EconS 501: ADVANCED MICROECONOMIC THEORY – I

(Ph.D. program in Economics)

PRACTICE EXERCISES ON
ADVANCED MICROECONOMIC THEORY - I

Felix Munoz-Garcia1
School of Economic Sciences
Washington State University

1
103G Hulbert Hall, School of Economic Sciences, Washington State University. Pullman, WA 99164-6210,
fmunoz@wsu.edu. Tel. 509-335-8402.
 Index

Chapter 1 - Preferences and Utility

1. Checking properties of preference relations-I.

2. Checking properties of preference relations-II.

3. Checking for completeness and transitivity.

4. Monotonicity and Strong monotonicity.

5. Utility representation of preferences.

6. Rational preference relation.

7. utility function and rational preference relation

8. Monotonic transformations.

9. Lexicographic preference relation.

10. Checking WARP.

11. Some applications of WARP-I

12. Some applications of WARP-II.

13. WARP-I.

14. WARP-II.

15. Lexicographic preferences and WARP.

16. UMP

17. Property of utility function-I

18. Transivity and completeness

19. Homogeneous and quasiconcave

20. Property of utility function-II

1
Chapter 2 - Demand Theory
1. Marshallian demand.

2. [Checking WalrasLaw].

3. [Substitution e¤ects]

4. [Income and substitution e¤ect with a quasi-linear utility function].

5. [Gi¤en and inferior goods].

6. [Relationship between WARP and CLD].

7. [Slutsky matrix].

8. [Slutsky matrix].

9. [Quasilinear Preferences].

10. [ULD].

11. [CES Utility function].

12. Slustky equation

13. [Gorman form for vi (p; wi )]

14. [Concavity of the support function]

Chapter 3 - Applications of Demand Theory

1. Taxes versus Subsidies.

2. Welfare changes.

3. Consumer Theory and Welfare.

4. Using the compensating variationto identify price changes.

5. Consumer theory.

6. Finding the compensating and equivalent variation with little information.

7. DWL.

8. Income tax and sales tax.

9. Cobb-Douglas preferences with three goods.

2
Chapter 4 - Aggregate demand
1. Social welfare maximization problem.

2. ULD-I.

3. Area variation measure approach.

4. Homothetic preferences and optimal wealth distribution rule.

5. WA

6. ULD-II.

7. Violation of ULD.

8. Application of aggregate demand

Chapter 5 - Production Theory

1. Property of production function.

2. Nonincreasing Returns to Scale.

3. Increasing returns to scale.

4. Average product and marginal product

5. Deriving the supply function from the pro.t function.

6. Law of Supply.

7. Properties of the pro.t function.

8. Duality Theorem.

9. Shephard.s lemma.

10. Applying Shephard.s lemma.

11. Deriving the cost function fromthe pro.t function.

12. Free-disposal.

13. Duality theorem when the set K is a singleton.

14. Aggregation in production.

15. Comparison of two technologies.

16. Pro…t maximization and Mergers.

3
Chapter 6 - Competitive Markets
1. Equilibrium in competitive markets.
2. Firm supply curve.
3. Entry and exit in perfectly competitive markets.
4. Per unit taxes vs. Ad valorem taxes.
5. Distribution of tax burden.
6. Ad valorem taxes.

Chapter 7 - Choice under Uncertainty

1. Independence axiom.
2. Independence axiom and convexity.
3. Neumann-Morgenstern utility function.
4. Some Regret theory.
5. An introductory example on risk aversion.
6. Does risk-aversion imply variance aversion?
7. Investing in two di¤erent assets.
8. Second order stochastic dominance.
9. What is the relationship between FOSD and dominance in means?
10. Coe¢ cient of risk aversion.
11. Concavity and Coe¢ cient of risk aversion.
12. Non-constant coe¢ cient of absolute risk aversion.
13. Purchasing insurance.
14. Purchasing health insurance.
15. Recovering the utility function from rA (x; ):
16. risk aversion.
17. Uncertainty about being audited by the IRS.
18. Uncertainty about the future.
19. Uncertainty in agricultural markets.
20. Uncertainty in production decisions.

4
Chapter 8 - Monopoly
1. Lerner Index.

2. Tax or subsidize a monopolist.

3. Monopolist with interdependent demands.

4. Monopoly: leasing vs. selling.

5. Discrimination or Nondiscrimination?

6. Multiplant Monopoly.

7. Cost reducing investment.

8. Intertemporal pricing.

9. Regulating a natural monopoly.

Chapter 9 - Externalities and Public Goods

1. Externalities.

2. Pollution and Income Level.

3. Regulating externalities under incomplete information.

4. The problem of the commons.

5. Private contributions to a public good.

6. Production and Externalities.

7. Positive and Negative Externalities.

8. Externalities.

5
 Chapter 1 - Preferences and Utility
End-of-Chapter Exercises - Answer key

1. [Checking properties of preference relations]. For each of the folllowing pref-

erence relations on R2 , check whether they are rational (complete and transitive),
monotone, convex, and locally nonsatiated (LSN).

(a) Bundle (x1 ; x2 )is weakly prefered to (y1 ; y2 ) , i.e., (x1 ; x2 ) % (y1 ; y2 ) if and only
if x1 > y1 and x2 > y2 .
1. Completeness. First note that intuitively, the pref. relation regular that a
bundle (x1 ; x2 )contains more of both conponents than another bundle (y1 ; y2 )
in order for the former to be preferable to the latter. Hence, if we take two
bundles, (x1 ; x2 ) and (y1 ; y2 ) 2 R2 , so that (x1 ; x2 ) contains more of only
the …rst component of the bundle, i.e., x1 > y1 but x2 < y2 , then clearly
(x1 ; x2 ) (y1 ; y2 ). In addition, (y1 ; y2 ) (x1 ; x2 ) since y1 > x1 and y2 > x2
are not satis…ed. This preference relation is then not complete, and as a
consequence, not rational.
2. Transitivity. Take three bundles (x1 ; x2 ),(y1 ; y2 ) and (z1 ; z2 ) 2 R2 satisfying
(x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ). Then, we must have that x1 > y1 and
x2 > y2 , and that y1 > z1 and y2 > z2 . It follows that x1 > z1 and x2 > z2 ,
so (x1 ; x2 ) % (z1 ; z2 ). Hence, the preference relation is transitive.
3. Monotonicity. Take any two bundles (x1 ; x2 ) and (y1 ; y2 ) 2 R2 with x1 > y1
and x2 > y2 . Then, (x1 ; x2 ) % (y1 ; y2 ). In addition, since x1 > y1 and
x2 > y2 , (y1 ; y2 ) (x1 ; x2 ). It follows that the preference must be strict, i.e.,
(x1 ; x2 ) (y1 ; y2 ), and hence the preference relation satis…es monotonicity,
i.e., x y if xi > yi for all i = f1; 2g.
4. Convexity. Take three bundles (x1 ; x2 ),(y1 ; y2 ) and (z1 ; z2 ) 2 R2 with (y1 ; y2 ) %
(x1 ; x2 ) and (z1 ; z2 ) % (x1 ; x2 ). Then it must be that y1 > x1 and y2 > x2 ,
and similarly that z1 > x1 and z2 > x2 . For all 2 [0; 1], it must be the case
that
y1 + (1 )z1 % x1 and y2 + (1 )z2 % x2
Intuitively, the linear combination of two points,yi andzi , both of them above
a third point,xi , also lies abovexi for all i = f1; 2g.(In fact, the linear combina-
tion, yi +(1 )zi ,lies above the minfyi ; zi g.) It follows that y+(1 )z % x,
and hence the relation is convex.
5. LNS. Take any (x1 ; x2 ) 2 R2 and " > 0. Let (y1 ; y2 ) x1 + 2" ; x2 + 2" so
that y1 > x1 and y2 > x2 . By the preference relation in this example, it
follows that y % x, but not x % y, so that y x. Since it is true for all " > 0
that
rh
" i2 h " i2 "
kx yk = x1 x1 + + x2 x2 + = p <"
2 2 2
and then the preference relation is locally nonsatiated.

1
(b) Bundle (x1 ; x2 )is weakly prefered to (y1 ; y2 ) , i.e., (x1 ; x2 ) % (y1 ; y2 ) if and only if
x1 > y1 1.
Let us …rst build some intuition on this preference relation. Take a bundle
(2; 1), you can take any other of course! Then, the upper contour set of this
bundle is given by

U CS% (2; 1) = f(x1 ; x2 ) % (2; 1) () x1 2 1g = f(x1 ; x2 ) : x1 1g

while the lower contour set is de…ned as

LCS% (2; 1) = f(2; 1) % (x1 ; x2 ) () 2 x1 1g = f(x1 ; x2 ) : x1 3g

Finally, the consumer is indi¤erent between bundle (2,1) and the set of bun-
dles where

IN D% (2; 1) = f(x1 ; x2 ) (2; 1) () 1 x1 3g

Graphically, all bundles in R2+ such that x1 1 belong to the set of bundles
weakly preferred to (2,1), all bundles such that x1 3 belong to the set of
bundles weakly preferred to (2,1), and those bundles in between, 1 x1 3,
are indi¤erent to (2,1). [For a graphical representation of the UCS, LCS and
IND sets, see the …gure below.]

Completeness. We need that for any pair of bundles (x1 ; x2 ) and (y1 ; y2 ), ei-
ther (x1 ; x2 ) % (y1 ; y2 ) or (y1 ; y2 ) % (x1 ; x2 ), or both (i.e., (x1 ; x2 ) (y1 ; y2 )).
Since this preference relation only depends on the …rst component of every
bundle, we have that, for every pair of bundles (x1 ; x2 ) and (y1 ; y2 ), either:
1. x1 y1 1, which implies that (x1 ; x2 ) % (y1 ; y2 ), or
2. x1 < y1 1, which implies y1 x1 1, and hence (y1 ; y2 ) % (x1 ; x2 ), or
3. y1 + 1 > x1 y1 1, which implies that (x1 ; x2 ) (y1 ; y2 ). Hence, this
preference relation is complete.

2
Re‡exivity. We need to show that, for every bundle (x1 ; x2 ), (x1 ; x2 ) %
(x1 ; x2 ). In this case, (x1 ; x2 ) % (x1 ; x2 ) implies that x1 x1 1, which
is indeed true for any x1 2 R+ .
Transitivity. We need to show that, for any three bundles (x1 ; x2 ), (y1 ; y2 )
and (z1 ; z2 ) such that

if (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ), then (x1 ; x2 ) % (z1 ; z2 )

This property does not hold for this preference relation. In order to show
that, notice that a bundle (x1 ; x2 ) is preferred to another bundle (y1 ; y2 )
if its …rst component, x1 , is larger than that of the other bundle, y1 , by
less than one unit, i.e., condition x1 y1 1 is equivalent to 1 y1 x1
This condition is depicted in the …gure below. A similar argument can be
extended to the comparison between two bundles (y1 ; y2 ) and (z1 ; z2 ), where
the former is preferred to the latter if and only if the distance between their
…rst components is smaller than one, i.e., 1 z1 y1 . Hence, for bundle
(z1 ; z2 ) to be preferred to (x1 ; x2 ), we need that the distance between their
…rst components is smaller than one, i.e., 1 z1 x1 . However, condition
1 z1 x1 does not necessarily holds for three bundles satisfying 1 y1 x1
(e.g., y1 x1 = 0:8) and 1 z1 y1 (e.g., z1 y1 = 0:6, implying z1 x1 =
0:8 + 0:6 = 1:4 > 1).

We can also show that this preference relation violates transitivity with a
counterexample. Consider the following three bundles (notice that the second
component of every bundle is inconsequential, since the preference ordering
only relies on a comparison of the …rst component of every vector):

(x1 ; x2 ) = (5; 4)
(y1 ; y2 ) = (6; 1)
(z1 ; z2 ) = (7; 2)

First, note that (x1 ; x2 ) % (y1 ; y2 ) since x1 y1 1 (i.e., 5 6 1). Addi-

tionally, (y1 ; y2 ) % (z1 ; z2 ) is also satis…ed since y1 z1 1 (i.e., 6 7 1).
However, (x1 ; x2 ) (z1 ; z2 ) since x1 z1 1 (i.e., 5 7 1). Hence, this
preference relation does not satisfy Transivity.

3
 Continuity. Continuity requires that both the upper and the lower contour
sets are closed, which is satis…ed given that they both contain their boundary
points.
Monotonicity. This property holds for any small increase > 0 in the amount
of good 1, x1 + . In particular, (x1 + ; x2 ) % (x1 ; x2 ) since for that we need
x1 + x1 1, and solving for yields 1, which holds for any > 0.
Weak Convexity. This property implies that the upper contour set must be
convex, that is, if bundle (x1 ; x2 ) is weakly preferred to (y1 ; y2 ), (x1 ; x2 ) %
(y1 ; y2 ), then the linear combination of them is also weakly preferred to
(y1 ; y2 ),

(x1 ; x2 ) + (1 ) (y1 ; y2 ) % (y1 ; y2 ) for any 2 [0; 1]

In this case, (x1 ; x2 ) % (y1 ; y2 ) implies that x1 y1 1; whereas (x1 ; x2 ) +

(1 ) (y1 ; y2 ) % (y1 ; y2 ) implies

x1 + (1 ) y1 y1 1 () x1 y1 1

which is true given that (x1 ; x2 ) % (y1 ; y2 ). Hence, this preference relation is
weakly convex.
(c) Bundle (x1 ; x2 )is weakly prefered to (y1 ; y2 ) , i.e., (x1 ; x2 ) % (y1 ; y2 ), if x1 y1 1
and x2 y2 + 1.
Let us …rst build some intuition on this preference relation. Take a bundle
(2; 1). Then, the upper contour set of this bundle is given by

U CS% (2; 1) = f(x1 ; x2 ) % (2; 1) () x1 2 1 and x2 1 + 1g

= f(x1 ; x2 ) : x1 1 and x2 2g

which is graphically represented by all those bundles in R2+ in the lower right-
hand corner (below x2 = 2 and to the right of x1 = 1). On the other hand,
the lower contour set is de…ned as

LCS% (2; 1) = f(2; 1) % (x1 ; x2 ) () 2 x1 1 and 1 x2 + 1g

= f(x1 ; x2 ) : x1 3 and x2 0g

which is graphically represented by all those bundles in R2+ in the left half of
the positive quadrant (above x2 = 0 and to the left of x1 = 3).Finally, the
consumer is indi¤erent between bundle (2,1) and the set of bundles where

IN D% (2; 1) = f(x1 ; x2 ) (2; 1) () 1 x1 3 and 0 x2 2g

[For a graphical representation of the UCS, LCS and IND sets, see the …gure

4
below.]

Completeness. From the above analysis it is easy to note that this property
is not satis…ed, since there are bundles in the area x1 > 3 and x2 2 where
our preference relation does not specify if they belong to the upper contour
set, the lower contour set, or the indi¤erence set. Another way to prove that
completeness does not hold is by …nding a counterexample. In particular, we
must …nd an example of two bundles such that neither (x1 ; x2 ) % (y1 ; y2 ) nor
(y1 ; y2 ) % (x1 ; x2 ). Let us take two bundles,

(x1 ; x2 ) = (1; 2) and (y1 ; y2 ) = (4; 6)

We have that:
1. (x1 ; x2 ) (y1 ; y2 ) since 1 4 1 for the …rst component of the bundle,
and
2. (y1 ; y2 ) (x1 ; x2 ) since 6 2 + 1 for the second component of the bun-
dle. Hence, there are two bundles for which neither (x1 ; x2 ) % (y1 ; y2 )
nor (y1 ; y2 ) % (x1 ; x2 ), which implies that this preference relation is not
complete.
3. [The graphical representation of the UCS, LCS and IND sets allows us
to detect that this preference relation is not complete, since we obtained
a region of (x1 ; x2 ) (for which x1 > 3 and x2 > 2) for which we cannot
determine a preference ordering between the bundle (2; 1) and the bundles
in that region.]
Re‡exivity. We need to show that, for every bundle (x1 ; x2 ), (x1 ; x2 ) %
(x1 ; x2 ). In this case, (x1 ; x2 ) % (x1 ; x2 ) implies that x1 x1 1, which
is indeed true for any x1 2 R+ , and x2 x2 + 1 which is also true for any
x 2 2 R+ .
Transitivity. We need to show that, for any three bundles (x1 ; x2 ), (y1 ; y2 )
and (z1 ; z2 ) such that

if (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ), then (x1 ; x2 ) % (z1 ; z2 )

5
 This property does not hold for this preference relation. In order to show
that, let us consider the following three bundles (that is, we are …nding a
counterexample to show that transitivity does not hold):
(x1 ; x2 ) = (2; 1)
(y1 ; y2 ) = (3; 4)
(z1 ; z2 ) = (4; 6)
First, note that (x1 ; x2 ) % (y1 ; y2 ) since x1 y1 1 (i.e., 2 3 1), and
x2 y2 + 1 (i.e., 1 4 + 1). Additionally, (y1 ; y2 ) % (z1 ; z2 ) is also satis…ed
since y1 z1 1 (i.e., 3 4 1), and y2 z2 + 1 (i.e, 3 4 + 1). However,
(x1 ; x2 ) (z1 ; z2 ) since x1 z1 1 (i.e., 2 4 1). Hence, this preference
relation does not satisfy Transivity.
Continuity. Continuity requires that both the upper and the lower contour
sets are closed, which is satis…ed given that they both contain their boundary
points.
Monotonicity. This property holds for any small increase > 0 in the amount
of good 1, x1 + . In particular, (x1 + ; x2 ) % (x1 ; x2 ) since for that we need
x1 + x1 1, and solving for yields 1, which holds for any > 0.
Weak Convexity. This property implies that the upper contour set must be
convex, that is, if bundle (x1 ; x2 ) is weakly preferred to (y1 ; y2 ), (x1 ; x2 ) %
(y1 ; y2 ), then the linear combination of them is also weakly preferred to
(y1 ; y2 ),
(x1 ; x2 ) + (1 ) (y1 ; y2 ) % (y1 ; y2 ) for any 2 [0; 1]
In this case, (x1 ; x2 ) % (y1 ; y2 ) implies that x1 y1 1 and x2 y2 + 1;
whereas (x1 ; x2 ) + (1 ) (y1 ; y2 ) % (y1 ; y2 ) implies
x1 + (1 ) y1 y1 1 for the …rst component, and
x2 + (1 ) y2 y2 + 1 for the second component.
which respectively imply
(x1 y1 ) 1, and
(x2 y2 ) 1
and since (x1 y1 ) 1 and (x2 y2 ) 1 by assumption, i.e., (x1 ; x2 ) %
(y1 ; y2 ), then both of the above conditions are true for any 2 [0; 1]. Hence,
this preference relation is weakly convex.
(d) Bundle (x1 ; x2 )is weakly prefered to (y1 ; y2 ) , i.e., (x1 ; x2 ) % (y1 ; y2 ) if and only
if max fx1 ; x2 g > max fy1 ; y2 g.
1. Completeness. Intuitively, this relation states that a bundle (x1 ; x2 )is prefered
to an alternative bundle (y1 ; y2 )if and only if the most abundant component
of the …rst bundle exceeds the most abundant component of the second bun-
dle. Hence, for any two bundles (x1 ; x2 ),(y1 ; y2 ) 2 R2 , either max fx1 ; x2 g >
max fy1 ; y2 g, or max fy1 ; y2 g > max fx1 ; x2 g, or both. It follows that either
(x1 ; x2 ) (y1 ; y2 ), or (y1 ; y2 ) (x1 ; x2 ), or both. Hence, this preference
relation is complete.

6
 2. Transitivity. Take three bundles (x1 ; x2 ), (y1 ; y2 ) and (z1 ; z2 ) 2 R2 satis-
fying (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ). Then, they must sratisfy
that, on one hand, max fx1 ; x2 g > max fy1 ; y2 g, and on the other hand,
max fy1 ; y2 g > max fz1 ; z2 g. Therefore, max fx1 ; x2 g > max fz1 ; z2 g, which
implies (x1 ; x2 ) % (z1 ; z2 ). Hence, the preference relation is transitive. Since,
it is also complete as shown above, this preference relation is rational.
3. Monotonicity. Take any (x1 ; x2 ) and (y1 ; y2 ) 2 R2 with x1 > y1 and x2 > y2 .
Then, max fx1 ; x2 g > max fy1 ; y2 g with strict inequality, and it follows that
(x1 ; x2 ) % (y1 ; y2 ). In addition, since max fx1 ; x2 g > max fy1 ; y2 g, we can
also say that (y1 ; y2 ) (x1 ; x2 ). It follows that (x1 ; x2 ) (y1 ; y2 ), and hence
the preference relation satis…es monotonicity, i.e., x y if xi > yi for all
i = f1; 2g.
4. Convexity. Take some (x1 ; x2 ),(y1 ; y2 ) and (z1 ; z2 ) 2 R2 with (y1 ; y2 ) %
(x1 ; x2 ) and (z1 ; z2 ) % (x1 ; x2 ). Therefore, it must be thatmax fy1 ; y2 g >
max fx1 ; x2 g, and similarly that max fz1 ; z2 g > max fx1 ; x2 g. However, the
convex combination of (y1 ; y2 ) and (z1 ; z2 ) with ,

max f y1 + (1 )z1 ; y2 + (1 )z2 g

is not necessarily higher than max fx1 ; x2 g. In order to see that, consider an
example in which max fy1 ; y2 g > max fx1 ; x2 g, and max fz1 ; z2 g > max fx1 ; x2 g,
such as (y1 ; y2 ) = (0; 4) and (x1 ; x2 ) = (3; 3) and (z1 ; z2 ) = (4; 0). Now, note
that the convex combination of (y1 ; y2 ) and (z1 ; z2 ) with , will give us values
between 0 and 4. For intermediate values of (such as = 21 ) we have that

1 1 1 1
max 0 + 4; 4 + 0 = max f2; 2g < max f3; 3g
2 2 2 2

Hence, the preference relation is not convex.

5. LNS. Take any (x1 ; x2 ) 2 R2 and " > 0. Let (y1 ; y2 ) x1 + 2" ; x2 + 2" so
that max fy1 ; y2 g > max fx1 ; x2 g. By the preference relation in this example,
it follows that y % x, but not x % y, so that y x. Since it is true for all
" > 0 that
rh
" i2 h " i2 "
kx yk = x1 x1 + + x2 x2 + = p <"
2 2 2
and then the preference relation is locally nonsatiated.
(e) Bundle (x1 ; x2 )is weakly prefered to (y1 ; y2 ) , i.e., (x1 ; x2 ) % (y1 ; y2 ) if and only
if max fx1 ; x2 g > min fy1 ; y2 g.
1. Completeness. Intuitively, this relation states that a bundle (x1 ; x2 )is prefered
to an alternative bundle (y1 ; y2 )if and only if the most abundant component of
the …rst bundle exceeds the minimal component of the second bundle. Hence,
for any two bundles (x1 ; x2 ),(y1 ; y2 ) 2 R2 , either max fx1 ; x2 g > min fy1 ; y2 g,
or min fy1 ; y2 g > max fx1 ; x2 g, or both. In the …rst case, we clearly have

7
 (x1 ; x2 ) (y1 ; y2 ). In the second, note that min fy1 ; y2 g > max fx1 ; x2 g
implies

max fy1 ; y2 g > min fy1 ; y2 g > max fx1 ; x2 g > min fx1 ; x2 g

and hence, max fy1 ; y2 g > min fx1 ; x2 g, which implies (y1 ; y2 ) (x1 ; x2 ).
Hence, this preference relation is complete.
2. Transitivity. Take three bundles (x1 ; x2 ), (y1 ; y2 ) and (z1 ; z2 ) 2 R2 satis-
fying (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ). Then, they must satisfy
that, on one hand, max fx1 ; x2 g > min fy1 ; y2 g, and on the other hand,
max fy1 ; y2 g > min fz1 ; z2 g. However, it can be the case that max fx1 ; x2 g <
min fz1 ; z2 g. For instance, consider (x1 ; x2 ) = (1; 0) and (y1 ; y2 ) = (3; 0),
which satis…es max fx1 ; x2 g > min fy1 ; y2 g; and (z1 ; z2 ) = (2; 2), which sat-
is…es max fy1 ; y2 g > min fz1 ; z2 g. However, note that max fx1 ; x2 g = 1 <
min fz1 ; z2 g = 2, which implies (x1 ; x2 ) (z1 ; z2 ), and we cannot say that
(x1 ; x2 ) % (z1 ; z2 ). Hence the preference relation is not transitive, and as a
consequence not rational.
3. Monotonicity. Take any two bundles (x1 ; x2 ) and (y1 ; y2 ) 2 R2 with x1 > y1
and x2 > y2 . Then, max fx1 ; x2 g > min fy1 ; y2 g and hence (x1 ; x2 ) % (y1 ; y2 ).
However, we can also have that max fy1 ; y2 g > min fx1 ; x2 g, which implies
(y1 ; y2 ) % (x1 ; x2 ). And as a consequence, x y. In order to see that, let us
consider the following example: (x1 ; x2 ) = (3; 1) and (y1 ; y2 ) = (2; 0). Indeed,
note that max fx1 ; x2 g = 3 > min fy1 ; y2 g = 0, but also max fy1 ; y2 g = 2 >
min fx1 ; x2 g = 1. Hence, this preference relation doesn’t satisfy monotonicity.
4. Convexity. Take three bundles (x1 ; x2 ),(y1 ; y2 ) and (z1 ; z2 ) 2 R2 with (y1 ; y2 ) %
(x1 ; x2 ) and (z1 ; z2 ) % (x1 ; x2 ). Therefore, it must be thatmax fy1 ; y2 g >
min fx1 ; x2 g, and similarly thetmax fz1 ; z2 g > min fx1 ; x2 g. However, the
convex combination of (y1 ; y2 ) and (z1 ; z2 ) with ,

max f y1 + (1 )z1 ; y2 + (1 )z2 g

is not necessarily higher than min fx1 ; x2 g. In order to see that, consider an
example in which max fy1 ; y2 g > min fx1 ; x2 g, and max fz1 ; z2 g > min fx1 ; x2 g,
such as (y1 ; y2 ) = (0; 4) and (x1 ; x2 ) = (3; 3) and (z1 ; z2 ) = (4; 0). Now, note
that the convex combination of (y1 ; y2 ) and (z1 ; z2 ) with , will give us values
between 0 and 4. For intermediate values of (such as = 21 ) we have that

1 1 1 1
max 0 + 4; 4 + 0 = max f2; 2g < min f3; 3g
2 2 2 2

Hence, the preference relation is not convex.

5. LNS. Consider (x1 ; x2 ) = (1; 0). To establish LNS we must …nd a pair
(y1 ; y2 ) 2 R2 such that (y1 ; y2 ) it is arbitrarilly close to (x1 ; x2 ), and (y1 ; y2 )
(x1 ; x2 ) strictly. In order to obtain (y1 ; y2 ) (x1 ; x2 ) we need that (y1 ; y2 ) %
(x1 ; x2 ) and (x1 ; x2 ) (y1 ; y2 ). By the preference relation in this example,
the …rst condition implies max fy1 ; y2 g > min fx1 ; x2 g, whereas the second

8
 condition implies max fx1 ; x2 g < min fy1 ; y2 g. However, min fx1 ; x2 g = 0
and max fx1 ; x2 g = 1. This implies that
max fy1 ; y2 g > 0 and 1 < min fy1 ; y2 g
Hence, max fy1 ; y2 g > min fy1 ; y2 g > 1. As a consequence, both coordinates
in (y1 ; y2 ) must exceed 1 for this condition to be ful…lled, and points to the
northeast of (1; 1) cannot be found to be arbitrarilly close to (x1 ; x2 ) = (1; 0).
Hence, this preference relation does not satisfy LNS.

2. [Checking properties of preference relations]. Consider the following preference

relation de…ned in X = R2+ : (x1 ; x2 ) % (y1 ; y2 ) if and only if
min f3x1 + 2x2 ; 2x1 + 3x2 g > min f3y1 + 2y2 ; 2y1 + 3y2 g

(a) For any given bundle (y1 ; y2 ), draw the upper contour set, the lower contour set,
and the indi¤erence set of this preference relation.
Take a bundle (2; 1). Then, the upper contour set of this bundle is given by
U CS% (2; 1) = f(x1 ; x2 ) % (2; 1)g
= fmin f3x1 + 2x2 ; 2x1 + 3x2 g > 7 min f8; 7gg
which is graphically represented by all those bundles in R2+ which are strictly
above both lines 3x1 + 2x2 = 7 and 2x1 + 3x2 = 7. (See …gure below). On the
other hand, the lower contour set is de…ned as
LCS% (2; 1) = f(2; 1) % (x1 ; x2 )g
= f7 > min f3x1 + 2x2 ; 2x1 + 3x2 gg
(See …gure below).

Finally, there are no bundles for which the consumer is just indi¤erent be-
tween bundle (2,1) and any other bundle (note that there are no bundles for
which the upper and lower contour set coincide). Hence,
IN D% (2; 1) = ?

9
(b) Check if this preference relation satis…es: (i) completeness, (ii) transitivity, (iii)
monotonicity, and (iv) weak convexity.
Completeness. For all (3x1 + 2x2 ) 2 R+ and (2x1 + 3x2 ) 2 R+ , the minimum
min f3x1 + 2x2 ; 2x1 + 3x2 g = a exists and a 2 R+ . Similarly, the minimum
min f3y1 + 2y2 ; 2y1 + 3y2 g = b exists and b 2 R+ . Therefore, we can easily
compare a and b, obtaining that either a > b which implies (x1 ; x2 ) % (y1 ; y2 ),
or a < b which implies (y1 ; y2 ) % (x1 ; x2 ), or both.
Transitivity. We need to show that, for any three bundles (x1 ; x2 ), (y1 ; y2 )
and (z1 ; z2 ) such that
if (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ), then (x1 ; x2 ) % (z1 ; z2 )
First, note that (x1 ; x2 ) % (y1 ; y2 ) implies
a min f3x1 + 2x2 ; 2x1 + 3x2 g > min f3y1 + 2y2 ; 2y1 + 3y2 g b
and (y1 ; y2 ) % (z1 ; z2 ) implies that
b min f3y1 + 2y2 ; 2y1 + 3y2 g > min f3z1 + 2z2 ; 2z1 + 3z2 g c
Combining both conditions we have that a > b > c, which implies
min f3x1 + 2x2 ; 2x1 + 3x2 g > min f3z1 + 2z2 ; 2z1 + 3z2 g
and we can conclude that this preference relation is Transitive.
Monotonicity. We need to show that, for a small increase > 0 in the amount
of good 1, x1 + , we have that (x1 + ; x2 ) % (x1 ; x2 ). That is, we need to
show that
min f3 (x1 + ) + 2x2 ; 2 (x1 + ) + 3x2 g > min f3x1 + 2x2 ; 2x1 + 3x2 g
rewriting the above inequality as
min fa0 ; b0 g > min fa; bg
we have that: (1) if x2 > x1 then b > a; or (2) if x2 > x1 then b0 > a0 .
Given that x1 2 (x2 ; x2 ), then we have that
3x1 + 2x2 < 2(x1 + ) + 3x2
and if x1 < x2 2 , then min fa; a0 g = a. On the other hand, if x1 2
(x2 2 ; x2 ), then min fa; b0 g = a and if min fb; b0 g = b.
Weak Convexity. This property implies that the upper contour set must be
convex, that is, if bundle (x1 ; x2 ) is weakly preferred to (y1 ; y2 ), (x1 ; x2 ) %
(y1 ; y2 ), then the linear combination of them is also weakly preferred to
(y1 ; y2 ),
(x1 ; x2 ) + (1 ) (y1 ; y2 ) % (y1 ; y2 ) for any 2 [0; 1]
In this case we must show that min fa; bg > min fc; dg implies
min f a + (1 ) c; b + (1 ) dg > min fc; dg

10
 1. First case: min fa; bg a, min fc; dg c and without loss of generality,
a > c.Therefore,

min f a + (1 ) c; b + (1 ) dg a + (1 )c

and a + (1 ) c > min fc; dg c. For this case, convexity is satis…ed.

