2, . . ., it implies that
1 γ
v2 (xt , kt ) = − .
β(ct−1 − γct−2 ) ct − γct−1

3
 Substituting these last two expressions into the envelope condition for kt yields
 
1 γ α−1 1 βγ
− = (αkt + 1 − δ) − ,
β(ct−1 − γct−2 ) ct − γct−1 ct − γct−1 ct+1 − γct
which can be combined with the constraint

kt+1 = ktα + (1 − δ)kt − ct

to form a system of two difference equations involving the variables ct and kt and the
parameters α, δ, γ, and β.

c. If this economy starts in its nontrivial steady state, ct = c∗ > 0 and kt = k ∗ > 0 for all
t = 0, 1, 2, . . ., where c∗ and k ∗ are positive constants. The two difference equations
derived above imply that the steady-state values c∗ and k ∗ must satisfy
 
1 γ ∗ α−1 1 βγ
− = [α(k ) + 1 − δ] −
β(1 − γ)c∗ (1 − γ)c∗ (1 − γ)c∗ (1 − γ)c∗
and
0 = (k ∗ )α − δk ∗ − c∗ .
Since the first of these two conditions simplifies to

1 = β[α(k ∗ )α−1 + 1 − δ],

it can be used to solve for

  1/(α−1)
∗ 1 1
k = −1+δ .
α β
The second condition can then be used to solve for
  α/(α−1)   1/(α−1)
∗ 1 1 1 1
c = −1+δ −δ −1+δ .
α β α β

Interestingly, although the difference equations for ct and kt in this version of the model with
habit formation are quite different from those in the standard neoclassical growth model,
the steady-state values c∗ and k ∗ are the same. These observations imply that while habit
formation affects the dynamic path towards the steady state, it does not affect the steady
state to which the economy eventually converges.

3. Real and Nominal Interest Rates

The representative consumer chooses contingency plans for bt , and Bt for t = 0, 1, 2, . . . and
At+1 for t = 1, 2, 3, . . . to maximize the expected utility function
∞  
X
t At Qt Bt
E0 βu − q t bt − . (3.3)
t=0
P t P t

4
subject to the constraint
At+1 = Pt+1 bt + Bt (3.2)
for all t = 0, 1, 2, . . . and all possible realizations of εt+1 = (Pt+1 , qt+1 , Qt+1 ), taking A0 and
ε0 as given.

a. The Bellman equation for this problem is

 
At Qt Bt
v(At , εt ) = max u − q t bt − + βEt v(Pt+1 bt + Bt , εt+1 ).
bt ,Bt Pt Pt
The first-order conditions for bt and Bt are therefore
 
0 At Qt Bt
−qt u − q t bt − + βEt [Pt+1 v1 (Pt+1 bt + Bt , εt+1 )] = 0
Pt Pt
and    
Qt 0 At Qt Bt
− u − q t bt − + βEt v1 (Pt+1 bt + Bt , εt+1 ) = 0,
Pt Pt Pt
and the envelope condition for At is
   
1 0 At Qt Bt
v1 (At , εt ) = u − qt b t − .
Pt Pt Pt

b. The constraints (3.2) and

At = Pt ct + Pt qt bt + Qt Bt (3.1)
imply that the first-order and envelope conditions can be written more simply as

qt u0 (ct ) = βEt [Pt+1 v1 (At+1 , εt+1 )],

 
Qt
u0 (ct ) = βEt v1 (At+1 , εt+1 ),
Pt
and  
1
v1 (At , εt ) = u0 (ct ).
Pt
Moreover, since the envelope condition must hold for all t = 0, 1, 2, . . ., it implies that
 
1
v1 (At+1 , εt+1 ) = u0 (ct+1 ).
Pt+1
Substituting this last expression into the two first-order conditions yields the simpler
expressions
qt u0 (ct ) = βEt u0 (ct+1 )
and     
Qt 0 1 0
u (ct ) = βEt u (ct+1 ) .
Pt Pt+1

5
c. With the real interest rate defined as
1
rt =
qt
and the intertemporal marginal rate of substitution mt+1 between t and t + 1 denoted
by
βu0 (ct+1 )
mt+1 = ,
u0 (ct )
the first-order condition for bt implies
1
= Et mt+1 .
rt
And with the nominal interest rate defined by
1
Rt =
Qt
and the gross rate of inflation denoted by
Pt+1
πt+1 = ,
Pt
the first-order condition for Bt implies
 
1 mt+1
= Et .
Rt πt+1
These last two results can be used to generalize the relationship between the nominal
interest rate, the real interest rate, and the inflation rate described by Irving Fisher.
In particular, since
     
