Economics 500, Fall 2020

Problem Set 4, Due Thursday, September 30

Question 1 Suppose U (x1 , x2 ) = xα1 1 xα2 2 is a Cobb-Douglas utility func-

tion. We use this to explore some questions of comparative statics in a
setting in which the calculations are straightforward. Begin by writing the
first-order conditions for utility maximization. Remember throughout that
you are free to take strictly increasing transformations of utility functions
whenever useful.

1.1 Solve the first-order conditions to find explicit solutions for the demand
functions, and then differentiate these functiosn to obtain comparative
static results.
Solution: We should be very familiar with the demand function now
after previous problem set. With Cobb-Douglas utility function, con-
sumers will spend constant share of their wealth on a good. Assuming
α1 + α2 = 1, we have
α1 w
x1 (p, w) = ,
p1
α2 w
x2 (p, 2) = .
p2
Therefore x1 is strictly decreasing in p1 and strictly increasing in w.
x1 does not depend on p2 .
When a closed-form solution is available, the complexity of conducting
comparative analysis is typically acceptable.

1.2 Now instead apply the implicit function theorem to the first-order
conditions and solve for the collection of comparative static results.
Solution: First, let’s set the stage. Since U is strictly concave we
know there is a unique global maximizer and KKT will also have the
global maximizer as the unique solution1 . Since when xi = 0 the
utility is 0 while any interior point brings positive utility, we know the
optimal solution is interior.
1
When we take derivatives on the first order condition to do comparative analysis, it
is very important to argue the solution of the first order condition is unique.
Next, it helps to first write the utility function as

U (x1 , x2 ) = α1 ln x1 + α2 ln x2 .

The first-order conditions are:

α1
+ λp1 = 0
x1
α2
+ λp2 = 0
x2
p1 x1 + p2 x2 − w = 0.

Now differentiate these first-order conditions in p1 to obtain:

 α1   dx  
− x2 0 p1 dp1
1
−λ

1
− αx22 p2   dx
 2  
 0 dp1  = 0 

2
dλ −x1
p1 p2 0 dp 1

Now we apply Cramer’s rule to obtain

−λ 0 p1
0 − αx22 p2
2
dx1 −x1 p2 0 α1 w
= =− .
dp1 |H| p21
The final equality requires some rather tedious algebra; it requires
using the first-order conditions to simplify. We are often interested
only in determining the sign of this derivative, which is easier, and so
it is typically fine to stop at that point (showed later).
Similarly, if we took our initial derivatives of the first-order conditions
with respect to p2 , we could find

0 0 p1
−λ − αx22 p2
2
dx1 −x2 p2 0
= =0
dp2 |H|
This is an easier calculation, since you can ignore the denominator.
Finally,
0 0 p1
0 − αx22 p2
2
dx1 1 p2 0 α1
= = .
dw |H| p1

2
 Clearly we could do the same for x2 . To illustrate how to simplify the
expression of dx1 /dp1 to a degree that is “just enough”. Let us redo
the calculation
−λ 0 p1
0 − αx22 p2
2
dx1 −x1 p2 0 −p1 α2 x1 x−2
2 + λp1 p2
= = .
dp1 |H| |H|
From second-order condition we know |H| ≥ 0 (actually we well as-
sume |H| > 0). Also λ < 0 from the FOC, thus this derivative is
negative.

0 0 p1
0 − αx22 p2
2
dx1 −1 p2 0 α2 p1 x−2
2
= = > 0.
dw |H| H

1.3 For a third approach, use the budget constraint to write utility as a
function of a x1 , p1 , p2 , and w. Then use monotone comparative statics
to derive comparative static results. Identify the extent to which your
answers to the three approaches agree or disagree.
Solution: To simplify calculation we first do a log transformation:

V (x1 , x2 ) = log U (x1 , x2 ) = α1 log x1 + α2 log x2 .

Plugging in the budget constraint we have:

w − p 1 x1
V (x1 , p1 , p2 , w) = α1 log x1 + α2 log( ),
p2
∂U p1
= α1 x−1
1 − α2 ,
∂x1 w − p 1 x1
∂2U αw
=− < 0,
∂x1 ∂p1 (w − p1 x1 )2
∂2U p1
= α2 > 0,
∂x1 ∂w (w − p1 x1 )2
∂2U
= 0.
∂x1 ∂p2

3
One might thus conclude that since U has decreasing difference in
(x1 , p1 ) and thus the demand function x1 (p, w) is decreasing in p1 .
This is almost correct but it is not a perfect argument because the
maximization problem is:

max V (x1 , p, w),

x1
s.t. 0 ≤ p1 x1 ≤ w.