2. Second case: min fa; bg a, min fc; dg d and without loss of generality,
a > d.Therefore,

min f a + (1 ) c; b + (1 ) dg a + (1 )c

and a + (1 ) c > min fc; dg d given that a > d and c > d. For this
case, convexity is satis…ed as well. Analogously for the other two cases.

3. [Checking for completeness and transitivity.]. Consider the preference relation

de…ned on a consumption set X R2+ : for all any two bundles x, x0 2 X, bundle x is
weakly preferred to x0 , i.e., x % x0 , if and only if

x1 x01 1 and x2 x02 + 1:

Check if this preference relation satis…es completeness and transitivity.

Completeness does not hold. It su¢ ces to consider two bundles, such as (1; 2)
and (4; 6) for which we have that (1; 2) (4; 6), given that 1 4 1 in the …rst
component of every bundle, and we also have that (4; 6) (1; 2) since 6 2 + 1
in the second component. Hence, it is easy to …nd pairs of bundles x, x0 2 X, for
which this preference relation does not conclude x % x0 nor x0 % x, proving that
completeness is not satis…ed.
Transitivity does not hold either since, for any three bundles x; x0 ; x00 2 R2+ , we
cannot guarantee that x % x0 and x0 % x00 implies x % x00 . You can take three
bundles, such as (2,1), (3,4) and (4,6) in order to show that:
– First, (2; 1) % (3; 4), given that 2 3 1 and 1 4 + 1,
– Second, (3; 4) % (4; 6), since 3 4 1 and 4 6 + 1,
– However, (2; 1) % (4; 6) does not hold, because 2 4 1 = 3.

4. Explain: Monotonicity and Strong monotonicity in preference relations.

Monotonicity states that increasing the amount of some commodities cannot hurt,
and increasing the amount of all commodities is strictly preferred. Formally, if
xk yk for all k, then x % y, and if xk > yk for all k, then x y. On the
other hand, strong monotonicity states that the consumer is strictly happier with
any additional amount of any comodity, that is, if xk yk for all k and x 6= y,
then x y. Then if the consumer strictly prefers bundle x to y, x y,.when his
preference relation satis…es strong monotonicity, it may be the case that bundle
x is not strictly preferred to y when his preferences only satisfy monotonicity.
That is why strong monotonicity is a more restrictive (“stronger”) assumption on
preferences than monotonicity.

11
 (a) Provide an example where a bundle x is (strictly) preferred to bundle y when
preferences satisfy strong monotonicity, but x is not necessarily preferred to y
under monotonocity.
Example: x = (1; 2) and y = (1; 1). Clearly, x y when preferences satisfy
strong monotonicity since some component in x is higher than the correspond-
ing component in y, i.e., xj yj for some j. However, we cannot state that
x y (strictly) when preferences only satisfy monotonicity, since xk > yk
does not hold for all k commodities.
(b) Local nonsatiation allows for the fact that some commodities may be “bads” in
the sense that the consumer would sometimes prefer less of them (like garbage or
noise). However, it is not possible for all goods to always be bads if preferences
satisfy local nonsatiation. Why?
If all goods are always bads, then the consumer would like to reduce his
consumption of them as much as possible (indi¤erence curves would be bowed-
in towards the origin). But if this is the case, then the consumer would be
totally satiated when his consumption of all goods reaches zero for all of them,
which contradicts local nonsatiation.

5. [Utility representation of preferences]. Consider a rational preference relation %.

Show that

if u(x) = u(y) implies x y, and

if u(x) > u(y) implies x y,

then u( ) is a utility function representing the preference relation %.

In order to show that u( ) is a utility function representing the preference relation

%, recall the de…nition of a utility function: A function u : X ! R is a utility
function representing % if, for all x; y 2 X,

x % y , u (x) > u (y)

First, we show that x % y =) u (x) > u (y). Then we show that x % y (=

u (x) > u (y).

(a) 1. Suppose that x % y. If, in addition, y % x then it must be that x y, and

the consumer is indi¤erent between x and y, which implies that u (x) = u (y).
If, instead, y x then x y and the consumer strictly prefers x to y, what
implies u (x) > u (y). Thus, we have shown that if x % y then u (x) > u (y),
for all x; y 2 X:
2. Suppose that u (x) > u (y). This implies that either u (x) > u (y) or u (x) =
u (y). In the …rst case, this implies that x y, i.e., x % y and y x. In the
second case, it implies that x y, i.e., x % y and y % x. Hence, we have
shown that if u (x) > u (y) then x % y, for all x; y 2 X:
3. Both parts of the proof show that the utility function u( ) represents the
preference relation% :

12
6. Prove that, if a preference relation % is rational, and if x y % z, then x z.

By contradiction, let us suppose that in a rational preference relation %, if x

y % z, then z % x. Note that if y % z and z % x, by transitivity, y % x. (Note
that we can use transitivity because the preference relation is rational) But our
last result (y % x) contradicts the …rst part of our initial assumption (x y % z).
Hence, we have reached a contradiction. Then, if x y % z, then x z.

7. Show that if f : R ! R is a strictly increasing function, and u : X ! R is a utility

function representing a rational preference relation %, then the function v : X ! R
de…ned by v(x) = f (u(x)) is also a utility function representing the same rational
preference relation %.

To show that v(x) = f [u (x)]is also a utility function representing the same ra-
tional preference relation%, we need to show
f [u (x)] > f [u (y)] () x % y
– Since f : R ! R is a strictly increasing function, then
f [u (x)] > f [u (y)] () u (x) > u (y)
and since the utility function u( ) represents the rational preference relation
%, we can conclude that u (x) > u (y) () x % y. Therefore, we can obtain
the same result for any function v : X ! R de…ned by v(x) = f (u(x)), where
f : R ! R is a strictly increasing function of the utility function u( ). As
a consequence, any function v : X ! R de…ned by v(x) = f (u(x)) is also a
utility function representing the same rational preference relation %.

8. [Monotonic transformations]. Let u(x) represent a consumer’s preference relation

% over the consumption set X = R2+ . Assume that utility levels are positive for any
positive consumption, i.e., u(x) 0 for all x 2 R2+ . For each of the following functions,
state whether or not they also represent the same preference relation %. [Hint: From
the previous exercise, we know that only monotonic transformations of the utility
function represent the original preference relation].

(a) f (x) = u(x) + [u(x)]2

This transformation of the utility function u(x) does represent the same pref-
erences, since it is monotonic in u(x). A simple way to show this is by checking
whether f (x) is weakly increasing in u(x) (i.e., f (x) is monotonic in u(x)).
Di¤erentiating f (x) with respect to u(x) we obtain 1 + 2u(x), which is always
positive since u(x) 0. Therefore, f (x) does represent the same preferences
as u(x).
(b) f (x) = u(x) [u(x)]2
This transformation of the utility function u(x) does not represent the same
preferences. Similarly to the previous exercise, let us di¤erentiate f (x) with
respect to u(x), obtaining 1 2u(x), which is positive if and only if u(x) 1=2.
Hence, we cannot conclude that function f (x) is a monotonic transformation
of u(x), and it hence does not represent the same preferences as u(x).

13
 Alternative Proof: Let us now prove it by a counterexample. First, suppose
that x is weakly prefered to y, x % y. Since u(x) is representing the prefer-
ence relation %, then x % y =) u(x) u(y). Secondly, say that for example,
u(x) = 2 and u(y) = 1. Clearly u(x) u(y), but the monotonic transforma-
tion f (x) = 2 and f (y) = 0 and then we have f (y) > f (x). Therefore, f (x)
does not represent the same preferences as u(x).
P
2
(c) f (x) = u(x) + xi
i=1

This function does not represent the same preferences. Let us prove it by
using a counterexample in which u(x) u(y) but f (x) < f (y). Note that
intuitively, this is the case for any utility function which assigns most of the
weight to the …rst component of the consumption bundle x 2 R2+ . A valid
counterexample is then the following: x = (2; 1), y = (1; 4), and a utility
function just “concerned” about the …rst component of every consumption
bundle u(x) = 2 and u(y) = 1. Therefore, f (x) = 2 + (2 + 1) = 5 and
f (y) = 1 + (1 + 4) = 6, leading to f (x) < f (y). Hence, f (x) does represent
the same preferences as u(x).
(d) f (x) = [u(x)]2 + b u(x) + c, where b; c > 0.
Di¤erentiating f (x) with respect to u(x) we obtain 2u(x) + b, which is always
positive since u(x) 0 and b > 0.

9. [Lexicographic preference relation]. Let us de…ne a lexicographic preference re-

lation in a continuouous consumption set X Y , where for simplicity both X = [0; 1]
and Y = [0; 1], as follows:

x1 > y1 , or if
(x1 ; x2 ) % (y1 ; y2 ) if and only if (1)
x1 = y1 and x2 > y2

(a) Show that % is a rational preference relation (i.e., it is complete and transitive).
1. Completeness. By de…nition, %is a complete preference relation if for all
(x1 ; x2 ),(y1 ; y2 ) 2 R2 , either (x1 ; x2 ) % (y1 ; y2 ) or (y1 ; y2 ) % (x1 ; x2 ), or both.
Hence, we need to show that

(x1 ; x2 ) (y1 ; y2 ) =) (y1 ; y2 ) % (x1 ; x2 )

Indeed, note that

y1 > x1 , and if
(x1 ; x2 ) (y1 ; y2 ) if (2)
x1 6= y1 or y2 > x2

First, note that we changed “or” for “and”, and viceversa, in order to con-
struct a full negation of the argument in (1). Now, note that (2) implies
(y1 ; y2 ) % (x1 ; x2 ). Therefore, we have shown that (x1 ; x2 ) (y1 ; y2 ) implies
(y1 ; y2 ) % (x1 ; x2 ). Hence, the preference relation is complete.

14
 2. Transitivity. Let us take three bundles (x1 ; x2 ),(y1 ; y2 ) and (z1 ; z2 ) 2 R2 with
(x1 ; x2 ) % (y1 ; y2 ):

x1 > y1 , or if
(x1 ; x2 ) % (y1 ; y2 ) if
x1 = y1 and x2 > y2

and (y1 ; y2 ) % (z1 ; z2 ), that is

y1 > z1 , or if
(y1 ; y2 ) % (z1 ; z2 ) if
y1 = z1 and y2 > z2

Hence, we need to check for transitivity in the four possible cases in which
(x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) % (z1 ; z2 ).
1. If x1 > y1 , and y1 > z1 , then x1 > z1 . As we know that x1 > z1 implies
(x1 ; x2 ) % (z1 ; z2 ). Hence, transitivity is checked in this case.
2. If (x1 = y1 and x2 > y2 ) and (y1 = z1 and y2 > z2 ), then (x1 = z1 and
x2 > z2 ). And we know that (x1 = z1 and x2 > z2 ) implies (x1 ; x2 ) %
(z1 ; z2 ), which validates transitivity.
3. If x1 > y1 , and (y1 = z1 and y2 > z2 ), then x1 > z1 . As we know that
x1 > z1 implies (x1 ; x2 ) % (z1 ; z2 ). Hence, transitivity is checked in this
case.
4. If y1 > z1 and (x1 = y1 and x2 > y2 ), then x1 > z1 . As we know that
x1 > z1 implies (x1 ; x2 ) % (z1 ; z2 ), which validates transitivity. We have
then checked all four cases under which (x1 ; x2 ) % (y1 ; y2 ) and (y1 ; y2 ) %
(z1 ; z2 ) may occur, and in all of them we obtained (x1 ; x2 ) % (z1 ; z2 ),
con…rming that this preference relation is transitive. Therefore, since the
preference relation is complete and transitive, we can conclude that it is
rational.
(b) Show that % cannot be represented by a utility fuction u : X Y ! R.
Let us work by contradiction. So, let us suppose that there is a utility function
u( ) representing this lexicographic preference relation %. Then, for any x1 2
X, the pair (x1 ; 1) is strictly preferred to the pair (x1 ; 0), i.e., (x1 ; 1) (x1 ; 0).
If there is a utility function u( ) representing this preference relation, then we
must have by de…nition

(x1 ; 1) (x1 ; 0) () u(x1 ; 1) > u(x1 ; 0)

On the other hand, from the Archideam property, we know that we can pick
a rational number r(x1 ) such that it lies in between u(x1 ; 1) and u(x1 ; 0).

u(x1 ; 1) > r(x1 ) > u(x1 ; 0)

Let take any x1 ; x2 2 X, and let us suppose without loss of generality that
x1 > x2 . Similarly to our above result, we then have that

u(x2 ; 1) > r(x2 ) > u(x2 ; 0)

15
 And since x1 > x2 , we have that
u(x1 ; 1) > r(x1 ) > u(x1 ; 0) > u(x2 ; 1) > r(x2 ) > u(x2 ; 0)
what implies
r(x1 ) > r(x2 )
Then, r( ) provides a one-to-one function from the set of real numbers, R
(which is uncountable) to the set of rational numbers, Q, which is countable.
But this is a mathematical impossibility (for a review of real and rational
numbers, see Simon and Blume’s Mathematics for Economists, pp. 848-849).
Thus, we conclude that there can be no utility function representing the
lexicographic preferences when they are de…ned over a continuous set X Y ,
where X = [0; 1] and Y = [0; 1].
(c) [Hard] Assume now that this preference relation is de…ned in a …nite consumption
set X = X1 X2 , where X1 = fx11 ; x12 ; :::; x1n g and X2 = fx21 ; x22 ; :::; x2m g.
[Hint: You can de…ne a function Ni (xij ) as the number of elements in sequence
Xi prior to element xij : Ni (xi ) = fy 2 Xi jy < xij g. Then de…ne a utility function
u(y1 ; y2 ) = mN1 (y1 ) + N2 (y2 ), for any m 2 R++ , and for any pair (y1 ; y2 ) 2
X1 X2 .]
1. Let us …rst de…ne a function Ni (xij ) as the number of elements in sequence
Xi prior to element xij :
Ni (xi ) = # fy 2 Xi jy < xij g , where Xi = fxi1 ; xi2 ; :::; xin g :
Then we de…ne a utility function u(y1 ; y2 ) = mN1 (y1 ) + N2 (y2 ), for any
m 2 R++ , and for any pair (y1 ; y2 ) 2 X1 X2 . In order to show that this
utility function indeed represents the lexicographic preference realtion (when
consumption sets are …nite), we need to show that
(y1 ; y2 ) % (z1 ; z2 ) =) u(y1 ; y2 ) u(z1 ; z2 ), and
(y1 ; y2 ) % (z1 ; z2 ) (= u(y1 ; y2 ) u(z1 ; z2 )
2. Let us …rst show that (y1 ; y2 ) % (z1 ; z2 ) =) u(y1 ; y2 ) u(z1 ; z2 ). In order to
show that, we need to show that
y1 > z1 , or if
if then mN1 (y1 ) + N2 (y2 ) mN1 (z1 ) + N2 (z2 )
y1 = z1 and y2 > z2
So, we …rst need to check if this inequality is satis…ed when y1 > z1 , and
when (y1 = z1 and y2 > z2 ).
1. Let us …rst check that y1 > z1 implies mN1 (y1 ) + N2 (y2 ) mN1 (z1 ) +
N2 (z2 ). Alternatively, we can rewrite this inequality as
m [N1 (y1 ) N1 (z1 )] + [N2 (y2 ) N2 (z2 )] 0
Let us analyze if this expression can ever be negative (we will examine
the in…mum values). Regarding the …rst term, we know that, if y1 > z1 ,
inf [N1 (y1 ) N1 (z1 )] = k (k 1) = 1, since N1 (y1 ) > N1 (z1 ) given that y1 > z1

16
 and hence, inf [m [N1 (y1 ) N1 (z1 )]] = m. Therefore, m [N1 (y1 ) N1 (z1 )]
m, and the …rst term is always weakly above m. Let us now focus on the
second term (note that now we are not impossing any conditions on y2
and z2 ):
inf [N2 (y2 ) N2 (z2 )] = inf N2 (y2 ) sup N2 (z2 ) = 0 (m 1) = 1 m
Intuition: inf N2 (y2 ) = 0 implies that there are no elements prior to y2
(that is, y2 is the …rst term of the …nite consumption set); in contrast,
sup N2 (z2 ) = m 1 means that z2 is the last element in the sequence of
lenght m, and hence all other m 1 elements in the sequence were prior
to z2 . Hence, N1 (y1 ) N1 (z1 ) 1 m, and the second term is always
weakly above 1 m. Combining the results of the …rst and second term,
we can conclude that
m [N1 (y1 ) N1 (z1 )] + [N2 (y2 ) N2 (z2 )] m (1 m) = 1
which is clearly above 0. (Recall that we needed to show m [N1 (y1 ) N1 (z1 )]+
[N2 (y2 ) N2 (z2 )] 0). Therefore, y1 > z1 indeed implies u(y1 ; y2 )
u(z1 ; z2 ).
2. Let us now check that (y1 = z1 and y2 > z2 ) also implies mN1 (y1 ) +
N2 (y2 ) mN1 (z1 ) + N2 (z2 ). Alternatively, we can rewrite this inequality
as
m [N1 (y1 ) N1 (z1 )] + [N2 (y2 ) N2 (z2 )] 0
First, note that y1 = z1 implies that N1 (y1 ) = N1 (z1 ). Second, note that
y2 > z2 implies that N2 (y2 ) N2 (z2 ). Therefore, the above inequality
becomes
0 + [N2 (y2 ) N2 (z2 )] 0
| {z }
0

which con…rms what we needed to show. Hence, (y1 = z1 and y2 > z2 )

indeed implies u(y1 ; y2 ) u(z1 ; z2 ).
3. Let us now show that (y1 ; y2 ) % (z1 ; z2 ) (= u(y1 ; y2 ) u(z1 ; z2 ). First, note
that if u(y1 ; y2 ) u(z1 ; z2 ), then it must be the case that mN1 (y1 ) + N2 (y2 )
mN1 (z1 ) + N2 (z2 ). Rearranging, we obtain
m [N1 (y1 ) N1 (z1 )] + [N2 (y2 ) N2 (z2 )] 0
Then, note that this inequality can be positive for two di¤erent reasons: (1)
because N1 (y1 ) > N1 (z1 ), which implies y1 > z1 ; or because (2) N1 (y1 ) =
N1 (z1 ) and N2 (y2 ) N2 (z2 ), which implies y1 = z1 and y2 > z2 . And we
know that, by de…nition, these two cases describe the lexicographic preference
relation
y1 > z1 , or if
(y1 ; y2 ) % (z1 ; z2 ) if
y1 = z1 and y2 > z2
Hence, (y1 ; y2 ) % (z1 ; z2 ) (= u(y1 ; y2 ) u(z1 ; z2 ). Since we have shown
this implication in both direction, then we have con…rmed that this utility
function indeed represents the lexicographic preference relation
(y1 ; y2 ) % (z1 ; z2 ) () u(y1 ; y2 ) u(z1 ; z2 )

17
10. [Checking WARP]. For each of the following demand functions, check whether they
satisfy the weak axiom of revealed preference (WARP).

(a) “Random demand”: For any pair of prices p1 and p2 and wealth w, the consumer
randomizes uniformly over all points in the budget frontier.
Let us prove it by a counterexample.

x2

Bp,w

x(p’,w’)

x(p,w)

Bp’,w’

x1
Random demand

WARP states that

if p x (p0 ; w0 ) w and x (p0 ; w0 ) 6= x (p; w) then p0 x (p; w) > w0
That is, if the new consumption bundle (chosen under the new prices and
wealth) would not have been a¤ordable under the old prices and wealth, then
it must be the case that the old consumption bundle is not a¤ordable under
the new prices and wealth. In this case, we …nd that
p x (p0 ; w0 ) w and x (p0 ; w0 ) 6= x (p; w) but p0 x (p; w) < w0
That is, the old consumption bundle is still a¤ordable under the new prices
and wealth. Hence, there exists a probability greater than zero that random
demand assigns bundles as the ones chosen in the …gure. As a consequence
we can conclude that random demand does not satisfy WARP.
(b) “Average demand”: The expected “random demand”given p1 , p2 and w.
First, note that if the consumer randomizes uniformly over all points in her
budget line, then the expected random demand is allocated at the midpoint
of the budget line.
x2
w/p2
Bp,w

x(p,w)

0.5w/p2

0.5w/p1 w/p1 x1

Average demand

18
 Let us now prove that WARP is satis…ed for average demand. Let us work
by contradiction, by assuming that average demand violates WARP. There
are two possibilities in which this violation might take place, as the following
two …gures illustrate.

x2 x2
w/p2 w/p2
Bp,w Bp,w
x(p’,w’) x(p’,w’)
x(p,w)

x(p,w)

Bp’,w’ Bp’,w’

w/p1 w’/p1' x1 x1 ' x1 w/p1 w’/p1' x1

Let us compute point x1 and x01 . Recall that these points have to be allocated
at the midpoint of the budget line. Hence,

1w 1 w0
x1 = and x01 =
2 p1 2 p01
0
therefore 2x1 = pw1 and 2x01 = wp0 . Moreover, we can see in both …gures that
1
x01 < x1 . Therefore, 2x01 < 2x1 , which implies

w0 w
0
<
p1 p1
0
But in both …gures we in fact see that wp0 > pw1 . Hence, we have reached a
1
contradiction, and average demand cannot violate WARP.
(c) “Conscopious demand”: For any p1 , p2 and w,
8 n o
> w
< pi if pwi = min pw1 ; pw2 and w
6= w
p1 p2
xi (p1 ; p2 ; w) = w
if pw1 = pw2 and i = 1
>
: pi
0 otherwise
w w
Let us divide it into two cases: Case 1, in which p1 p2
, and Case 2, in which
w
p1
> pw2 .
w w
Case 1 p1 p2
: The demand function to the following expression
8 w
if pw1 6= pw2
< p1
w
xi (p1 ; p2 ; w) = if pw1 = pw2
p1
:
0 otherwise

19
 x2

Bp,w

x(p,w)

w/p1 x1

w w
Case 2 p1
> p2
: The demand function to the following expression

w
p2
if pw1 6= pw2
xi (p1 ; p2 ; w) =
0 otherwise

x2

x(p’,w’)

w’/p2'

Bp’,w’

x1

Finally, in this …gure we represent both cases simultaneously.

x2

Bp,w

x(p’,w’)

w’/p2'

x(p,w)

Bp’,w’

w/p1 x1

1. Let us now check if conscopious demand satis…es WARP. As mentioned in

part (a), WARP states that

if p x (p0 ; w0 ) w and x (p0 ; w0 ) 6= x (p; w) then p0 x (p; w) > w0

But as we can see in the …gure that deals with both cases, we have that
p0 x (p; w) < w0 . Hence, conscopious demand violates WARP.

20
11. [Some applications of WARP] The budget of a given consumer is entirely spent on
the following two goods, during the months of September and October.

Sept Oct
p1 $3 $8
p2 $4 $6
x1 4 3
x2 3 4

Let x0 = (x01 ; x02 ) denote his consumption of goods 1 and 2 during September, and
x1 = (x11 ; x12 ) his consumption during October. Similarly, let p0 = (p01 ; p02 ) be the vector
of prices for goods 1 and 2 during September, and p1 = (p11 ; p12 ) be his vector of prices
during October.

(a) Determine if his consumption bundle during October, x1 , is a¤ordable at Septem-

ber prices p0 , i.e., check if p0 x0 < p0 x1 .
His consumption bundle in September, under September prices, costs

p0 x0 = (4 3) + (3 4) = 24

and if the consumer were to buy his October consumption bundle x1 under
September prices, p0 , he would have to pay

p0 x1 = (3 3) + (4 5) = 25

Hence, the October bundle x1 is not a¤ordable at September prices p0 .

(b) Determine if his consumption bundle during September, x0 , is a¤ordable at Oc-
tober prices p1 , i.e., check if p1 x0 < p1 x1 .
His consumption bundle in October, under October prices, costs

p1 x1 = (8 3) + (6 4) = 48

and if the consumer were to buy his September consumption bundle x0 under
the new October prices, p1 , he would have to pay

p1 x0 = (8 4) + (6 3) = 50

Hence, the September bundle x0 is not a¤ordable at October prices p1 , the

consumer cannot keep consuming his initial consumption bundle x0 after the
price change.
(c) Based on your answers on parts (a) and (b), do his preferences violate the weak
axiom of revealed preference (WARP)?
Since these two bundles are never simultaneously available (they are never
a¤ordable under a given month’s prices), this consumer’s choice does not
violate WARP.

21
12. [WARP for undergrads] Darrell has a monthly income of $60. He spends his money
making telephone calls (measured in minutes of calls) and on other composite good y,
with price py = $1. His mobile phone company o¤ers him two plans: Plan A: pay no
monthly fee and make calls for $0.50 per minute, Plan B: pay a $20 monthly fee and
make calls for $0.20 per minute.

(a) Graph Darrell’s budget constraint under each of the two plans, with the number
of phone calls (good x) in the horizontal axis and the composite good (good y)
in the vertical axis.
Let x denote the number of phone calls, and y denote spending on other
goods. Under Plan A, Darrell’s budget line is JLM.

60 L

40
J
K

200 x
120

The expression of BLA is 0:5x + y = 60. Under Plan B, Darrell’s budget

line is KLN in the …gure. The expression of BLB is 0:2x + y = 40. In order
to know the point at which these two budget lines intersect each other, we
must rewrite one of them, in order to insert it in the other budget line. For
example, BLB can be rewritten as y = 40 0:2x. Inserting it into BLA we
obtain: 0:5x + (40 0:2x) = 60. Solving for x, we obtain that x = 66:67.
Hence,
y = 40 0:2x = 40 (0:2 66:67) = 26:67
Therefore, BLA and BLB intersect at point L, or about x = 67.
(b) If Plan A is better for him, what is the set of baskets he may purchase if his
behavior is consistent with the WARP?
If we know that Darrell chooses Plan A, his optimal bundle must lie on the
line segment JL. No point between L and M would be optimal under this plan
because then Darrell could have chosen a point under Plan B, between L and
N, that would have given him more minutes, and left him with more money
to buy other goods. However, we cannot exclude point L itself (Darrell could,
for instance, have perfect complements preferences with an “elbow”at point
L). Thus, if Darrell chooses Plan A his optimal basket could be anywhere
between J and L.
(c) What baskets might he purchase if Plan B is better for him?

22
 Similarly, if he chose Plan B then his optimal basket must lie between L and
N. Any point between L and K would be dominated by a point under Plan
A between L and J. Thus, if Darrell chooses Plan B his optimal basket could
be anywhere between L and N, including either of these points.

13. [WARP and rationality] Consider a given choice rule C(: ) de…ned over a budget set
B, and assume it satis…es the weak axiom of revealed preference (WARP). Does this
choice structure guarantee a rational preference relation? [Hint: it is su¢ cient to …nd
an example of choice rule satisfying WARP, but which would not satisfy transitivity
(one of the conditions for rationality)].

Let us prove it using an example of choice rule satisfying WARP, but which
does not satisfy transitivity. Consider a consumption set X = fx; y; zg, and the
following budget set
B = ffx; yg ; fy; zg ; fx; zgg
Now, let us a consider the following choice rule

C (fx; yg) = fxg

C (fy; zg) = fyg

C (fx; zg) = fzg
Note that this choice rule satis…es WARP. However, note that such a choice rule
implies
C (fx; yg) = fxg
then x y and y z
C (fy; zg) = fyg
But if the preference relation was rational (satisfying transitivity as a consequence)
then we would have
x y and y z, then x z
But x z contradicts the choice rule C (fx; zg) = fzg, since this implies z x.
Therefore, we can conclude that a choice rule which satis…es WARP does not
guarantee a rational preference relation.

14. [WARP] Let (B; C( )) be a choice structure where B includes all nonempty subsets of
X, i.e., C(B) 6= ? for all sets B 2 B. We de…ne the choice rule C( ) to be distributive
if, for any two sets B and B 0 in B,

C(B) \ C(B 0 ) 6= ? implies that C(B) \ C(B 0 ) = C(B \ B 0 )

In words, the elements that choice rule C( ) selects both when facing set B and when
facing set B 0 , C(B) \ C(B 0 ), coincide with the elements that choice rule would select
when confronted with the elements that belong to both sets B \ B 0 , i.e., C(B \ B 0 ).
Show that, if choice rule C( ) is distributive, then choice structure (B; C( )) does not
necessarily satisfy the weak axiom of revealed preference. (A counterexample su¢ ces.)

23
 One possible counterexample is with X = fx; y; zg and

B = ffxg; fyg; fzg; fx; yg; fx; zg; fy; zg; fx; y; zgg

. Let C( ) be given by

Cfxg = fxg, Cfyg = fyg and Cfzg = fzg, and

Cfx; yg = fyg, Cfx; zg = fxg, Cfy; zg = fyg and Cfx; y; zg = fxg

First, note that the weak axiom is not satis…ed. In particular, while x and y both
belong to fx; yg and to fx; y; zg, this individual selects y 2 Cfx; yg (and does
not select x) but changes his choice to x when his set of available options also
includes z, i.e., Cfx; y; zg = fxg. Where the weak axiom fails, the intersection
between the chosen sets is empty: Cfx; y; zg \ Cfx; yg = ?, and so the fact that
C( ) is distributive does not have any e¤ect in this comparison.
For completeness, note that

C (fxg) \ C (fx; zg) = C (fxg) = fxg,

C (fxg) \ C (fx; y; zg) = C (fxg) = fxg,
C (fx; zg) \ C (fx; y; zg) = C (fx; zg) = fxg,
C (fyg) \ C (fx; yg) = C (fyg) = fyg,
C (fyg) \ C (fy; zg) = C (fyg) = fyg,
C (fx; yg) \ C (fy; zg) = C (fyg) = fyg.