1 mt+1 1 1 1
= Et = Et + covt mt+1 , ,
Rt πt+1 rt πt+1 πt+1
the traditional “Fisher equation” linking the nominal interest rate to the real interest
rate and the inflation rate must be modified in two ways to account for the effects of
uncertainty. First, the relationship must be re-expressed in terms of the reciprocals of
the nominal interest rate, the real interest rate, and the inflation rate; and, second,
the relationship must account for the co-variance between the intertemporal marginal
rate of substitution and the inflation rate. Suppose in particular that people begin to
worry that the economy will fall into a deflationary recession. In this recession, the
intertemporal marginal rate of substitution will be high, because the marginal utility of
consumption will be high, and the reciprocal of the inflation rate will be high, because
the inflation rate will be low, so that the covariance term in the last expression will be
positive. The expression then indicates that the nominal interest rate will be Rt lower
than what the real interest rate and the expected rate of inflation might otherwise
suggest and, indeed, nominal interest rates in the United States have, in recent years,
been quite low, perhaps reflecting fears of renewed deflation and recession.

6
 Midterm Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2013

Thursday, October 31, 1:30 - 2:45pm

This exam has three questions on three pages; before you begin, please check to make sure
that your copy has all three questions and all three pages. The three questions will be will
be weighted equally in determining your overall exam score.

1. The Kuhn-Tucker Theorem

Consider the problem of choosing values for two variables, x and y, to maximize the objective
function
F (x, y) = xy,
subject to the constraints
8 ≥ x2 + y 2 ,
x ≥ 0,
and
y ≥ 0.

a. Set up the Lagrangian for this constrained optimization problem. Recall from our
conversations in class that this can be done in a number of different ways: feel free to
use whichever formulation of the Lagrangian you find most convenient.
b. Next, write down the first-order conditions, constraints, nonnegativity conditions, and
complementary slackness conditions that, according to the Kuhn-Tucker theorem, must
be satisfied by the values of x∗ and y ∗ that solve this problem, together with the
associated values of the Lagrange multiplier (or multipliers, if you use more than one).
Again, as we discussed in class, there are a number of ways to do this, but make
sure that your statement of the Kuhn-Tucker conditions is consistent with your chosen
definition of the Lagrangian.
c. Finally, use your results from above to find the numerical values of x∗ and y ∗ that solve
the problem.

2. Expenditure Minimization

Suppose that a consumer has the utility function

1/2 1/2
U (c1 , c2 ) = 2c1 + 2c2 ,

defined over consumption of two goods, where c1 is consumption of the first good and c2 is
consumption of the second good. Let p1 be the price of the first good and p2 be the price of

1
the second. Suppose that the consumer solves an expenditure minimization problem: choose
c1 and c2 to minimize the cost of achieving a level of utility that is greater than or equal to
Ū . Assuming that the consumer operates in perfectly competitive markets that allow him
or her to buy as many or as few units of each good as he or she likes at the given prices p1
and p2 , this problem can be stated mathematically as
1/2 1/2
min p1 c1 + p2 c2 subject to 2c1 + 2c2 ≥ Ū .
c1 ,c2

a. Set up the Lagrangian for this constrained optimization problem. Recall from our
conversations in class that this can be done in a number of different ways: feel free to
use whichever formulation of the Lagrangian you find most convenient.

b. Next, write down the first-order conditions, constraints, nonnegativity conditions, and
complementary slackness conditions that, according to the Kuhn-Tucker theorem, must
be satisfied by the values of c∗1 and c∗2 that solve this problem, together with the
associated value of the Lagrange multiplier. Again, as we discussed in class, there are
a number of ways to do this, but make sure that your statement of the Kuhn-Tucker
conditions is consistent with your chosen definition of the Lagrangian.

c. Finally, use your results from above to derive expressions for the optimal values of c∗1
and c∗2 as functions of the parameters p1 , p2 , and Ū . In doing this, you can assume
that the prices of the two goods are always strictly positive.

3. Production and Inventories

Consider a firm that produces yt units of output during each period t = 0, 1, . . . , T in order
to add to its stock of inventories xt , which evolves according to

xt + yt ≥ xt+1

for all t = 0, 1, . . . , T . The firm starts out with zero inventories, so that x0 = 0, but must
have at least x∗ > 0 units in inventory by the end of the final period T , so that xT +1 ≥ x∗ .
Suppose that the firm’s total costs during each period t = 0, 1, . . . , T equal the sum of the
cost of producing new output, given by (a/2)yt2 , where a > 0 is a parameter, and the cost of
storing its existing inventory, given by bxt , where b > 0 is another parameter. Thus, if r ≥ 0
denotes the constant interest rate, the present discounted value of the firm’s costs over all
periods t = 0, 1, . . . , T is
T  t h 
X 1 a 2 i
yt + bxt .
t=0
1+r 2
Assume that the firm acts to minimize these discounted costs while still producing at least
x∗ units of output over the T + 1 periods so as to satisfy the constraint xT +1 ≥ x∗ .

