Fall 2013 John Rust

Economics 425 Georgetown University

Solutions to Problem Set 2

1. Gradient is orthogonal to the indifference curve Let u : RN → R be a differentiable function (for
concreteness you can think of this as a utility function). Let I = {x ∈ RN |u(x) = u} be a level set of the
function (in the utility function case, an indifference curve). Prove that if x ∈ I then hx, ∇ui = 0 where
∇u is the gradient of u.

Answer This proposition is not quite correct as stated. The correct statement is that if x is
any point in the (shifted) tangent hyperplane to the indifference curve at a point x0 ∈ RN then
h∇u(x0 ), xi = 0. What is the tangent hyperplane? It is the generalization of a tangent line in the
case N = 2. You can visualize this as a line in the (x1 , x2 ) plane that is tangent to the utility func-
tion indifference curve at a particular point x0 = (x1,0 , x2,0 ) ∈ R2 , i.e. the line both a) touches the
indifference curve at x0 and b) has the same slope as the indifference curve at the point x0 . Notice
that a line is equivalent to R1 , which is one dimension less than the dimension of the space R2 . In
RN a hyperplane is a linear subspace of dimension N − 1, i.e. one dimension less than the total
dimension of the overall space, N. It can be shown that any hyperplane can be represented as the
set of all x that are orthogonal to some vector a ∈ RN , i.e. any hyperplane consists of all x ∈ RN
in the set H(a) given by
H(a) = {x ∈ RN |ha, xi = 0}. (1)
Since l(x) = ha, xi is a linear mapping from RN to R, the hyperplane is the null space of this linear
mapping (this is a term you would know if you took linear algebra), and so the hyperplane – the
null space — is just the set of all x ∈ RN that are orthogonal (perpendicular) to the fixed vector
a ∈ RN .
So a tangent hyperplane to a function u : RN → R will be a set of points on a hyperplane of RN
that satisfies a) the (shifted) hyperplane touches the function u(x) at x0 , and b) the hyperplane has
the same slope as u(x) at the point x0 (i.e. it is tangent to u(x) at the point x0 ).
How do we write the intuition of tangency down mathematically and show that implies that the
gradient of u at x0 is orthogonal (i.e. perpendicular to) the tangent hyperplane to u at x0 ? First let
us use a bit more specific notation for the indifference curve to a (differentiable) function u at a
point x0 ∈ RN : we will use the notation Iu (x0 ) = {x ∈ RN |u(x) = u(x0 )} to denote this indifference
curve, which is generally a curved manifold of RN , i.e. a “surface” in RN that is one dimension
less than N the dimension of the full space RN . Thus when N = 2, an indifference curve is a
curved line in R2 and when N = 3 we get an indifference surface which is curved surface in R3
and so forth (note a surface in R3 is “locally” two dimensional, i.e. it is two dimensional instead
of three-dimensional, i.e. not a “solid”).
Now by Taylor’s Theorem, if u(x) is differentiable at a point x0 we can form a linear approxima-
tion to the function u(x) at a point x0 via the linear function l(x) given by

l(x) = u(x0 ) + h∇(x0 ), (x − x0 )i. (2)

Notice that this linear approximation l(x) satisfies a) l(x0 ) = u(x0 ) (i.e. the linear approximation
to u(x) touches u at the point x0 , and b) ∇l(x) = ∇u(x0 ), i.e. the slope of the linear function l(x)

1
 equals the slope of the nonlinear function u(x) at the point x0 (which is just the gradient, ∇u(x0 )).
So now we can define the tangent plane to the function u(x) at the point x0 as the indifference
curve for the function l(x) at the point x0

Il (x0 ) = {x ∈ RN |l(x) = l(x0 ) = u(x0 )} (3)

Clearly the indifference curve of a linear function will just be a linear space, i.e. a hyperplane,
rather than a curved surface or manifold which is what an indifference curve of a nonlinear func-
tion typically is.
So now it just boils down to showing that any point x in the tangent hyperplane is orthogonal to
the gradient of u(x) at x0 . But by the definition of the tangent hyperplane, x ∈ Il (x0 ) if and only if
we have
l(x) = u(x0 ) + h∇u(x0 ), (x − x0 )i = l(x0 ) = u(x0 ). (4)
But subtracting u(x0 ) on each side of the equation above, we see this is equivalent to

h∇u(x0 ), (x − x0 )i = 0. (5)

This is almost the result we want, since equation (5) tells us that the point (x − x0 ) is orthogonal
(perpendicular) to ∇u(x0 ). But notice that the indifference curve Il (x0 ) does not generally go
through the origin, i.e. it is generally not the case that 0 ∈ Il (x0 ) (make sure you understand why
this is the case). However we can define a parallel shift of this indifference curve by subtracting
x0 from every point in Il (x0 ). That is, we can define the tangent hyperplane as a parallel shift of
the tangent plane Il (x0 ), so we define this tangent hyperplane by Tl (x0 ) as follows

Tl (x0 ) = {x ∈ RN |x = (y − x0 ) for some y ∈ Il (x0 )}. (6)