15. [Lexicographic preferences and WARP]. Does the lexicographic preference re-
lation induce a choice structure that satis…es the weak axiom of revealed preference
(WARP)? [Hint: prove it in two steps. First, show whether the lexicographic preference
relation is a rational preference relation, and then show that every rational preference
relation implies a choice structure which satis…es WARP].

We want to prove this statement in two steps:

Lexicographic % =) % is rational
=) Any rational % induces a choice rule which satis…es WARP

The …rst step [Lexicographic % =) %is rational] was already shown in another
exercise of this chapter.
Let us prove here the second step [%is rational=)Any rational %induces a choice
rule which satis…es WARP].
– Let us …rst recall what is the de…nition of a choice rule which satis…es WARP:

if for some B 2 B with x; y 2 B, we have x 2 C(B),

then for any B 0 2 B with x; y 2 B 0 , we must have x 2 C(B 0 )

– First take some B 2 B. Let us assume that x; y 2 B, and that x 2 C (B; %).
Hence, x % y.

24
 – To check whether WARP is satis…ed, suppose that for some B 0 2 B and
x; y 2 B 0 , we have that y 2 C (B 0 ; %). This implies that y % z for any
z 2 B0.
– But we already had that x % y. Hence, by transitivity (note that here we
are allowed to use transitivity because the preference relation is rational) we
have that
x % y and y % z, then x % z
and x % z for all z 2 B 0 implies that x 2 C (B 0 ; %).
– Therefore, WARP is satis…ed. Indeed:

if for some B 2 B with x; y 2 B, we have x 2 C(B),

then for any B 0 2 B with x; y 2 B 0 , we must have x 2 C(B 0 )

16. For each of the following utility functions on R2+ , determine the Walrasian demand
correspondence x(p; w) = (x1 (p; w); x2 (p; w)) and the indirect utility function v(p; w).
First draw some indi¤erence curves and — if possible— compute the marginal rate of
substitution M RS12 (x).

Existence: First, note that in all three cases the budget set is compact (it is closed,
since the bundles in the frontier are available for the consumer, and bounded, given
by the bundles which lead the consumer to totally spend his budget). Additionally,
all utility functions are continuous. Therefore, we can apply Weierstrass theorem
to conclude that these UMPs have a solution.
Binding constraints: We know preferences are locally non-satiated, then the bud-
get constraint will be binding (see MWG Proposition 3.D.2 for a formal proof).
Additionally, we can easily check that these utility functions are increasing in both
x1 and x2 , which implies local non-satiation. Hence, we can assume thereafter
that the budget constraint is binding.

(a) u(x) = x31 x42

1. This utility function is a Cobb-Douglas utility function, with smooth indi¤er-
ence curves as discussed in class. Regarding the marginal rate of substitution
between goods x1 and x2 , M RSx1 ;x2 , we have
@u(x)
@x1 3x21 x42 3x2
M RSx1 ;x2 = @u(x)
= 3 3
=
4x1 x2 4x1
@x2

2. The UMP is given by

max u(x) = x31 x42
x1 ;x2

subject to
p1 x1 + p2 x2 w
x 1 ; x2 0
As mentioned above, the budget constraint will be binding. Furthermore,
since the utility from consuming zero amounts of either of the goods is zero,

25
 i.e., u (0; ) = u ( ; 0) = 0, and the consumer’s wealth is strictly positive,
w > 0, then it can never be optimal to consume zero amounts of either
of the goods. Therefore, we do not need to worry about the nonnegativity
constraints x1 ; x2 0. The Lagrangian of this UMP is then

L = x31 x42 [p1 x1 + p2 x2 w]

The …rst order conditions are:

@L
= 3x21 x42 p1 = 0
@x1
@L
= 4x31 x32 p2 = 0
@x2
Solving for on both …rst order conditions, we have

3x21 x42 4x3 x3

= 1 2
p1 p2
3x2 4x1
() =
p1 p2
Using now the budget constraint (which is binding), we have
w p2 x2
p1 x1 + p2 x2 = w () x1 =
p1 p1
and substituting this expression of x1 in the above equality, we have

3x2 4 w p2 x2 4w
= () x2 =
p1 p2 p1 p1 7 p2

and solving similarly for x1 , we obtain

3 w p1 x1 4x1 3w
= () x1 =
p1 p2 p2 p2 7 p1

Hence, the Walrasian demand function is

3w 4w
x(p; w) = ;
7 p1 7 p2

And the indirect utility function v(p; w) is given by

3 4
3w 4w 33 44 w7
v(p; w) = =
7 p1 7 p2 77 p31 p42

(b) u(x) = 3x1 + 4x2

26
In order to draw indi¤erence curves for this utility function, just consider some
…xed utility levels, and then solve for x2 . Note that the resulting expressions
are functions of x1 only, and importantly, they are linear in x1 . Intuitively,
this indicates that both goods can be substituted at the same rate, regardless
of the amount of goods the consumer owns of every good (we usually refer
to this cases by saying that goods are perfect substitutes). The M RSx1 ;x2
con…rms this intuition, since it is constant for any amount of x1 and x2 ,
@u(x)
@x1 3
M RSx1 ;x2 = @u(x)
=
4
@x2

The UMP in this case is given by

max u(x) = 3x1 + 4x2

x1 ;x2

subject to
p1 x1 + p2 x2 w
x 1 ; x2 0
As mentioned above, the budget constraint will be binding. The nonnega-
tivity constraints will not necessarily be binding however. Therefore, we face
a maximization problem with inequality constraints, and hence we must use
Kuhn-Tucker conditions. First, we set up the Kuhn-Tucker style Lagrangian
of this UMP

L = 3x1 + 4x2 1 [p1 x1 + p2 x2 w] + 2 x1 + 3 x2

The Kuhn-Tucker conditions are:

@L
= 3 1 p1 + 2 =0
@x1
@L
= 4 1 p2 + 3 =0
@x2
p1 x1 + p2 x2 w
x 1 ; x2 0
1 [p1 x1 + p2 x2 w] = 0, and
2 x1 = 0 and 1 x2 =0

27
Here I included all Kuhn-Tucker conditions you would need to specify in this
type of maximization problems with inequality constraints. However, some of
them can be eliminated, since we know that the budget constraint is binding,
and p1 x1 + p2 x2 = w. Additionally, solving for 1 in the …rst two expressions,
we have
3+ 2 4+ 3
= (3)
p1 p2
Now we are ready to consider the solutions that can arise in the four possible
cases in which the nonnegativity constraints can be met. These cases are

1. 2 = 0and 3 =0 i.e., x1 > 0and x2 >0

2. 2 = 0and 3 6= 0 i.e., x1 > 0and x2 =0
3. 2 6= 0and 3 =0 i.e., x1 = 0and x2 >0
4. 2 6= 0and 3 6= 0 i.e., x1 = 0and x2 =0

CASE 1. Interior solution, x1 > 0 and x2 > 0, i.e., 2 = 0 and 3 = 0. This

implies that equation (1) becomes
3 4 p1 3
= () =
p1 p2 p2 4
Hence, we can only have an interior solution when the price ratio is exactly
3
4
. In such case the budget line totally overlaps a indi¤erence curve with
the same slope M RS = 43 , and the consumer can choose any consumption
bundle on the budget line.
x2
IC3
IC1
IC2

BL (dashed)

x1

CASE 2. Lower corner solution, x1 > 0 and x2 = 0, i.e., 2 = 0 and 3 > 0.

Since wealth is fully spent on good 1, we have that x1 = pw1 . In order to
determine when is this solution optimal, we can use expression (1), and the
fact that 2 = 0, to obtain
3 4+ 3 p2
= () 3 =3 4
p1 p2 p1
and since 3 > 0 we get that this solution is optimal whenever,
p2 p1 3
3 =3 4 > 0 () <
p1 p2 4

28
 Graphically, this happens when the linear indi¤erence curves are steeper than
the budget line, i.e., M RS > pp12 .
x2
IC1 IC2 IC3 IC4

BL

x1

CASE 3. Upper corner solution, x1 = 0 and x2 > 0, i.e., 2 > 0 and 3 = 0.

Since wealth is fully spent on good 2, we have that x2 = pw2 . In order to
determine when is this solution optimal, we can use expression (1), and the
fact that 3 = 0, to obtain
3+ 2 4 p1
= () 2 =4 3
p1 p2 p2
and since 2 > 0 we get that this solution is optimal whenever,
p1 p1 3
2 =4 3 > 0 () >
p2 p2 4
Graphically, this happens when the linear indi¤erence curves are ‡atter than
the budget line, i.e., M RS < pp12 .
CASE 4. Corner solution, x1 = 0 and x2 = 0, i.e., 2 > 0 and 3 > 0. This
solution would violate the budget constraint for positive levels of wealth, i.e.,
p1 x1 + p2 x2 < w. However, the utility function is monotone, and therefore
such budget constraint should be binding (the consumer should spend his
entire wealth).
SUMMARY. Now we have determined the Walrasian demand correspondence
8
>
> 0; pw2 if pp12 > 34
<
x(p; w) = any (x1 ; x2 ) 2 R2+ s.t. p1 x1 + p2 x2 = w if pp12 = 34 ;
>
>
: w
; 0 if p1 < 3
p1 p2 4

And substituting this into the utility function, we get the indirect utility
function
4 pw2 if pp12 3
4
v(p; w) =
3 pw1 if pp12 < 34

17. Consider the utility function Yn

u(x) = xi i ;
i=1
where x denotes a vector of n di¤erent goods x 2 Rn+ , and i > 0. Check if u(x)
satis…es: (1) additivity, (2) homegeneity of degree k, and (3) homotheticity.

29
First, note that this utility function is just a generalization of the Cobb-Douglas
utility function to n goods. Indeed, for n = 2 goods
Yn
u(x) = xi i = x1 1 x2 2 , where 1 ; 2 > 0
i=1

The utility function is not additive since the marginal utility of additional amounts
of good k, uk (x), is
k
Yn
uk (x) = xi i
xk i=1

and therefore, depends on the amounts of other goods consumed, i.e., @u@xk (x)
j
6= 0.
[For a utility function to be additive, we should have obtained that the marginal
utility of good k is independent on other goods, i.e., @u@xk (x)
j
= 0]
– This can also be con…rmed for the case of n = 2 goods, u(x) = x1 1 x2 2 , where
the marginal utility of good 1 is
1 1
u1 (x) = 1 x1
1
x2 2 = x1 1 x2 2
x1
which depends on both x1 and x2 . For this utility function to be additive, we
should have obtained that u1 (x) does not depend on x2 . A similar argument
applies to u2 (x).
Let us now check its degree of homogeneity. Increasing its argument by a common
factor , we obtain
X
n X
n

Yn i Yn i
i
u( x) = ( xi ) = i=1 xi i
= i=1 u(x)
i=1 i=1

X
n
Therefore, utility function u(x) is homogeneous of degree i. (In the case
i=1
of a Cobb-Douglas utility function u(x) = x1 1 x2 2 , we have that the degree of
X
n=2
homogeneity is i = 1 + 2 ).
i=1
Finally, the utility function satis…es homotheticity. In order to show that, let us
…rst …nd the marginal rate of substitution between any two goods k 6= l
k xl
M RSl;k
l xk
As a consequence, the M RSl;k between two goods k 6= l only depends on the
ratio of these two goods that the consumer enjoys, but is independent on any
@M RS
other good h. Therefore @xh l;k 6= 0, for any three di¤erent goods k 6= l 6= h,
and the M RSl;k (the slope of the consumer’s indi¤erence curves) only depends on
the proportion of good l and k that the individual consumes. Graphically, this
implies that the slope of the indi¤erence curve coincide along any ray from the
origin.

30
18. [Transivity and completeness] Consider the set of alternatives X = N , i.e., X =
f1; 2; 3; :::g. Are the following relations transitive? Are they complete?

(a) "is at least as great as" ( ); (b) "is equal to"(=); (c) "is strictly greater than
(>); and (d) "is divisible by"(j):
(a) Transitive, complete.
We want to show that for all x; y andz 2 f1; 2; 3; :::g, ifx y and y z, then
x z
ex: x = 3; y = 2; z = 2 ) x z )transitive For any x and y 2 f1; 2; 3; :::g,
we have eitherx y or y x or both ) complete
(b) Transitive, incomplete.
For all x, y and z 2 f1; 2; 3; :::g , if x = y and y = z, then x = z ) transitive
Forx = 1 and y = 2, x 6= y and y 6= x ) incomplete
(c) Transitive, incomplete.
For all x, y and z 2 f1; 2; 3; :::g, if x > y and y > z, then x > z ) transitive
For x = 1 and y = 1, x y and y x ) incomplete
(d) Transitive, incomplete.
For all x, y and z 2 f1; 2; 3; :::g, if x=y = n; y=z = m, wheren and m 2
f1; 2; 3; :::g, then x=z = m n, wherem n 2 f1; 2; 3; :::g ) transitive Forx = 3
andy = 7, x - y and y - x ) incomplete.

19. [Homogeneous and quasiconcave] Suppose u(x1 ; x2 ) and v(x1 ; x2 ) are utility func-
tions.

(a) Prove that if u(x1 ; x2 ) and v(x1 ; x2 ) are both homogeneous of degree r, then
s(x1 ; x2 ) u(x1 ; x2 ) + v(x1 ; x2 ) is homogeneous of degree r.
Answer: Whenever it holds that tr u(x1 ; x2 ) = u(tx1 ; tx2 ) and tr v(x1 ; x2 ) =
v(tx1 ; tx2 )for all r > 0, it must also hold that tr s(x1 ; x2 ) u(tx1 ; tx2 )+v(tx1 ; tx2 )
tr u(x1 ; x2 ) + tr v(x1 ; x2 ).
(b) Prove that if u(x1 ; x2 ) and v(x1 ; x2 ) are quasiconcave, then m(x1 ; x2 ) u(x1 ; x2 )+
v(x1 ; x2 ) is also quasiconcave.
Answer: Forming a convex combination of the two functions u and v and com-
paring with m(xt ) satis…es the de…nition of quasiconcavity:
When u(xt ) minftu(x1 ) + (1 t)u(x2 )g
and v(xt ) minftv(x1 ) + (1 t)v(x2 )g
so m(xt ) minfu(xt ) + v(xt )g = [t(u(x1 ) + v(x1 )) + (1 t)(u(x2 ) + v(x2 ))].

20. Check if the utility function u(x1 ; x2 ) = x1 x2 where ; > 0 satis…es the following
properties:

(a) Local non-satiation;

(b) Decreasing marginal utility for both goods 1 and 2;
(c) Quasiconcavity.
(d) Homogeneous

31
(e) Homothetic.

(a) Local non-satiation (LNS). When working with a di¤erentiable utility function we
can check LNS by checking that the marginal utility from additional amounts of
goods 1 and 2 are non-negative,

@u(x1 ; x2 ) 1 x1 x2
= x1 x2 = > 0 if and only if >0
@x1 x1

@u(x1 ; x2 ) 1 x1 x2
= x 1 x2 = > 0 if and only if >0
@x2 x2
(b) Decreasing marginal utility. We need to show that the marginal utilities we found
above are nonincreasing. That is,

@ 2 u(x1 ; x2 ) 2
= ( 1)x1 x2 0 if and only if 1
@x21

@ 2 u(x1 ; x2 ) 2
= ( 1)x1 x2 0 if and only if 1
@x22
(c) Quasiconcavity. Let us …rst simplify the expression of the utility function by
applying a monotonic transformation on u(x1 ; x2 ), since any monotonic transfor-
mation of a utility function maintains the same preference ordering. That is, if a
utility function is quasiconcave, any monotonic transformation of it will also be
quasiconcave. In this case, we apply

z1 = ln u( ) = ln x1 + ln x2

We now need to …nd the bordered Hessian matrix, then …nd its determinant, if
this determinant is greater than (or equal to) zero, then this utility function is
quasiconcave; otherwise it is quasiconvex (see Appendix M.D. in MWG, or Simon
and Blume’s Mathematics for Economists). The bordered Hessian matrix is

0 z1 z2 0 x1 x2
z1 z11 z12 = x1 x21
0
z2 z21 z22 x
0 x22
2

( + )
and the determinant of this matrix is x21 x22
0 for all x1 ; x2 2 R+ , which
implies that u( ) is quasiconcave.
(d) Homogeneous. It is easy to show that this Cobb-Douglas utility function repre-
sents preferences which are homogeneous of degree + in x1 and x2 ,
+ +
u( x1 ; x2 ) = ( x1 ) ( x2 ) = x1 x2 = u(x1 ; x2 )

32
(e) Homothetic preferences. First note that
1
x1 x2
M RS1;2 = 1
x 1 x2

scaling up all goods by a factor t, we obtain

1 1+ 1 1
(tx1 ) (tx2 ) t x1 x2 x1 x2
M RS1;2 = 1
= + 1 1
= 1
(tx1 ) (tx2 ) t x 1 x2 x 1 x2

which shows that the M RS1;2 does not change when we scale up all goods by a
common factor t, i.e., the slope of the indi¤erence cuve at a given point is not
changed.
A few remarks on Homothetic preferences. When preferences are homothetic,
the MRS between the two goods is just a function of the consumption ratio
between the goods, xx12 , but it does not depend on the absolute amounts
consumed. As a consequence, if we double the amount of both goods, the
MRS does not change.
Recall that this type of preferences induce wealth expansion paths that are
straight lines from the origin, i.e., if we double the wealth level of the in-
dividual, then his wealth expansion path (the line connecting his demanded
bundles for the initial and the new wealth level) are straight lines. A corollary
of this property is that the demand function obtained from homothetic pref-
erences must have an income-elasticity equal to 1, i.e., when the consumer’s
income is increased by 1%, the amount he purchases of any good k must
increase by 1% as well.
Examples of preference relations that are homothetic: Cobb-Douglas (as in
the previous example), preferences over goods that are considered substitutes,
preferences over goods that are considered complements and CES preferences.
In contrast, quasilinear preference relations are not homothetic.

33
 Chapter 2 - Demand Theory
End-of-Chapter Exercises - Answer key

1. A consumer with convex and monotonic preferences consumes two goods, x1 and x2 .
1
Her utility function is u (x1 ; x2 ) = x1 x22 .

(a) Find the Marshallian demand function for goods 1 and 2.

Since we are asked about the conditions under which the consumer demands
non-negative amounts of the goods in the next part of this exercise, we can
start assuming interior solutions. Using the shortcut M RS = pp12 , and plug-
ging the (binding) budget constraint, we have
2 w (1 2 ) w
x1 (p; w) = , and x2 (p; w) =
p1 p2
(b) What restrictions on would ensure that the consumer demands non-negative
amounts of the goods (interior solution)?
In order to have x1 (p; w) 0, we just need 0, since w > 0 by de…nition.
On the other hand, x2 (p; w) 0 is satis…ed if and only if 1 2 0, that is,
if 1=2. Combining both conditions, we can guarantee that the consumer
demands non-negative amounts of both goods when 2 [0; 1=2].

2. [Checking Walras’Law]. State conditions under which the following demand func-
tions satisfy Walras’ Law. [Note: K is the number of goods, and p i represents the
vector of all prices di¤erent from pi ]
" #
P
K
(a) xi (pi ; p i ; w) = p1i i + iw + ij pj
j=1

First, recall that for Walras’ Law to be satis…ed, we need that wealth is
entirely spent on consumption of all K goods. That is, for strictly positive
vectors of prices (p >> 0) and positive wealth levels (w > 0), p x = w, or
P
K
alternatively, pi xi = w. Summing over all K goods,
j=1

X
K X
K X
K X
K X
K
xi (pi ; p i ; w) pi = i+ iw + ij pj
i=1 i=1 i=1 i=1 j=1

Rearranging, we can see that Walras’Law is satis…ed if

X
K X
K X
K X
K X
K
xi (pi ; p i ; w) pi = i+ ij pj + w i
i=1 i=1 i=1 j=1 i=1
| {z } | {z }
=0 =1

P
K P
K P
K P
K
That is, Walras’Law hold if i + ij pj = 0, and if i = 1.
i=1 i=1j=1 i=1

1
 w
P
K
(b) xi = pi i+ i log w + ij log pj
i=1
Rearranging, " #
X
K
pi xi = w i + i log w + ij log pj
i=1
Summing over all K goods,
" K #
X K X X
K X
K
pi xi = w i + i log w + ij log pj
i=1 i=1 i=1 i=1

P
K P
K P
K
Hence, Walras’Law is satis…ed if i + i log w + ij log pj = 1
i=1 i=1 i=1
1 2
(c) xi = pi
[ i + iw + iw ]
Rearranging, and summing over all K goods,
X
K X
K X
K X
K
2
pi xi = i+w i+w i
i=1 i=1 i=1 i=1

Hence, Walras’Law is satis…ed if

X
K X
K X
K X
K
2
pi xi = i+w i+w i+
i=1
|i=1 {z i=1
} |i=1
{z }
=0 =1

3. [Substitution e¤ects] Consider an individual with utlity function u(x1 ; x2 ; x3 ) for

three goods, where the cross-price elasticities are null, i.e., "ij = 0 for any two goods i
@x2
s23
and j. Show that the ratio of substitution e¤ects s13
is equal to @w
@x1 .
@w

@h2 (p;u0 )
Substitution e¤ect s23 = @p3
can be found by using the Slutsky equation,

@h2 (p; u0 ) @x2 (p; w) @x2 (p; w) @x2 (p; w)

= + x3 (p; w) = x3 (p; w)
@p3 @p3 @w @w
where the second equality uses the fact that "23 = 0, implying that the derivative
@x2 (p;w) (p;u0 )
@p3
= 0. Similarly, substitution e¤ect s13 = @h1@p 3
can be found by using
the Slutsky equation,
@h1 (p; u0 ) @x1 (p; w) @x1 (p; w) @x1 (p; w)
= + x3 (p; w) = x3 (p; w)
@p3 @p3 @w @w
where the second equality also uses the property that cross-price elasticities are
null, i.e., "13 = 0 and thus @x1@p(p;w)
3
= 0. Hence, the ratio of substitution e¤ects ss23
13
reduces to
@x2 (p;w) @x2 (p;w)
s23 @w
x3 (p; w)
= @x1 (p;w) = @x1@w
(p;w)
:
s13 x3 (p; w) @w @w

2
4. [Income and substitution e¤ect with a quasi-linear utility function] Michael’s
preferences pover soda (good x) and other goods (composite good, y) are given by
U (x; y) = 2 x + y. His income is $10. Assume that the price of the composite good
is 1, py = $1.

(a) What is Michael’s optimal basket when the price of sodas is px = $0:5? Label it
basket A.
At an interior optimum,
p 1
M Ux px x 1
= () = px () x =
M Uy py 1 (px )2

When the price of soda is px = 0:5 Michael buys x = 4 sodas. We can …nd
the number of units of good y (the composite good) by using the budget line
px x + py y = I, hence 0:5 4 + y = 10, which implies y = 8 units of the
composite good.

(b) What is his optimal basket when the price of sodas drops to px = $0:2? Label it
basket C.
Similarly to the previous part,
1 1
x= 2 = = 25
(px ) (0:2)2

and for the case of the composite good, 0:2 25 + y = 10, which implies y = 5

3
 units of the composite good.

(c) The “decomposition”budget line, BLd , is tangent to the initial utility level. Label
this tangency point “basket B” and …nd the amount of goods x and y that this
consumer will choose at basket B.
First, we know that the all points in the “decomposition” budget line, BLd ,
must give the same utility
p level that basket A. At basket A, x = 4 and y = 8
which
p implies U1 = 2 4 + 8 = 12. Therefore, basket B must satisfy that
2 x + y = 12. Secondly, we know that the slope of the decomposition budget
line at B must be the same as the slope of the …nal budget line at C.
p 1
M Ux px x 0:2
= () =
M Uy py 1 1
which implies that x = 25 and y = 2. Basket B is then (x; y) = (25; 2).

(d) Find the income and substitution e¤ects of a decrease in the price of sodas.
The substitution e¤ect is the change in the quantity of sodas purchased as the
consumer moves from the initial basket A (where he consumes 4 cans of sodas)
to the decomposition basket B (where he consumes 25). The substitution
e¤ect on sodas is therefore 25 4 = 21 sodas.

4
 The income e¤ect is the change in the quantity of sodas purchased as the
consumer moves from the decomposition basket B to the …nal basket C. Be-
cause he consumes the same number of sodas at B and C, the income e¤ect
is zero.
(e) Represent in a graph basket A, basket C, and the decomposition budget line BLd ,
basket B, and the income and substitution e¤ects.
See …gure in part (c) of the exercise.
(f) Calculate the compensating variation of this price decrease. Explain your reason-
ing with a graph, di¤erent from the one you did above, and determine which the
exact size of the compensating variation is.
The compensating variation is the di¤erence between the consumer’s income
($10) and the income he would need to purchase the decomposition basket B
at the new price of sodas ($0.20). Since at basket B he buys 25 cans of sodas
and 2 units of the composite good, she would need

$0:2 25 + $1 2 = $7

The consumer would be willing to have his income reduced from $10 to $7 if
the price of sodas falls from $0.50 to $0.20. Thus, the compensating variation
is $3.

(g) Calculate the equivalent variation of the price decrease. Explain your reasoning
with a graph, di¤erent from the one you did above, and determine which the exact
size of the equivalent variation is.
The equivalent variation is the measure of how much additional income a
consumer would need before a price reduction to be as well o¤ as after the
price decrease. Hence, we …rst need to determine the location of basket E
(see …gure). First, we know that basket E lies on the …nal
p indi¤erence curve
U2 , which has a value of 15. Therefore, at basket E, 2 x + y = 15. Second,
we know that at basket E, the slope of the …nal indi¤erence curve must equal
the slope of the initial budget line,
p 1
M Ux px x 0:5
= () =
M Uy py 1 1

which reduces
p to x = 4 units. When we substitute this value of x into the
equation 2 x + y = 15 we …nd that y = 11. Then, at basket E the consumer
purchases 4 cans of soda and 11 units of the composite good. To purchase
basket E at the initial price of $0.50, he should need an income of

$0:5 4 + $1 11 = $13

The equivalent variation is the di¤erence between this amount ($13) and her
income ($10), or $3. Thus, the compensating variation and the equivalent

5
 variation are equal

Note: If there is no income e¤ect, the compensating variation and the equiv-
alent variation of a price change will be equal to each other, and also equal
to the change in the cosumer surplus, i.e., change in consumer surplus is the
area behind the Walrasian demand curve x = (p1)2 between prices px = $0:5
x
Z0:5 h i0:5
and px = $0:2, that is (p1)2 = 1
px
= 3. However, if income e¤ects are
x 0:2
0:2
strictly positive, the compensating and the equivalent variation will give us
di¤erent measures of the monetary value that a consumer would assign to a
reduction in the price of a good. Moreover, these measures will not generally
coincide with the change in the consumer surplus.

5. [Gi¤en and inferior goods] Consider a consumer in a two-good world.

(a) Show graphically that a Gi¤en good must be inferior. Use the Slutsky equation
to help your discussion.
The Walrasian demand of a Gi¤en good decreases when its price decreases,
@xl (p;w)
@pl
> 0, i.e., price and demand move in the same direction. Moreover, we
know that by de…nition of the Slutsky equation, the substitution e¤ect must
be nonpositive, sll 0, where

@xl (p; w) @xl (p; w)

sll = + xl (p; w)
@pl @w

Hence, since @xl@p

(p;w)
l
> 0 (because good l is a Gi¤en good), and sll 0
(because the substitution e¤ect is always nonpositive), then the income e¤ect
must be negative, @xl@w(p;w)
< 0. And we know that, by de…nition, those goods
for which the income e¤ect is negative (i.e., their Walrasian demand decreases

6
 as the consumer income increases), are denoted as inferior goods.

(b) Show graphically that a inferior good does not need to be a Gi¤en good. Use the
Slutsky equation to help your discussion.
Let us show it by providing an example. We must …nd an example of a good
for which the total e¤ect is negative, @xl@p
(p;w)
l
< 0 (i.e., the good is not a Gi¤en
good), but for which the income e¤ect is negative, @xl@w
(p;w)
< 0 (inferior good).
That is, we just need to check if the Slutsky equation can hold (if sll 0),

@xl (p; w) @xl (p; w)

sll = + xl (p; w)
@pl @w
@xl (p;w) @xl (p;w)
when both @pl
< 0 and @w
< 0. Clearly, sll 0 can easily be satis…ed
@xl (p;w) @xl (p;w)
when both @pl
< 0 and @w
< 0.

6. [Relationship between WARP and CLD] The …gure below illustrates the change
in budget line Bp;w to Bp0 ;w (pivoting outwards), as a result of the decrease in the
price of good 1, maintaining both the price of good 2 and wealth constant. Then, the
consumer receives a wealth compensation (changing his wealth level from w to w0 ) that

7
 guarantees he can still a¤ord his initial consumption bundle, x(p; w). Show that, if
Walrasian demand satis…es the Weak Axiom of Revealed Preference (WARP), then

(a) x(p0 ; w0 ) cannot lie on segment A, but it must lie on segment B.

Let us start checking that x(p0 ; w0 ) cannot lie on segment A. Applying WARP,
…rst note that both x(p; w) and x(p0 ; w0 ) are a¤ordable under initial prices
and wealth, Bp;w . However, in the second step of WARP, we see that x(p; w)
is a¤ordable under Bp0 ;w0 , which constitutes a violation of WARP. Hence,
x(p0 ; w0 ) cannot lie on segment A.
Let us now check if x(p0 ; w0 ) can lie on segment B. In the …rst step of WARP,
we see that x(p; w) is a¤ordable under initial prices and wealth, Bp;w , but
x(p0 ; w0 ) is not. Hence, the premise of WARP does not hold, and as a conse-
quence WARP is not violated if x(p0 ; w0 ) lies on segment B.
(b) What conclusions can you infer from your results in part (a) about the slope of
the Walrasian demand function? What about the slope of the Hicksian demand
function?
From the previous result, we can conclude that x(p0 ; w0 ) must contain more
of good 1 (note that graphically, bundle x(p0 ; w0 ) lies to the right hand side of
bundle x(p; w)). Then, a decrease in the price of good 1 (when we appropri-
ately compensate for wealth e¤ects) leads to an increase in the consumption
of good 1. This is the Compensated Law of Demand (CLD), and it implies
that the Hicksian (compensated) demand curve must be negatively sloped (a
decrease in prices leads to an increase in the consumption of that good).
However, from this result we cannot guarantee that the uncompensated law of
demand (ULD) is satis…ed. Therefore, we cannot conclude that the Walrasian
demand curve is also negatively sloped. It can be negatively or positively
sloped.