The easiest way to solve this problem using methods that we discussed in class is to convert
it into a constrained maximization problem by multiplying the firm’s costs by −1. Consider,

2
therefore, the problem of choosing sequences {yt }Tt=0 for output and {xt }Tt=1
+1
for the inventory
stock in order to maximize
T  t h 
X 1 a 2 i
− yt − bxt .
t=0
1+r 2

subject to the constraints

xt + yt ≥ xt+1
for all t = 0, 1, . . . , T ,
x0 = 0,
and
xT +1 ≥ x∗ > 0.

Strictly speaking, this problem only makes sense if the firm produces positive amounts of
output yt in each period t = 0, 1, . . . , T . It is possible to show, however, that this will be
true so long as the final inventory requirement x∗ is large enough. For simplicity, you can
assume that this condition holds, so that there is no need to explicitly impose nonnegativity
constraints on yt or xt when solving the problem below.

a. Write down the Hamiltonian for this problem. Recall from our conversations in class
that this can be done in a number of different ways ways: feel free to use whichever
formulation of the Hamiltonian you find most convenient.

b. Next, write down the first-order condition and the pair of difference equations that,
according to the maximum principle, must be satisfied by the sequences of values that
solve the firm’s problem. Again, as we discussed in class, there are a number of ways
to do this, but make sure that your statement of the optimality conditions is consistent
with your chosen definition of the Hamiltonian.

c. Finally, use your results to answer the following question. Assuming that, as indicated
above, the firm’s output yt is always strictly positive, the cost parameters a > 0 and
b > 0 are both strictly positive, and the interest rate r ≥ 0 is nonnegative, will the firm
optimally choose to have production yt that is rising, falling, or constant over time?

3
 Solutions to Midterm Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2013

Thursday, October 31, 1:30 - 2:45pm

1. The Kuhn-Tucker Theorem

The problem is finding values for two variables, x and y, to maximize the objective function

F (x, y) = xy,

subject to the constraints

8 ≥ x2 + y 2 ,
x ≥ 0,
and
y ≥ 0.

a. One definition of the Lagrangian for this problem is

L(x, y, λ, µ, φ) = xy + λ(8 − x2 − y 2 ) + µx + φy.

b. The Kuhn-Tucker theorem implies that there exist values λ∗ , µ∗ , and φ∗ that, together
with the values of x∗ and y ∗ that solve the problem, satisfy the first-order conditions

L1 (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) = y ∗ − 2λ∗ x∗ + µ∗ = 0,

and
L2 (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) = x∗ − 2λ∗ y ∗ + φ∗ = 0,
the constraints
L3 (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) = 8 − (x∗ )2 − (y ∗ )2 ≥ 0,
L4 (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) = x∗ ≥ 0,
and
L5 (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) = y ∗ ≥ 0,
the nonnegativity conditions
λ∗ ≥ 0,
µ∗ ≥ 0,
and
φ∗ ≥ 0,

1
 and the complementary slackness conditions
λ∗ [8 − (x∗ )2 − (y ∗ )2 ] = 0,
µ∗ x∗ = 0,
and
φ∗ y ∗ = 0.