The problem here is that the feasible set is also changing with (p, w)
thus we can’t directly apply the result learned in the class. There are
now two ways to proceed.
First, one is often interested in comparative statics around a neighbor-
hood of the existing values of (p1 , p2 , w). If so, one can often restrict
(p, w) to come from set [p1 , p1 ] × [p2 , p2 ] × [w, w] and then restrict x1
to an interval such that w − [1 x1 ] remains positive. ni this case, we
have the results we need. Could we always do this, in which case we
would never have to worry about the sets from which the variables are
drawn? Not necessarily. A typical utility maximization problem max-
imizes over the set px ≤ w, which varies as do the exogenous variables.
This is a problem because the maximizer lies on the boundary of this
set. To avoid this complication, we have substituted the constraint
into the objective. This allows us to proceed as described if and only
if we are confident the maximizer is interior.
To be more ambitious, we can we prove the following generalization,
which just includes a very small modification. To be more reader
friendly we focus on R2 , but it is easy to generalize to arbitrary lattice.

Lemma 1. Consider a maximization problem in R2 :

x(y) = argmax u(x, y).

x∈X(y)

Suppose

– X(y) ⊂ R is decreasing in y in the strong-set order. Namely for

y1 > y2 and x1 ∈ X(y1 ), x2 ∈ X(y2 ) we have

min{x1 , x2 } ∈ X(y1 ), max{x1 , x2 } ∈ X(y2 ).

– u(x, y) has decreasing difference in (x, y).

4
Then the solution of the problem x(y) is decreasing in y in the strong-
set order.

Proof. To prove the lemma, consider any pair y1 ≥ y2 and x1 ∈ x(y1 ),

x2 ∈ x(y2 ). We want to show:

x̄ = max{x1 , x2 } ∈ x(y2 ), x = min{x1 , x2 } ∈ x(y1 ). (1)

Suppose this is not true. We know x̄ ∈ x(y1 ) ⊂ X(y1 ) and x ∈ x(y2 ) ⊂

X(y2 ). Since X(y) is decreasing in y in the strong-set order, we know
x̄ ∈ X(y2 ) and x ∈ X(y1 ). Thus in conclusion x̄, x ∈ X(y1 ) ∩ X(y2 ).
After taking care of the feasibility, we use the same trick in the class to
induce contradiction. Since equation 1 is not true, we have two cases:

– x̄ ̸∈ x(y2 ). Since x ∈ x(y2 ), this implies:

u(x̄, y2 ) − u(x, y2 ) < 0,

=⇒ u(x̄, y1 ) − u(x, y1 ) ≤ u(x̄, y2 ) − u(x, y2 ) < 0,
=⇒ x̄ ̸∈ x(y1 ).

This is a contradiction.
– x ̸∈ x(y1 ). Since x̄ ∈ x(y1 ), this implies:

u(x̄, y1 ) − u(x, y1 ) > 0,

=⇒ u(x̄, y2 ) − u(x, y2 ) ≥ u(x̄, y1 ) − u(x, y1 ) > 0,
=⇒ x ̸∈ x(y2 ).

This is a contradiction.

Now we turn back to the problem. Notice that the feasible set for x1
is
w
X1 (p, w) = [0, ].
p1
X1 (p, w) is increasing in w and decreasing in p1 (in the sense of strong
set order). Thus we conclude: x1 (p, w) is decreasing in p1 and increas-
ing in w, it (is increasing and decreasing and thus) does not depend
on p2 .

5
1.4 Revisit 1.1-1.3 for the case of a Cobb-Douglas utility function of an ar-
bitrary number L of goods rather than just two goods. Explain in each
case why the extension to L goods is straightforward or problematic.
Solution:

– The direct method is straightforward to generalize, since we have

the identical formula for demand function in the L good case:
xi = αi w/pi .
– The implicit theorem approach can be generalized, but the cal-
culations are sufficiently intricate that you can reasonably hope
to establish only the signs of the results. With some algebra, you
can do so. The log transformation helps here, as it makes the
matrix of second derivatives contain many zero entries.
– See 1.7 for more discussion.