7. [Slutsky matrix] Consider a setting with three goods (L = 3) and a consumer whose
consumption set is in R3 ..Suppose that his demand function x(p; w) is
p2 p1 w
x1 (p; w) = ; x2 (p; w) = ; x3 (p; w) =
p3 p3 p3

8
(a) Show that x(p; w) = (x1 (p; w); x2 (p; w); x3 (p; w)) is homogeneous of degree zero
in prices and wealth, (p; w).
p2 p2
x1 ( p; w) = p3
= p3
= x1 (p; w)
p1 p1
x2 ( p; w) = p3
= p3
= x2 (p; w)
w w
x3 ( p; w) = p3
= p3
= x3 (p; w)
(b) Show that x(p; w) satis…es Walras’law.
Recall Walras’ Law: for strictly positive vectors of prices (p >> 0) and
P
3
positive wealth levels (w > 0), p x = w, or alternatively, pi xi = w. Hence,
i=1

X
3
pi xi = p1 x1 (p; w) + p2 x2 (p; w) + p3 x3 (p; w) =
i=1
p2 p1 w
= p1 + p2 + p3
p3 p3 p3
p1 p2 p2 p1 + p3 w
= =w
p3
Therefore, Walras’Law is satis…ed.
(c) Show that x(p; w) violates the weak axiom of revealed preference (WARP).
Let us use a counterexample.

w = 1 p = (1; 1; 1) x (p; w) = (1; 1; 1)

w0 = 2 p0 = (1; 1; 2) x (p; w) = 21 ; 21 ; 1

We know that WARP is satis…ed if for any pair of prices and wealth (p; w)
and (p0 ; w0 ),

if p x (p0 ; w0 ) w and x (p0 ; w0 ) 6= x (p; w) then p0 x (p; w) > w0

In our example:
32
1
p x (p0 ; w0 ) = [1; 1; 1] 4 1 5 = 1 1+1=1 w (where w = 1)
1

Additionally, x (p; w) 6= x (p0 ; w0 ). Finally,

2 1 3
5= 1 1
2
p0 x (p; w) = [1; 1; 2] 4 1
2
+2=2
2 2
1

But p0 x (p; w) = w0 , and hence p0 x (p; w) > w0 is not satis…ed. Therefore,

WARP is violated.
(d) Show that v S (p; w) v = 0 for all v 2 R3 .

9
Let us …rst recall the Slutsky matrix:
2 3
s11 s12 ::: s1N
6 s21 s22 ::: s2N 7
6 7
S(p; w) = 6 .. .. . . .. 7
4 . . . . 5
sN 1 sN 2 ::: sN N

where every component sik is de…ned as

@xi (p; w) @xi (p; w)

sik = + xk (p; w) (Slutsky equation)
@pk @w
Hence, Slutsky equation informs us about what is the change in the demand
for good i after the price of good k has varied, given that we change the
consumer’s wealth so that he can a¤ord her initial consumption bundle at
the new prices. Let us now …nd each of the components of the Slutsky matrix
for this particular exercise.

@x1 (p; w) @x1 (p; w)

s11 = + x1 (p; w) = 0 + 0 = 0
@p1 @w
@x1 (p; w) @x1 (p; w) 1 1
s12 = + x2 (p; w) = +0=
@p2 @w p3 p3
@x1 (p; w) @x1 (p; w) p2 p2
s13 = + x3 (p; w) = 2
+0=
@p3 @w p3 p23
@x2 (p; w) @x2 (p; w) 1 1
s21 = + x1 (p; w) = +0=
@p1 @w p3 p3
@x2 (p; w) @x2 (p; w)
s22 = + x2 (p; w) = 0 + 0 = 0
@p2 @w
@x2 (p; w) @x2 (p; w) p1 p1
s23 = + x3 (p; w) = 2
+0=
@p3 @w p3 p23
@x3 (p; w) @x3 (p; w) 1 p2 p2
s31 = + x1 (p; w) = 0 + = 2
@p1 @w p3 p3 p3
@x3 (p; w) @x3 (p; w) 1 p1 p1
s32 = + x2 (p; w) = 0 + =
@p2 @w p3 p3 p23
@x3 (p; w) @x3 (p; w) w 1 w
s33 = + x3 (p; w) = 2
+ =0
@p3 @w p3 p3 p3

Therefore, the Slutsky matrix is

2 1 p2
3
0 p3 p23
6 1
0 p1 7
S(p; w) = 4 p3 p23 5
p2 p1
p23 p23
0

Now, we need to compute v S (p; w) v = 0 for all v 2 R3 . First, we will …nd

10
 v S (p; w):

1 p2
v S (p; w) = v1 0 v2 v3 ;
(1 3) (3 3) p3 p23
| {z }
(1 3)

1 p1 p2 p1
v1 + v2 0 v3 ; v1 + v2 + v3 0
p3 p23 p23 p23
Simplifying,
v2 p2 v1 p1 p2 p1
= v3 ; v3 ; v 1 + v 2
p3 p23 p3 p23 p23 p23
Now we can compute v S (p; w) v:
2 3
v1
v2 p2 v1 p1 p2 p1 4 v2 5 =
v3 ; v3 ; v1 2 + v 2 2
p3 p23 p3 2
p3 p3 p3
v3
v1 v2 v1 v3 p2 v2 v1 v2 v3 p1 v3 v1 p2 v3 v2 p1
= + + =0
p3 p23 p3 p23 p23 p23
Therefore, v S (p; w) v = 0 for all v 2 R3 , and we state that this demand
function has a negative semide…nite Slutsky matrix.
Remark: Note that this exercise constitutes a perfect example to show that
the result of Proposition 2.F.2 in MWG cannot be stated in the opposite
direction: a demand function with a negative semide…nite Slutsky matrix
does not need to satisfy WARP. That is,

v S (p; w) v = 0 for all v 2 R3 ; x (p; w) satis…es WARP

8. [Slutsky matrix] An econometrician estimates a Slutsky matrix for three goods at

prices (p1 ; p2 ; p3 ) = (1; 2; 6) and income w = 27. Unfortunately, due to a computer
failure he is only able to retrieve some of the matrix. The partial Slutsky matrix that
he is able to retrieve is 2 3
10 a b
S(p; w) = 4 c 4 d 5
3 e f
His colleague, the microeconomist, assures him, however, that the rest of the entries
of the Slutsky matrix can be calculated under the assumption that consumers are
rational, their Walrasian demand x(p,w) is di¤erentiable, homogeneous of degree zero,
and satis…es Walras’law.

(a) Under these hypothesis, which are the remaining entries?

By symmetry, we know that b = 3, a = c and d = e (recall the symmety can
be applied on L > 2 goods because of rational preferences). We also know
that the Slutsky matrix must satisfy p S(p; w) = 0. Hence,

10p1 + cp2 + 3p3 = 0 () 10 1 + c 2 + 3 6 = 0

11
 And thus, c = 4, and by symmtry, a = c = 4. Similarly,
ap1 4p2 + ep3 = 0 () 4 1+4 2+e 6=0
Therefore, e = 2, and by symmetry d = 2. Finally,
bp1 + dp2 + f p3 = 0 () 3 1 + 2 2 + f 6 = 0
7
Which implies f = 6
:
(b) Does the resulting matrix posess all the properties of a substitution matrix? [Hint:
you must check that (1) p S(p; w) = 0, (2) negative semide…nite v S(p; w) v 0,
and (3) symmetric.
First we check p S(p; w) = 0:
2
3
10 4 3
p S(p; w) = p1 p2 p3 4 4 4 2 5
7
3 2 6
= [ 10 8 + 18; 4 8 + 12; 3 + 4 7]
= [0; 0; 0]
Note that we multiply a (1 3) vector of prices times a (3 3) Slutsky matrix,
and hence we must obtain a (1 3) vector of zeros.
Secondly, we check that S(p; w) is negative semide…nite. Note that: (1) the
principal minors of order one are all negative (elements in the main diagonal);
(2) the principal minors of order two have all positive determinants; and
(3) the single principal minor of order 3 (the entire matrix) has a negative
determinant. [For more on useful matrix properties we will use, see appendix
M.D. in MWG, or some of the examples at section 16.2 of Simon and Blume’s
Mathematics for Economists].
Third, the matrix is clearly symmetric.

9. [Quasilinear Preferences]. Consider a quasilinear preference relation in R2 , and

assume that it is quasilinear with respect to good 1. As we saw in class, all indi¤erence
curves are then parallel displacements to each other along the axis of good 1, and good
1 is desirable. Show that the wealth e¤ects for good 2 are eliminated, i.e., all additional
income is spent in good 1.

Let us …rst recall some properties of quasilinear preferences:

x2
IC1 IC2 IC3 IC4

y y+αe1

x x+αe1

x1

12
 Good 1 is desirable. That is, for any bundle x 2 ( 1; +1) RL+ 1 , we have that

x + e1 x, where > 0 and e1 = (1; 0; :::; 0)

All indi¤erence curves are parallel displacements to each other along good 1 axis.
That is, for all bundles x; y 2 ( 1; +1) RL+ 1 , we have that if the consumer
is indi¤erent between bundles x and y, x y, then he is also indi¤erent between
bundles
x + e1 y + e1 for > 0 and e1 = (1; 0; :::; 0)
We need to show that the wealth e¤ects for all the remaining goods (other than
good 1) are eliminated. That is, all additional wealth is spent on good 1. More
formally, for every price vector p 2 RL++ , for wealth w 0, for every consumption
bundle x 2 ( 1; +1) RL+ 1 , for all > 0 and e1 = (1; 0; :::; 0), we need to
show that, if x = x (p; w) is the consumer’s demand vector at prices p and wealth
w, then his demand vector when prices are unchanged but his wealth increases in
, x(p; w + ), is given by x + e1 . That is,

x + e1 = x(p; w + )

Proof. For any given price vector p 2 RL++ , for wealth w 0, for every consump-
tion bundle x 2 ( 1; +1) RL+ 1 , we have that bundle x is a¤ordable at (p; w),
p x w, and bundle x + e1 is a¤ordable at (p; w + ), i.e., p (x + e1 ) w + .
Let us now take a new consumption bundle y 2 ( 1; +1) RL+ 1 , such that it
is also a¤ordable at (p; w + ),

p y w+

Hence, consumption bundle y e1 is a¤ordable at (p; w);

p (y e1 ) w

But we previously stated that bundle x was also a¤ordable at (p; w). Therefore,

x % z for all z 2 Bp;w

Then, x % y e1 given that y e1 2 Bp;w , which implies

x + e1 % y for all y 2 Bp;w+

Hence, all additional wealth is spent on good 1. As a consequence, x + e1 =

x(p; w + ). Thus, quasilinear preference relations eliminate the wealth e¤ects in
all the remaining goods, i.e., for all l = 2; 3; :::; L goods, we have that

xl (p; w) = xl (p; w + )

10. [ULD] Show that if the demand function xi (p; w) satis…es the uncompensated law of
demand (ULD), then Dp xi (p; w) is negative semide…nite. [Hint: Totally di¤erentiate
xi (p; w) with respect to p and w, and use the uncompensated law of demand to cancel
some of the terms. Justify].

13
 If xi (p; w) satis…es the uncompensated law of demand (ULD), then we have that
for any change in prices dp, Dp xi (p; w) is negative semide…nite, that is we need
to show
dp Dp xi (p; w)dp 0
First, totally di¤erentiating the Walrasian demand function xi (p; w) we obtain

dxi (p; w) = Dp xi (p; w)dp + Dw xi (p; w)dw

But given that we are assuming the ULD property, we know that any change
in prices is not going to be compensated by a change in the wealth level of the
individual (in order to maintain him on the original indi¤erence curve). Hence,
dw = 0, and we can rewrite the above expression as

dxi (p; w) = Dp xi (p; w)dp

Premultiplying both sides of this expression by dp, we have

dp dxi (p; w) = dp Dp xi (p; w)dp

And since by the ULD property we know that dp dxi (p; w) 0, we can conclude
that dp Dp xi (p; w)dp 0, which allows us to conclude that Dp xi (p; w) is negative
semide…nite.

11. [CES Utility function] Consider the following utility function with constant elasticity
1
of substitution (CES): u(x1 ; x2 ) = [ 1 x1 + 2 x2 ] . Show that:

(a) When = 1, indi¤erence curves are linear (goods 1 and 2 are perfect substitutes).
If = 1, then u(x1 ; x2 ) = 1 x1 + 2 x2 . Then, the utility function becomes
linear (goods are perfect substitutes), and the equation of a given indi¤erence
curve would be given by
U U
x2 = x1
2 2
(b) When ! 0, the utility function comes to represent the same preferences as the
(generalized) Cobb-Douglas utility function, u(x1 ; x2 ) = x1 1 x2 2 , where 1 + 2 =
1.
1
e(x) = ln u(x) =
Let u ln [ 1 x1 + 2 x2 ]. Then,

ln [ 1 x1 + 2 x2 ] 0
e(x) = lim
lim u =
!0 !0 0
Hence, we need to use l’Hopital’s rule, as follows,
@ ln[ 1 x1 + 2 x2 ]
@ @ ln [ 1 x1 + 2 x2 ]
lim @
= lim =
!0
@
!0 @

1
lim [ 1 ln(x1 )x1 + 2 ln(x2 )x2 ] =
!0 1 x1 + 2 x2

14
 ln(x1 ) + 2 ln(x2 )
1 [x 1 x 2 ]
= 1 2
1+ 2 1+ 2
Recall that so far we have been dealing with lim u e(x), and in fact we are
!0
e(x) = ln u(x), we have that lim u(x) =
interested in lim u(x). So, given that u
!0 !0
x1 1 x2 2 :
(c) When ! 1, indi¤ference curves become “right angles,” that is, this utility
function becomes in the limit a Leontie¤ utility function given by u(x1 ; x2 ) =
min fx1 ; x2 g
1
We need to show that if x2 x1 , then lim [ 1 x1 + 2 x2 ] = min fx1 ; x2 g =
! 1
x1 :
Suppose, without loss of generality, that x2 x1 , then x2 x1 since 0 < <
1. Hence, 2 x2 2 x1 given that 2 0. Thus, adding 1 x1 on both sides
of the inequality,
1 x1 + 2 x2 1 x1 + 2 x1

() 1 x1 + 2 x2 ( 1 + 2 ) x1

and since 0 < < 1.

1 1
[ 1 x1 + 2 x2 ] [( 1 + 2 ) x1 ] (1)
| {z } | {z }
B C

If x1 0 and x2 0, then
1 x1 1 x1 + 2 x2

and since 0 < < 1.

1 1
[ 1 x1 ] [ 1 x1 + 2 x2 ]

and from expression (1), we have that

1 1 1
[ 1 x1 ] [ 1 x1 + 2 x2 ] [( 1 + 2 ) x1 ] (2)
| {z } | {z } | {z }
A B C

We can use the “Squeezing Theorem”to obtain the limits of terms A and C,
in order to obtain the limit of term B, which lies between A and C. Let us
…rst …nd the limit of term A,
1
lim [ 1 x1 ] = x1
! 1

Let us now …nd the limit of term C,

1
lim [( 1 + 2 ) x1 ] = x1
! 1

Hence, since the limits of both terms A and C coincide (both of them are x1 ),
the limit of term B must also be x1 . That is,
1
lim [ 1 x1 + 2 x2 ] = x1
! 1
1
which is exaclty what we needed to show: if x2 x1 , then lim [ 1 x1 + 2 x2 ] =
! 1
min fx1 ; x2 g = x1 .

15
12. Suppose that demand and utility functions are di¤erentiable, and preferences are ho-
mothetic. Show that:
1
(a) Dp xi (p; wi ) = Si (p; wi ) x (p; wi )
wi i
xi (p; wi )T
From the Slustky equation we know that

Si (p; w) = Dp xi (p; wi ) + Dw xi (p; wi )xi (p; wi )T

Futhermore, given homothetic preferences, we know that the Walrasian de-

mand is going to be de…ned by xi (p; wi ) = i wi (just a proportion of the
individual’s wealth, no consideration about prices!). Hence, solving for i we
obtain i = xi (p;w
w
i)
. Additionally, the e¤ect of an increase in the individual’s
wealth is given by Dw xi (p; wi ) = i . Combining both results we have that
xi (p; wi )
Dw xi (p; wi ) =
w
Plugging this result back into the Slustky equation,
xi (p; wi )
Si (p; w) = Dp xi (p; wi ) + xi (p; wi )T
w
and rearranging terms
xi (p; wi )
Dp xi (p; wi ) = Si (p; w) xi (p; wi )T
w
(b) dp Si (p; wi ) dp = 0 when dp = p (when the price change dp is proportional to
the initial price level).
From part (a) we know that
xi (p; wi )
Si (p; w) = Dp xi (p; wi ) + xi (p; wi )T
w
Pre- and post-multiplying every term by dp = p, we have
xi (p; wi )
p Si (p; w) p = p Dp xi (p; wi ) p+ p xi (p; wi )T p
w
and canceling out the ’s on both sides, we obtain the following expression,
where we will especially focus on terms 1 and 2
xi (p; wi )
p Si (p; w) p = p Dp xi (p; wi ) p + p xi (p; wi )T p
| {z } | {zw }
Term 1
Term 2

Term 1: From Cournot aggregation, we know that

p Dp xi (p; wi ) + xi (p; wi )T = 0 () p Dp xi (p; wi ) = xi (p; wi )T

Term 2: From Engel aggregation, we have that

p Dw xi (p; wi ) = 1

16
 which in our case is p xi (p;w
w
i)
= 1. Substituting the values of Terms 1 and 2
into the above Slustky equation,

p Si (p; w) p = xi (p; wi )T p + 1 xi (p; wi )T p

Hence, p Si (p; w) p = 0. We can then conclude that when price changes dp

are proportional to the initial price level, dp = p, then dp Si (p; wi ) dp = 0.
(c) dp xi (p; wi ) > 0 when dp = p (when the price change dp is proportional to the
initial price level).
By Walras’law, we know that

p xi (p; wi ) = w
dp
If dp is proportional to the price level, dp = p, then p = . Substituting
for this value of p in the above expression
dp
xi (p; wi ) = w for any >0

() dp xi (p; wi ) = w
And since both > 0 and w > 0, then w > 0, and thus dp xi (p; wi ) > 0
when change in prices are proportional to the initial price level, i.e., dp = p.

13. [Gorman form for vi (p; wi )] Prove that if the indirect utility function vi (p; wi ) admits
the Gorman form (i.e., it can be represented as a linear combination of the individuals’
wealth, as follows),
vi (p; wi ) = ai (p) + b(p)wi
then all consumers exhibit parallel, straight wealth expansion paths at any price vector
p. [Hint: Use Roy’s identity].

To know the form of the wealth expansion paths, we need to know how the
Walrasian demand xi (p; w) responds to changes in wealth levels. Since we have
the consumer’s indirect utility function vi (p; wi ), we can use Roy’s identity in
order to …nd his Walrasian demand xi (p; w)
@vi (p;wi )
@p a0i (p) + b0 (p)wi
xi (p; w) = @vi (p;wi )
=
b(p)
@w

Now, we want to know how the consumer’s Walrasian demand xi (p; w) responds
to changes in wealth levels.
@xi (p; w) b0 (p)b(p) 0 b0 (p)
= =
@w [b(p)]2 b(p)
0
Hence, the wealth expansion paths are given by the expression bb(p)
(p)
. In order to
check that wealth expansion paths are straight, we just have to check that their
curvature does not change in w. Observing the above expression, we clearly see

17
 that wealth expansion paths have a slope which is constant in w. Additionally,
note that the slope of the wealth expansion paths coincide across customers (no
subscripts in the slope). Hence, the only di¤erence in wealth expansion paths
a0i (p)
across consumers must be in the origin of this line, given by - b(p) , which varies
across consumers.

(a) Show also that, if preferences admit Gorman-form indirect utility functions, vi (p; wi ) =
ai (p) + b(p)wi , with the same b(p) for all individuals, then preferences admit ex-
penditure function, ei (p; ui ), of the form

ei (p; ui ) = c(p)ui + di (p)

By the Duality theorem, we have

ei (p; vi (p; wi )) = wi

Isolating wi from the above expression of the Gorman-form indirect utility

function vi (p; wi ) = ai (p) + b(p)wi , we obtain
vi (p; wi ) ai (p)
wi =
b(p)
Hence,
vi (p; wi ) ai (p)
ei (p; vi (p; wi )) =
b(p)
And rearranging terms,
1 ai (p)
ei (p; vi (p; wi )) = vi (p; wi )
b(p) b(p)
or more compactly,

ei (p; vi (p; wi )) = c(p) vi (p; wi ) di (p)

where c(p) = b(p) 1

for all i, and di (p) = ab(p)
i (p)
. Therefore, we obtained that, if
preferences admit Gorman-form indirect utility function with the same b(p)
for all individuals, then preferences admit expenditure functions of the form
ei (p; vi (p; wi )) = c(p) vi (p; wi ) di (p).

14. [Concavity of the support function] We know that the, given a non-empty, closed
set K, its support function, K (p), is de…ned by

K (p) = inf fp xg for all x 2 K and p 2 RL

Hence, the value of this support function, K , satis…es K p x for all x in set K.
Given this de…nition, prove the concavity of the support function. That is, show that

K ( p + (1 ) p0 ) > K (p) + (1 ) K (p0 )

for every p; p0 2 RL and for any 2 [0; 1].

18
 First, from the de…nition of the support function we know that, for a given price
vector p, and for every element x in the set K,

K (p) p x, then K (p) p x, for all 2 [0; 1] (3)

And similarly for any other price vector p0 ,
0
K (p ) p0 x, then K (p
0
) p0 x, for all 2 [0; 1] (4)
Similarly,

K (p) p x, then (1 ) K (p) (1 ) p x, for all 2 [0; 1] (5)

0
K (p ) p0 x, then (1 ) K (p
0
) (1 ) p0 x, for all 2 [0; 1] (6)
Summing up expressions (1) and (4), we have

K (p) + (1 ) K (p) p x + (1 ) p0 x
which can be simpli…ed to

K (p) + (1 ) K (p) [ p + (1 ) p0 ] x
and by the de…nition of the support function, we know that K( p + (1 ) p0 ) =
[ p + (1 ) p0 ] x. Therefore,

K (p) + (1 ) K (p) K( p + (1 ) p0 )
and hence the support function K (p) is concave.

15. Consider a function F (p; w) representing the inverse of the indirect utility function
v(p; w), that is
1
F (p; w)
v(p; w)
where the indirect utility function satis…es the usual properties, and v(p; w) 6= 0.

(a) Use function F (p; w) to …nd the Walrasian demand of good i, i.e., xi (p; w).
Using Roy’s identity, we have
@v(p;w)
@pj
xj (p; w) = @v(p;w)
:
@w

Using the identify from this question, we can rewrite the right-hand side of
this expression as
1
@ F (p;w)
@pj
xj (p; w) = 1
@ F (p;w)
@w
rearranging,
@F (p;w)
@pj @F (p;w)
2
[F (p;w)] @pj
xj (p; w) = @F (p;w)
= @F (p;w)
@w
2 @w
[F (p;w)]

19
 (b) Let sj denote the share of income that the consumer spends on good j. Show
that sj can be expressed as follows
@F (p;w) pj
@pj F (p;w)
sj = @F (p;w) w
@w F (p;w)

pj xj (p;w)
The consumer’s budget share of good j is sj = w
. Using the expression
of xj (p; w) we found in part (a), we obtain
@F (p;w)
pj xj (p; w) pj @pj pj
sj = = xj (p; w) @F (p;w)
w w w
@w

F (p;w)
and multiplying this ratio by F (p;w)
, we have
@F (p;w) pj
@pj F (p;w)
sj = @F (p;w) w
@w F (p;w)

as we needed to show.

16. Measuring welfare changes through the expenditure function]. A consumer

1=2 1=2
has a utility function u (x1 ; x2 ) = x1 x2 , where good x1 is the consumption of alco-
holic beverages, and x2 is her consumption of all other goods. The price of alcohol is
p > 0, and the price of all other goods is normalized to 1.

(a) Set the consumer’s expenditure minimization problem. Find …rst order conditions,
and …nd his optimal consumption of x1 and x2 .
The consumer’s minimization problem is
h i
1=2 1=2
min p1 x1 + p2 x2 + U x1 x2
x1 ;x2

And the …rst order conditions are

1 1=2 1=2
p1 = x x2
2 1
1 1=2 1=2
p2 = x x
2 1 2
p2 x2
x1 =
p1
Substituting in the constraint, we have
1=2
p2 x2
U= (x2 )1=2
p1
And solving for x2 in this expression, we …nd the Hicksian demand for good
2,
1=2
p1
x2 = U
p2

20
 And similarly for x1 , we …nd the Hicksian demand for good 1,
1=2
p2
x1 = U
p1

(b) Substituting your results from part (a) into your objective function, …nd the
expenditure function e(p1 ; p2 ; U ) for this consumer.
Substituting x2 and x1 into p1 x1 + p2 x2 ,
1=2 1=2
p2 p1
e(p1 ; p2 ; U ) = p1 U + p2 U
p1 p2
= 2U (p1 p2 )1=2

For convenience, we can denote p1 = p be the price of alcohol, and p2 = 1 be

the price of all other commodities. Hence, the expenditure function can be
rewritten as
e(p; U ) = 2U (p)1=2
(c) Let us now consider a proposal to reduce the price of alcohol from p = $2 to
p = $1 per unit. If the new utility enjoyed by the consumer after the price change
is U = 100,
1. What is his minimum expenditure in order to reach U = 100 when p = $2?
And when p = $1? [Hint: Use the expression of the expenditure function
e(p1 ; p2 ; U ) you found in part (b)]
His minimum expenditure in order to reach U = 100 when p = $2 is

e($2; 100) = 2 100 (2)1=2 = 282:84

And when p = $1 is

e($1; 100) = 2 100 (1)1=2 = 200

2. What is then the maximum amount that this consumer would be willing to
pay for this price reduction?
The maximum amount that this consumer would be willing to pay for
this price reduction is the di¤erence in the minimum expenditure he need
to maintain the same utility level (U=100). This is the EV. That is,

e($2; 100) e($1; 100) = 82:84

17. [Measuring welfare changes when preferences are quasilinear] Show that the
compensating and the equivalent variation coincide when the utility function is qua-
silinear with respect to the …rst good (and we …x p1 = 1). [Hint: use the de…nitions
of the compensating and equivalent variations in terms of the expenditure function
(not the hicksian demand). In addition, recall that if u(x) is quasilinear, then we can
express it as u(x) = x1 + (x 1 ), and rearranging x1 = u(x) (x 1 ) where x 1
represents all the reamining goods, l = 2; 3; :::; L.]

21
 From the de…nition of the compensating and the equivalent variation, we know
that
CV p0 ; p1 ; w = e p1 ; u1 e p1 ; u0
EV p0 ; p1 ; w = e p0 ; u1 e p0 ; u0
Moreover, we know that if u(x) is quasilinear, then we can express it as
u(x) = x1 + (x 1 ) () x1 = u(x) (x 1 )
where x 1 represents all the reamining goods, l = 2; 3; :::; L. Therefore, the
expenditure function becomes
X
L X
L
e (p; u) = pi xi = p1 x1 + pk xk = x1 + p 1 x 1
|{z}
i=1 k=2
$1 | {z }
p 1 x 1

where we use the fact that p1 = 1. Substituting x1 from the above expression of
the quasilinear utility function, we have
e (p; u) = u(x) (x (p )) + p x
| {z 1 1 } 1 1
x1

Using this expression for the expenditure function into the above de…nition of the
compensating variation,
CV p0 ; p1 ; w = e p1 ; u1 e p1 ; u0
= u1 (x) (x 1 (p 1 )) + p 1 x 1
0
u (x) (x 1 (p 1 )) + p 1 x 1
1 0
= u (x) u (x)
And similarly for the de…nition of the equivalent variation,
CV p0 ; p1 ; w = e p0 ; u1 e p0 ; u0
= u1 (x) (x 1 (p 1 )) + p 1 x 1
0
u (x) (x 1 (p 1 )) + p 1 x 1
1 0
= u (x) u (x)
Therefore, for quasilinear utility functions, the compensating and the equivalent
variation give us the same measures of the monetary value, u1 (x) u0 (x), that
a consumer would assign to a reduction (or increase) in the price of a good.
Recall that this is because quasilinear utility functions do not provoke any income
e¤ects, and when income e¤ects are zero, the equivalent variation coincides with
the compensating variation, and they both coincide with the area variation.

18. Consider a function F (p; w) representing the inverse of the indirect utility function
v(p; w), that is
1
F (p; w)
v(p; w)
where the indirect utility function satis…es the usual properties, and v(p; w) 6= 0.

22
 (a) Use function F (p; w) to …nd the Walrasian demand of good i, i.e., xi (p; w).
Using Roy’s identity, we have
@v(p;w)
@pj
xj (p; w) = @v(p;w)
:
@w

Using the identify from this question, we can rewrite the right-hand side of
this expression as
1
@ F (p;w)
@pj
xj (p; w) = 1
@ F (p;w)
@w
rearranging,
@F (p;w)
@pj @F (p;w)
2
[F (p;w)] @pj
xj (p; w) = @F (p;w)
= @F (p;w)
@w
2 @w
[F (p;w)]

(b) Let sj denote the share of income that the consumer spends on good j. Show
that sj can be expressed as follows
@F (p;w) pj
@pj F (p;w)
sj = @F (p;w) w
@w F (p;w)

pj xj (p;w)
The consumer’s budget share of good j is sj = w
. Using the expression
of xj (p; w) we found in part (a), we obtain
@F (p;w)
pj xj (p; w) pj @pj pj
sj = = xj (p; w) @F (p;w)
w w w
@w
F (p;w)
and multiplying this ratio by F (p;w)
, we have
@F (p;w) pj
@pj F (p;w)
sj = @F (p;w) w
@w F (p;w)

as we needed to show.

19. Consider a consumer with the following expenditure function

e(p; u0 ) = g(p) + u0 f (p)
where functions g(p) and f (p) depend on the price vector p alone. Show that the
income elasticity of any good i converges to one when the consumer’s wealth level
tends to in…nity, i.e., lim "xi ;w = 1.
w!1

The income-elasticity of good i is given by "xi ;w = @x i w

@w xi
. We hence need to …rst
…nd the Walrasian demand associated to this expenditure function. In order to
…nd it, we need to do it in two steps: …rst, use the expenditure function e(p; u0 )
to obtain the indirect utility function v(p; w); and second, use the indirect utility
function v(p; w) to obtain the Walrasian demand of good i, xi (p; w).