Notice that since x0 ∈ Il (x0 ) (x0 is just the point of tangency), it follows that x = (x0 − x0 ) = 0 ∈
Tl (x0 ), so the tangent hyperplane includes the point 0 ∈ RN , and we say that the tangent hyperplane
passes through the origin. Since any point y ∈ Il (x0 ) satisfies the condition h∇u(x0 ), (y − x0 )i = 0,
it follows that the point x = (y − x0 ) ∈ Tl (x0 ) satisfies h∇u(x0 ), xi = 0. That is, we have shown
that any point in the tangent hyperplane to the function u(x) at a point x0 ∈ RN is orthogonal to
the gradient ∇u(x0 ). It should also be clear why Tl (x0 ) is a hyperplane. As we discussed above,
a hyperplane is any set of points in RN that satisfy ha, xi = 0 for some a ∈ RN . But we have just
shown that Tl (x0 ) is the set of all x ∈ RN such that h∇u(x0 ), xi = 0 so letting a = ∇u(x0 ) it follows
that Tl (x0 ) is indeed a hyperplane.
2. Lagrangian saddlepoint solution for constrained optimization problems Consider the following
constrained optimization problem

max u(x) subject to g(x) ≥ 0, and x≥0 (7)

x

where u : Rn → R is a continuous function and g : Rn → Rm are m constraint functions which are also
continuous functions of x. Define the Lagrangian L (x, λ) : R( n + m) → R by

L (x, λ) = u(x) + λ′ g(x) (8)

where
m
λ′ g(x) = hλ, g(x)i = ∑ λ j g j (x). (9)
j=1

2
Definition (x∗ , λ∗ ) is a saddlepoint of L if and only if
L (x∗ , λ∗ ) ≥ L (x, λ∗ ) ∀x ≥ 0
L (x∗ , λ∗ ) ≤ L (x∗ , λ) ∀λ ≥ 0
Theorem If (x∗ , λ∗ ) is a saddlepoint of L then x∗ solves the constrained optimization problem (7).
Prove this theorem. Hint: use the method of proof by contradiction.

Proof I proved this in class, perhaps not in every detail but I used a proof by contradiction to
show that if (x∗ , λ∗ ) is a saddlepoint of L (x, λ) then x∗ must solve the constrained optimization
problem (7). The more difficult thing to prove is the converse to this Theorem, i.e. if x∗ solves
the the constrained optimization problem (7), then there exists a λ∗ ≥ 0 such that (x∗ , λ∗ ) is a
saddlepoint to the Lagrangian L (x, λ). I did not ask you to prove this converse result and it is not
true in general without more assumptions. It can be proven under the stronger assumptions that
{x|g(x) ≥ 0} is a convex set and u(x) is quasiconcave. Then we can appeal to a theorem called the
separating hyperplane theorem to prove the existence of a Lagrange multiplier vector λ∗ such that
(x∗ , λ∗ ) is a saddlepoint of L (x, λ). However this is beyond the level of this class and I emphasize
it is not something I asked for or expected any of you to prove.
3. Prove that if x∗ is an interior solution that maximizes the consumer’s problem below, the indif-
ference curve at x∗ is tangent to the budget line.
max u(x) subject to hp, xi ≤ y (10)
x≥0

where u : Rn → R is a continuously differentiable utility function and p ∈ Rn are positive prices of the n
goods entering the consumer’s utility function.

Answer We need a slightly stronger assumption for this result to hold, namely we need to assume
that more is always better which is mathematically equivalent to ∇u(x) > 0 for any x ≥ 0 (where
the vector inequality ∇u(x) > 0 means that each component of ∇u(x) is strictly greater than zero,
so that ∂x∂ i u(x) > 0 for i = 1, . . . , n). Suppose the conditions to the converse to theorem above
holds, i.e. there exists a λ∗ ≥ 0 such that (x∗ , λ∗ ) is a saddlepoint to the Lagrangian L (x, λ). Since
u(x) is differentiable and (x∗ , λ∗ ) is a saddlepoint, x∗ must maximize L (x, λ∗ ) in x. Since x∗ is
interior — i.e. x∗ > 0 — it follows that the the gradient of L (x, λ∗ ) with respect to x must be
identically 0 at x = x∗ , i.e. we must have
∂
L (x, λ∗ ) = ∇u(x∗ ) − λ∗ p = 0, (11)
∂x
where 0 is interpreted as the zero vector in RN . Since ∇u(x∗ ) > 0 (by the “more is always better”
assumption) and p > 0, it follows that λ∗ > 0. Now we already showed in problem 1 above that
the tangent hyperplane to the indifference curve of u at x∗ , Iu (x∗ ), is perpendicular (orthogonal)
to ∇u(x∗ ) (i.e. it is the set of all x ∈ RN satisfying h∇u(x∗ ), xi = 0). Now we show that the
shifted budget line (the budget hyperplane) is also orthogonal to ∇u(x∗ ), but if both the tangent
hyperplane to the indifference curve and the shifted budget line are both orthogonal to ∇u(x∗ ),
then these must be the same hyperplane and thus parallel and thus, it follows that the budget line
(or plane) must be tangent to the indifference curve Iu (x∗ ). That is, we can define a shifted version
of the budget line by the linear function l(x) given by
l(x) = u(x∗ ) + hλ∗ p, (x − x∗ )i (12)