1.5 Now consider a general utility function U (x1 , x2 ). Use the implicit
function theorem to calculate dx1 /dp1 . Then use the implicity function
theorem again to find dx1 /dw. Comparing your two results, explain
how dx1 /dp1 depends on the sign of the income effect, and explain how
your expression for dx1 /dp1 is an example of the Slutzky equation.
Assume throughout that you are dealing with a demand function with
interior solutions.
Solution:
We have
−λ u12 p1
0 u22 p2
dx1 −x1 p2 0 1  2 
= = λp2 − x1 (p2 u12 − p1 u22 ) ,
dp1 |H| |H|

where we have |H| ≥ 0, from the second-order conditions. Similarly,

0 u12 p1
0 u22 p2
dx1 −1 p2 0 1
= = [(p2 u12 − p1 u22 )] ,
dw |H| |H|

where we have |H| ≥ 0, from the second-order conditions. The expres-

sion for dx1 /dp1 is the Slutzky equation. We see that the first term is

6
 the substitution effect, which is negative, while the second term con-
tains the income effect. dx1 /dp1 can be positive only if the income
effect is negative and sufficiently strong.

1.6 Continuing with the setting of 1.5, use the budget constraint to write
utility as a function solely of x1 . Identify what it means for this func-
tion to have increasing differences, in both discrete and differential
form. Using the latter, derive a relationship between the income effect
and increasing differences.
Solution: Define
w − p 1 x1
V (x1 , p1 , p2 , w) = U (x1 , ).
p2
V has increasing difference in x1 , w if for any x1 ≥ x2 and w1 ≥ w2 we
have

V (x1 , w1 ) − V (x2 , w1 ) ≥ V (x1 , w2 ) − V (x2 , w2 ).

Or in the differential form:

∂2V
≥ 0.
∂x1 ∂w
We can define similar notion for other variables. It is easy to show
that above definition is equivalent to
1 p1
U12 (x1 , x2 ) − 2 U22 > 0.
p2 p2

Now look at your answer to 1.5, and in particular look at the numerator
of the income effect you calculated there. You will see that for this
numerator to be positive, and hence for good 1 to be a normal good,
is equivalent to V having increasing differences.

1.7 We have so far considered comparative static results that vary a sin-
gle exogenous variable. Consider the Cobb-Douglas utility function
of L goods from 1.4. Use the budget constraint to remove one of the
endogenous variables and write utility as a function of the L − 1 re-
maining endogenous variables and the exogenous variables. For what

7
combinations of variables does this function exhibit strictly increasing
differences? What is the implied comparative statics?
Solution: Define the utility function V as (after log transformation)
P
w − i<L xi pi
V (x1 , ..., xL−1 , p1 , ..., pL , w) = U (x1 , ..., xL−1 , ).
pL
Knowing that this is Cobb-Douglas utility function, we can take a log
transformation and drop a −αL pL term to consider
X X
V = aL log(w − p i xi ) + αi log xi .
i<L i<L

We calculate the second order derivatives:

∂V αi p
= − αL Pi ,
∂xi xi w − i<L pi xi
∂2V α p
= PL i > 0, ∀i,
∂xi ∂w (w − i<L pi xi )2

Thus V has strictly increasing difference in (x, w), where x = (x1 , ..., xL−1 ).
Now consider prices:

∂2V −αL pi xj
= < 0,
(w − i<L pi xi )2
P
∂xi ∂pj
∂2V −αL pi xj α
= − PL < 0,
(w − i<L pi xi )2 w − i≤L pi xi
P
∂xi ∂pi

Thus V has strictly decreasing difference in (x, pi ).

This initially looks like it is not an encouraging development, since we
need increasing differences to apply our monotonic comparative statics
results. However, there are two ways we can proceed. First, there is a
version of the monotone comparative statics result that applies when
there are decreasing differences in the variables (x, t). The alternative
provides insight into why we should expect such a result. We can think
of the exogenous variables as being the negative of prices rather than
prices, and this will get us back to increasing differences. This should
be intuitive. We know the Marshallian demands for the Cobb-Douglas
utility function are negatively slope, or dxi /dp)i < 0. The monotone
comparative statics result yields positive derivatives. To match these,

8
we can view the negative slope of the Marshallian demand as the
statement that xi increases as −pi increases, which is then something
we can get out of our monotone comparative statics result..
We need one more condition to apply the full monotone comparative
statics result, which is that dV /dvi dxj > 0. We need this whether
we are working with increasing differences or decreasing differences.
This is essentially a condition indicating that the exogenous variables
tend to move together. However, this fails. As a result, we do not
immediately obtain useful monotone comparative statics results from
this approach to the Cobb-Douglas utility function.