23
 First step. From e(p; u0 ) to v(p; w). Recall that e(p; v(p; w)) = w, which in this
exercise entails
g(p) + [v(p; w) f (p)] = w;
w g(p)
and solving for v(p; w), we …nd v(p; w) = f (p)
.
Second step. From v(p; w) to xi (p; w). Since we have the consumer’s indirect
utility function, we can now use Roy’s identity to obtain
@v(p;w) gi (p)f (p) fi (p)[w g(p)]
@pi f (p)2
xi (p; w) = @v(p;w)
= 1 =
@w f (p)
fi (p)
= gi (p) + [w g(p)]
f (p)
@g(p) @f (p)
where, for compactness, we denoted gi (p) @pi
and similarly fi (p) @pi
.
Once we have obtained the Walrasian demand xi (p; w), we can …nd the income
elasticity of good i, as follows

@xi w fi (p) w
"xi ;w = = fi (p)
=
@w xi f (p) gi (p) + [w g(p)]
f (p)
f i(p)w
=
gi (p)f (p) + fi (p)w fi (p)g(p)

and taking the limit of this ratio when w ! 1, we obtain that lim "xi ;w = 1.
w!1

20. Let u : RL++ ! R be a di¤erentiable, strictly concave, locally nonsatiated utility

function. We restrict ourselves in what follows to a price-wealth vectors (p; w) : RL+1
++
for which the utility maximization problem

max u(x; p; w)
x 0

has a solution.

(a) Brie‡y argue that the solution to the above UMP, denoted by x~(p; w) , is unique
and de…ned by the …rst order conditions.
Answer: Uniqueness follows from strict quasiconcavity (and strict quasiconcavity
is implied from strict concavity). The FOCs characterize the solution because the
domain of the utility function u( ) is an open set.
(b) Verbally de…ne the marginal utility of wealth, to be denoted ~ (p; w) . State and
prove its relation to the Lagrange multiplier
Answer: The marginal utility of wealth is the partial derivative of the indirect
utility function with respect to wealth, and it coincides with the Lagrangian mul-
tiplier of the UMP, as can be proved by a standard application of the Envelope
Theorem.

24
 (c) We now partition the set of L goods into three subsets: f1; :::; L1 g, fL1 +1; :::; L2 g,
and fL2 + 1; :::; Lg . A consumption vector x = (x1 ; :::; xL ) is similarly par-
titioned into the three subvectors x1 = (x1 ; :::; xL1 ), x2 = (xL1 +1 ; :::; xL2 ) and
x3 = (xL2 +1 ; :::; xL3 ), thus allowing us to express the consumption vector x as
the triplet x = (x1 ; x2 ; x3 ). The price vector p is similarly partitioned into
p1 = (p1 ; :::; pL1 ), p2 = (pL1 +1 ; :::; pL2 ) and p3 = (pL2 +1 ; :::; pL3 ), thus allowing
for a similar representation p = (p1 ; p2 ; p3 ). We assume that the utility function
u is separable in the form

u(x) = u1 (x1 ) + u2 (x2 ) + u3 (x3 )

(for J = 1; 2; 3) uJ is di¤erentiable, strictly concave and locally nonsatiated and

adopts the following interpretation: our household, comprised of Adam and Eve,
consumes three types of goods: x1 is the bundle of goods consumed in Paradise
1, x2 is the bundle consumed in Paradise 2, and x3 are all other goods consumed
by these individuals. Our household has solved the UMP at some point in the
past, but they have forgotten some details. Next answer the following questions,
trying to solve a marital dispute between Adam and Eve about how to determine
their optimal consumption bundles.
1. Before engaging in their two-paradise vacation, Adam says: “All we have to
remember are our planned expenditures in Paradise 1,w1 , and in Paradise 2,
w2 . Then, when we are in Paradise 1, we observe its prices p1 and solve

max
1
u1 (x1 ) subject to p1 x1 w1 , (1.1)
x

with solution denoted x~~1 (p1 ; w1 ), and when we are in Paradise 2, we observe
its prices p2 and solve a similar problem

max
2
u2 (x2 ) subject to p2 x2 w2 ,
x

with solution denoted x~~2 (p2 ; w2 ).” Then, Adam claims, our solutions will
coincide with what we had originally planned, i.e., x~~J pJ ; wJ = x~J pJ ; wJ ,
for all J = 1; 2. Is he right? Explain.
Answer: Yes, Adam is right, as long as he correctly remembers w1 as w1 =
p1 x~1 (p; w) and w2 as w2 = p2 x~2 (p; w). Consider Paradise 1. Suppose,
by contradiction, that Adam is wrong so that bundle x1 where x1 6= x~1 (p; w)
solves the UMP
max u1 (x1 ) subject to p1 x1 w1
x1

Then x satis…es u (x1 )

1 1
u1 (~
x1 (p; w)), because x~1 (p; w) clearly satis…es
the budget constraint p1 x1 w1 , i.e., u(x1 ; x~2 (p; w)) x(p; w)), while
u(~
1 2
(x ; x~ (p; w)) satis…es the constraint p x w , contradicting the fact that
x~(p; w) is the unique solution to UMP.
2. The wife interjects: “What are you talking about!1 We just need to remember
the marginal utility of wealth, . Then, when we arrive at Paradise 1, we can
1
It looks like they already ate the apple and started to be disrespectful to one another.

25
 simply solve the unconstrained maximization problem
max
1
u1 x 1 p 1 x1
x

with solution x^1 (p1 ; ), and when we are in Paradise 2 we solve the analogous
problem
max
2
u2 x 2 p 2 x2
x
2 2
with solution x^ (p ; ).”Doing that, she claims, we will also get the equality
x^J (pJ ; ) = x~J (pJ ; ), for all J = 1; 2. Is Eve right? Explain.
Answer: Yes, Eve is right too, as long as she correctly remembers as being
~ (p; w). Consider Paradise 1: when she solves the system of L1 equations in
L1 unknowns
@u1 (x1 )
= pj , for every good j = 1; :::; L1 ; (1.2)
@xj
for = ~ (p; w) given, she must …nd the solution x1 = x~1 (p; w), because
x~1 (p; w) satis…es problem (1.2) for the same pj , for all goods j = 1; :::; L1 ,
that is
ru1 (x1 ) = ru1 (~
x1 (p; w))
Note that x1 6= x~1 (p; w) is impossible given the strict concavity of u1 . In
other words, if we let rx1 = x~1 (p; w) x1 denote the di¤erence between these
consumption vectors, and such a di¤erence is not zero, rx1 6= 0, then the
strict concavity of u1 implies that ru1 (x1 )4x1 > u1 (~
x1 (p; w)) u1 (x1 ).and
ru1 (~
x1 (p; w)) 4x1 > u1 (x1 ) u1 (~
x1 (p; w))
which yields the contradiction 0 > 0 when ru1 (x1 ) = ru1 (~
x1 (p; w)).
Intuitively, rewritting (1.2), we obtain
@u1 (x1 )
@xj
= , for every good j = 1; :::; L1 ;
pj
implying that the wife’s approach equates the marginal utility per dollar spent
in good j to the marginal utility of wealth, .
(d) Let there be a large number of Paradises. Are the arguments of Adam and Eve
in part (c) of the exercise extensive to this context?
Answer: Yes, their arguments are applicable to this environment as well, as long
as the utility function is separable Paradise by Paradise, as it did in the case of
two Paradises. But with multiple Paradises Adam’s approach requires remember-
ing several expenditures per island, wj , whereas Eve’s approach only requires to
remember one number, namely, the marginal utility of wealth, .
(e) Check Adam’s and Eve’s approaches for Paradise 1 in the following parametric
example.
X
L X
L
u (x) = i ln xi ; where i = 1; i > 0; for all goods i
i=1 i=1

26
Answer: The Walrasian demands associated with this Cobb-Douglas utility func-
tion are
w
x~1 (p; w) = i , for all goods i = 1; :::; L (1.3)
pi
In addition, we can obtain the lagrange multiplier from the …rst-order equality
@u
@xi
= pi , i.e. xii = pi ; which by (1.3) yields

i
w = pi
i pi

, or
^1 (p; w) = 1
w
PL w
We can furthermore compute the indirect utility function v(p; w) = i=1 i ln i pi .
1
Once we obtained v(p; w), we can …nd ^ (p; w) as follows

XL
1X
L
^ 1 (p; w) = @v = i
1 i
= i =
1
:
w
@w i=1 i pi pi w i=1 w

which indeed coincides with the expression we found above using a more direct
approach.

27
 Chapter 3 - Applications of Demand Theory
End-of-Chapter Exercises - Answer key

1. [Taxes versus Subsidies] The trucking industry has recently complained about the
high price of gas. Consider a representative trucker with utility function

u(q1 ; q2 ) = ln q1 + q2 ,

where q1 denotes gallons of gas and q2 is a numeraire representing all other goods. The
price of q2 is therefore normalized to one, p2 = 1, while the price of gas is p1 (1 + t),
where t 2 [0; 1] represents a speci…c tax per gallon of gas. The income of this trucker
is given by m > 0.

a) Find the Walrasian demand for q1 and q2 , denoting them as q1W and q2W , and distinguish
the case in which m > 1 and that when m 1.

The representative trucker’s utility maximization problem is

max u(q1 ; q2 ) = ln q1 + q2
q1 ;q2
subject to p1 (1 + t)q1 + q2 = m

Plugging the constraint into the objective function, the UMP can be simpli…ed to

max u(q1 ; q2 ) = ln q1 + [m p1 (1 + t)q1 ]

q1

Taking …rst-order conditions with respect to q1 we obtain the Walrasian demand

for good 1
1
q1W (p; m; t) =
p1 (1 + t)
and plugging this result into the budget constraint, we …nd the Walrasian demand
for good 2
1
q2W (p; m; t) = m p1 (1 + t) =m 1
p1 (1 + t)
Note that there is a corner solution when m 1 in which only good 1 is consumed.
We can hence summarize the Walrasian demand correspondence as follows
8
< 1
; m 1 if m > 1, and
W W p1 (1+t)
q1 (p; m; t); q2 (p; m; t) =
: m
; 0 if m 1
p1 (1+t)

Note that in the interior solution the demand for gas (good 1) does not depend
on income m, i.e., its income e¤ect is zero.

b) Find the associated indirect utility function, v q1W ; q2W .

1
 Plugging the above results into the trucker’s utility function, we obtain
8
< ln 1
+ m 1 if m > 1, and
p1 (1+t)
v1 (p; m; t) =
: m
ln p1 (1+t) if m 1

After months of lobbying and violent demonstrations from the union of truckers, the gov-
ernment is considering implementing either of the following policies: (1) reduce the tax on
gas, from t to t0 = t ; or (2) maintain the tax at t but give a subsidy of S dollars to the
trucker equal to the tax revenue collected by the tax on gas.

c) Let us …rst consider that the trucker’s income satis…es m > 1, i.e., the trucker is
relatively rich. Find the trucker’s indirect utility function if the government implements
the …rst policy, v I q1W ; q2W , and if the government implements the second policy,
v II q1W ; q2W . Under which conditions does the trucker prefer the …rst policy?

Since m > 1 then we are at the interior solution. If the …rst policy is implemented,
reducing the tax rate to t0 = t , the trucker’s indirect utility function becomes

1
v I q1W ; q2W = ln +m 1
p1 (1 + t )

If, instead, the second policy is implemented, then the tax rate is unaltered, but
tax revenue
1 t
R = tp1 q1W = tp1 =
p1 (1 + t) 1+t
is given to the trucker in the form of a subsidy, yielding an indirect utility function
of
subsidy
z }| {
1 t
v II q1W ; q2W = ln +m 1+
p1 (1 + t) 1+t
since we know that any increase in his wealth is only used to increase the amount
of q2 being consumed (the amount of good q1 does not increase in income).
We can now compare v I q1W ; q2W and v II q1W ; q2W , obtaining that

1+t t
v I q1W ; q2W > v II q1W ; q2W () ln >
1+t 1+t

which implies that if the tax reduction is su¢ ciently high, the trucker prefers
the …rst to the second policy. Indeed, solving for , we …nd that > , where
t
(1 + t) 1 e 1+t .
The following …gure depicts cuto¤ , for di¤erent values of t. Hence, the region of
(t; )-pairs above cuto¤ describe settings in which the trucker prefers the …rst

2
 policy, while for (t; )-pairs below cuto¤ the trucker prefers the second policy.

As an example, if the initial tax on gas is t = 0:5, then we obtain that ln 0:50:5 >
0:5
1+0:5
holds if the tax reduction, , satis…es > 0:14. If the initial tax is higher,
t = 0:8, then we obtain that ln 0:80:8 > 1+0:80:8
holds only if > 0:29. Intuitively,
as the tax rate becomes higher, the trucker needs a larger tax reduction, , in
order to make the …rst policy preferable.

d) Let us now consider that the trucker’s income satis…es m 1, i.e., the trucker is rela-
tively poor. Find the trucker’s indirect utility function if the government implements
the …rst policy, v I q1W ; q2W , and if the government implements the second policy,
v II q1W ; q2W . Under which conditions does the trucker prefer the …rst policy?

m
Since m 1 then we are at the corner solution, which is q1W ; q2W = p1 (1+t) ;0 .
If the …rst policy is implemented, reducing the tax rate to t0 = t , the trucker’s
indirect utility function becomes
m
v I q1W ; q2W = ln :
p1 (1 + t )
If, instead, the second policy is implemented, then the tax rate is unaltered, but
m t
tax revenue R = tp1 p1 (1+t) = 1+t m is given to the trucker in the form of a
t
subsidy, yielding a total income of m + 1+t m = 1+2t
1+t
m. Note that m 1 does
1+2t
not guarantee that 1+t m 1, so with this new income, the trucker is still in a
corner solution if 1+2t
1+t
m 1, or moves to an interior solution if 1+2t
1+t
m > 1, i.e.,
1+2t
where 1+t m > 1 > m. Let us analyze each case separately:
1+2t
CASE 1: When 1+t
m 1, Walrasian demands are given by
m(1 + 2t)
q1W (p; m; t); q2W (p; m; t) = ;0
p1 (1 + t)2
yielding an indirect utility function of
1st m(1 + 2t)
v II q1W ; q2W = ln
p1 (1 + t)2

3
We can now compare the indirect utility function of the trucker in the …rst and sec-
st
ond policy, i.e., v I q1W ; q2W and v II 1 q1W ; q2W , which implies that the trucker
prefers the …rst policy if
m m(1 + 2t) (1 + t)2
ln > ln () ln >0
p1 (1 + t ) p1 (1 + t)2 (1 + t )(1 + 2t)
You were not required to provide a more explicit solution for the minimal cuto¤
of that satis…es this condition but, in case you are interested, the value of
that solves the above inequality is > e , where
t(1 + t)
e
1 + 2t
The following …gure depicts cuto¤ e for di¤erent values of t. As indicated in
part (c) of the exercise, (t; )-pairs above cuto¤ e describe settings in which the
trucker prefers the …rst policy (a tax reduction), while for (t; )-pairs below cuto¤
e the trucker prefers the second policy (a subsidy).

1+2t
CASE 2: When 1+t
m > 1, Walrasian demands are now both interior, and given
by
1 1 + 2t
q1W (p; m; t); q2W (p; m; t) = ; m 1
p1 (1 + t) 1 + t
yielding an indirect utility function of
2nd m 1 + 2t
v II q1W ; q2W = ln + m 1
p1 (1 + t) 1+t
We can now compare the indirect utility function of the trucker in the …rst and sec-
nd
ond policy, i.e., v I q1W ; q2W and v II 2 q1W ; q2W , which implies that the trucker
prefers the …rst policy if
m m 1 + 2t
ln > ln + m 1
1+t p1 (1 + t) 1+t

4
 m(1 + t) 1 + 2t
() ln > m 1
1+t 1+t
which holds if is su¢ ciently high.
You were not required to provide a more explicit solution for the minimal cuto¤
of that satis…es this condition but, in case you are interested, the value of
that solves the above inequality is > b , where
t
b 1+t (1 + t)me1+m( 1+t 2)

where, in the case that the trucker’s income is m = 0:8, yields the following …gure.
The intuition behind the …gure is similar to that in part (c) above.

As an example, if the initial tax rate is t = 0:5 and the trucker’s income is m = 0:8,
then the above condition holds as long as > 0:37. Similarly as in part (c) of the
exercise, if the tax rate is higher, t = 0:8, then the above condition holds as long
as > 0:56, suggesting a similar intuition.

2. [Welfare changes] Consider an individual with utility function u(q1 ; q2 ) = q12 + q2

1, where q1 (q2 ) denotes the units of good 1 (good 2, respectively) this individual
consumes. His income level is denoted by w 2 R+ , and prices are both strictly positive,
i.e., p = (p1 ; p2 ) 2 R2++ .

(a) Determine this individual’s Walrasian demand, and his associated indirect utility
function.
Before starting the exercise, note that this individual’s indi¤erence curves are
strictly concave in the fq1 ; q2 g space. Indeed, solving this individual’s utility
function for q2 , we obtain q2 (q1 ) = q12 + u + 1. Hence,
q20 (q1 ) = 2q1 < 0 and q200 (q1 ) = 2<0
i.e., indi¤erence curves have a negative slope that becomes more negative as
q1 increases (bowed away from the origin). Furthermore,
p its vertical intercept
is (0; u + 1) and its horizontal intercept is ( u + 1; 0).

5
 Walrasian demands. Such a concavity makes us anticipate that the con-
sumer’s Walrasian demands will be corner points, i.e., pw1 ; 0 and 0; pw2 .
If we compare the utility levels associated to each of these bundles, we ob-
tain that the consumer prefers pw1 ; 0 to 0; pw2 if and only if u pw1 ; 0 >
2 p21
u 0; pw2 , which implies wp2 1> w
p2
1, i.e., w > p2
. We can hence express
1
the Walrasian demands as
8
< w p21
p1
;0 if w > p2
, and
q1W (p; w); q2W (p; w) = p21
(1)
: 0; pw2 if w p2

p2
Intuitively, if w > p12 good 2 is expensive (relative to the price of good 1
and the individual’s wealth level), leading the consumer to buy only good 1.
Otherwise, the consumer only buys good 2. Given (1), the indirect utility
function becomes
( 2 p2
w
p21
1 if w > p12 , and
v1 (p; w) = w p2
(2)
p2
1 if w p12

(b) Determine this individual’s Hicksian demand, and his associated expenditure func-
tion.
Because of concavity on the consumer’s indi¤erence curves, hicksiand de-
mands must also be in corner points. Hence, whenpq1 = 0 we must have that
q2 = u + 1; while if q2 = 0 we must have p that q1 = u + 1. We therefore have
two corner solutions: (0; u + 1) and u + 1; 0 . Since they both lie on the
same indi¤erence curve, the consumer chooses the less expensive of the two
corner solutions. In particular,
p the cost of buying bundle (0; u + 1) is larger
than that of bundle u + 1; 0 if
p p2
p1 0 + p2 (u + 1) > p1 u + 1 + p2 0 () u > 12 1
p2
We hence obtain the Hicksian demands
( p p21
u + 1; 0 if u > p22
1, and
(h1 (p; u); h2 (p; u)) = p21 (3)
(0; u + 1) if u p22
1

As a consequence, the expenditure function is

( p p2
p1 u + 1 if u > p21 1, and
e(p; u) = 2
p21 (4)
p2 (u + 1) if u p22
1

Consider now that this individual’s income level is w = 6, and the initial vector of market
prices is p0 = (4; 3). If both prices increase by 50%, determine:

(c) The compensating variation of this price increase. Interpret.

6
 We now have all the elements to provide a monetary evaluation of the welfare
change associated to the price increase. The initial price vector and wealth
p2 2
(p0 ; w) = (4; 3; 6) satisfy condition w > p12 , i.e., 6 > 43 = 5:33. Therefore,
Walrasian demands in this setting are q1W (p; w) = pw1 = 64 = 32 and q2W (p; w) = 0.
p2 2
When prices increase to p1 = 6; 92 , condition w < p12 holds, i.e., 6 < 9=26
= 8,
implying that Walrasian demands after the price increase are q1W (p; w) = 0 and
q2W (p; w) = pw2 = 9=2 6
= 43 . We hence need to evaluate the welfare change that
arises from “jumping” from the corner bundle 23 ; 0 to the other corner bundle
0; 43 , as depicted in the following …gure.

q2

Final equil. after

the price change:
 4 
 0, 3 
 

u0
u1

q1
Initial equil. before  3 
the price change:  2 , 0 

Walrasian demand before and after the price increase.

p
level is e(p0 ; u0 ) = p01 u0 + 1, where
The expenditure function at the initial price q
m2
u0 = p21
1 = 45 . Therefore, e(p0 ; 54 ) = 4 5
+ 1 = 6 = w. Similarly, at the
4
q
…nal price level, the expenditure function becomes e(p1 ; 54 ) = 6 54 + 1 = 9.1 As
a consequence, the compensating variation (CV) is
5 5
CV = e p1 ; e p0 ; =9 6 = $3
4 4
Intuitively, the income level should be increased in $3, to w + CV = $9, in order
to guarantee that the consumer can still reach his initial utility level of u0 at the
new (higher) price level, i.e., we need to increase his income level by 50%.
1 p2 p2
Note that the expression of e(p; u) depends on whether condition u > p12 1 or u 1
p22
1 holds.
2
This condition is, however, una¤ected by a proportional increase in the price of both goods. Therefore,
p the
cheapest bundle that reaches utility u0 is 23 ; 0 before and after the price increase. Indeed, p11 u0 + 1 =
q
6 54 + 1 < 92 54 + 1 = p12 u0 + 1 , implying that 32 ; 0 is still cheaper than 0; 43 at the new prices.

7
(d) The change in consumer surplus associated to this price increase. Interpret.

The change in CS, also referred as Area Variation (AV ) in MWG, as a consequence
of the “jump”from one corner solution to another, is given by
p02 =3 p11 =6
z }| { z }| {
Z p
3 2 Z 6 Z 9=2
w w
AV = dp1 + p 0dp1 + dp2
4 p1 3 2 3 p2
p
9 2
= 6 ln ' 2:78
8
Explanation of three elements in AV. (As usual, we need to increase the price of
each good separately.) We …rst increase p1 in the …rst two terms, and p2 in the
third term. Here is a more detailed explanation (The following …gures illustrate
the Walrasian demand for good 1 and 2):

– Given an income level w = 6 and a price for good 2 of p2 = 3, condition

p2 p2 p p
w > p21 holds if 6 > 31 , i.e., for all p1 < 3 2 ' 4:24. Hence if p1 < 3 2, the
Walrasian demand for good 1 is q1W (p; w) = pw1 .
p
– If, instead, p1 3 2 holds, the Walrasian demand for good 1 collapses to
zero, i.e., q1W (p; w) = 0.
– Once p1 is at the …nal price level p1 = $6, we can increase p2 . In particular,
p2
we use condition w ? p21 , to determine that 6 > p62 for all p2 2 3; 92 , implying
2

that the demand for good 2 is q2W (p; w) = pw2 .

3. [Consumer Theory and Welfare] Consider a representative consumer in an econ-

omy with J goods, j = 1; 2; :::; J. Since we are mainly interested in this individual’s
consumption of goods 1 and 2, we group all the remaining goods j = 3; 4; :::; J as good
zero. The price of good zero is p0 = 1 (the numeraire). The prices of goods 1 and 2
are p1 and p2 , and income is m > 0. This consumer’s preferences are represented by
utility function
1 1
u(q1 ; q2 ; q0 ) = q14 q24 + q0

8
(a) Find the Walrasian demands and the associated indirect utility function. Invert
the indirect utility function to obtain the expenditure function.
UMP: In order to solve this problem, we use a standard argument for addi-
tively separate utility functions: de…nte eR (p; m) p1 q1W + p2 q2W to be the
amount of money spent on purchasing the Walrasian demand of goods 1 and
2. Then, the pair q1W ; q2W must solve the auxiliary problem
1 1
max q14 q24 (1)
q1 ;q2

subject to p1 q1 + p2 q2 = eR (p; m)
Taking FOCs and solving for q1 and q2 , we obtain
eR eR
q1W (p; eR ) = and q2W (p; eR ) =
2p1 2p2
which do not depend on the overall income of the individual, m, but on the
amount of income he spends on good 1 and 2 alone, eR . Expressions q1W (p; eR )
and q2W (p; eR ) yield an associated utility level of
1=4 1=4 1=2
R R 1 1 eR
v (p; e ) =
p1 p2 2
which can be interpreted as the indirect utility function of the auxiliary max-
imization problem (1).
Given these results for goods 1 and 2, we can analyze good 0. In particular,
the Walrasian demand for good 0, q0W , and the amount of income spent on
goods 1 and 2, eR (p; m), must solve

max v R (p; eR ) + q0
q0 ;eR

subject to eR (p; m) + q0 = m
Furthermore, since q0 = m eR (p; m), the above program can be simpli…ed
to the following maximization problem (with only one argument):

max g(eR ; p) = v R (p; eR ) + m eR (p; m)

eR

Taking FOCs with respect to eR , we obtain

1=4 1=4
1 1
@g(eR ; p) p1 p2
= p p 1 (2)
@eR 2 2 eR
and SOCs
1=4 1=4
1 1
2 R
@ g(e ; p) p1 p2
= p <0 (3)
@eR 2 4 2 (eR )3=2
showing that the objective function g(eR ; p) is strictly concave.

9
 Therefore, from the FOCs in (2), the value of eR (p; m) that maximizes
g(eR ; p) is e (p) = 8pp11 p2 . This implies that:
– When m > p1 , we have that
8 p1 p2

p1 p1
8 p1 p2 1 8 p1 p2 1
q1W = = p 3 and q2W = = p
2p1 16 p1 p2 2p2 16 p1 p32
for goods 1 and 2, and the rest of income, q0W = m 8pp11 p2 , on good 0.
(Interior solution).
– By contrast, when m p1 , no income is spent on good 0, q0W = 0,
8 p1 p2
but only on goods 1 and 2, that is
m m
q1W = and q2W =
2p1 2p2
at a corner solution.
Hence, the Walrasiand demand correspondence can be summarized as
8
>
< p1 ; p1 ; m 8pp1 p if m > 8pp11 p2 , and
W W W 16 p31 p2 16 p1 p32 1 2
q1 ; q2 ; q0 =
>
: m
; m ; 0 if m 8pp11 p2 .
2p1 2p2

Note that, at the interior solution, the Walrasian demands of goods 1 and 2
do not depend on income, implying that these goods do not exhibit income
e¤ects, since all income e¤ect is absorbed by the numeraire good.
From the above demands, it is easy to obtain the associated indirect utility
function 8
< m + 8pp11 p2 if m > 8pp11 p2 , and
v(p; m) = m2
1=4
: if m 8pp11 p2 .
4p1 p2

Note that in order to obtain the expenditure function e(p; u), we just need
to invert the indirect utility function v(p; m), yielding
(
u 8pp11 p2 if u > 4pp11 p2 , and
e(p; u) = p
2u2 p1 p2 if u 4pp11 p2 .

(b) Consider that the price vector increases from p0 = (p01 ; p02 ) = (1; 1) to p1 =
(p11 ; p12 ) = (2; 1), i.e., only the price of good 1 doubles. Let us next use the
equivalent variation (EV) to evaluate the loss in welfare that the consumer su¤ers
from the increase in the price of good 1.
1. What is the EV when income satis…es m > 18 , i.e., the consumer is relatively
rich?
In this case, the consumer is at the interior solution both before and after
the price change. In particular,
1 1
u0 = v(p0 ; m) = m + and u1 = v(p1 ; m) = m + p
8 8 2

10
 and the corresponding expenditure functions are
1 1
e(p0 ; u0 ) = u0 = m and e(p1 ; u1 ) = u1 p =m
8 8 2
and
1 1 1 1 1 1
e(p0 ; u1 ) = u1 = m+ p and e(p1 ; u0 ) = u0 p = m+ p
8 8 2 8 8 2 8 8 2
Therefore, the equivalent variation (EV) is
p
0 0 0 1 1 1 2 1
EV = e(p ; u ) e(p ; u ) = m m+ p = p
8 2 8 8 2
where note that we de…ne the EV as the negative of the standard de…ni-
tion in MWG, since in this case we measure a loss in consumer welfare.
2. What is the EV when income satis…es 81 > m > 8p1 2 , i.e., the consumer is
moderately rich?
In this case, the initial equilibrium before the price change is at a cor-
ner solution, while the equilibrium after the price change is interior. In
particular,
1=4
m2 1
u0 = v(p0 ; m) = and u1 = v(p1 ; m) = m + p
4 8 2
and the corresponding expenditure functions are
r
m2 1 1
e(p0 ; u0 ) = 2 = m and e(p1 ; u1 ) = m + p p =m
4 8 2 8 2
and
s
1 1 1
e(p0 ; u1 ) = m + p =2 m+ p
8 2 8 8 2
1=4
1 0 m2 1 m 1=2 1
and e(p ; u ) = p = p
4 8 2 2 8 2
Therefore, the the equivalent variation (EV) is

1 1 1 1
EV = e(p0 ; u0 ) e(p0 ; u1 ) = m m+ p = p
8 2 8 8 8 2
3. What is the EV when income satis…es 8p1 2 > m, i.e., the consumer is poor?
In this case, the equilibrium is at a corner solution, both before and after
the price change. In particular,
1=4 1=4
0 0 m2 1 1 m2
u = v(p ; m) = and u = v(p ; m) =
4 8

11
 and the corresponding expenditure functions are
r r
m 2 m2
e(p0 ; u0 ) = 2 = m and e(p1 ; u1 ) = 2 2 =m
4 8
and r
0 1 m2 m
e(p ; u ) = 2 =p
8 2
Therefore, the the equivalent variation (EV) is
m 1
EV = e(p0 ; u0 ) e(p0 ; u1 ) = m p = 1 p m
2 2
4. [Using the compensating variationto identify price changes.] The preferences
of some consumer can be represented as: u(x1 ; x2 ) = min fx1 ; x2 g. We have been
informed that only the price of the good 2 has changed, from p02 to p12 , but we have
not informed about by how much did it change. We know, however, that the amount
of income that has to be transferred to the consumer in order for him to recover his
initial utility level is:
p02 w
dollars
p01 + p02
where w is the initial income, and p01 and p02 are the initial prices of goods 1 and 2,
respectively. Can you provide some information about the size of the price change, i.e.,
the di¤erence between p02 and p12 ?
p0 m
According to the information, p02+p0 is the amount of income that, at the new
1 2
price ratio, has to be transferred to the consumer in order to recover his initial
utility level, which is the de…nition of CV. Then,
p02 m
CV =
p01 + p02
Since we can calculate the CV as:

CV = v (P1 ; P0 ; m 0 ) v (P1 ; P1 ; m 1 )

Or using the expenditure function as follows:

CV = e(P1 ; u 0 ) e(P0 ; u 0 )

From the last homework assignment, we know that under this utility function
u(x1 ; x2 ) = min fx1 ; x2 g, Walrasian demands are:
m
x1 (p; m) = x2 (p; m) =
p1 + p2
thus the Indirect Utility Function is:
m m m
v(p; m) = min ;: =
p1 + p2 p1 + p2 p1 + p2

12
 Using the identity v(p; e(p; u0 )) = u0 into the previous Indirect Utility Function
we have that
e(p; u0 )
= u0 ;
p1 + p2
and solving e(p; u0 ) = (p1 + p2 )u0 . On the other hand, we also know that:
m
e(P; v (P0 ; m)) = e(P; u 0 ) =(p1 + p2 )
p10 + p20

We can use this expression to calculate CV, since we know that the price of good
1 does not change,

m m (p12 p02 ) m
CV = e(P1 ; u 0 ) e(P0 ; u 0 ) = (p11 + p12 ) (p01 + p02 ) = :
p1 + p02
0
p01 + p02 p01 + p02
p02 m
We can …nally check that this expression coincides with the equality CV = p01 +p02
given in the introduction of the exercise,

(p12 p02 ) m p02 m

0 0
= 0 0
() p12 p02 = p02
p1 + p2 p1 + p2

And rearranging, we obtain p12 = 2p02 , implying that the price of good 2 has
doubled.