c. Suppose first that the solution to this problem has x∗ = 0. Then the first-order condition
for x requires that
y ∗ + µ∗ = 0.
But this last expression, coupled with the constraint that y ∗ ≥ 0 and the nonnegativity
condition that µ∗ ≥ 0, can only hold if y ∗ = µ∗ = 0 as well. Now the first-order
condition for y requires that φ∗ = 0 and the complementary slackness condition
λ∗ [8 − (x∗ )2 − (y ∗ )2 ] = 8λ∗ = 0
requires that λ∗ = 0 as well. Thus, the combination of values (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) =
(0, 0, 0, 0, 0) satisfies all of the Kuhn-Tucker conditions and provides one candidate for
a solution to the original problem.
Suppose next that the solution to the problem has x∗ > 0. In this case, the first-order
condition for y ∗ , rewritten as
x∗ + φ∗ = 2λ∗ y ∗
implies that y ∗ > 0 must hold as well, since if instead y ∗ = 0, this equation would
require that φ∗ < 0, which is ruled out by the nonnegativity condition for that Lagrange
multiplier. Since x∗ > 0 and y ∗ > 0 must both hold, complementary slackness requires
that µ∗ = 0 and φ∗ = 0 must hold as well. Now the first-order condition for x implies
that
y∗
λ∗ = ∗
2x
while the first-order condition for y implies that
x∗
λ∗ = ,
2y ∗
but, together, these conditions imply that (x∗ )2 = (y ∗ )2 . Moreover, since λ∗ = 1/2 > 0,
the complementary slackness condition
λ∗ [8 − (x∗ )2 − (y ∗ )2 ] = 0
requires that
8 = (x∗ )2 + (y ∗ )2 ,
and hence that (x∗ )2 = (y ∗ )2 = 4. Thus, the combination of values (x∗ , y ∗ , λ∗ , µ∗ , φ∗ ) =
(2, 2, 1/2, 0, 0) satisfies all of the Kuhn-Tucker conditions and provides another candi-
date for a solution to the original problem. But x∗ = y ∗ = 2 yields a higher value for
the objective function than x∗ = y ∗ = 0, implying that the solution to the problem
sets x∗ = y ∗ = 2.

2
2. Expenditure Minimization

The consumer solves

1/2 1/2
min p1 c1 + p2 c2 subject to 2c1 + 2c2 ≥ Ū .
c1 ,c2

a. The Lagrangian for this problem can be defined as

1/2 1/2
L(c1 , c2 , λ) = p1 c1 + p2 c2 − λ(2c1 + 2c2 − Ū ).

b. The Kuhn-Tucker theorem implies that there exists a value λ∗ that, together with the
values of c∗1 and c∗2 that solve the consumer’s problem, satisfies the first-order conditions
L1 (c∗1 , c∗2 , λ∗ ) = p1 − λ∗ (c∗1 )−1/2 = 0
and
L2 (c∗1 , c∗2 , λ∗ ) = p2 − λ∗ (c∗2 )−1/2 = 0,
the constraint
2(c∗1 )1/2 + 2(c∗2 )1/2 ≥ Ū
the nonnegativity condition
λ∗ ≥ 0,
and the complementary slackness condition
λ∗ [2(c∗1 )1/2 + 2(c∗2 )1/2 − Ū ] = 0.

c. The first-order conditions imply that

(c∗1 )1/2 = λ∗ /p1
and
(c∗2 )1/2 = λ∗ /p2 .
Since the prices of the two goods are strictly positive and the utility function is strictly
increasing, the constraint will always bind at the optimum. Hence, using the expres-
sions for (c∗1 )1/2 and (c∗2 )1/2 just derived,
Ū = 2(c∗1 )1/2 + 2(c∗2 )1/2 = 2λ∗ /p1 + 2λ∗ /p2 ,
or  −1  
∗ Ū 1 1 Ū p 1 p2
λ = + = .
2 p1 p2 2 p 1 + p2
Substituting this expression for λ∗ back into the earlier ones for (c∗1 )1/2 and (c∗2 )1/2
yields
  2
Ū p2
c1 =
2 p 1 + p2
and   2
Ū p1
c2 =
2 p 1 + p2

3
3. Production and Inventories

The firm chooses sequences {yt }Tt=0 for output and {xt }Tt=1
+1
for the inventory stock in order
to maximize
T  t h 
X 1 a i
− yt2 − bxt .
t=0
1+r 2
subject to the constraints
xt + yt ≥ xt+1
for all t = 0, 1, . . . , T ,
x0 = 0,
and
xT +1 ≥ x∗ > 0.

a. The Hamiltonian for this problem can be defined as

 t h 
1 a i
H(xt , πt+1 ; t) = max − yt2 − bxt + πt+1 yt .
yt 1+r 2

b. The maximum principle implies that the solution to the firm’s dynamic optimization
problem is characterized by the first-order condition for the value of yt that solves the
static and unconstrained optimization problem stated in the definition of the Hamil-
tonian,  t
1
− ayt + πt+1 = 0,
1+r
and the pair of difference equations
 t
1
πt+1 − πt = −Hx (xt , πt+1 ; t) = b
1+r
and
xt+1 − xt = Hπ (xt , πt+1 ; t) = yt .

c. The first-order condition for yt implies that

 t
1
πt+1 = ayt
1+r
and  t−1
1
πt = ayt−1 .
1+r
Substituting these expressions into the difference equation linking πt+1 to πt yields
 t  t−1  t
1 1 1
ayt − ayt−1 = b
1+r 1+r 1+r

4
or, more simply,
yt − (1 + r)yt−1 = b/a.
Since a > 0, b > 0, r ≥ 0, and output yt is always strictly positive, this last question
implies that

0 < b/a = yt − (1 + r)yt−1 = yt − yt−1 − ryt−1 ≤ yt − yt−1 .