Question 2 This problem examines the concepts of compensating

variation, equilibrating variation, and consumer surplus.

2.1 Which demand function is steeper—the Hicksian or Marshallian

demand function? Explain how your answer changes depending
on whether the good in question is a normal good or an infe-
rior good, and provide an appropriate argument supporting your
result. (You may assume throughout that you are dealing with
demand functions, interior solutions, and so on.)
Solution: Fix a price-wealth pair (p̄, w̄), and let ū = v(p̄, w̄).
Then
x(p̄, w̄) = h(p̄, ū).
From the Slutksy equation,
∂x1 (p̄, w̄) ∂h1 (p̄, ū) ∂x1 (p̄, w̄)
= − x1 (p̄, w̄).
∂p1 ∂p1 ∂w
If the good is normal at (p̄, w̄), then
∂x1 (p̄, w̄) ∂h1 (p̄, ū)
≤ ≤ 0,
∂p1 ∂p1
so Walrasian demand for good 1 is more sensitive to the price
p1 than is the Hicksian demand. Since price is on the vertical
axis, this means that the Hicksian demand is steeper than the
Walrasian demand.
If the good is inferior at (p̄, w̄), then
∂x1 (p̄, w̄) ∂h1 (p̄, ū)
≥
∂p1 ∂p1

9
 Assuming the good is not Giffen good, it follows that Walrasian
demand for good 1 is less sensitive to the price p1 than is the
Hicksian, and hence the Walrasian demand is steeper.

2.2 Give an example in which the equilibrating variation is strictly

larger than the consumer surplus, which is strictly larger than
the equilibrating variation. Give an example in which these three
are equal, and an example in which the equilibrating variation
is strictly smaller than the consumer surplus, which is strictly
smaller than the equilibrating variation. Explain how your results
are related to your findings in 2.1 concerning income effects and
the relative slopes of Hicksian and Marshallian demand functions.
Solution: We first do some general calculation before presenting
the example. Consider a 2 good economy and suppose prince
decreases from p0 = (p01 , p2 ) to p1 = (p11 , p2 ). The initial wealth
level is w and denote u0 = v(p0 , w) and u1 = v(p1 , w). As the
price changes from p0 to p1 , the agent’s utility changes from u0
to u1 . The equilibrating variation is

EV (p0 , p1 , w) = e(p0 , u1 ) − e(p0 , u0 )

= e(p0 , u1 ) − w = e(p0 , u1 ) − e(p1 , u1 )
Z p0
1
= h1 (p1 , p2 , u1 )dp1 .
p11

the compensating variation is

CV (p0 , p1 , w) = e(p1 , u1 ) − e(p1 , u0 )

= w − e(p1 , u0 ) = e(p0 , u0 ) − e(p1 , u0 )
Z p0
1
= h1 (p1 , p2 , u0 )dp1 .
p11

and the consumer surplus is

Z p01
0 1
CS(p , p , w) = x1 (p1 , p2 , w)dp1 ,
p11
Z p01
= h1 (p1 , p2 , v(p1 , p2 , w))dp1 .
p11

10
Since h(p, u) = x(p, e(p, u)) and e is always strictly increasing
in u, we know h1 (p, u) is strictly increasing in u if and only if
x1 (p, w) is strictly increasing in w. Also v(p1 , p2 , w) is strictly
decreasing in p1 (as long as it is interior solution). Since p01 > p11
we know u0 < u1 .
When the good 1 is normal good, namely x1 (p, w) is strictly in-
creasing in w, we have

h1 (p1 , p2 , u1 ) > h1 (p1 , p2 , v(p1 , p2 , w)) > h1 (p1 , p2 , u0 ).