3
 Notice that a) l(x) touches the indifference curve Iu (x∗ ) since we have l(x∗ ) = u(x∗ ), and b) the
slope of l(x) is the same as the slope of u(x) at x∗ . To see this latter point, note that the slope of
l(x) is just its gradient, which is ∇l(x) = λ∗ ∇p. However by equation (11) we have λ∗ p = ∇u(x∗ ).
So it follows that the slope (gradient) of l(x), which is just the slope of the (shifted) budget line,
equals the slope of the indifference curve at u(x∗ ) which is ∇u(x∗ ). In other words, the budget
line is tangent to the indifference curve at the optimal bundle x∗ .

4. Firm Profit Maximization Problem Consider a firm whose production function has 2 outputs, y1
and y2 and 2 inputs, x1 and x2 . Suppose that its production function is given by
 2 1/2  2 1/2
y1 + 4y22 = x1 + x2 2 (13)

a. Does this production function have increasing, decreasing, or constant returns to scale? (Hint: if
you double both inputs x1 and x2 can you double, more than double, or less than double both of
the outputs y1 and y2 ?)

Answer: We can write a general production function as F(y, x) ≤ 0. The general definition of constant
returns to scale is that if (y, x) is feasible to produce, i.e. if F(y, x) ≤ 0, then for any positive scalar
λ ≥ 0 we have that it is also feasible to produce (λy, λx), i.e. we need to check that F(λy, λx) ≤ 0.
For this problem it is easy to see that there are constant returns to scale since
1/2  2 1/2
F(y, x) = y21 + 4y22 − x1 + x2 2

(14)

and it is easy to check that for any λ ≥ 0 we have F(λy, λx) ≤ 0.

b. Suppose for a moment that we fix input levels so that x1 = x2 = 5. Plot the output possibility
frontier, i.e. plot (in (y2 , y1 ) space) the set of feasible combinations of y1 and y2 that can be
produced using inputs x1 = x2 = 5.

Answer: If x1 = x2 = 5 then the production function constraint tells us that the set of feasible outputs
(y1 , y2 ) that can be produced is
n  2 2 1/2
 √ o
P = (y1 , y2 )| y1 + 4y2 ≤ 50, y1 ≥ 0, y2 ≥ 0 . (15)

2
√ is an ellipse and the production possibility frontier is the graph of the equation y1 +
This region
2
4ys = 50 and is plotted in figure 1 below. We can rewrite the equationi for the production
possibility frontier as s
50 − y21
y2 = . (16)
4

4
 Figure 1: Output Possibility Frontier for inputs x1 = x2 = 5

Production Possibility Frontier for Problem 7−b

4

3.5

2.5
y2

1.5

0.5

0
0 1 2 3 4 5 6 7 8 9 10
y
1

c. Continuing the previous question, if the output prices are p1 = 6 for y1 and p2 = 16 for y2 , and if
we assume that the inputs x1 and x2 are fixed at 5, what combination of outputs (y∗1 , y∗2 ) maximize
the firm’s revenue? If we were to increase x1 by a small amount, on the margin, by how much
would the revenues of the firm increase (i.e. how much does revenue increase for an increase of
amount ε, some small positive number, in input x1 )?
Answer: We want to maximize revenues subject to the fixed input constraints that x1 = x2 = 5. The La-
grangian for this problem is
p q 
L (y1 , y2 , λ) = p1 y1 + p2 y2 + λ 2 2 1/2 2 2
5 + 5 ) − y1 + 4y2 (17)

I am going to leave it to you to write the first order conditions and solve them to get the optimal
outputs (y∗1 , y∗2 ) which are given by
√
∗ 50
y1 = r
p2
1 + 4p22
1
√
50
y∗2 = r (18)
16p21
4 + p2
2

Plugging in the prices p1 = 6 and p2 = 16 into these formulas, we get

y∗1 = 4.2426
y∗2 = 2.8284 (19)

5
 and the maximized value of revenue is
√
R∗ = p1 y∗1 + p2 y∗2 = 6 × 4.2426 + 16 × 2.8284 = 10 50 = 70.71. (20)

The final part of this question is to compute by how much revenue increases if input x1 = 5
increases by a small amount ε. This can be computed as

∂R∗ ∂y∗1 ∂y∗2

 
ε = p1 + p2 ε. (21)
∂x1 ∂p1 ∂p2

Using equation (18) we can compute the derivatives of y∗1 and y∗2 with respect to x1 to get:

∂R∗ ∂y∗ ∂y∗

 
ε = p1 1 + p2 2 ε
∂x1 ∂p1 ∂p2
1 ∗
= R ε
√10
= 50ε
= 7.071ε, (22)

since ∂y∗1 /∂x1 = ∂y∗2 /∂x1 = y∗1 x1 /50 = y∗2 x1 /50 = 1/10 when x1 = 5. We can also use the Lagrange
multiplier from the Lagrangian to get this. From the first order condition for y∗1 we get
√ p1 √ 6
λ∗ = 50
∗ = 50 = 10.00. (23)
y1 4.2426
√
However λ∗ measures the effect of relaxing the constraint, K 50, but we are interested in mea-
suring the effect of increasing x1 by ε. Using the chain rule,
q
∂ x21 + x22 x1 5
=q =√ (24)
∂x1 2
x +x 2 50
1 2

when x1 = x2 = 5. Thus, the effect of an increase of ε in x1 on revenues using the Lagrange

multiplier is q
∂R ∗ ∂ x21 + x22 λ∗ 5 √
= λ∗ = √ = 50 = 7.071. (25)
∂x1 ∂x1 50
So we get the same answer regardless of which route we take to computing it. That’s reassurring!

d. Now consider what the optimal level of inputs should be in order to produce the (y∗1 , y∗2 ) combi-
nation that you computed in part c. If the price of the inputs are w1 = 4 for x1 and w2 = 6 for
x2 , what is the cost-minimizing level of√inputs that can produce (y∗1 , y∗2 )? (Hint: recall that when
x1 = x2 = 5 we have [x1 2 + x2 2 ]1/2 = 50. √So you need to minimize total costs w1 x1 + w2 x2
subject to the constraint that [x1 2 + x2 2 ]1/2 = 50). If we needed to increase output y1 by a small
amount, say by .1, approximately how much would it cost the firm to do this?

Answer: This answer is the same as the answer to problem 6: given that the input isoquants are concave
rather than convex to the origin, the optimal policy is to use the cheaper of the two inputs. Thus,
let w denote the cheaper of the two input prices and let x = denote the quantity of the cheaper input

6
 that was used to produce the goods. In this case good 1√is cheaper since its price is p1 = 4 and
good 2’s price is p2 = 6. So the level of x1 used is x1 = 50 = 7.071. To compute the additional
cost of increasing y1 by a small amount we need to consider the different ways we could increase
y1 . One way would be to decrease the production of y2 to increase the production of y1 , leaving
the input requirement x unchanged. The other way is to assume that we increase the level of input
1, x. But this results in joint production of not only y1 but also y2 . To measure the incremental cost
properly, we have to deduct the increased revenues resulting from the use of the addition inputs.
Computing the incremental cost the first way (i.e. holding x constant and treating the cost as the
opportunity cost of lost sales of good 2). Totally differentiating the production function constraint
we bet
y dy 4y dy
q 1 1 + q 2 2 = 0. (26)
2 2
y1 + 4y2 2 2
y1 + y2
and solving for dy2 /dy1 we get
dy2 y1
=− (27)
dy1 4y2
so the opportunity cost in terms of lost output of y2 from increasing y1 by a small amount ε is ε/4
of the ratio y1 /y2 . If the firm is maximizing revenues, then we saw above that the optimal y∗1 and
y∗2 are produced in the ratio of 4p1 /p2 . With p1 = 6 and p2 = 16, then we have
dy2 y∗ p1 6
= − 1∗ = − = − . (28)
dy1 4y2 p2 16
Since the price of good 2 is p2 = 16, the cost to the firm of increasing y1 by ε units is
dy∗2
p2 ε = −6ε. (29)
dy∗1
You should show that the incremental cost will be the same if the firm decides to produce more
of y1 by increasing its input level x. However be careful to deduct the extra revenue from sales of
increased amount of output of good 2 from the cost of the extra inputs!

e. Now step back and look at the firm overall. Is the production plan (y∗1 , y∗2 , x∗1 , x∗2 ) that you computed
in parts c and d above a profit maximizing production plan for this firm? Why or why not?

Answer: Since the firm’s production function has constant returns to scale, the profit maximizing scale of
operations is not well-defined: if the firm can make positive profits at a given scale of operations,
then it could increase profits without bounds by scaling up its levels of inputs and outputs simul-
taneously. If the prices are such that the firm earns zero profits at one scale of operation, then it is
not hard to show that it will earn zero profits at any other scale of operations, so in either case, the
scale of the firm is not well defined. So the only thing we can do is to determine the optimal com-
bination of outputs for any arbitrarily fixed scale of operations.
√ We can fix the scale by setting the
cheaper of the two inputs, x to a given level such as x = 50. With this (arbitrary normalization),
the optimal outputs computed in part a. are optimal, and thus part of a profit maximizing plan,
provided profits are positive. Profit is
√ √
Π∗ = p1 y∗1 + p2 y∗2 − wx = R∗ − wx = 10 50 − w 10. (30)
√
Thus as long as w ≤ 50 = 7.071 the firm makes positive profits, and if it could, √ it would want to
expand its scale of operations without bound to drive its profits to ∞. If w = 50 the firm makes

7
 √
0 profits regardless of its scale of operations. If w < 50 then the firm would make a loss at any
scale of operations, so its best course of action is to shut down.
Another way to solve this problem is to write down the Lagrangian to the firm’s full profit maxi-
mization problem. The Lagrangian is
q q 
L (y1 , y2 , x1 , x2 , λ) = p1 y1 + p2 y2 − w1x1 − w2x2 + λ 2 2 2 2
w1 + w2 − y1 + 4y2 (31)

Actually the algebra becomes a lot easier if we square both sides of the production function
constraint and write it as
y21 + 4y22 = x1 2 + x2 2 (32)
This is equivalent to the original production function constraint and the Lagrangian for the profit
maximization problem with this simpler but equivalent version of the constraint is

L (y1 , y2 , x1 , x2 , λ) = p1 y1 + p2 y2 − w1x1 − w2 x2 + λ x21 + x22 − y21 − 4y22 .