Evidently, the firm finds it optimal to have output increasing over time, with yt > yt−1
for all t = 0, 1, . . . , T . The cost of storing inventories counteracts the firm’s desire to
minimize the convex costs of production by smoothing output over time, so that the
firm produces more output in later periods than in earlier periods.

5
 Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2013

Due Tuesday, December 17 at 12 noon

This exam has two questions on five pages; please check to make sure that your copy has
both questions and all five pages. Question one has five parts and question two has four
parts. Each part of each question will be weighted equally in determining your overall exam
score, so that question one is worth 25 points in total and question two is worth 20 points
in total.

This is an open-book exam, meaning that it is fine for you to consult your notes, textbooks,
and other written references when working on your answers to the questions. I expect you
to work independently on the exam, however, without discussing the questions or answers
with anyone else, inside or outside of the class: the answers you submit must be yours and
yours alone.

1. Dynamic Pricing by an Industry Leader

Consider a firm that is the “industry leader.” This firm acts partly like a monopolist, in the
sense that it perceives itself as facing a downward-sloping demand curve and sets the price
for its output accordingly. A number of other firms produce the same good, however, and
expand or contract the amount they produce and sell in a way, described in detail below,
that the leading firm must take as given.

This industry leader produces output at constant marginal cost denoted by c > 0. The
demand for its output is described by

d[p(t), x(t)] = a − bp(t) − x(t),

where a > 0 and b > 0 are positive parameters, p(t) is the leader’s price, and x(t) measures
the amount of output that the other firms produce and sell, thereby subtracting from the
amount that the leader can sell. Hence, the leader’s profits at each time t ∈ [0, ∞) are

[p(t) − c][a − bp(t) − x(t)].

Assume that all other firms in the industry “follow the leader” by setting the same price for
their output. This induces the followers to produce and sell more when the leader raises its
price and to produce and sell less when the leader lowers its price. Suppose, in particular,
that x(t) evolves according to
ẋ(t) = k[p(t) − p̄] (1)
where, as usual, ẋ(t) = dx(t)/dt denotes the derivative of x(t) with respect to t and where
k > 0 and p̄ > 0 are also positive parameters.

1
The firm must therefore choose a continuously differentiable function p(t) describing the
time path for its price to maximize the present discounted value of its profits over the
infinite horizon, Z ∞
e−rt [p(t) − c][a − bp(t) − x(t)]dt
0

where r > 0 is the constant rate of interest, subject to the constraint shown in (1), which
must hold for all t ∈ [0, ∞), taking the initial value x(0) as given.

This problem does not map directly into the form of the general, continuous-time dynamic
optimization problem that we studied in class, because the constraint in (1) is depicted as an
equality and because, if we allowed the industry leader to “freely dispose of market share,”
the version of (1) that would be written as an inequality would take the form

ẋ(t) ≥ k[p(t) − p̄],

indicating that ẋ(t) could be larger that the value implied by the strict equality in (1) if
the leader simply decided to allow the followers to take away additional demand at time t.
Using the change of variables
y(t) = −x(t), (2)
however, the industry leader’s problem can be restated as one of choosing continuously
differentiable functions p(t) for t ∈ [0, ∞) and y(t) for t ∈ (0, ∞) to maximize
Z ∞
e−rt [p(t) − c][a − bp(t) + y(t)]dt
0

subject to the constraints

−k[p(t) − p̄] ≥ ẏ(t) (3)
for all t ∈ [0, ∞), taking y(0) as given.

In all that follows, you can assume for simplicity that the leading firm will always choose a
positive price for its output, so that there is no need to explicitly impose an nonnegativity
constraint on p(t), and you can assume in addition that the constraint in (3) will always
bind when the leading firm behaves optimally.

a. Set up the Hamiltonian for the industry leader’s problem. Recall from our conversations
in class that this can be done in a number of different ways: feel free to use whichever
formulation of the Hamiltonian you find most convenient.

b. Next, write down the first-order condition and the pair of differential equations that,
according to the maximum principle, must be satisfied by the choices of p(t) and y(t)
that solve the leader’s problem. Again, as we discussed in class, there are a number
of ways to do this, but make sure that your statement of the optimality conditions is
consistent with your chosen definition of the Hamiltonian.