Therefore EV > CS > CV . Similarly, it is easy to show that

when the good is inferior good, EV = CS = CV . And when
∂w x1 = 0, EV = CS = CV .
Thus we simply need to construct three utility function where
the solution is interior where ∂w x1 is positive, negative and 0.
* ∂w x1 > 0. We know Cobb-Douglass function would work.
* ∂w x1 = 0. Many reasonable quasi-linear utility function will
work. For example, consider

U (x1 , x2 ) = log x1 + x2 .

For any fixed p, as long as w is large enough (w > p2 ). we

will get a interior solution that satisfies
1 1 p2
= =⇒ x1 (p, w) = ,
p1 x1 (p, w) p2 p1
=⇒ ∂w x1 = 0.

* It is actually hard to give a very satisfying and easy form of

utility function where ∂w x1 < 0. In general, a good technique
when seeking counterexamples is to work backward. First
identify what you need, then infer what will deliver this. In
this case, you need an income expansion path that causes the
quantity of x1 to decrease as income increases. Draw such
a path. Then insert indifference curves that are piecewise
linear, with a kink at the desired income expansion path, so
as to ensure that utility-maximizing bundles lie along this
path. Then attach utility numbers to the indifference curves.
If you are more ambitious, consider

U (x1 , x2 ) = log x1 + x22 /2.

11
Plug in the budget constraint we get
w − p 1 x1 2
V (x1 ) = log x1 + ( ) /2,
p2
1 p 1 w − p 1 x1
V ′ (x1 ) = − ,
x1 p 2 p2
1 p2
V ′′ (x1 ) = − 2 + 12 .
x1 p 2
Thus V is concave in [0, p2 /p1 ] then convex in [p2 /p1 , ∞].
Also V (0) = −∞,
p2 p1 w
V ′ ( ) = (2 − ) < 0
p1 p2 p2
if w > 2p2 . In this case, the maximum value of V in [0, p2 /p1 ]
is obtained by interior FOC:
1 p1 w − p1 x1
V ′ (x1 ) = − = 0,
x1 p2 p2
p1 x21 − p1 wx1 + p22 = 0.
This equation has only one solution in [0, p2 /p1 ]:
p
∗ p1 w − p21 w2 − 4p21 p22
x1 =
2p21
On the other hand, the maximum of V in [p2 /p1 , ∞) is max{V (p2 /p1 ), V (w/p1 )}.
Since p2 /p1 ∈ [0, p2 /p1 ], V (p2 /p1 ) ≤ V (x∗1 ), we know the op-
timal value of utility maximization is:
max V (x1 ) = max{V (x∗1 ), V (w/p1 )}.V (w/p1 )
x1 ∈[0,w/p1 ]

fixing p we know x∗1 < 1 and

V (x∗1 ) ≥ V (p2 /p1 ) = log(p2 /p1 ) + (w/p2 − 1)2 ,
V (w/p1 ) = log(w/p1 ).
Thus when w is sufficiently large V (x∗1 ) > V (w/p1 ), and x∗1
is the optimal demand in this case. Now let us calculate how
x∗1 will change with w.
p
∂(p1 w − p21 w2 − 4p21 p22 ) 2p2 w
= p1 − p 2 1 ,
∂w 2 p1 w2 − 4p21 p22
p1 w
= p1 − p < 0.
w2 − 4p22
Thus x1 is indeed an inferior good.

12
2.3 Suppose you are considering a policy that will lower the price of
good 1. Implementing the policy requires that you incur a per
capita cost K. You would like to implement the policy if and
only if the per capita benefit to consumers exceeds K. Suppose
you calculate the equilibrating variation and the compensating
variation of the policy change (assume consumers are identical, so
that you need do this only once) and find that the former exceeds
K, while that latter falls short of K. One of your examples in
2.2 should assure you that this can be the case How would you
evaluate the policy?
Solution: If you are confident of your welfare measures, then the
answer is likely to depend on the details of the policy. Suppose
the policy maker has K on hand, and is deciding whether to
pay the agents this amount or instead implement the policy. In
this case, the (candidate) wealth change comes at the old prices,
so EV ≥ K tells us that the consumers would be better off by
implementing the policy.
Suppose instead that the policy maker is deciding whether to
implement the policy and then afterwards charge the agents for
its cost K. In this case, the wealth change comes at the new low
prices, so the relation CV ≤ K tells us that the policy makes
consumers worse off.
More generally, you should recognize that you can never mea-
sure the compensating and equilibrating variations precisely. In-
stead, the best you can obtain are necessarily imprecise estimates.
These welfare measures should then be interpreted as providing
support for a policy only if they are safely larger than a cost.
In many cases, the difference is an order of magnitude. If the
measures are so close as to straddle the cost, then one should
conclude that the welfare ranking is unclear, and should invoke
other arguments for or against the policy.
2.4 Let’s suppose instead that upon considering a potential policy
that will reduce the income of each person i in the economy by
amount Ki . When assessing the benefits, you find that the sum
(across people in the economy) of the compensating variations,
as well as the corresponding sum of equilibrating variations, is
larger than the total cost. However, some people in the economy
realize higher benefits than others, while some incur higher costs
than others. Is there a collection of lump sum taxes and subsidies