(33)

Now, recall that we want to maximize the Lagrangian over the variables (y1 , y2 , x1 , x2 ) but to min-
imize it over the λ variable. The first order conditions for the maximization of L (y1 , y2 , x1 , x2 , λ)
with respect to (y1 , y2 , x1 , x2 ) are

∂
L (y1 , y2 , x1 , x2 , λ) = p1 − 2λy1 ≤ 0
∂y1
∂
L (y1 , y2 , x1 , x2 , λ) = p2 − 8λy2 ≤ 0
∂y2
∂
L (y1 , y2 , x1 , x2 , λ) = w1 + 2λx1 ≤ 0
∂x1
∂
L (y1 , y2 , x1 , x2 , λ) = w2 + 2λx2 ≤ 0. (34)
∂x2
I have written the first order conditions as inequalities in equation (34) to account for the possi-
bility of corner solutions. We have already been alerted to this possibility in part b above, where
we plotted the isoquants in (x1 , x2 ) space and showed they were concave and thus the input cost-
minimizing bundle would be to let either x1 = 0 or x2 = 0 depending on which of the two inputs
is more expensive. When w1 6= w2 we can see from the first order conditions for the Lagrangian
in (34) above that it is impossible to have an interior solution in (x1 , x2 ) (i.e. where x1 > 0 and
x2 > 0 simultaneously). To see this, if there was an interior solution, then we would be able to
take a second derivative of the Lagrangian to check the second order conditions. We would find
the following:

∂2
L (y1 , y2 , x1 , x2 , λ) = −2λ ≤ 0
∂2 y1
∂
L (y1 , y2 , x1 , x2 , λ) = −8λ ≤ 0
∂y2
∂
L (y1 , y2 , x1 , x2 , λ) = 2λ ≥ 0
∂x1
∂
L (y1 , y2 , x1 , x2 , λ) = 2λ ≥ 0. (35)
∂x2

8
The second order conditions tell us that while there can be an interior solution for (y1 , y2 ) (since
the second derivative of the Lagrangian with respect to y1 and y2 are both negative if λ > 0), there
cannot be an interior solution for (x1 , x2 ) when λ > 0 since then the second order conditions for
x1 and x2 are positive indicating that these would constitute a local minimum of the Lagrangian,
but we are looking for x1 and x2 that maximize the Lagrangian. So we conclude that if λ > 0 then
the only possible solution is for x1 = x2 = 0 and the first does not produce anything and earns zero
profit. However it is also possible that λ = 0. In this case the second order condition in equation
(35) above would be zero, which is not necessarily any contradiction, but the first order conditions
in equation (34) will no longer hold.
So this is a very tricky problem where the Lagrangian approach is not that useful. To see what
the general solution is we need to take another tack, which is to use the insight from part b above
that the firm will set x2 = 0 if w2 > w1 and set x1 = 0 if w1 > w2 . Let x denote the amount of the
cheaper input that the firm uses and let w = min[w1 , w2 ]. For a fixed level of inputs we can solve
a conditional profit maximization problem namely

max p1 y1 + p2 y2 − wx subject to: x2 = y21 + 4y22 (36)

y1 ,y2

Call the solution to this problem Π(p1 , p2, w|x) the conditional profit function since it is con-
ditional on the firm restricting its inputs to the level x. What we want to do now is solve this
conditional profit maximization problem and find an expression for Π(p1 , p2 , w|x). Then we can
do a second stage optimization to find the optimal level of the input x to maximize profits. The
Lagrangian for the first stage conditional profit maximization problem is

L (y1 , y2 , λ) = p1 y1 + p2 y2 − wx + λ x2 − y21 − 4y22 .

(37)

The first order conditions are

∂
L (y1 , y2 , x1 , x2 , λ) = p1 − 2λy1 = 0
∂y1
∂
L (y1 , y2 , x1 , x2 , λ) = p2 − 8λy2 = 0. (38)
∂y2
Now we can solve these equations to get y1 = p1 /2λ and y2 = p2 /8λ. We can substitute these
into the production function constraint x2 = y21 + 4y22 to solve for λ
    
2 p1 2 p2 2
x = + (39)
2λ 8λ

Solving for λ we get q

p21 + p22 /4
λ= . (40)
2x
Substituting this equation for λ into the equations for y1 and y2 above we get
xp1
y1 = q
p21 + p22 /4
xp2
y2 = q . (41)
2
4 p1 + p2 /42