2
 c. A problem in interpreting the optimality conditions as they are given by the maximum
principle is that those conditions make reference to an additional function, which we
called either π(t) or θ(t) in our in-class discussions, that gets introduced in defining the
Hamiltonian and that is closely related to the Lagrange multiplier on the constraint
(3) that we would have introduced if we had decided to solve the problem using the
Kuhn-Tucker theorem instead. By using the change of variables in (2) in reverse and
by manipulating the optimality conditions you obtained in part (b), above, derive a
pair of differential equations linking ṗ(t) and ẋ(t) to expressions involving only the
functions p(t) and x(t) and the model’s parameters a, b, c, k, p̄, and r.

d. Use the pair of differential equations you just derived, in part (c), to find steady-state
values of p(t) and x(t) that will prevail when the industry leader chooses these variables
optimally and when ṗ(t) = ẋ(t) = 0.

e. Draw a phase diagram that shows that, for any given initial value of x(0), there is a
unique value of p(0) such that, starting from x(0) and p(0), x(t) and p(t) both converge
to their steady-state values. In doing this, you can assume that the model’s parameters
are such that the steady state values of x(t) and p(t) are both strictly positive.

2. Real Business Cycles

Consider a stochastic economy in which time is discrete, the horizon is infinite, and output Yt
gets produced using capital Kt and labor Lt according to the aggregate production function

Yt = Zt Ktα Lt1−α ,

with 0 < α < 1. In this model, Zt represents a random shock to aggregate productivity that
will give rise to “real,” meaning non-monetary, business cycle fluctuations in output. This
shock is assumed to follow a first-order autoregressive process in its natural logarithm, so
that
ln(Zt+1 ) = ρ ln(Zt ) + εt+1 ,
where the parameter ρ, assumed to lie between zero and one, measures the degree of serial
correlation in the shock and the innovation εt+1 is independently and identically distributed
(iid) over time, with Et εt+1 = 0.

During each period t = 0, 1, 2, . . ., a social planner chooses the amount of labor Lt to be

allocated to production, then divides the economy’s output up into an amount Ct to be
consumed and an amount It to be invested, subject to the aggregate resource constraint

Zt Ktα Lt1−α ≥ Ct + It .

The timing of the realization of the productivity shock is such that, when Lt , Ct , and It are
chosen, the value of Zt is known but the value of Zt+1 is still viewed as random.

As in the optimal growth example we studied in class, physical capital is assumed to de-
preciate fully in use during each period; as you are about to see, that makes it possible to

3
find closed-form solutions for the model’s key variables even in this stochastic version of the
model and even with a variable labor supply. Thus, next period’s capital equals this period’s
investment:
Kt+1 = It
for all t = 0, 1, 2, . . ..

The model’s specification is completed by assuming that the economy has a representative
consumer/worker, with preferences over consumption and leisure in each period described
by the expected utility function
∞
X
E0 β t [ln(Ct ) + θ ln(1 − Lt )], (4)
t=0

where the discount factor β lies between zero and one and where the total amount of time
available within each period is normalized to equal one, so that θ > 0 measures the weight
on leisure 1 − Lt relative to consumption Ct in the single-period utility function.

The social planner’s problem can now be stated as one of choose contingency plans for Ct ,
t = 0, 1, 2, . . ., Lt , t = 0, 1, 2, . . ., and Kt , t = 1, 2, 3, . . . to maximize the expected utility
function in (4) subject to the constraints

Zt Ktα L1−α
t ≥ Ct + Kt+1 , (5)

for all t = 0, 1, 2, . . ., taking the initial values of K0 and Z0 as given.

Once more, in solving this problem, you can assume that it is not necessary to explicitly
impose nonnegativity constraints on any of the variables and that the aggregate resource
constraint in (5) will always bind when quantities are chosen optimally.

a. Write down the Bellman equation for the social planner’s problem.

b. As a first step in solving this problem, guess that the value function takes the time-
invariant form
v(Kt , Zt ) = F + G ln(Kt ) + H ln(Zt ),
where F , G, and H are constant coefficients that depend on the model’s parameters.
Then, using this guess, write down the first-order conditions and the envelope condition
that help describe the social planner’s optimal choices of Ct , Lt , and Kt .

c. Use your results from part (b), above, to find solutions that show how the optimal
values of Ct , Lt , and Kt+1 = It chosen during each period t = 0, 1, 2, . . . depend on the
realized value Zt of the productivity shock, the beginning-of-period capital stock Kt ,
and the model’s parameters α, ρ, β, and θ.

d. Use your results from part (c), above, to answer the following questions:

4
i. Does this real business cycle model account for the procyclicality of consumption,
that is, the strong tendency seen in actual economies around the world for ag-
gregate consumption Ct and aggregate output Yt to move up and down together
during booms and recessions?
ii. Does this model account for the procylicality of investment, that is, the strong
tendency seen in actual economies around the world for aggregate investment
It and aggregate output Yt to move up and down together during booms and
recessions?
iii. Does this model account for the procylicality of employment, that is, the strong
tendency seen in actual economies around the world for aggregate time working
Lt and aggregate output Yt to move up and down together during booms and
recessions?