13
that will make everyone better off? The first step in addressing
this question is to formulate the question more precisely; then
provide either a proof or a counterexample.
Solution: As the question indicates, the first step is to make the
question more precise. It is typically the case when embarking on
a research project of constructing a model that much of the work
consists of formulating the question. How might we proceed here?
First, what is the policy we are talking about? The question
indicates that it has reduced people’s income. It must have done
something more than that, since the benefit indicates that it has
been beneficial. So let’s assume (so that we can use things we’ve
learned in class) the benefits have come from changing price from
p0 to p1 . Next, we are told that the sum of the equilibrating
and compensating variations exceeds the sum of the costs. What
are these costs? Does this refer to the decreases in income, or
something else? Most people would view a decrease in income as
a cost, and so we can interpret the costs as the decreased incomes.
Without the policy, the price vector is p0 and each person has
wealth wi0 . Denote vi0 = v(p, wi0 ) as his/her initial payoff. Af-
ter implementing the policy, the price vector changes to p1 and
each person has wealth wi1 . Denote vi1 = v(p, wi1 ). The problem
indicates
X X
ei (p1 , vi1 ) − ei (p1 , vi0 ) > Ki ,
i i
X X
0
ei (p , vi1 ) − ei (p 0
, vi0 ) > Ki
i i

We are interested to know if there exists a transfer scheme ti such

that, for every person i, we have

ti + ei (p1 , vi1 ) ≥ ei (p1 , vi0 )

and X
ti = 0.
Why is this the relevant criterion? After the policy change, per-
son i was income wi1 = ei (p1 , vi1 ). If ti + ei (p1 , vi1 ) ≥ ei (p1 , vi0 ),
then the tax/subsidy scheme leaves this person with enough in-
come to obtain utility vi0 , which is what we need. The constraint
that the transfers sum to zero indicates that, after the original

14
 adjustments to income caused by the policy, we require any fur-
ther rearrangements of income to balance, to that we are neither
injecting income into the economy nor extracting income from
the economy.
To see that this is feasible, for each i set

ti = ei (p1 , vi0 ) − ei (p1 , vi1 ),

which just satisfies our first requirement. Then we have

X X X
ti = ei (p1 , vi0 ) − ei (p1 , vi1 ) = − Ki < 0,
i

allowing us to hit all of our utility targets with some room for
subsidies left
P over.1 Indeed, we could make such transfers as long
as we had i ei (p , vi ) − ei (p1 , vi0 ) ≥ 0.
1

One thing we see here is that the criterion for adopting the policy,
that the sum of equilibrating or compensating variations exceeds
the costs, is both unduly conservative and rather odd, since (ie
we indeed make the requisite transfers, which is a big “if”) all
we need is that the sum of the variations exceed zero. At the
same time, the criterion that aggregate benefits exceed aggregate
costs seems an eminently sensible criterion. To reconcile these
views, we note that the criterion adopted by the question does
compare benefits and costs, but it effectively double counts the
costs. We are evaluating the first term in the compensating and
equilibrating variations at vi1 , which already takes into account
the decrease in income. Given this, we do not also have to ask
that the difference in the sum of variations exceed the reduction
in income. If one had to motivate this criterion, one might return
to the idea that we use the welfare measures to motivate a policy
only if they show clear improvements, or one might note that
actual policy decisions often follow seemingly sensible but odd
rules.

Question 3 We consider here some questions of aggregation. A

person i’s utility function is said to be of Gorman form if it is given
by vi (p, wi ) = ai (p) + bi (p)wi .