9
 Now, substituting these formulas for the optimal outputs results in the following formula for the
conditional profit function Π(p1 , p2 , w|x)
q 
Π(p1 , p2 , w|x) = x 2 2
p2 + p2 /4 − w . (42)

Now, if the term in brackets in equation (42) is strictly positive, the firm would want to increase
production without bound, and drive the input level x to infinity. But infinite inputs and infinite
profits is not a legitimate solution. If the term in brackets is zero, then the firm makes zero profits
regardless of the scale of production x and it does not care what value x would be, and technically
there are a continuum of profit maximizing solutions in this case. If the term in brackets in
equation (42) is negative, then the firm wants to set x = 0 and so it does not produce anything and
makes zero profits.

f. Super bonus question: If you answered in part e that the production plan (y∗1 , y∗2 , x∗1 , x∗2 ) computed
in parts c and d is not a profit maximizing plan, then find the profit maximizing production plan.

Answer: I described the profit maxmizing plan and showed that the optimal revenue-maximizing
√ √ output
combination from part c. is profit maximizing provided w is below 50. If w = 50 then the
firm gets
√ zero profits as any scale of production also yields 0 profits. However if w is strictly less
than 50 then as I showed above, the firm√ would want to expand its scale without bound, so then
the answer in part c is not optimal. If w < 50 then the firm would make 0 profits at any positive
scale of production, so its best course of action is to shut down.

5. Bertrand Duopoly Problem Consider those regions in the Washington DC area where households
have a choice between two cable tv/internet providers: Comcast and Starpower. Assume that these com-
panies do not engage in price discrimination, but rather provide cable/internet using a simple single per
month pricing scheme. Assume also that there are no switching or hookup costs, so that customers can
switch from Starpower to Comcast or vice versa (or to not have cable) at zero cost. We now consider
the pricing problem faced by these two competing customers, treating their services as imperfect substi-
tutes in the minds of the consumers in the Washington DC area. Thus, a household in this area has the
following television “mode” choices:

1. No pay TV (i.e. watch broadcast TV, or don’t watch TV or use broadband)

2. cable TV/broadband (via Comcast)

3. cable TV/broadband (via Starpower)

Of course, it is possible for some households to subscribe to both Starpower and Comcast simultane-
ously, but I assume that this is too expensive relative to the incremental value of having both hooked up,
so that virtually no households would subscribe to both at the same time. Thus, I have limited house-
holds to the 3 possible choices given above, which I assume are mutually exclusive and exhaustive
(having ruleed out the possibility of subcribing to both Comcast and Starpower).
Assume that Starpower and Comcast choose their prices independently and without any collusion as
part of a Nash equilibrium in which each tries to maximize its profits, treating the price of its opponent
as given. Initially I ignore the presence of explicit or implicit regulatory constraints. I assume that in
the DC area where these two companies provide overlapping coverage there are N households. Let
Pc (pc , ps ) denote the fraction of these N households who choose Comcast, and Ps (pc , ps ) be the fraction

10
who choose Starpower. The remaining fraction, 1 − Pc (pc , ps ) − Ps (pc , ps ) either watch broadcast TV
(which has a price of $0 per month), or do not watch TV or need broadband internet at all (god forbid!). It
is convenient to start with a simple logit representation for the market shares for Comcast and Starpower:

exp{ac + bc pc }
Pc (pc , ps ) =
1 + exp{ac + bc pc } + exp{as + bs ps }
exp{as + bs ps }
Ps (ps , ps ) = (43)
1 + exp{ac + bc pc } + exp{as + bs ps }

A more advanced approach would derive these market shares from a household level demand study,
using micro data to estimate the consumer choices and accounting for other demographic variables,
including household income y, and the characteristics of the “outside alternative”, i.e. the characteristics
of free to air TV. I assume these market shares are “reduced forms” consistent with the results of a micro
level study. This initial “reduced form” approach requires specification of 7 pieces of information in
order to predict the prices, profits, and market shares for Comcast and Starpower:

1. the number of households N in the “overlap region” served by both Comcast and Starpower,

2. the 4 market share coefficients (ao , bo , a f , b f )

3. the 2 marginal cost parameters (kc , ks )

Given suggested values for these 7 parameters, your job is to compute the Bertrand Nash equilibruium
outcome, i.e. the prices that Comcast and Starpower will charge, their profits, and their equilibrium
market shares.
Let kc and ks denote the marginal costs (i.e. costs which depend on the number of their subscribers)
of providing their cable service. Then the Nash equilibrium, profit maximization conditions determining
the prices (p∗c , p∗s ) (where the ∗ superscripts denote their Nash equilibrium values) are given by

p∗c = argmax(pc − kc )NPc(pc , p∗s )