5. [Consumer theory] In this question we consider utility functions of the additively

separable form
X
L
u(x1 ; x2 ; :::; xL ) = uj (xj ) (5)
j=1

where u0j (xj ) > 0 for every good j = 1; 2; :::; L. Throughout this question, we assume
that all utility maximization problems (where the price vector p and the wealth level
w are strictly positive) have unique, interior solutions, and that all functions are (twice
continuously) di¤erentiable. Except for part (a), we consider a single consumer.

a) Suppose that all consumers have preferences representable by the same utility
function, of the form described in expression (1). Can we be sure that a positive
representative consumer exists for an unrestricted domain of wealth vectors? If
YES, justify your claim. If NO, provide a counterexample.
NO. You can …nd several examples of separable utility functions whose wealth
p
expansion path is not straight. Here is one example: u(x1 +x2 ) = x1 +ln x2 .
Taking FOCs with respect to x1 and x2 , we obtain
1
p = p1
2 x1

1
= p2
x2

13
 x2 p1
=) p =
2 x1 p2
p1 p
Solving for x2 , we …nd the equationx2 = 2 p2 x1 , which represents the rela-
tionship between the utility-maximizing value of x1 and x2 , i.e., the wealth
expansion path. This equation is depicted in the following …gure.

The wealth expansion path, is, hence, not a straight line. Thus, no representa-
tive consumer exists. Note: several popular preferences (such as Stone-Geary,
Cobb-Douglas, quasilinear forL = 2, CES) can be represented by separable
utility functions of the form (1), and also by indirect utility functions of
the Gorman form, in which case a positive representative consumer exists.
But expression (1) by itself is insu¢ cient to guarantee the existence of a
positive representative consumer, even with identical utility functions for all
consumers.
b) Write the L+1 …rst order conditions of the utility maximization problem, denoting
its solution by
e(p; w) = (e
x eL (p; w))
x1 (p; w); :::; x
i.e., a vector with one component for each good, and its (positive) Lagrange
multiplier by (p; w).
Taking FOCs with respect to xj we obtain L equations of the form
uj 0 (~
xj (p; w)) = (p; w)pj
and taking FOC with respect to the Lagrange Multiplier we obtain the budget
constraint evaluated at the utility-maximizing bundle
p x~1 (p; w) = w
Note that equationuj 0 (~
xj (p; w)) = (p; w)pj implies that (p; w) > 0:
c) By di¤erentiating the just obtained …rst-order equalities with respect to the para-
meters, obtain (L+1)2 equalities involving the partial derivatives of the Walrasian
demand function.
Di¤erentiating the above equation,uj 0 (~ xj (p; w)) = (p; w)pj , with respect to
the price of every good, p1 ; p2 ; :::; pL , and income,w, we obtain:
00 @ x~j @
uj = pj ;
@pk @pk

14
 j = 1; 2; :::L; : k = 1; 2; :::L; : j 6= k =) L(L 1)equations
intuitively representing the cross-price e¤ects.And
00 @ x~k @
uk = pk + ; k=1,:::,L ) (L equations),
@pk @pk
intuitively representing the own-price e¤ects. And
@ x~j
00 @
uj = pj , j=1,:::,L ) (L equations),
@w @w
representing how the consumer’s FOCs are a¤ected by a marginal change in
his income.
We can now di¤erentiate the budget constraint, p x~1 (p; w) = w, with respect
to the price of every good, p1 ; p2 ; :::; pL , and income,w, we obtain:
X
L
@ x~j
pj = x~k ; k=1,:::,L ) (L equations),
j=1
@pk

representing how the budget constraint is a¤ected by a marginal change in

the price of one of the goods. And
X
L
@ x~j
pj = 1 ) (1 equations),
j=1
@w

representing how the budget constraint is a¤ected by a marginal change in

income. Hence, the total number of equations is: L2 L + L + L + L + 1 =
L2 + 2L + 1 = (L + 1)2 :, as required.
d) Show that
@v(p; w)
= (p; w),
@w
where v(p; w) is the indirect utility function, and provide an interpretation of this
result.
The indirect utility function of this consumer is just given by his utility func-
tion evaluated at the utility-maximizing bundle found in exercise (b):
X
v(p; w) = uj (~
xj (p; w)):
j

Therefore, di¤erentiating with respect to w, we obtain

@v X 0 @ x~j 0
X @ x~j X @ x~j
= uj and since uj = pj from exercise (b), = pj = pj :
@w j
@w j
@w j
@w
P @ x~ @v
And since from exercise (c), we know that pj @wj = 1 we obtain that @w
=
j
:(you can alternatively, use the Envelope Theorem to show this result). As
usual, the interpretation of the Lagrange multiplier is the marginal (indirect)
@v
utility of wealth, @w :It couldn’t be more explicit from the result you just
obtained!

15
e) Using the equalities obtained in exercise (c), show that, if all J goods exhibit
dminishing marginal utility, i.e., u00j (xj ) < 0 for all j = 1; 2; :::; L, then
@ (p; w)
< 0. (6)
@w
In addition, show that all goods must be strictly normal. Provide a verbal inter-
pretation of expression (2).
P @ x~
By the result of part (c), pj @wj = 1 , we know that the income e¤ect must
j
be positive, i.e., @@w
x
~k
> 0 , for at least one good k. (Recall our discussion in
class about not all goods being inferior.) Then, expression (2) requires the
00 @x
~
RHS of uk @@w x
~k @
= @w pk for good k to be negative as long as @wj > 0for all
00
goods j 2 f1; :::; Lg, since uk < 0(diminishing marginal utility holds for all
goods). Intuitively, inequality @ @w (p;u)
< 0says that, for the utility function
@v(p;w)
in expression (1) to satisfy @w = (p; u) , the marginal utility of wealth,
despite positive, must be decreasing.
f) Using the equalities obtained in (c), it can also be shown that, if u00j (xj ) < 0 for
all goods j = 1; 2; :::; L, then
" #
@ @ @e
xk
= ek + @
x (7)
@pk @w @w
@w

for all k = 1; 2; :::; L. (You don’t need to show this result.) Use expression (3)
and the equations you found in exercise (b) to …nd an expression, involving @e xk
@w
@e
x
and @wj , for o¤-diagonal terms sjk (p; w), j 6= k, of the Slutsky matrix.
For completeness, here we show how to obtain expression (3), but recall
that this (rather long) derivation was not required in the exam. Using the
@x
~ @
derivative from exercise (c), u00 j @pjj = @pk
pj , we can rearrange it, obtaining
@x
~j pj
@pk
= @
@pk 00 ;Similarly, using
uj
the derivative in exercise (c), u00 k @@px~kk = @
p ,
@pk k

we can solve for @@px~kk to obtain

@ x~k @ pk
= 00 + 00 ;
@pk @pk uk uk
which substituted into our derivatives of the budget line (also from exercise
P
L
@x
~ P
L
@ pj
c), pj @pkj = x~k , yields pj [ @pk u
00 ]+pk 00 =
u
x~k , or
j=1 j=1 j k

@ X pj 2
L

00 = x~k -pk 00 (1.10)

@pk j=1 uj uk
@x
~ @
Finally, from our last derivative in exercise (c), u00 j @wj = p,
@w j
we can solve
@x
~
for @wj to obtain
@ x~j @ pj
= 00 ,
@w @w uj

16
 P
L
@x
~
Our derivative of the budget line (also from exercise c), pj @wj = 1, yields
j=1

X
L
@ pj @ X pj 2
L
pj 00 =1, or 00 =1 (1.11)
j=1
@w uj @w j=1 uj

From (1.10) and (1.11) we obtain

@
@pk @x @ @x @x pk
= ; i. e., = x~k -pk 00 = x~k - 00
x~k -pk u00 @w @pk @w uk @w uk
k

@x
~ @
or, using our derivativeu00 j @wj = p
@w j
from exercise (c), we can alternatively
@ @x~k
pk
express u00 k asu00 k = @w
@x~k pk . Therefore, u00 k
becomes up00kk = @w
@ : ! Hence,
@w @w
@
@pk
can be alternatively expressed as
" # " #
@x
~k @x
~k
@ @ @w @ @w
= x~k @
= x~k + @
@pk @w @w
@w @w

which coincides with expression (3). (Notice, however, that you didn’t have
to show this result, i.e., the origin of expression in exercise (f).)
From expression (3), we can now actually start answering the question in
exercise (f) of the exam. Speci…cally, using the derivatives we obtained in
@x
~ @ @ @ u00 j @ x
~j
exercise (c), u00 j @pkj = @pk
p j , we can solve for @pk
to obtain @pk
= pj @pk
(which
@ @x
~
coincides with the LHS of expression 3), and we can use u00 j @wj = @w pj , and
00
u j @x ~j
@ @
solve for @w to obtain @w = pj @w (which coincides with the …rst term in the
RHS of expression 3). We, thus, rewrite expression (3) becomes
00 00
" #
@x
~k
uj @ x~j uj @ x~j
= x~k - @w
@
; j 6= k
pj @pk pj @w @w

u00 j
Cancelling pj
on both sides of the equality, and rearranging yields

@ x~j @ x~j @ x~j @ x~k

= x~k -@
@pk @w @w @w @w

And further rearranging, we obtain

@ x~j @ x~j @ x~j @ x~k
+ x~k = @
@pk @w @w
@w @w
@x
~j @x
~
where the LHS coincides with Slutsky equation sjk @pk
+ x~k @wj for any two
goods j 6= k. Hence, we obtain
@ x~j @ x~k
sjk = @
@w
@w @w

17
 (a) What can you say about the sign of sjk (p; w), j 6= k? Interpret in terms of goods
being net substitutes vs. net complements in consumption.
@
The expression we found in exercise (f), sjk , is positive, given that @w < 0
from expression (2) in exercise (e), and the fact that all goods are normal
@x
~
(also shown in exercise (e)), i.e. @wj > 0and @@w
x
~k
> 0. Thus, all goods are net
substitutes.
(b) To check that the above condition

u00j (xj ) < 0 for all j = 1; 2; :::; L;

is indispensable for the above results, consider the following example with only
two goods
1 2
u(x1 ; x2 ) = ax1 x + x2 .
2 1
Do the results of exercise (e) hold in this case? Explain.
The example does not satisfy decreasing marginal utility, i.e., u00 j (xj ) <
0;becauseu00 2 (x2 ) = 0: Its FOC include

a~
x1 (p; w) = (p; w)p1

1 = (p; w)p2 ;
Therefore, (p; w) = 1
p2
;with @@w
x
~1
= 0, contrary to the property of goods being
@x
~j
normal, > 0;found in exercise (e).
@w
@x
~j @ x
~k
Note: Some of you tried to use sjk (p; w) = @w @w @ here, but such an
@w
@
expression is not valid in this example. In particular, @w appears in its de-
@ 1
nominator, while @w =0 here, since (p; w) = p2 is independent on wealth.
@x~1 @ x
~2
In fact, the product @w @w is zero, while the o¤-diagonal Slutsky entry is

@ x~1 @ x~1 @ x~1 @ x~1

s12 (p; w) + x~2 = (because = 0)
@p2 @w @p2 @w
p1
= >0
p2 2

6. [Finding the compensating and equivalent variation with little information.]

Consider a consumer who, facing a initial price vector p0 2 Rn++ , purchases an n-
dimensional bundle x 2 Rn+ with an income of w dollars. Assume that the price of all
goods experience a common increase measured by factor > 1.

(a) Compute the compensating variation (CV) of this price increase.

Using the expenditure function, the CV is

CV = e(p1 ; u0 ) e(p0 ; u0 )

18
 where p1 and p0 denote the …nal and initial price vector, respectively, and
u0 represents the utility level that the consumer achieves at the initial price-
wealth pair (p0 ; w). In this exercise, we are informed that p1 = p0 , implying
that the above expression for CV can be rewritten as
CV = e( p0 ; u0 ) e(p0 ; u0 )
Recall now that the expenditure function is homogeneous of degree one in
prices, i.e., e( p0 ; u0 ) = e(p0 ; u0 ), and that the consumer spends w dollars.
These properties reduce the CV to
CV = e( p0 ; u0 ) e(p0 ; u0 ) =
= e(p0 ; u0 ) e(p0 ; u0 ) =
| {z } | {z }
w w
= w w = w( 1)
(b) Compute the equivalent variation (EV) of this price increase.
Using the expenditure function, the EV is
EV = e(p1 ; u1 ) e(p0 ; u1 )
where u1 represents the utility level that the consumer achieves at the …nal
price-wealth pair (p1 ; w). In this exercise, we are informed that p1 = p0 , or
p0 = 1 p1 , implying that the above expression for EV can be rewritten as
1
EV = e(p1 ; u1 ) e p1 ; u0

Recall now that the expenditure function is homogeneous of degree one in

prices, i.e., e 1 p1 ; u0 = 1 e (p1 ; u0 ), and that the consumer spends w dollars,
which means e(p1 ; u1 ) = e(p0 ; u0 ) = w. These properties reduce the EV to
1
EV = e(p1 ; u1 ) e p1 ; u0 =
1
= e(p0 ; u0 ) e p1 ; u0 =
| {z } | {z }
w w
1 1
= w w=w 1

7. Consider a household that is seen to purchase quantities of just two goods, bread and
cheese. Denote quantities of bread by x and quantities of cheese by y. The household
comprises two individuals; Andrew, whose preference relation can be represented by the
utility function uA (x; y) = x and Brenda, whose preference relation can be represented
by the utility function uB (x; y) = y .

(a) Derive the uncompensated demand functions for both Andrew and Brenda and
their indirect utility functions
max uA (x; y) = x
x;y

19
 s.t. px x + py y wA
Answer: Andrew will spend all wealth on x and buy no y (since good y doesn’t
show up in Andrew’s utility function) Similarly Brenda maximizes her utility by
spending all her wealth on y (since good x doesn’t enter in her utility function).
Thus:
wA
xA (px ; py ; wA ) =
px
yA (px ; py ; wA ) = 0
xB (px ; py ; wB ) = 0
wB
yB (px ; py ; wB ) =
px
(b) The households’wealth w is divided evenly between Andrew and Brenda. Suppose
that you observe the aggregate demands of this household and you interpret it as
if it came from just a single consumer. Find the demands of the supposed single
consumer.
Answer: If a positive representative consumer exists, its demand will equal
aggregate demand, that is:
w
x(px ; py ; w) = xA (px ; py ; w=2) + xB (px ; py ; w=2) = +0
2px
w
y(px ; py ; w) = yA (px ; py ; w=2) + yB (px ; py ; w=2) = 0 +
2py
Recall in lectures that the equivalent variation of a change in prices and income
from(p0 ; w0 ) to(p1 ; w1 ) is de…ned as:
EV = e(p0 ; v(p1 ; w1 )) e(p0 ; v(p0 ; w0 ))
If w0 = w1 and the change in prices are caused by the imposition of commodity taxes
then the deadweight loss (DWL) or excess burden of the taxes is given by:
X
L
DW L = EV tl xl (p1 ; w0 )
l=1

where tl = p1l p0l .

c. Brie‡y explain why this measure may be viewed as a deadweight loss to (social)
economic e¢ ciency.
P
Answer: Ll=1 tl xl (p1 ; w0 )is the tax revenue raised by the tax. EV is the equiva-
lent change in wealth that would leave the individual as well o¤ as the price change
induced by the tax. So EV is how much tax revenue the government could have
raised from the individual through a lump-sum (i.e. non-distortionary) tax, that
would leave her as well o¤ as under the commodity tax. So the di¤erence (that
is, the di¤erence between how much the government could have raised through
a lumpsum tax and how much it actually raised through distortionary taxes)
measures the “excess burden” associated with the distortionary commodity tax
relative to the non-distortionary lumpsum tax on wealth.

20
d. Suppose that the household initially faces pricesp0 = (1; 2) and has wealthw0 =
300 . Then a speci…c tax of 2 is imposed on bread (i.e. good x) that leads to its
price rising to 3 (with the price of cheese, i.e. goody, and the households’wealth
both remaining unchanged). Calculate the DWL under the false assumption that
the household demands come from just one consumer.
Answer: From part (b) we havex(px ; py ; w) = w=2px andy(px ; py ; w) = w=2py
. Notice that this is the demand system corresponding to an individual with
Cobb-Douglas (or equivalently, log-linear) utility,u(x; y) = xy (or equivalently,
ln x + ln y).
v(px ; py ; w) = x(px ; py ; w) y(px ; py ; w)
w2
=
4px py
From the identity:
v(px ; py ; e(px ; py ; u)) u
We obtain:
p
e(px ; py ; u) = 2 px py u
So,
EV = e(p0 ; v(p1 ; w1 )) e(p0 ; v(p0 ; w0 ))
q
= 2 p0x p0y v(p1x ; p1y ; w) w0
s
2(300)2
=2 300
4 6
p
= ( 3 3) 100
And XL
tl xl (p1 ; w0 ) = 2 x(3; 2; 300) = 2 50 = 100
l=1
So p
DW L = (3 3) 100 100
p
= (2 3) 100
26:795
e. Using the individuals’indirect utility functions derived in part (a) calculate the
two individual DWLs, Explain why does or does not equal DWL.
Answer: Now from part (a) we have:
wA
vA (px ; py ; wA ) = xA (px ; py ; wA ) =
px
So:
eA (px ; py ; uA ) = px uA
Similarly,
wB
vB (px ; py ; wB ) = xB (px ; py ; wB ) =
py

21
 eB (px ; py ; uB ) = px uB
So:
EVA = eA (p0 ; vA (p1 ; wA
1
)) e(p0 ; vA (p0 ; wA
0
))
150
= 150 = 100
3
EVB = eB (p0 ; vB (p1 ; wB
1
)) e(p0 ; vB (p0 ; wB
0
))
2 150
= 150 = 0
2
Since Andrew is the only individual who consumers bread, the amount of tax on
bread that he pays must be 100, while Brenda is paying no tax at all since she
consumes no bread. Thus:

DW LA = 100 100 = 0

DW LB = 0 0=0
So,
DW LA + DW LB = 0 < DW L
Notice that since Andrew only consumes bread, a tax on bread is equivalent
to a lumpsum tax for him, and so this generates no deadweight loss. And of
course since Brenda does not consume bread, the tax on bread does not a¤ect
her consumption or her utility. But given the (arbitrary) equal division of wealth
rule for this household, aggregate demand appears as if it is coming from a Cobb-
Douglas utility function. For such preferences, a tax on bread is distortionary and
so generates a deadweight loss, but this is only an artifact of the equal division of
wealth rule and does not correspond to any underlying social planner allocating
the household’s wealth in order to maximize some social welfare function of the
individuals’ utilities. Thus we cannot attach any normative signi…cance to the
“welfare” measures generated from this positive representative consumer. What
appears to be a distortionary tax for the household is actually a non-distortionary
tax, equivalent to a lumpsum for Andrew.

8. Consider an individual with Cobb-Douglas preferences u(x1 ; x2 ) = (x1 x2 )0:5 , where x1

and x2 denote the amounts consumed of goods 1 and 2, respectively. The prices of
these goods are p1 > 0 and p2 > 0, respectively, and this individual’s wealth is w > 0.
The government needs to collect a large amount of money to …nance a new Health Care
plan, and contemplates two options: (1) introduce an income tax equivalent to 40%
of individuals’ wealth; or (2) charge a sales tax over the price of good 1 (e.g., fuels)
which would imply an increase in the price of good 1 fromp1 to p1 (1 + t), collecting
the same dollar amount as with the income tax. Using the indirect utility function
of this individual under option 1 (income tax) and option 2 (sales tax), explain which
tax produces a smaller utility reduction to this individual (i.e., which tax is preferred
by this individual). You can accompany your discussion with an intuitive explanation
and/or a …gure if necessary.

22
Answer: We need to calculate the e¤ect that both types of taxes on this individual’s
utility level, for a given total tax revenue. In order to …nd the individual’s utility level
in equilibrium (after he decides how to distribute his wealth among di¤erent goods)
we must use this individual’s indirect utility function. First, recall that the Walrasian
demands in the context of a Cobb-Douglas utility function as that given in this exercise
are x1 (p; w) = 2pw1 and x2 (p; w) = 2pw2 , respectively. Inserting these Walrasian demands
into the individual’s utility function we obtain the indirect utility function
w
v(p; w) = p
2 p1 p2

Taking into account that the tax revenue must be 40% of the individual’s wealth, i.e.,
T = 52 w, we need to …nd the increase in p1 that generates this tax revenue. If the price
of good 1 increases as a result of a sales tax, i.e., (1 + t)p1 , then Walrasian demand
w
becomesx1 (p; w) = 2(1+t)p1
, implying that tax revenue is

w
T = tp1 x1 (p; w) = tp1
2(1 + t)p1

The condition that tax authorities generate the same revenue charging 40% of the
individual’s wealth, 25 w , and using a sales tax as described above, implies that

2 w
w = tp1
5 2(1 + t)p1

which implies that t = 4. If we use an income tax, then the individual’s wealth after
taxes becomes w0 = w 52 w = 35 w, inducing an after tax indirect utility function

w0 3w
v IncT ax (p; w) = p = p
2 p1 p2 10 p1 p2

If, in contrast, we use a sales tax on good 1, for an amount of tp1 = 4p1 , i.e., p01 =
(1 + t)p1 = 5p1 , the new Walrasian demand for good 1 is x10 = 2 w5p1 = 10p
w
1
, implying
that the individual’s indirect utility function in this case is
w w
v SalesT ax (p; w) = p 0 = p
2 p 1 p2 2 5p1 p2

Comparing the indirect utility function of this individual under the income tax and
the sales tax, we obtain that

v SalesT ax (p; w) < v IncomeT ax (p; w)

That is, the individual prefers to bear an income tax rather than a sales tax that
increases the price of good 1.
Interpretation: A sales tax modi…es the optimal consumption bundle through two
e¤ects. On one hand, the sales tax produces a downward shift in the budget set
(maintain the price ratio una¤ected) re‡ecting the fact that the individual becomes
poorer in purchasing power terms, ultimately reducing his utility level. On the other

23
 0
hand, the fact that good 1 becomes relatively more expensive, i.e., pp21 = 5p p2
1
6= pp12 ,
implying that the individual’s utility is further reduced again. Note that the …rst e¤ect I
just described could be also interpreted as the income e¤ect associated to the imposition
of a sales tax, while the second e¤ect described above refers to the substitution e¤ect
that a sales tax brings about. In contrast, the income tax solely produces the …rst
of the above e¤ects, namely, a downward shift in the individual’s budget line (in a
parallel fashion, since good prices are una¤ected), reducing the consumer’s purchasing
power. The income tax does not produce the second e¤ect (substitution e¤ect), and
as a consequence, the income tax reduces individual’s utility in a smaller amount.
Note that this result can be extended to the case in which all goods are taxed (not
only good 1, as in the previous analysis). In this case, prices increase to p01 = (1 + t1 )p1
and p02 = (1 + t2 )p2 , respectively, where note that we allow sales taxes on goods 1 and
2 to di¤er. These sales taxes induce Walrasian demands of
w w
x01 (p; w) = and x02 (p; w) =
2(1 + t1 )p1 2(1 + t2 )p2

And total tax collection becomes

w w
T = t1 p1 x1 + t2 p2 x2 = t1 p1 + t2 p2
2(1 + t1 )p1 2(1 + t2 )p2

And since the total tax collection using the sales and the income tax must coincide,
T = 25 w , implying that

w w 2
t1 p1 + t2 p2 = w
2(1 + t1 )p1 2(1 + t2 )p2 5

Simplifying this expression, we obtain t1 + t1 + 6t1 t2 = 4, and solving for t1 and t2 , we

t1
havet1 = 4 + 1+6t 1
where t1 2 (0; 4) . Then, the maximum utility that the individual
can reach with the sales tax is
w
v SalesT ax (p; w) = q ; where t1 2 (0; 4)
4 t1
2 p1 (1 + t1 )p2 (1 + 1+6t1
)

Comparing this expression with the indirect utility function under the income tax,
v IncT ax( p; w), we still obtain that v SalesT ax (p; w) < v IncT ax (p; w). That is, the individual
still prefers the income to the sales tax.

9. Consider the three good setting in which the consumer has utility function

u(x) = (x1 b1 ) (x2 b2 ) (x3 b3 )

where b1 ; b2 ; b3 > 0 represent the minimal amounts of goods 1, 2 and 3 that this
individual must consume at every period in order to remain alive (e.g., calories, water
and shelter).

(a) Why can you assume that + + = 1 without loss of generality? Do so for the
rest of the problem.

24
 De…ne 0 0 0
u(x) = u(x)1=( + + )
= (x1 b1 ) (x2 b2 ) (x3 b3 )
with
0 0 0 0 0 0
= =( + + ); = =( + + ); = =( + + ): Then + + =1
and u~( )represents the same preferences asu( ),because the functionu ! u1=( + + ) is
a monotone transformation. Thus we can assume without loss of generality
that + + = 1
(b) Write down the …rst-order conditions for the UMP, and derive the consumer’s
Walrasian demand and indirect utility functions. [This system of demands is
known as the “linear expenditure system,” and it is due to Stone (1954), and
the above utility function is usually referred as the Stone-Geary utility function.]
[Hint: Use another monotone transformation, u(x) = ln u(x) of the given utility
function u(x)].
Use another monotone transformation of the given utility function
ln u(x) = ln(x1 b1 ) + ln(x2 b2 ) + ln(x3 b3 )
And therefore the Lagragian of this UMP is:
L(x1 ,x2 ; x3 ; ) = ln(x1 b1 )+ ln(x2 b2 )+ ln(x3 b3 ) (p1 x1 +p2 x2 +p3 x3 w)
The …rst order conditions of the UMP yields
@L
= p 1 = 0 , p1 x1 = + p 1 b1
@x1 x1 b1
@L
= p 2 = 0 , p 2 x 2 = + p 2 b2
@x1 x 2 b2
@L
= p 3 = 0 , p 3 x 3 = + p 3 b3
@x1 x 3 b3
@L
= p1 x1 + p2 x2 + p3 x3 w = 0
@x1
Plugging the information from the …rst FOCs into the last FOC, we obtain =
1
w p1 b1 +p2 b2 +p3 b3
, or simply = w 1 pb where we denote
p b = p1 b1 + p2 b2 + p3 b3 . The Walrasian demand functions therefore become

x1 (p; w) = b1 + (w p b)
p1

x2 (p; w) = b2 + (w p b)
p2
x3 (p; w) = b3 + (w p b)
p3
Plugging these demand functions into u( ) we obtain the indirect utility function

v(p; w) = (w p b)
p1 p2 p3

25
(c) Verify that the Walrasian demand functions x(p; w) obtained in part (b) satisfy
homogeneity of degree zero in prices.
To check the homogeneity of degree zero of the Walrasian demand function, we
increase all prices and wealth by a common factor

x( p; w) = (b1 ; b2 ; b3 ) + ( w p b) ; ;
p1 p2 p3

= (b1 ; b2 ; b3 ) + (w p b) ; ; = x(p; w)
p1 p2 p3
to check Walras law,

p x(p; w) = p b + (w p b) p1 + p2 + p3
p1 p2 p3
= p b + (w p b)( + + )=w
The demand function is unique: for a given price vector and wealth, the consumer
demands a particular amount of every good.
(d) Verify that the indirect utility function v(p; w) obtained in part (b) satis…es:
a. Homogeneity of degree zero,
To check homogeneity of the indirect utility function, we scale all prices and
wealth by a common factor

v( p; w) = ( w p b)
p1 p2 p3

1 ( + + )
= (w p b)
p1 p2 p3

= (w p b) = v(p; w)
p1 p2 p3
b. Increasing in wealth,
To check the that v(p; w) is increasing in wealth,

@v(p; w)
= >0
@w p1 p2 p3
c. Decreasing in prices,
To check that v(p,w) is decreasing in prices,
@v(p; w)
= v(p; w) <0
@p1 p1
@v(p; w)
= v(p; w) <0
@p2 p2
@v(p; w)
= v(p; w) <0
@p3 p3

26
 d. Quasiconvex in prices.
In order to prove the quasiconvexity, it is su¢ cient to prove that for any
v 2 R and w > 0, the set p 2 R3 : v(p; w) vgis convex. Consider:

ln v(p; w) = ln(w p b) + ln + ln + ln ln p1 ln p2 ln p3

Since the logarithmic function is concave, the set

p 2 R3 : ln(w p b) ln p1 ln p2 ln p3 v

is convex for every v 2 R.Since the other terms, ln + ln + ln , do

not depend on p, this implies that the set fp 2 R3 : ln v(p; w) vgis convex.
Hence so is fp 2 R3 : v(p; w) vg.
(e) Let us now restrict our analysis to a utility function with only two goods,

u(x) = (x1 b1 ) (x2 b2 ) ,

where + = 1. Are the preferences represented by this utility function homo-

thetic? [Hint: …nd the share of income spent on each good (i.e., budget shares)]
This preference relation is not homothetic. In order to see why, let us …rst
…nd the M RS1;2 and then scale all the amounts of good 1 and 2 by a common
factor t.
1
M U1 (x1 b1 ) (x2 b2 ) x2 b2
M RS1;2 (x1 ; x2 ) = = 1
=
M U2 (x2 b2 ) (x1 b1 ) x1 b1

Scaling up all goods by a common factor, the MRS becomes

tx2 b2
M RS1;2 (tx1 ; tx2 ) =
tx1 b1

which does not coincide with M RS1;2 (x1 ; x2 ). Therefore, the slope of this
individual’s indi¤erence curve change its slope as this individual consump-
tion of both goods increases in the same factor t, for a given proportion of
goods 1 and 2 (i.e., for a given ray from the origin). Finally, note that if
b1 = b2 = 0 (intuitively, when there is no minimal amount of goods 1 and 2
that the individual must consume in order to survive), then the above utility
function becomes a Cobb-Douglas utility function, i.e., and this utility func-
tion is homothetic. This is easy to check by making b1 = b2 = 0 in both
M RS1;2 (x1 ; x2 )andM RS1;2 (tx1 ; tx2 ) , since we obtain the same value in both
cases.