3.1 Show (using Roy’s identity) that if person i has such a utility
function, then the income-expansion path for good ℓ is linear.

15
 Now argue that if every person i in an economy has a utility
function whose indirect utility function is given by vi (p, wi ) =
ai (p) + b(p)wi (notice that the subscript on b has disappeared, so
we are assuming that this term is common across people), then
their income-expansion paths are parallel. Show that this implies
that the aggregate demand x(p, w1 , . . . , wI ) can be written as
x(p, w), where w is the sum of the individual wealth levels.
Solution: If agent i has indirect utility vi (p, wi ) = ai (p)+bi (p)wi ,
then Roy’s identity tells us agent i’s demand for commodity l is

xi (p, w) = −bi (p)−1 (∇ai (p) + wi ∇bi (p)).

Therefore the income expansion path is linear2 . If the bi take the

common form b, then the wealth expansion paths have common
direction vector and hence are parallel. In this case, the aggregate
demand is
X X
x(p, w1 , ..., wI ) = xi (p, wi ) − b(p)−1 (∇ai (p) + wi ∇b(p))
i i
X X
−1
= −b(p) ∇ai (p) − ( wi )b(p)−1 ∇b(p)
i i
X
−1
= −b(p) ∇ai (p) − wb(p)−1 ∇b(p)
i

3.2 In light of the previous result, it is interesting to know when the

indirect utility function has the Gorman form. Show that this is
the case for Cobb-Douglas preferences.
Solution: Suppose the utility function takes the form
Y α X
u(x1 , ..., xL ) = xl l , αl = 1.
l l

We know that xl (p, w) = αl w/pl .

3.3 Show that if preferences are homothetic, then the indirect util-
ity function has Gorman form. Hence, if individual demands are
2
Economists typically abuse the word “linear” where a mathematician would say
“affine”.

16
 homothetic and identical, then there exists a representative con-
sumer. Provide an example of preferences that are not homoth-
etic but that are represented by a utility function of the Gorman
form.
Solution: We have shown in problem set 2 that when the pref-
erence is homothetic, the demand function is linear in w and
thus satisfies Gorman form. Here we use the quasi-linear utility
function as an example again:

U (x1 , x2 ) = log x1 + x2 .

Since negative consumption is allowed, the optimal solution is

interior and satisfies
1 1 p2
= =⇒ x1 (p, w) = ,
p1 x1 (p, w) p2 p1
w
x2 (p, w) = − 1.
p2
This is exactly the Gorman form.
3.4 Suppose we are concerend with the rule wi (p, w) = λi w, where w
is aggregate income, wi (p, w) gives person i’s income as a function
of aggregate income, and λi for i = 1, . . . , I is a collection of
positive numbers that sume to one. Specify a utility function for
each persion in the economy and a Bergstrom-Samuelson social
welfare function for which there exists a representative consumer.
Solution: Suppose each agent i has a Cobb-Douglass utility
function:
X
Ui (x1 , ..., xL ) = αl log xl ,
l
P
where αl = 1. The demand function satisfies xil = αl w/pl .
The indirect utility function for agent i is
X
vi (p, wi ) = αl (log αl + log wi − log pl ).
l

Consider a weighted linear Bergstrom-Samuelson social welfare

function
X
W (v1 , ..., vI ) = λi vi .
i

17
We first show that the proposed wealth-allocation rule is the op-
timal allocation under the proposed social welfare function:
X
V (p, w) = max λvi (p, wi ),
w1 ,...,wI
i
X
s.t. wi ≤ w.
i

It is easy to see this problem is equivalent to

X
max λi log wi ,
w1 ,...,wI
i
X
s.t. wi = w.
i

Since Inada condition is satisfied, the optimal solution is interior.

Then first order condition implies
λi λj
= ,
wi wj
=⇒ wi = λi w.

Then without knowing the specific form of V (p, w), we know from
the theorem introduced in the class that V (p, w) is an indirect
utility function derived from some utility function U . This U can
be viewed as the utility function of the representative agent.
In this example, we can actually provides closed form expression
of these two:
XX
V (p, w) = αl (log αl + log(λi w, ) − log pl )
i l
X X
=N αl (log αl − log pl + log w) + log λi ,
l i
X X
U (x1 , ..., xL ) = log λi + N αl log xl ,
i i
αl w
xl (p, w) = .
pl

18