5
 Solutions to Final Exam

EC720.01 - Math for Economists Peter Ireland

Boston College, Department of Economics Fall 2013

Due Tuesday, December 17 at 12 noon

1. Dynamic Pricing by an Industry Leader

The industry leader chooses continuously differentiable functions p(t) for t ∈ [0, ∞) and y(t)
for t ∈ (0, ∞) to maximize
Z ∞
e−rt [p(t) − c][a − bp(t) + y(t)]dt
0

subject to the constraints

−k[p(t) − p̄] ≥ ẏ(t)
for all t ∈ [0, ∞), taking y(0) as given.

a. The present-value Hamiltonian for the industry leader’s problem can be defined as

H(y(t), π(t); t) = max e−rt [p(t) − c][a − bp(t) + y(t)] − π(t)k[p(t) − p̄].
p(t)

b. With the Hamiltonian defined as above, the maximum principle implies that the choices
of p(t) and y(t) that solve the leader’s problem must satisfy the first-order condition

e−rt [(a + bc) − 2bp(t) + y(t)] − π(t)k = 0

and the differential equations

π̇(t) = −Hy (y(t), π(t); t) = −e−rt [p(t) − c]

and
ẏ(t) = Hπ (y(t), π(t); t) = −k[p(t) − p̄].

c. To simplify the optimality conditions from part (b), above, replace y(t) with −x(t) and
differentiate the first-order condition with respect to t to obtain

π̇(t)k = −re−rt [(a + bc) − 2bp(t) − x(t)] − e−rt [2bṗ(t) + ẋ(t)].

Substituting the differential equations π̇(t) = −e−rt [p(t) − c] and ẋ(t) = −ẏ(t) =
k[p(t) − p̄] into this last expression yields

−e−rt [p(t) − c]k = −re−rt [(a + bc) − 2bp(t) − x(t)] − e−rt 2bṗ(t) − e−rt k[p(t) − p̄]

1
 or, more simply,

k(p̄ − c) − r[(a + bc) − 2bp(t) − x(t)]

ṗ(t) = ,
2b
which can be combined with
ẋ(t) = k[p(t) − p̄]
to obtain a system of two differential equations in the two unknown functions p(t) and
x(t).

d. When ṗ(t) = ẋ(t) = 0, the second of the two differential equations derived in part (c),
above, requires that
p(t) = p∗ = p̄.
in any steady state. Hence, the first differential equation implies that

k(p̄ − c) = r[(a + bc) − 2bp̄ − x(t)]

or
x(t) = x∗ = (k/r)(c − p̄) + (a + bc) − 2bp̄
in any steady-state as well. Assuming that the model’s parameters are such that the
expression on the right-hand side of this last expression is strictly positive, the model
has a unique steady state in which the industry leader’s price equals p∗ = p̄ and the
remaining firms produce and sell x∗ units of output.

e. The differential equation

ẋ(t) = k[p(t) − p̄]
implies that, in a phase diagram with x(t) on the horizontal axis and p(t) on the vertical
axis, the ẋ(t) = 0 locus will be a horizontal line at p(t) = p̄. Above this line x(t) will
be increasing, and below this line, x(t) will be decreasing. Meanwhile, the differential
equation
k(p̄ − c) − r[(a + bc) − 2bp(t) − x(t)]
ṗ(t) =
2b
implies that the ṗ(t) = 0 locus will be the straight, downward-sloping line described
by the equation

(k/r)(c − p̄) + (a + bc) x(t) x∗ + 2bp̄ x(t)

p(t) = − = − .
2b 2b 2b 2b
If, as assumed above, the model’s parameters are such that the steady-state value x∗
is strictly positive, then the y-intercept of this line lies above p̄. Moreover, p(t) will be
increasing above the ṗ(t) = 0 locus and decreasing below the ṗ(t) = 0 locus. Putting
all this information together yields a phase diagram that shows that, starting from any
initial value of x(0), there is a unique value of p(0) such that, starting from x(0) and
p(0), x(t) and p(t) both converge to their steady-state values.