27
 Chapter 4 - Aggregate Demand
End-of-Chapter Exercises - Answer key

1. Verify that the social indirect utility function, v (p; w), which is the optimal value of
the social welfare maximization problem:

max W (u1 (x1 ) ; u2 (x2 ) ; :::; uI (xI ))

w1 ;:::;wI
!
X
subject to p xi w
i

where the social welfare function W ( ) is increasing in every individual utility level,
satis…es the properties of the indirect utility function v (p; w):

(a) Homogeneous of degree zero

(b) Increasing in w:
(c) Non-increasing p:
(d) Quasiconvex. [Hint: use the de…nition of quasiconvexity we found for the indi-
vidual’s indirect utility function].

a) Homogeneous of degree zero

We need to show that v ( p; w) = v (p; w). Hence,
!
X
p xi ( p; w) w for all >0
i

And since the Walrasian Demand xi (p; w) is homogeneous of degree zero, we have
that,
xi ( p; w) = xi (p; w)
Using this fact into the above expression,
!
X
p xi ( p; w) w (1)
i
!
X
() p xi (p; w) w (2)
i
!
X
() p xi (p; w) w (3)
i

Since the original constraint (3) is equivalent to the constraint agragated by (inequal-
ity (1)), and since the objective function hasn’t been changed at all, the maximization
problem is going to lead to the same optimal value. Hence v (p; w) = v ( p; w), and
therefore the indirect utility function satis…es Homogeneity of degree zero.

1
b) Increasing in w.
Let’s take w0 such that w0 w:Let’s denote by (w1 ; w2 ; :::; wI ) the solution to the
maximization problem given (p; w).
Let v (p; w) = w (vi (p; w1 ) ; v2 (p; w2 ) ; :::; vI (p; wI )) :We also have:
X X
wi w =) wi w0
i i

So, by the de…nition of

v (p; w0 ) : v (p; w) = W (vi (p; w1 ) ; v2 (p; w2 ) ; :::; vI (p; wI )) v (p; w0 )

2. Non-increasing in p
Let us take a price vector p0 such that p0 p. Let us (w1 ; w2 ; :::; wI ) be the solution to
the maximization problem given (p0 ; w). Hence,

v (p0 ; w) = W (v1 (p01 ; w1 ) ; v2 (p01 ; w2 ) ; :::; vI (p01 ; wI ))

As price vector p0 satis…es p0 p, we have that

vi (p0 ; wi ) vi (p; wi ) for all i

Since W ( ) is increasing in the utility levels of every individual,

vi (p0 ; wi ) vi (p; wi ) () W (vi (p0 ; wi ) ; :::; vI (p0 ; wI )) w (vI (p; wI ) ; :::; vI (p; wI ))

Then, by the de…nition of v (p; w) we have,

w (v1 (p0 ; w1 ) ; v2 (p0 ; w2 ) ; :::; vI (p0 ; wI )) v (p; w)

Therefore,

v (p0 ; w) v (p; w) for all p0 p

Hence v (p; w) is non-increasing in p.

d) Quasiconvex
We know that a function is quasiconvex when its lower contour set is convex. (Alter-
natively, a function is quasiconvex if its upper contour set is concave). Then, we need
to show that the set

I = f(p; w) : v (p; w) v for all v 2 R+ g is convex

For this set to be convex we need that, for any two price-wealth pairs (p; w) and
(p0 ; w0 ) that belong to the lower contour set, i.e., (p; w) 2 I and (p0 ; w0 ) 2 I , their
linear combination satis…es

v (p; w) + (1 + ) v (p0 ; w0 ) v

2
 for all 2 [0; 1] ; all v 2 R+ and all p >> 0 and w > 0:
Proof. Using the de…nition of the lower contour set of v(p; w), we have that

v(p; w) v () v(p; w) v; and

v(p0 ; w0 ) v () (1 ) v(p0 ; w0 ) (1 )v
Adding up, we obtain that

v(p; w) + (1 )v(p0 ; w0 ) v + (1 )v

which simpli…es into

v(p; w) + (1 )v(p0 ; w0 ) v
which implies that the lower contour set of v(p; w) is convex.

2. Suppose that all consumers have neutral preferences % de…ned as RL+ (with individual
demand functions denoted by x~ (p; w) and that individual wealth is uniformly distrib-
uted on the interval [0; w]. Then the aggregate demand function
Z w
x (p) = x~ (p; w) dw
0

satis…es the unrestricted law of demand property.

(a) Prove the above claim.

Answer: From similar exercises we know that if a demand function x (p; w)
satis…es the uncompensated law of demand property (ULD) [dp dx (p; w) 0],
then we must also have that Dp xi (p; w) is negative semide…nite (NSD). That
is,
dp dxi (p; w) 0 () dp Dp xi (p; w) dp 0
Hence in this case, (the aggregate demand is x (p), in order to prove ULD
we only need to show that Dp x (p) is NSD. That is, we just need to show:
v Dp x (p) v 0:
Proof. First we di¤erentiate the aggregate demand function with respect to
prices, Z w
Dx (p) = Dp x~ (p; w) dw
0
Then we pre- and post-multiply by v 6= 0,
Z w
v Dx (p) v = v Dp x~ (p; w) v dw (1)
0

We can now use the Slutsky equation,

S (p; w) = Dp x~ (p; w) + Dw x~ (p; w) x~ (p; w)T

3
Hence,
Dp x~ (p; w) = S (p; w) Dw x~ (p; w) x~ (p; w)T
Substituting this expression into (1) we obtain:
Z w Z w
v Dx (p) v = v S (p; w) v dw v Dw x~ (p; w) x~ (p; w)T v dw
0 0
Rearranging the second term:
Z w Z w
v Dx (p) v = v S (p; w) v dw (v Dw x~ (p; w)) (v x~ (p; w)) dw
0 0
| {z } | {z }
(1) (2)

Recall that we need to show that v Dx (p) v 0: In order to do so,

we
Rw will focus separately on both terms of the right hand side: Term (1)
0
v S (p; w) v dw is negative semide…nite (we were talking about this in
class di¤erent times).
Regarding term (2),
(v Dw x~ (p; w)) (v x~ (p; w))
Di¤erentiating [v x~ (p; w)]2 with respect to w,
d [v x~ (p; w)]2
= 2 [v x~ (p; w)] [v Dw x~ (p; w)]
dw
And multiplying 12 on both sides:
1 d [v x~ (p; w)]2
= [v x~ (p; w)] [v Dw x~ (p; w)] ,
2 dw
Then term (2) becomes:
Z w Z w
1 d [v x~ (p; w)]2
(v Dw x~ (p; w)) (v x~ (p; w)) dw = dw =
0 0 2 dw
Z w
1 d [v x~ (p; w)]2 1 w
= dw = [v x~ (p; w)]2 0 =
2 0 dw 2
2 2 32 3
1 1
= 4[v x~ (p; w)]2 4v x~ (p; 0)5 5 = [v x~ (p; w)]2 0
2 | {z } 2
0
Hence term (2) is positive semide…nite. Hence, combining terms (1) and (2)
we obtain,
Z w Z w
v Dx (p) v = (v S (p; w) vdw) (v Dw x~ (p; w)) (v x~ (p; w)) dw
|0 {z } |0 {z }
(1) 0 (2) 0

Therefore, v Dx (p) v 0, and as a consequence Dp (p) is NSD. Thus, the

aggregate demand function
Z w
x (p) = x~ (p; w) dw satis…es ULD
0

4
(b) Demonstrate that the proof fails if instead wealth is distributed uniformly on the
interval [w; w], where w > 0:
Answer: When wealth is distributed uniformly over the interval [w; w], where
w > 0 the above property doesn’t hold (i.e., aggregate demand functions x (p)
as de…ned in part a) doesn’t satisfy the ULD property) because
v Dx (p) v is not NSD.
Using the de…nition of the aggregate demand function of the previous exercise
(but changing the interval over which the wealth is distributed) we have:
Z w
x (p) = e (p; w) dw
x
w

Taking derivatives with respect to prices, and pre- and post- multiplying by
v, Z w
v Dx (p) v = v Dp xe (p; w) v dw
w

And expressing Dp xe (p; w) in terms of the Slustsky equation (just as we did

in the previous part of the exercise) we have:
Z w Z
1 w d [v x e (p; w)]2
v Dx (p) v = v S (p; w) v dw dw
w 2 w dw
| {z } | {z }
(1) (2)

We know that term (1) is NSD from part (a). Let us now analyze term (2)
using the expression we found in part (a):
Z
1 w d [v x e (p; w)]2 1 w
dw = e (p; w)]2 w =
[v x
2 w dw 2
1
= e (p; w)]2 [v x
[v x e (p; w)]2
2
If some good in the consumption bundle is inferior, then may have that:
e (p; w) < x
x e (p; w)
(an increase in wealth implies a decrease in the amount demanded). This
implies that
v x e (p; w)
e (p; w) < v x

() [v x e (p; w)]2 < [v x

e (p; w)]2
1
() [v xe (p; w)]2 [v x e (p; w)]2 < 0
2
So, if some good in the consumption bundle is inferior, we may have:
1
[v xe (p; w)]2 [v x
e (p; w)]2 < 0
2
which makes term (2) to be NSD. Hence we cannot say that v Dx (p) v 0:
As a consequesce, we cannot guarantee that x (p) satis…es ULD property.

5
3. Construct an example in which the area variation measure approach incorrectly ranks
p0 and p1 . [Hint: Let the change from p0 to p1 involve a change in the price of more
than one good.]

We consider an example of a consumer who faces the choices over two goods
and whose preference % and demand function x (p1 ; p2 ; w) satisfy the following
conditions:
(1 ")
For every p1 2 [1; 2] ; x (p1 ; 1; 2) = ; 1+" ;
p1

(1 + ")
for every p2 2 [1; 2] ; x (1; p2 ; 2) = 1 "; ;
p2
(1 ") (1 + ")
and ; 1+" 1 ";
2 2
By using the …gure below, you can convince yourself, perhaps with some applica-
tion of the weak axiom, that there actually exists such a preference.

De…ne p0 = (2; 1) and p1 = (1; 2). Then

x p0 ; 2 = ((1 ") =2; 1 + ") ;

x p1 ; 2 = (1 "; (1 + ") =2) .

Thus x (p0 ; 2) x (p1 ; 2). However, the area variation measure following the
price-change path
p0 ! (1; 1) ! p1 is
[2;1] ![1;1] ![1;2]

6
 Z 1 Z 2
0 1
AV p ; p ; 2 = x1 (p1 ; 1; 2) dp1 + x2 (1; p2 ; 2) dp2
2 1
Z 2 Z 2
(1 ") (1 + ")
= dp1 + dp2
1 p1 1 p2
= [(1 ") ln p1 ]2p1 =1 + [(1 + ") ln p2 ]2p1 =1
= 2" ln 2 > 0.

Hence the area variation measure ranks p1 (new prices) over p0 (old prices).

4. Suppose that consumers all have homothetic preferences represented by utility func-
tions homogeneous of degree one. Consider now the social welfare function

X
I X
I
W (u1 ; u2 ; :::; uI ) = i ln ui with i > 0 and i =1
i=1 i=1

Show that the optimal wealth distribution rule (for the maximization problem 4.D.1
we discussed in class) is
wi (p; w) = i w
that is, show that the optimal wealth distribution rule assigns a constant proportion of
wealth to every individual, irrespective of the price level. [Note that this implies that
in the homothetic case analyzed here, the aggregate demand

X
I
x(p; w) = xi (p; i w)
i=1

can be viewed as originating from the normative respresentative consumer generated

by this social welfare function].

The social welfare maximization problem is now written as

X
max i ln vi (p; wi )
(w1 ;:::;wI )
i
X
subject to wi w
i

The …rst-order conditions are that there exists > 0 such that
@W @vi
=
@ui @wi

for every i, where all derivatives are evaluated at a solution (w1 ; :::; wI ). By the
de…nition of W ( ) ;
@W i
= :
@ui vi (p; wi )

7
 Using Euler’s theorem, since vi (p; wi ) is homogeneous of degree one in wi , we
then know that
@vi (p; wi ) @vi (p; wi ) vi (p; wi )
vi (p; wi ) = wi () = :
@wi @wi wi
Hence the left-hand sides of the above …rst-order conditions equal
i vi (p; wi ) i
=
vi (p; wi ) wi wi
P P
for every i. Thus wi = i
. Since i i = 1 and i wi = w; w = 1 . Hence
wi (p; w) = i w.

5. Suppose that we know not only p0 ; p1 ; and x0 but also x1 = x(p1 ; w). Show that
if (p1 p0 ) x1 > 0, then the consumer must be worse o¤ at price-wealth situation
x(p1 ; w) than at (p0 ; w). Interpret this test as a …rst-order approximation to the
expenditure function at p1 . Also show that an alternative way to write this test is
p0 (x1 x2 ) < 0, and depict the test for the case where L = 2 in (x1; x2 ) space. [Hint:
Locate the point x0 on the set x 2 RL+ : u (x) = u0 .]

If (p1 p0 ) x1 > 0, then w > p0 x1 . The local non-satiation implies that x0 is

preferred to x1 . Hence the consumer must be worse o¤ at (p1 ; w).
As for the interpretation in terms of the …rst-order approximation, since e (p; u)
is concave in p,
e p0 ; u1 e p1 ; u1 + re p1 ; u1 p0 p1 :
Since re (p1 ; u1 ) (p0 p1 ) < 0; e (p0 ; u1 ) < e (p1 ; u1 ) = w. Thus u0 = v (p0 ; w) >
u1 .
Finally, (p1 p0 ) x1 > 0 if and only if w > p0 x1 , which, in turn, is equivalent
to p0 (x1 x0 ) < 0. This test is depicted in the picture below:

8
6. Prove that if the Walrasian demand xi (p; wi ) satis…es the uncompensated law of de-
mand (ULD), then Dp xi (p; wi ) is negative semidi…nite. That is,

dp Dp xi (p; wi )dp 0 for all dp:

Answer: By the de…nition of a directional partial derivate.

1
Dp x(p; w)dp = lim [x(p + dp; w) x(p; w)]
!0

Hence
1
dp Dp x(p; w)dp = dp lim [x(p + dp; w) x(p; w)]
!0
1
= lim dp [x(p + dp; w) x(p; w)]
!0

But the ULD property implies that

dp [x(p + dp; w) x(p; w)] 0 for all > 0:

Hence, by taking the limit ! 0, we obtain dp Dp x(p; w)dp 0. Thus Dp x(p; w)is
negative semide…nite.

7. Provide a graphical illustration of a preference relation that produces a Walrasian

demand that does not satisfy the ULD property. Explain.

Answer: In the …gure below, good 1 is a Gi¤en good, since a price decrease of
good 1 entails a negative total e¤ect on its Walrasian demand, i.e., a decrease in
p1 actually reduces x1 or, alternatively, @x1
@p1
> 0, which yields a positively sloped
Walrasian demand for good 1.

X2

X1

This example shows that the ULD property is actually not derived from the utility
maximization. It is, instead, a restriction on preferences.

8. Suppose there are two consumers,1 and 2, with utility function over two goods, 1and
p p
2, of u1 (x11 ; x21 ) = x11 + 4 x21 and u2 (x12 ; x22 ) = x12 + 4 x22 . The two consumers
have identical wealth levels w1 = w2 = w=2

(a) Calculate the individual demand functions and the aggregate demand function.

9
 Answer: When deriving demand from the …rst-order conditions of utility
maximization, we ignore the nonnegativity constrains. Later on, we show
that, for prices and wealths under consideration, the demands are always
interior.
Taking …rst-order conditions, we …nd that the Walrasian demand for individ-
ual 1 is
w w w w p1 p21
x1 p; = x11 p; ; x21 p; = 4 ;4
2 2 2 2p1 p2 p22

and that of individual 2 is

w w w p21 w p1
x2 p; = x12 p; ; x22 p; = 4 ; 4
2 2 2 p22 2p2 p2

Hence
w w
x(p; w) = x1 p; + x2 p;
2 2
2
w p1 p w p1 p2
= 4 + 4 12 ; 4 + 4 12
2p1 p2 p2 2p2 p2 p2
(b) Compute the individual Slutsky matrice Si p; w2 and the aggregate Slutsky matrix
S(p; w). [Hint: Note that for this two-good example, only one element of each
matrix must be computed to entire matrix.] Show that dp S(p; w)dp < 0 for all
dp 6= 0 not propotional to p. Conclude that aggregate demand satis…es the weak
axiom.
Answer: Denote the (l; k) entry of the Slutsky matrix Si (p; w) of consumer
@x (p; w ) @x21 (p; w
2) 8p2
i by Slki (p; w). Since 21@w1 2 = 0, s221 p; w2 = @p2
= p31 . Hence, by
2
8p11
symmetry of the Slutsky matrix, we obtain s221 p; w2 = s121 p; w2 = p22
and, as a consequence, s111 p; w2 = 8
p2
. Thus,
" 8 8
#
w p2 p22
S1 p; = 8p1 8p21
2 p22 p32

" 8p22
#
8p2
w p31 p21
S2 p; = 8p2 8
2 p21 p1

we can also derive the Slutsky matrix S(p; w) of the aggregate demand func-
tion: " #
w 6 6p22 w 6p1 6p2
4p12 p2 3
p1 4p1 p2
+ p22 + p12
S(p; w) = w 6p1 6p2 w 6 6p21
4p1 p2
+ p2 + p2 4p2 p1 p3
2 1 2 2

2
In addition, we know that if dp 2 R , dp 6= 0, and dp is not proportional
to the price vector, p, then dp S(p; w)dp < 0. Thus, the aggregate demand
function x(p; w) satis…es the weak axiom.

10
 P w
(c) Compute the matrix C(p; w) = i Si p; 2 S(p; w) for price p1 = p2 = 1.
Show that itis positive semidi…nite if w > 16 and that it is negative semidi…nite is
8 < w < 16. In fact, Argue that in the latter case, dp C(p; w)dp < 0 for some dp
(so that C(p; w) is not positive semidi…nite). Conclude that the matrix C(p; w)
being positive semidi…nite is not necessary for the weak axiom to be satis…ed.
Answer: By substituting p = (1; 1), we obtain
X w w
4 4 w
C(p; w) = Si p; S(p; w) = 4 4
w w
2 4 4 4
4
i

Thus, if w > 16, then it is positive semidi…nite, and if 8 < w < 16, then
it is negative semidi…nite. For example, if w = 12 and v = (1; 1), then
v C(p; w)v = 1. Thus C(p; w) is not positive semidi…nite. We saw in
part (b) of the exercise that the aggregate demand function satis…es the weak
axiom. We can thus conclude that in order for an aggregate demand function
to satisfy the weak axiom, it is not necessary that the matrix C(p; w) is
positive semide…nite.
(d) For each of the two cases w > 16 and 8 < w < 16, draw a picture in the (x1 ; x2 )
plane depicting each consumer’s consumption bundle and his wealth expansion
path for the prices p1 = p2 = 1.
Answer: Here is a …gure depicting the wealth expansions paths of the two con-
sumers for p = (1; 1). They intersect each other at (4; 4) because x1 (1; 1; 8) =
x2 (1; 1; 8) = (4; 4).

X2
consumer2

4
consumer1

4 X1

Note that if 8 < w < 16, the Engel curves resemble those of the right-hand

11
…gure; while if w > 16, the Engel curve resemble those of the left-hand …gure.

12
 Chapter 5 - Production Theory
End-of-Chapter Exercises - Answer key

1. Consider a single-output technology with production function f : R2+ ! R+ , given

1=4 1=4
by f (z) = 23=4 z1 z2 .

(a) Check if the production function has nonincreasing, nondecreasing, or constant

returns to scale.

– Nonincreasing returns to scale. If the production function satis…es nonin-

creasing returns to scale, for all inputs z 2 R2+ and for all 2 [0; 1], we must
have f (z) f ( z), which is this exercise implies
1=4 1=4
23=4 z1 z2 23=4 ( z1 )1=4 ( z2 )1=4

Simplifying the right-hand side, we obtain

1=4 1=4 1=4 1=4
23=4 z1 z2 1=2
23=4 z1 z2

which is satis…ed for all 2 [0; 1], and hence this production function satis…es
nonincreasing returns to scale.
– Nondecreasing returns to scale. If the production function satis…es nonde-
creasing returns to scale, for all inputs z 2 R2+ and for all 2 [0; 1], we must
have f (z) f ( z), which in this exercise implies
1=4 1=4 1=4 1=4
23=4 z1 z2 1=2
23=4 z1 z2

and we know that this cannot hold for any 2 [0; 1]. Then this production
function cannot exhibit nondecreasing returns to scale.
– Constant returns to scale. If the production function satis…es constant re-
turns to scale, it must satisfy nonincreasing and nondecreasing returns to
scale. Since this production function does not satisfy both, it cannot exhibit
constant returns to scale.
– Note: This production function is a standard Cobb-Douglas production func-
tion f (z) = Az1 z2 . It could be good to remember that when + < 1 the
production function has nonincreasing returns to scale, when + > 1 it
has nondecreasing returns to scale, and when + = 1 it exhibits constant
returns to scale.

b. Let w 2 R2++ denote the input price vector and p > 0 the output price. Determine
for each output level q > 0 the cost function c(w; q) and the conditional factor
demand z(w; q).

1
– We …rst need to …nd the conditional factor demand (solving the cost min-
imization problem, CMP, of the …rm), and afterwards we can compute the
…rm’s cost function.
CMP : min w z
z 0

1=4 1=4
s.t. 23=4 z1 z2 q,
z 0

First, note that z = 0 can be ruled out. Indeed, from the production function
we know that output would be zero when either of the inputs are zero, i.e.
1=4 1=4
z1 = 0 or z2 = 0. Secondly, the production function constraint 23=4 z1 z2
q must be binding since the production function is increasing in both inputs
and inputs are costly (they are not free). Thus, we can solve for z1 in this
constraint,
1=4 1=4 1 q4
23=4 z1 z2 = q () z1 =
8 z2
and substitute this result into the previous cost minimization problem,

1 q4
min w1 z1 + w2 z2 = w1 + w2 z2
z2 8 z2

The …rst order condition with repect to z2 is

1 q4
w1 + w2 = 0
8 z22

and solving for z2 , r

1 w1
z2 = q 2
2 2w2
Substituting z2 into the expression for z1 we found above,
r
1 q4 1 q4 1 2 w2
z1 = =) z1 = q = q
8 z2 8 1 q 2 w1 2 2w1
2 2w2

Therefore, the conditional factor demand is

r r
1 2 w2 1 2 w1
z(w; q) = q ; q
2 2w1 2 2w2

As a consequence, the cost function is

c(w; q) = w1 z1 (w; q) + w2 z2 (w; q)

r r
1 2 w2 1 2 w1
= w1 q + w2 q
2 2w1 2 2w2
1 2p
= q 2w1 w2
2

2
c. Verify Shephard’s lemma.

– Let us …rst recall Shephard’s lemma: If the production set is closed and
satis…es the free-disposal property, and the conditional factor demand z(w; q)
consists of a single point z, then the cost function c(w; q) is di¤erentiable with
respect to w at w, and this derivative is

@c(w; q)
= zl
@wl
Hence, in order to verify Shephard’s lemma, we must …rst check that the
production set Y is closed, and that it satis…es the free disposal property.
– Closedness. The production set associated with the production function is
given by
Y = ( z; q) 2 R3 : q f (z) and z 2 R2+
and for convenience, we can rewrite this set as

Y = y 2 R3 : y 1 0 \ y 2 R3 : y 2 0 \ y 2 R3 : y 3 f ( y1 ; y2 )

which is the intersection of three closed sets (the …rst two representing inputs,
and the third representing output), and as a consequence it is closed. [Recall
that the intersection of …nitely many closed sets is also closed.]

q
(z’, f(z’))
(z, f(z))

(z’, q’)

Z’ Z z

Free-disposal. Let ( z; q) 2 Y and ( z 0 ; q 0 ) ( z; q). This means that either

0 0 0
z1 z1 , or z2 z2 , or q q; or all. In order to show that the free-disposal
property is satis…ed, we must show that ( z 0 ; q 0 ) also belongs to the production
set Y . Since the production function f ( ) is weakly increasing in inputs, we …nd
that
q 0 q f (z) f (z 0 )
that is, ( z 0 ; q 0 ) also belongs to the production set Y .
Hence, the production set Y is closed and satis…es free-disposal, so all conditions
for Shephard’s lemma are satis…ed. [Note that there is an additional condition,
which states that conditional factor demand correspondences consist of a single

3
 point (they are functions); and this was clearly satis…ed in our exercise]. We can
thus determine the conditional factor demand function from the cost function,
p r
@c(w; q) @ 12 q 2 2w1 w2 1 2 w1
= = q
@w1 @w1 2 2w2
p r
@c(w; q) @ 12 q 2 2w1 w2 1 2 w2
= = q
@w2 @w2 2 2w1

(a) d) Determine the pro…t function (p; w).

To determine the pro…t function, we can solve the pro…t maximization problem
using the cost function,
1 2p
max pq q 2w1 w2
q 2
s.t. :q 0
The Kuhn-Tucker conditions are
p
p q 2w1 w2 + = 0
q = 0

when q > 0 in an interior solution, then = 0; and using the …rst condition, we
get the optimal quantity,
p
q =p
2w1 w2
And the pro…t arising from production this output is
1 p
(p; w) = pq (q )2 2w1 w2
2
p2
= p
2 2w1 w2
Again we can see that since p > 0 and w > 0, the pro…t from producing q is
always positive, and it is therefore never optimal to set q = 0 (which gives zero
pro…ts).
Su¢ ciency: We must check that the cost function is strictly convex in q,

@c(w; q) p @ 2 c(w; q) p
= q 2w1 w2 and = 2w1 w2 > 0
@q @q 2
Hence we can conclude that the optimal output q found above is the correct one,
and the pro…t function (p; w) associated with this output level is also correct.

4
2. [Nonincreasing Returns to Scale] Consider a single-output technology with pro-
duction function f : RL 1 ! R satisfying f (0; 0; :::; 0) = 0. The associated production
set is
Y = ( z; q) 2 RL : q f (z) and z 2 RL 1
Recall that the production set has Y has nonincreasing returns to scale if

for every y 2 Y , and for every 2 [0; 1] , y also belongs to Y (1)

Alternatively, other textbooks de…ne nonincreasing returns to scale over the production
function by stating that the production function f ( ) has nonincreasing returns to scale
if
for every z 2 RL 1 and for every t > 1, f (tz) t f (z) (2)
That is, multiplying all inputs z1 ; z2 ; :::; zL 1 by a factor t > 1, we obtain an output
that is lower than t times the output given by z. We now show that both de…nitions
are indeed equivalent.

(a) Assume that

Y = ( z; q) 2 RL : q f (z) and z 2 RL 1

has nonincreasing returns to scale. Show for every z 2 RL 1

and for every 2
[0; 1], f (z) f ( z).
Given that y = ( z; f (z)) 2 Y , where f (z) = q, expression (1) about nonin-
creasing returns to scale can be rewritten as

for all ( z; f (z)) 2 Y and 2 [0; 1] we have that ( z; f (z)) 2 Y

That is, if the bundle ( z; f (z)) belongs to the production set Y , then its
downward scale ( z; f (z)) since 2 [0; 1], also belongs to the production
set Y . Therefore, f (z) f ( z).is satis…ed, which is exactly what we had
to show.
(b) Show that for every z 2 RL 1 and for every t > 1, f (tz) t f (z). [Hint: Use
your result from the previous part, and make z 0 = tz and 0
= 1t ].
Applying the result of the previous part, we have that

for every z 2 RL 1
and 2 [0; 1] , we have f (z) f ( z)

Let us take z 0 = tz and 0 = 1t , given that t 1, we can achieve to have

2 [0; 1] in order to be able to apply our result from section (a),

for all z 0 2 RL 1
and 0
2 [0; 1] , we have 0
f (z 0 ) f ( 0z0)

Substituting now for z 0 = tz and 0

= 1t ,

1 1
for all z 2 RL 1
and 2 [0; 1] , we have f (tz) f tz
t t
1
() f (tz) f (z) () f (tz) tf (z)
t

5
 Hence, we have shown that

for all z 2 RL 1
and 2 [0; 1] , we have f (tz) tf (z)

Therefore, we have shown that the de…nition of nonincreasing returns to scale

in (1) implies the de…nition in (2). In the next section of this exercise we show
the converse.
(c) Note that combining parts (a) and (b), we have shown that (1) implies (2). Hence,
now we must prove the converse. Assume that (2) holds. Show that Y satis…es
(1). Distinguish two cases: = 0, and 2 (0; 1].

– Let us recall what the de…nition of nonincreasing returns to scale in (2) states:

for every z 2 RL 1
and for every t > 1, f (tz) t f (z)

Assuming that (2) holds, then we must show that the following is also satis…ed

for every z 2 RL+ 1

and 2 [0; 1] then f (z) f ( z)

– First case = 0. Let us take ( z; q) 2 Y and recall that ( z; q) 2 Y

for any 2 [0; 1] as we showed in part (a). If = 0, then ( z; q) =
(0; 0; :::; 0) 2 Y since by de…nition the output of employing zero amount of
all inputs is also zero, i.e., f (0; 0; :::; 0) = 0:
– Second case 2 (0; 1]. By the de…nition of the production set Y , ( z; q) 2 Y
means that f (z) q. As a consequence, f (z) q. Taking z 0 = z and
t0 = 1 and inserting them into (2), we have

f (t0 z 0 ) t0 f (z 0 )

Substituting,

1 1
f z f ( z) () f (z) f ( z)

which is exactly what the de…nition of nonincreasing returns to scale in (1)

states.