pc
p∗s = argmax(ps − ks )NPs (p∗c , ps ) (44)
ps

Note that I have treated the cost of the programming content that Comcast and Starpower purchase as
fixed costs, Fc and Fs that do not depend on the number of customers and thus do not enter into the
determination of the the equilibrium prices (p∗o , p∗f ). This would change if Comcast and Starpower paid
per subscriber royalty fees to HBO, ESPN, and the other providers of their programming content. These
fees would then be embodied in the marginal cost parameters (kc , ks ).
Figure 1 shows an illustrative Bertrand-Nash equilibrium calculated for a particular choice of the 5
parameters given above. Notice that the number of households N is just a multiplicative constant in the
profit functions for Comcast and Starpower and thus, in actuality, the equilibrium is fully determined by
the 6 parameters (ac , bc , as , bs , kc , ks ).
Your job is to try to calculate the equilibrium, writing the necessary programs to calculate the equi-
librium in your favorite programming language. Once you calculate the equilibrium, prepare a plot of
the equilibrium as done above and determine whether or not the equilibrium appears to be unique (inthe
diagram above, it is clear that there is a unique “stable” equilibrium).
Also, I want you to compare the Bertrand-Nash duopoly outcome with the two possible monopoly
outcomes:

11
 1. Comcast has a monopoly in the DC area

2. Starpower has a monopoly in the DC area

Figure 1: Example of a Bertrand-Nash Equilibrium

Under the monopoly scenario, customers have only two options: 1) watch broadcast TV (or dont’
watch), 2) subscribe to cable. If Comcast is the monopolist, assume that the share of the DC households
it could obtain if it charged price pc is given by

exp{ac + bc pc }
Pc (pc ) = (45)
1 + exp{ac + bc pc }

and if Starpower is the monopolist and charged price ps it would get the following share of DC house-
holds
exp{as + bs ps }
Ps (ps ) = (46)
1 + exp{as + bs ps }
Thus, I assume that the same set of market share or “demand” coefficients (ac , bc ) and (as , bs ) hold in
the monopoly case as in the duopoly case.
Your job is to compute the monopoly and duopoly outcomes, and predict by how much cable prices
would go up in the DC area if Comcast or Starpower gained monopoly control of this market.
To get you started I have provided a Gauss file setup.gpr that contains parameter values that
you can use to compute the duopoly and monopoly outcomes, and a Gauss procedure, br c.g, which
computes the “best response function” for Comcast, i.e.

pc = brc (ps ) = argmax(p − kc )NPc(p, ps ) (47)

p

12
which is Comcast’s optimal price given that Starpower charges a price of ps . With this hint you should
be able to program the other pieces and compute the solution to this problem. You do not need to do
your programming in Gauss: I have used Gauss only as an illustration to get you started.

Answer Using Matlab files that I have posted along with these answers on the Econ 425 web site,
I obtain the following solutions. For the Bertrand duopoly, the Matlab function equil.m calculates
Bertrand duopoly prices of pc = 173.89 and ps = 346.30. Comcast achieves a market share of 21.82%
and Starpower has a market share of 71.12% and 0.76% of the people in this market do not use either
Comcast or Starpower. The profits for Comcast are $4.888 million and the profits for Starpower are
$24.630 million.
If Starpower is the only cable company, it would charge a price of $632.72 and would earn $53.27
million in profits. It would serve 84.2% of the households, but due to the high monopoly price, 15.8%
of the households go without cable.
In the case where Comcast is the only cable company, it would charge a price of $490.78 and earn
profits of $36.578 million. It would serve 74.5% of the market. Because of the perceived lower quality of
Comcast’s servicea, it cannot manage to charge as high of a monopoly price as Starpower can, and more
consumers decide to go without cable when Comcast is the monopolist compared to when Starpower is
the monopolist.
In the Bertrand case, the higher quality of service that Starpower provides it customers enables
it to charge a significantly higher price and obtain a significantly larger market share than Comcast
can obtain. However the competition between the two firms drives dow the prices to consumers by a
significant amount, and the lower prices induces virtually all households to subscribe to cable.

6. Intertemporal utility maximization with certain lifetimes. Suppose a person has an additively
separate, discounted utility function of the form
T
V (c1 , . . . , cT ) = ∑ βts u(ct ) (48)
t=1

where βs is a subjective discount factor and u(ct ) is an increasing utility function of consumption ct in
period t. Let the market discount factor is βm = 1/(1 + r) where r is the market interest rate.

a. If βs = βm show that the optimal consumption plan in a market where there are no borrowing
constraints (i.e. the consumer has unlimited ability to borrow and lend subject to an intertemporal
budget constraint) is to have a constant consumption stream over time, i.e. c1 = c2 = · · · = ct =
ct+1 = · · · = cT .

b. If βs < βm will the optimal consumption stream be flat, increasing over time, or decreasing over
time, or can’t you tell from the information given?

c. How does your answer to part b change if I tell you that the utility function u(c) is convex in c?

Answers: The answer to this question is in my lecture notes. See pages 23 onward in the lecture notes
on intertemporal choice. For part c, note that if the utility function is convex, then u′′ (c) > 0 and the
answers to parts b is reversed, iv βs < βm , then optimal consumption will be increasing over time,
the opposite of the case if utility is concave (diminishing marginal utility), in which case consumption
would be decreasing over time.