2
For a more general and thorough analysis of this model, including an argument that confirms
that trajectories along the stable manifold are, in fact, optimal, see the article by Darius
W. Gaskins, “Dynamic Limit Pricing: Optimal Pricing under Threat of Entry,” Journal of
Economic Theory Vol.3 (1971): pp.306-322.

2. Real Business Cycles

The social planner chooses contingency plans for Ct , t = 0, 1, 2, . . ., Lt , t = 0, 1, 2, . . ., and

Kt , t = 1, 2, 3, . . . to maximize the expected utility function
∞
X
E0 β t [ln(Ct ) + θ ln(1 − Lt )]
t=0

subject to the constraints

Zt Ktα L1−α
t ≥ Ct + Kt+1
for all t = 0, 1, 2, . . ., taking as given the initial values of K0 and Z0 and the stochastic law
of motion
ln(Zt+1 ) = ρ ln(Zt ) + εt+1 ,
for the aggregate productivity shock.

a. The Bellman equation for this problem can be written as

v(Kt , Zt ) = max ln(Ct ) + θ ln(1 − Lt ) + βEt v(Zt Ktα Lt1−α − Ct , Zt+1 ).

Ct ,Lt

3
b. Using the conjecture that the value function takes the time-invariant form

v(Kt , Zt ) = F + G ln(Kt ) + H ln(Zt ),

the Bellman equation becomes

F + G ln(Kt ) + H ln(Zt ) = max ln(Ct ) + θ ln(1 − Lt ) + βF

Ct ,Lt

+βG ln(Zt Ktα L1−α

t − Ct ) + βHρ ln(Zt ).

The first-order conditions for Ct and Lt are

1 βG
= (1)
Ct Zt Kt Lt1−α − Ct
α

and
θ (1 − α)βGZt Ktα L−αt
= , (2)
1 − Lt Zt Ktα L1−α
t − C t

and the envelope condition for Kt is

G αβGZt Ktα−1 L1−α

t
= . (3)
Kt Zt Ktα L1−α
t − Ct
where F , G, and H are constant coefficients that depend on the model’s parameters.

c. Although there are many other ways to derive these same solutions, start here by
rearranging the terms in (1) to obtain
 
1
Ct = Zt Ktα L1−α
t . (4)
1 + βG

Next, rewrite (3) as

GZt Ktα Lt1−α − GCt = αβGZt Ktα Lt1−α ,

which, in light of (4), implies

1
= 1 − αβ, (5)
1 + βG
and hence
α
G= . (6)
1 − αβ
Combining (4) and (5) yields

Ct = (1 − αβ)Zt Ktα L1−α

t . (7)

Substitute this last expression into (2) to obtain

θ (1 − α)βGZt Ktα L−α

t
= ,
1 − Lt αβZt Ktα L1−α
t

4
 which simplifies to
θ (1 − α)G
= ,
1 − Lt αLt
and can then be used together with (6) to find
1−α
Lt = . (8)
1 − α + θ(1 − αβ)
Equation (8) provides the desired solution for employment Lt . Substituting (8) into
(7) yields the solution for Ct :
 1−α
1−α
Ct = (1 − αβ) Zt Ktα . (9)
1 − α + θ(1 − αβ)
Finally, substituting (7) into the aggregate resource constraint and using (8) again
yields the solution for
 1−α
1−α
Kt+1 = It = αβ Zt Ktα . (10)
1 − α + θ(1 − αβ)
Equations (8)-(10) can be used to construct sequences for the optimal values of Ct ,
Lt , and Kt , starting from any initial value K0 and given any realized sequences of
aggregate productivity shocks Zt . And while it is not necessary to find these values in
order to obtain the results just derived, it is also possible to find the solutions
  
1 1
H=
1 − βρ 1 − αβ
and
    
1 αβ 1−α
F = ln(1 − αβ) + ln(αβ) + ln(L) + θ ln(1 − L) ,
1−β 1 − αβ 1 − αβ
where L is the constant value of Lt from (8).
d. To help interpret the results from above, note first that (8) implies that employment Lt
is constant, while (9) and (10) can be written more simply as
Ct = (1 − αβ)Yt
and
It = αβYt ,
showing that it is optimal in this model to divide aggregate output Yt up into the
constant fraction 1 − αβ to be consumed and the constant fraction αβ is be invested.
It then follows that:
i. The model accounts for the procyclicality of consumption, since Ct and Yt move up
and down together.
ii. The model also accounts for the procyclicality of investment, since It and Yt move
up and down together.
iii. But this simple model cannot account for the procyclicality of employment, since
it implies that Lt is constant.