13
 7. Consumption and Taxes Suppose a consumer has a utility function u(x1 , x2 ) = log(x1 ) + log(x2 ) and
an income of y = 100 and the prices of the two goods are p1 = 2 and p2 = 3.
a. In a world with no sales or income taxes, tell me how much of goods x1 and x2 this consumer will
purchase.
answer Notice that the utility function is a monotonic transformation of a Cobb-Douglas utility function
1/2 1/2
l(x1 , x2 ) = x1 x2 , so demands are x1 (p1 , p2 , y) = y/2p1 and x2 (p1 , p2 , y) = y/2p2 . With these,
it is very easy to answer this question. x1 = 100/(2 ∗ 2) = 25 and x2 = 100/(2 ∗ 3) = 16.66667.
b. Now suppose there is a 10% a sales tax on good 1. That is, for every unit of good 1 the person
buys, he/she has to pay a price of p1 (1 + .1) = 2.2, where the 10% of the price, or 20 cents, goes
to the government as sales tax. How much of goods 1 and 2 does this person buy now?
answer With the tax in place, the price of good 1 increases to 2.2 so quantities demanded are x1 = 100/(2∗
2.2) = 22.727273 and x2 = 100/(2 ∗ 3) = 16.66667. The total taxes the person pays are .2x1 =
.2100/(2 ∗ 2.2) = 4.54.
c. Suppose instead there is a 5% income tax, so that the consumer must pay 5% of his/her income
to the government. If there is no sales tax but a 5% income tax, how much of goods 1 and 2 will
the consumer consume?
answer With a 5% income tax, the consumer has after-tax income equal to $95 (100(1−τ) where τ = .05).
So the consumption of goods 1 and 2 is given by x1 = 95/(2 ∗ 2) = 23.75 and x2 = 95/(2 ∗ 3) =
15.8333.
d. Which would the consumer prefer, a 10% sales tax on good 1, or a 5% income tax? Explain your
reasoning for full credit.
answer With the sales tax, the consumer consumes less of good 1 and more of good 2, and pays less
in tax overall. With the income tax the consumer consumes more of good 1 but less of good 2
and pays more overall in tax ($5.00 versus $4.54). But the only way to see which alternative
the consumer prefers is to plug the consumption bundles into his/her utility function and see
which one give more utility. The utility under the sales tax is log(22.727273) + log(16.66667) =
5.93699764. The consumer’s utility under the income tax is log(23.75) + log(15.8333) = 5.9297
so the consumer prefers the sales tax to the income tax.
e. How big would the sales tax on good 1 have to be for the government to get the same revenue
as a 5% income tax? Which of the two taxes would the consumer prefer in this case, or is the
consumer indifferent because the consumer has to pay a total tax of $5 (5% of $100) in either
case?
answer Now we want to set the sale tax rate α so that we raise tax revenue of $5, the same revenue that
we collect under an income tax of 5%. The equation for the necessary tax rate is
100
5=α (49)
2(2 + α)
Solving this for α we get α = 2/9 = .22222. Under this tax rate, consumption of good 1 falls to
100
x1 = 2(2+α) = 22.5 and the tax revenue collected is 22.5 ∗ 2/9 = 5. Now the person’s utility under
the sales tax is log(22.5) + log(16.66667) = 5.926926, so that now, the consumer slightly prefers
to have the income tax over the sales tax.

14
 8. Supply and Demand Problem The supply for corn is given by

S = 10 + 5p + .05R (50)

where R is the amount of rainfall. The demand for corn is given by

D = 5Y .2 p−.5 (51)

where Y is the per capita income.

a. What is the equation for the equilibrium price of corn, assuming this is a competitive market?

answer We find the price that sets supply equal to demand. Equilvalently, we seek a price p that sets
Excess demand E(p) = D(p) − S(p) to zero, where

E(p) = 5Y .2 p−.5 − (10 + 5p + .05R) (52)

b. Solve for the equilibrium price and quantity in this market, using numerical methods (e.g. New-
ton’s method) if necessary, or by any means possible to get numerical answers.

answer Suppose we set Y = 1000 and R = 20. Then the equation we want to solve is

E(p) = 5 1000.2 p−.5 − 10 − 5p − 1 = 0.

 
(53)

I programmed this function in Matlab as the file ed.m which is posted on the Econ 625 website
along with these answers. You can use Newton’s method to solve this equation. I was a bit lazy
and instead used the Matlab fsolve command to solve this equation, that is, I did fsolve(@ed,2)
(so that my initial guess for a solutions was p = 2). fsolve returned the solution p = 1.2964 and
checking, ed(1.2964)= 1.4087e−10 .

c. Derive a formula for d p/dR, i.e. the effect of an increase in rainfall on the price of corn.

Answer This is an exercise in the use of the implicit function theorem. See my lecture notes on this posted
on the Econ 425 website. But the answer is
dp .05
=− < 0. (54)
dR 2.5y.2 p−1.5 + 5

d. Derive a formula for d p/dY , i.e. the effect of an increase in per capita income on the price of
corn.

Answer Using the implicit function theorem again,

∂
dp ∂y E(p, y) y−.8 p−.5
=− ∂ = > 0. (55)
dy ∂p E(p, y)
2.5y.2 p−1.5 + 5

15