3. [Increasing returns to scale]. Show that if a production function f : RL 1

!R
satis…es increasing returns to scale

for every z 2 RL 1
and for every t > 1, f (tz) > t f (z)
z0
then it also satis…es increasing average product. [Hint: Use t = z
, and then multiply
both sides of the inequality f (tz) > t f (z) by z10 ].

6
 0
Let t = zz , so we have z 0 z, since t 1 then the de…nition of increasing returns
to scale becomes
z0 z0
f z > f (z)
z z
1
by multiplying both sides now by z0
, we have

1 z0 1 z0
f z > f (z)
z0 z z0 z

which reduces to
f (z 0 ) f (z)
>
z0 z
which exactly represents incrasing average product.

4. Give a counterexample of a production function which satis…es increasing average prod-

uct, but does not satisfy increasing marginal product. [For simplicity, you can use a
production function f : R ! R producing a single output with a single input, so that
you can represent it in R2 . Then, note that average product at a given point is graph-
ically represented by the slope of the rays connecting the origin and the production
function at that point. In contrast, the marginal product is simply represented by the
slope of the production function at that point].

The property of increasing returns to scale (which implies increasing average prod-
uct) is shown in the slope of the blue rays connecting points of the production
function f (z) with the origin: the slope of these rays increases in z. But as we
can see, the slope of the production function, f 0 (z), (in red in the …gure) is in fact
decreasing in some intervals, indicating decreasing marginal product. Hence, this
example shows that increasing returns to scale (increasing average product) does
not imply increasing marginal product.

f(z)

f’(z)

5. [Deriving the supply function from the pro…t function] Consider the following
pro…t function that has been obtained form a technology that uses a single input:

(p; w) = p2 w

where p is the output price, w is the input price and is a parameter value.

7
(a) Check if the pro…t function satis…es: (1) non-decreasing in output price p, (2)
non-decreasing in input prices w, (3) homogeneous of degree one, (4) convex in
prices p and w. In particular, determine for which values of these properties
are satis…ed (some properties might be satis…ed for all values of , while others
might hold only for certain values of ).
The pro…t function has to be homogeneous of degree one. Thus,

( p; w) = (p; w)

In this case we have:

2+
( p; w) = ( p)2 ( w) = p2 w (3)

and, on the other hand,

(p; w) = p2 w (4)
since, by homogeneity, expressions (2) and (3) must coincide. Then,
2+
p2 w = p2 w

which implies that 2 + = 1. That is, we need = 1. As a consequence,

the pro…t function that we obtain is
p2
(p; w) =
w
In this case the pro…t function satis…es the following the properties:
1. Continuous: this property holds for every value of w =
6 0.
@ (p;w) 2p
2. Non decreasing in the output price, p: @p = w 0
2
3. Non increasing in the factor prices: @ @w
(p;w)
= wp 2 0
4. Homogeneous of degree 1: if = 1 then ( p; w) = (p; w)
5. Convex in prices, factor prices and output prices:
@ 2 (p;w) @ 2 (p;w) 2 2p
@p2 @p@w w w2
@ 2 (p;w) @ 2 (p;w) = 2p 2p2 =0
@w@p @w2 w2 w3

Since the Hessian is a positive semi-de…nite matrix, the function (p; w) is

convex.
(b) Calculate the supply function of the product, q(p; w), and the demand for inputs,
z(p; w).
Using Hotelling’s Lemma we can …nd the supply function,
@ (p; w) 2p
q(p; w) = =
@p w
and the conditional factor demand correspondence
@ (p; w) p2
z(p; w) = = 2
@w w

8
6. [Law of Supply] Consider a …rm with production set Y and a price vector p 2 RL++ .
Suppose the price of one commodity l 2 f1; 2; :::; Lg is increased by an amount " > 0.
Let y 0 2 Y be a pro…t maximizing at the old prices and y 1 2 Y at the new prices.
Prove that yl1 > yl0 , i.e., the net supply of commodity l weakly increases as its price
increases.

First note that since y 0 is pro…t maximizing at the old prices p0 , then

p0 y 0 p0 y 1

and since y 1 is pro…t maximizing at the new prices p1 , then

p1 y 1 p1 y 0

Adding these inequalities, we have

p0 y 0 + p1 y 1 p0 y 1 + p1 y 0

Rearranging,
p1 p0 y1 p1 p0 y0
p1 p0 y1 y0 0 (Law of Supply)
Note that only the price of one commodity (good l) is increased (p1l p0l ) > 0,
while that the price of all the other commodities remain constant, i.e. (p1k p0k ) =
0 for any good k 6= l. Moreover, we know that the price change for good l
is (p1l p0l ) = ". Then the previous expression from the law of supply can be
reduced to
p1l p0l yl1 yl0 = " yl1 yl0 0
and since " > 0, then yl1 yl0 , i.e., the net supply of commodity l weakly increases
as its price increases.

7. [Properties of the pro…t function] The pro…t function, (p), is de…ned as

(p) = max fp y j y 2 Y g

or alternatively, (p) > p y for every y 2 Y .

(a) Show that the pro…t function (p) is convex in prices.

We need to show that, for any 2 [0; 1] and p 2 RL++ ,

( p + (1 )p0 ) (p) + (1 ) (p0 )

Proof. From the de…nition of the pro…t function we know that (p)is the
maximal ofp y, so we have

(p) p y, for any y 2 Y and p >> 0, and

(p0 ) p0 y, for any y 2 Y and p0 >> 0

9
 And similarly, using that the pro…t function is homogeneous of degree 1 in
prices, we can take any 2 [0; 1],

(p) p y, for any y 2 Y and p >> 0, and

(1 ) (p0 ) (1 ) p0 y, for any y 2 Y and p0 >> 0

Adding up the two previous inequalities,

(p) + (1 ) (p0 ) p y + (1 ) p0 y = [ p + (1 ) p0 ] y

where the right-hand side [ p + (1 ) p0 ] y coincides with the pro…t function

for price level p + (1 ) p0 , i.e., ( p + (1 ) p0 ). Hence,

(p) + (1 ) (p0 ) ( p + (1 ) p0 )

and therefore we can conclude that the pro…t function is convex in prices.
(b) Prove the Hotelling’s lemma using the Duality theorem. [Hint: easy, just rewrite
the duality theorem changing labels for Hotelling’s Lemma.]
Hotelling’s lemma states that if the output function evaluated at prices p,
y(p), consists of a single point, then the pro…t function (p) is di¤erentiable
at the price level p; and moreover such derivative is rp . (p) = y(p).
Let us …rst express the pro…t function as the support function that, for every
price vector p, chooses the in…mum of p ( y), i.e., instead of choosing the
max of p y, we rede…ne it as the inf of p ( y).

(p) = inf fp ( y) jy 2 Y g

In order to emphasize the similarities with the Duality Theorem, we reproduce

it here again: Let K be a nonempty and closed set, and let K (p) be its
support function,that is K (p) = inf fp xjx 2 Xg , . Then, there exists a
unique element x in the set K such that K (p) = p x if and only if K (p)
is di¤erentiable at p. Moreover, rp K (p) = x.
Therefore, given that we have noticed that the pro…t function can be ex-
pressed as a support function, we can rewrite Hotelling’s lemma as a direct
application of the Duality Theorem.(just changing labels!): Let Y be a non-
empty and closed (production) set, and let Y (p) be its support function.
Then, there exists a unique production function y (p) in the set Y such
that (p) = p y (p) if and only if (p) is di¤erentiable at p. Moreover, this
derivative is rp (p) = y (p).
(c) Show that, if the output function y(p) is di¤erentiable at p, then
1. Dp y(p) is a symmetric and positive semide…nite matrix.
[2 points] Symmetry: This comes from two sources. First, own-substitution
e¤ects are nonnegative, i.e., @y@pl (p)
l
0 for all l. Second, using Young’s
@yl (p) @yk (p)
Theorem, cross-substitution e¤ects are symmetric, i.e., @pk
= @pl
for
all l 6= k:

10
 [2 points] Positive semide…nite matrix : It is the mathematical expression
of the law of supply: quantities respond in the same direction as price
changes. That is given that there are no wealth e¤ects,
(p0 p) (y 0 y) 0
2. Dp y(p) p = 0.
This result is an inmediate consequence of the fact that y(p) is homoge-
neous of degree zero, H(0). Indeed, note that,
y( p) = y (p) for all p 2 RL+ and for all >0
which implies that y( p) y (p) = 0. Taking derivatives with respect to
, and evaluating this derivate at prices p = p, we obtain Dp y(p) p = 0.

8. [Duality Theorem].Prove the duality theorem when the set K is …nite. [Hint: Note
that the minimizer x 2 K in the support function can be either unique or multiple.
Both cases are straightforward. You can help your discussion with a …gure for the case
of a unique and multiple minimizer(s).]

If there is a unique minimizer, then a small change in prices would lead to the
same minimizer, and the Duality Theorem would still hold, i.e., a small change
in prices, rp K (p), leads to the same minimizer, x, rp K (p) = x.
If there are multiple minimizers, then a small change in prices would not necessar-
ily lead to the same minimizer (it would depend on the direction of the change),
and therefore rp K (p) 6= x. For a graphical illustration, take a set K with just
two bundles in R2+ , e.g. (0; 1) and (0; 5).
x2
p p’

(0,1) (0,5) x1

Note that for a certain price vector the …rst bundle may be cost-minimizing,
whereas when prices are changed the second bundle may become cost-minimizing
(“jump”).

9. [Shephard’s lemma]State and prove Shephard’s lemma. [Hint: Use the Duality
Theorem].

Shepard’s lemma: Let us assume that the production set Y satis…es free-disposal
and is closed. If the conditional factor demand correspondence z(w; q) resulting
from the cost minimization problem consists of a single element, then the …rm’s
cost function c(w; q) is di¤erentiable with respect to w evaluated at w = w, and
this derivative is rw c(w; q) = z(w; q).

11
 Proof. [Here we prove it using the Duality Theorem, but recall that it can also
be proved by using …rst order conditions, and by the Envelope Theorem]. The
…rm’s cost function can be understood as the support function of the feasible set
F (y) 0,
c(w; q) = inf fw q : z 2 F (y)g
And hence we can denote this expression as the support function Y (w). Applying
the Duality Theorem to this support function, we have:
Let Y be a nonempty and closed production set Y , and let Y (w) = c(w; q)
be its support function. Then, there exists a unique conditional factor demand
correspondence z(w; q) 2 Y such that c(w; q) = w z(w; q) if and only if Y (w)
is di¤erentiable at w; moreover, rw c(w; q) = z(w; q). This is exactly what the
Shephard’s lemma states.

10. [Applying Shephard’s lemma] Suppose that a …rm can produce a single unit of a
good by employing any set of N workers from a superset of M workers (N workers
are hired, and M N are not, where M > N ). There is no other input. Worker i’s
reservation wage is wi > 0. Assuming that w1 6= w2 6= ::: 6= wM , establish an analog
of the Shephard’s lemma. [Hint: First assume that w1 < w2 < ::: < wN < wN +1 <
::: < wM , where the …rm hires only those N workers with the lowest reservation wage.
Then, state the …rm’s conditional demand correspondence of labor, zi (w), and the
…rm’s cost function c(w). Finally, check if the derivative of the latter with respect to
wages coincides with the conditional demand of labor.]

Let us …rst assume that w1 < w2 < ::: < wN < wN +1 < ::: < wM , where the …rm
hires only those N workers with the lowest reservation wage. Then, the …rm’s
conditional demand correspondence, zi (w; 1), (given that the …rm produces q = 1
unit of output) of the labor services of worker i is given by
1 if i N , since i = f1; 2; :::; N g are hired
zi (w; 1) =
0 if i > N , since i = fN + 1; :::; M g are not hired
Then, …rm’s cost function is just given by the sum of the wages of all workers who
are hired,
c(w; 1) = w1 + w2 + ::: + wN
Therefore, the derivative of this cost function with respect to the price of factors
(labor) is
@c(w; 1) 1 for every worker i N
=
@wi 0 for every worker i > N
Hence, Shephard’s lemma is satis…ed, since
@c(w; 1)
= zi (w; 1)
@wi
11. [Deriving the cost function from the pro…t function] Consider a pro…t-maximization
problem (PMP) that produces the following pro…t function
p2 p2
(p; w; r) = + ,
4w 4r

12
 where w 2 R+ denotes the wage rate, r 2 R+ represents the interest rate and p 2 R+
denotes the price of the single output that the …rm produces. Obtain the expression
of its associated cost function c(w; r; q).

First, the unconditional factor demand correspondences, l(p; w; r) and k(p; w; r),
(unconditional because they don’t depend on q) by applying Hotelling’s lemma

@ (p; w; r) p2
l(p; w; r) = = , and
@w 4w2
@ (p; w; r) p2
k(p; w; r) = = 2.
@r 4r
We can …nd the supply correspondence y(p; w; r) in a similar fashion

@ (p; w; r) 2p 2p p(w + r)
y(p; w; r) = = + =
@p 4w 4r 2wr
2wrq
And solving for p we obtain p = w+r
.
We can now plug p = 2wrq w+r
into the unconditional factor demand correspondences
in order to obtain the conditional factor demand correspondences, lc (w; r; q) and
k c (w; r; q) (using the notation in NS), as follows
2wrq 2
c w+r r2
l (w; r; q) = = q 2 , and
4w2 (w + r)2

2wrq 2
c w+r w2
k (w; r; q) = = q2
4r2 (w + r)2

which we can use to …nd the cost function

c(w; r; q) = wlc (w; r; q) + rk c (w; r; q) =

r2 2 w2 wr 2
= w 2
q +r 2
q2 = q
(w + r) (w + r) w+r

12. [Free-disposal] Consider a one-input one-output production technology. Show that

if the production set Y satis…es the condition of free-disposal, then the corresponding
production function f ( ) should be (weakly) increasing.

From the free-disposal property of the production set Y we have that: if the pro-
duction vector y = ( z; q) belongs to the production set Y , then the production
vector y 0 = ( z 0 ; q 0 ) where more units of inputs are needed z 0 z and less
units of output are produced q 0 q also belongs to production set Y .
Threfore, if ( z; f (z)) 2 Y and z 0 z, we have that ( z 0 ; f (z)) 2 Y . It
follows that
( z 0 ; f (z)) ( z 0 ; f (z 0 ))

13
 And then, f (z) f (z 0 ), which implies that production function f ( ) is weakly
increasing.
f(z)
(-z’, f(z’))

f(z’)

(-z’, f(z))

z
-z’

13. [Duality theorem when the set K is a singleton] Prove the duality theorem when
the set K is a singleton.

Given a non-empty and closed set K, and given its support function K( ), then
we have that
fthere is a unique x 2 K s.t. K (p) = p xg () f K( ) is di¤erentiable at pg ,
and moreover, we have that rp K (p) = x. [Recall the de…nition of the support
function: K (p) p x, for every x 2 K. Thus, the support function K (p)
selects, for every p, the element of the set K, x 2 K, that makes the value of p x
minimal. That is, for every p, the value that the support function gives, K (p),
guarantees that all other p x are higher.]

(a) Prove the duality theorem when the set K is a singleton. [Hint: prove each
direction of the above equivalence relationship, and then prove that rp K (p) = x.
For the …rst direction, it is useful to use the de…nition of the derivative using limits.
For the second direction, use contradiction].
First we want to show that
fthere is a unique x 2 K s.t. K (p) = p xg =) f K( ) is di¤erentiable at pg
(5)
then we will show that
fthere is a unique x 2 K s.t. K (p) = p xg (= f K( ) is di¤erentiable at pg
(6)
And …nally we will show that
rp K (p) =x (7)
Proof of (1). Given that set K is a singleton only containing element x,
i.e., K = fxg, then we need to check that the support function K ( ) is
di¤erentiable at p. Using the de…nition of the derivative
n
k K (p ) K (p)k kpn x p xk
lim = lim =
n
p !p kpn pk n p !p kpn pk
kpn pk
lim x=x
n
p !p kpn pk

14
 Proof of (2). Let us work by contradition. So, assume that the support
function K ( ) of the singleton set K is di¤erentiable at p, but there exists
more than one element in the set K, x0 2 K where x0 6= x, for which K (p) =
p x0 . But if this is true, then set K has more than one element and is not a
singleton, which is a contradiction.
Proof of (3). From the …rst step of this proof we know that the support
function evaluated at p = p, K (p), is di¤erentiable in prices. Moreover, from
the second step we know that such di¤erentiability implies that K (p) = p x.
Hence, we can di¤erentiate the support function with respect to prices and
obtain that rp K (p) = x.

14. [Aggregation in production] Consider a economy with J …rms, with production

sets Y1 ; Y2 ; :::; YJ ., where each Yj is nonempty, closed, and satis…es the free disposal
property. Assume also that every supply correspondence yj (p) is single valued, and
di¤erentiable in prices, p >> 0. Furthermore, let Y be the aggregate production set,
( )
XJ
Y = Y1 + Y2 + ::: + YJ = y 2 RL : y = yj
j=1

P
for some yj 2 Yj and where j = 1; 2; :::; J. [Note that y = Jj=1 yj , where every yj is just
a feasible production plan of …rm j, but not necessarily …rm j’s supply correspondence
yj (p)]. Show that:

(a) The pro…t function of the aggregate production set Y , (p) coincides
PJ with the
sum of the individual pro…t functions for all …rms in the economy, j=1 j (p):
Note that if we take any collection of production plans yj 2 Yj for every
P
…rm j = 1; 2; :::; J, then Jj=1 yj 2 Y . Because (p) is the pro…t function
associated with the aggregate production plan Y , we therefore have
!
XJ XJ
(p) p yj = p yj
j=1 j=1
| {z }
PJ
j=1 j (p)

PJ
Hence, it follows that (p) j=1 j (p). Now we must show the statement
in the other direction. Consider any y 2 Y . By the de…nition of the aggregate
production
P set Y , there are yj 2 Yj ; for every …rm j = 1; :::; J, such that
j yj = y. So
!
X X X
p yj = p yj j (p) for all y 2 Y .
j j j
| {z }
(p)=p y
P
Thus
P (p) j j (p). Together, these two inequalities imply that (p) =
j (p).

15
 (b) The supply correspondence of the aggregate production set Y , y (p), coincides
with the
PJ sum of the individual supply correspondence for every …rm in this econ-
omy, j=1 yj (p):
P
For the second
P equality, we must show that j yj (p) y (p) and that
y (p) j yj (p). For the former relation, consider any set of individual
production plans yj 2 yj (p) ; for every …rm j = 1; :::; J. Then
!
X X X
p yj = p yj = j (p) = (p) ,
j j j

where
P the last equality followsP from part (i) of the proposition. Hence,
j yj 2 y (p), and therefore, j yj (p) y (p).
P
In the other direction, take any y 2 y (p). Then y = j yj for some
yj 2 Yj ; for every …rm j = 1; :::; J. Since
!
X X
p yj = (p) 2 j (p)
j j

and, for every …rm j, we have p yj j (p), it must be thatP

p yj = j (p) for
every …rm j. Thus yj 2 yj (p)P for all …rms j, and so y 2 j yj (p). Thus,
we have shown that y (p) j yj (p).

(c) What is the intuition behind these two results?

The result tells us that if …rms are maximizing their individual pro…ts facing
output price p and factor prices w, then their supply Pbehavior maximizes
aggregate (joint) pro…ts. But this means that if q = j yj is the aggregate
output produced by the …rms, the the total cost of production is exaclty
equal to c(w; q), the value of the aggregate cost function (the cost function
corresponding to the aggregate production set Y ). Thus, the allocation of the
production of output level q among the …rms is cost minimizing. In summary,
if …rms maximize pro…ts taking prices as given, then the sum of their pro…ts
coincides with the maximum pro…ts they could achieve by jointly maximizing
their pro…ts coordinating their actions.
(d) Are your results (or their interpretation) dependent on any convexity hypothesis
about the production sets Y1 ; Y2 ; :::; YJ .?
No, as the proofs in parts (a) and (b) illustrate, we do not need to use
convexity in any step of the proof.

15. A …rm can produce one output q using two inputs called z1 and z2 by means of two
di¤erent technologies. Technology 1 is represented by the production function q =
minfz1 ; z2 g for all z1 ; z2 0. Technology 2 is represented by the production function
q = z31 + z32 for all z1 ; z2 0. Input prices are w1 ; w2 0.

(a) Does Technology 1 exhibit constant returns to scale? What about Technology 2?

16
 Answer: For Techonology 1, q = min fz1 ; z2 g, we have that either q = z1 (if
z1 z2 ), or q = z2 (if z1 z2 ). In the …rst case (when z1 z2 ), if we increase all
inputs by a common factor , we obtain an output of min f z1 ; z2 g = z1 ; = q,
for any 0. (See …gure) Similarly, in the second case (when z1 z2 ), increasing
all inputs by the same factor yields min f z1 ; z2 g = z2 ; = q. We can thus
conclude that increasing all inputs by a common factor induces an increase in
output of exactly , and therefore Technology 1 exhibits constant returns to scale.

Technology 1 Technology 2

For Technology 2, an increase in all inputs by a common factor yields f ( z1 ; z2 ) =

z1
3
+ 3z1 = z1
3
+ z31 = f (z1 ; z2 ), and thus this technology also exhibits con-
stant returns to scale. Note that in this case, the associated isoquant becomes
z2 = 3q z1 , as depicted in the above …gure.
(b) Derive the cost function for both of them. [Hint: You do not need to set up the
Lagrangian, using a nice …gure accompanying with your explanation is enough].
Answer: Technology 1 represents that the …rm uses inputs 1 and 2 in constant
proportions (i.e., the …rm’s isoquant show a kink at z1 = z2 ). This implies that, if
the …rm wants to minimize costs, it will do so by selecting the lowest isocost that
“touches” the kink of isoquant associated to production level q. (Note that we
say that the isocost “touches”the isoquant rather than saying that it is “tangent”
to the isoquant because in this case the isocost cannot be tangent to the kink,
since the slope of the isoquant is not well de…ned at the kink.)

17
Hence, the …rm uses the same amount of both inputs, z1 = z2 . In addition, the
…rm uses one unit of each input in order to produce one unit of output. Therefore,
the cost function of reaching output level q is:
c(w; q) = w1 z1 + w2 z1 = (w1 + w2 )z1 = (w1 + w2 )q

Technology 2 represents a linear production function. The …rm’s isoquants are

therefore straight lines. We can hence anticipate that the …rm can operate at
either of the corners (using only input 1 or only input 2), or at a continuum of
cost-minimizing input combinations. Let us next analyze each case separately.
1. If w1 > w2 , then isocost lines are steeper than the isoquant associated to
production level q. (You can draw a …gure in order to con…rm this graphically).
Thus, the …rm uses input 2 only, i.e., the …rm only uses the cheapest input since
both inputs are perfectly substitutable in the production process. Hence, when
the …rm reaches production level q, it must be that q = 30 + z32 , or z2 = 3q. This
implies that the …rm’s cost function is c(w; q) = 3qw2 .

2. If w1 < w2 , then isocost lines are ‡atter than the isoquant associated to
production level q. Thus, the …rm uses input 1 only, i.e., the …rm only uses the
cheapest input since both inputs are perfectly substitutable in the production
process. Hence, when the …rm reaches production level q, it must be that q =
z1
3
+ 03 , orz1 = 3q. This implies that the …rm’s cost function is c(w; q) = 3qw1 .

18
 3. Finally, if w1 = w2 , then isocost lines have the same slope as the isoquant
associated to production level q. Graphically, the isocost associated to the lowest
cost that reaches production level q totally overlaps the isoquant representing that
production level. The …rm can therefore choose any input combination (z1 ; z2 )
along the isocost. Intuitively, this implies that the …rm is choosing any linear
combination between the extreme case in which it only uses input 2 (as in case 1
above) and the other extreme case where it only uses input 1 (as in case 2 above).
Hence, the cost function of the …rm is just a linear combination of the above two
cases: c(w; q) = 3qw2 + (1 )3qw1 .

(c) Suppose that input prices satisfy w1 < w2 . Suppose also that the …rm wants
to produce some amount of output q. For which values of w1 will the …rm use
technology 1, and for which values of w1 will the …rm use technology 2?
Answer: This exercise asks for a comparison of the costs under Technologies 1
and 2. Since the …rm’s costs under Technology 2 depend on input prices, we must
divide our analysis into 3 cases:
1. If w1 < w2 , we know from exercise (b) that Technology 2 implies costs of
c(w; q) = 3qw1 . Comparing this cost with that under Technology 1, c(w; q) =
(w1 + w2 )q, yields that the …rm prefers to use Technology 1 if its associated costs
are lower, i.e., 3qw1 > (w1 + w2 )q, or simply w1 > w22 . That is, the …rm prefers
Technology 1 if w2 > w1 > w22 . If, instead, w1 < w22 < w2 , then the …rm prefers
Technology 2.
2. [In this question you didn’t need to consider cases 2 and 3 below but, for
completeness, I include them here.] If w1 > w2 , we know from exercise (b)
that Technology 2 yields costs ofc(w; q) = 3qw2 . Comparing this cost with that
under Technology 1, c(w; q) = (w1 + w2 )q, we obtain that the …rm prefers to use
Technology 1 if its associated costs are lower, i.e., 3qw2 > (w1 + w2 )q, or simply
w2 > w21 . That is, the …rm prefers Technology 1 if w1 > w2 > w21 . If, instead,
w2 < w21 < w1 , then the …rm prefers Technology 2.
3. If w1 = w2 ; we know from exercise (b) that Technology 2 yields costs of
c(w; q) = 3qw2 + (1 )3qw1 . Comparing this cost with that under Technology
1, c(w; q) = (w1 + w2 )q, we obtain that the …rm prefers to use Technology 1
if its associated costs are lower, i.e., 3qw2 + (1 )3qw1 > (w1 + w2 )q, or

19
 simply w1 > w22(1 3 3) . [This result is dependent on the precise value of , i.e.,
the particular combination of inputs that the …rm selects along the isocost line.
Nonetheless, if we consider that, for instance, is close to 1, then all our discussion
from point (1) above can be extended to this case as well.]

16. [Pro…t maximization and Mergers] An economy is comprised of N identical …rms

(N > 1) producing a single output by means of a single input. For every …rm i =
1; : : : ; N , the amount of input that …rm i needs in order to produce qi units of output
depends not only on qi but also on the output of all …rms, according to the expression

c(qi ; q i ) = (qi + q i ) qi (8)

P
where q i j6=i qj represents the aggregate output of all other …rms, and parameters
and satisfy > 0 and + > 0. In addition, assume that c(0; 0) = 0, and let the
price of the single input be normalized to one.

(a) Interpret expression (1) in words for the various cases de…ned by the signs of
and of . [Hint: Interpret = 0 and > 0. Then, for the case in which = 0,
analyze the case in which > 1 and < 1]
Answer: There is a production externality,P because the physical cost of …rm
i depends on the aggregate output q = N j=1 qi through the term(qi + q i )
The parameter re‡ects the externality. If > 0, then the externality is
negative, because an increase in aggregate output increases the cost of …rm
i: perhaps all these …rms are using an exhaustible natural resource, or there
is some form of congestion among them. If, on the contrary, < 0 , then the
externality is positive (because, say, knowledge spillovers). Note that …rm i
contributes to these e¤ects not only on the other …rms but also on itself, and
that the externality across …rms depends only on aggregate output q, and not
in the manner in which q is allocated among …rms.
The value = 0corresponds to the traditional case of absence of production
externalities.
When = 0, the cost function becomes c(qi + q i ) = qi : In this case, the
parameter depicts the scale economies within the …rm, with > 1corresponding
to increasing returns to scale (the cost function is strictly convex), and <
1corresponding to increasing returns to scale (cost function strictly concave).
When 6= 0, this interpretation is not exact, because qi appears in the term
(qi + q i ) , and needs to be made conditional to keeping aggregate output
constant.
(b) Suppose that the N …rms merge into a single …rm with N plants, each with
the technology de…ned by expression (1) above. Find the cost function C(q),
de…ned as the minimum amount of input that the merged …rm needs in order to
produce q units of output. In other words, C( ) is the physical cost function of
the merged …rm. We assume that the merged …rm is a price taker. Separately
analyze the merged …rm’s cost-minimization problem in the case in which < 1,
= 1, and > 1. You can accompany your analysis with a …gure of the …rm’s
cost-minimization problem in each case.

20
Answer: C(q) is the value function of the problem
XN XN
min (q1 + ::: + qN ) (qi ) subject to qi q; (2.2)
(q1 ;:::;qN ) i=1 i=1

Its solutions are those of the problem

XN XN
min q (qi ) subject to qi q (2.3)
(q1 ;:::;qN ) i=1 i=1

Where, for a given output level q, the …rst term in the objective function,q
, operates as a constant.
Case 1: < 1. For …xed q, < 1entails increasing returns to scale, i.e.,
economies of scale within each plant. Intuitively, the merged …rm will then
concentrate its production in one plant. Figures 2.1-2.3 below depict, for a
given isoquantq1 + q2 = q (see straight line in the middle of every …gure),
and a given set of isocost curves, which isocost curve reaches the isoquant
q1 + q2 = qat the lowest cost. (The shaded area in the …gures represents
combinations of q1 and q2 that yield aggregate output levels above q, i.e.,q1 +
q2 q .) Importantly, isocost lines become bowed in (away) from the origin
when economies (diseconomies, respectively) of scale arise.
Formally, the objective function in Problem (2.3) is strictly quasiconcave,
with corner solutions, see Figure 2.1 below (in the left-hand side), i. e., any
vector of the form qi = q, qj = 0 for j 6= i solves Problem (2.3), and hence,
Problem 2.2. The value function of Problem (2.2), or, in other words, the
physical cost function of the merged …rm is then
+
C(q) = q (q + 0 + ::: + 0) = q

P
Case 2: = 1. Problem (2.2) reduces now to the minimization of N i=1 qi
PN
subject to i=1 qi = q i.e., any allocation of the output q among the N …rm
minimizes costs: see Figure 2.2. The physical cost function of the merged

21
 +
…rm is then C(q) = q q = q .

Case 3: > 1. This case corresponds to increasing returns to scale, i.e.,

diseconomies of the scale within each …rm. Intuitively, the merged …rm will
then allocate output equally among the (identical) plants. Formally, the
objective function in Problem (2.3) is strictly quasiconvex, with a unique
interior solution. (see the next …gure). Because of symmetry, qi = Nq (the
…rst-order equalities of Problem 2.2) are (qi ) 1 = : Hence, the physical
cost function of the merged …rm is then C(q) = q N ( Nq ) = N 1 q +

(c) Write the merged …rm’s pro…t-maximization problem (as a function of input price,
$1, and a given output price, p), i.e., P M P [1; p]. For which values of parameters
, and p does the problem have a solution? Explain, and solve the P M P [1; p]
for these values, denoting the solution by q M , where superscript M indicates the
merged …rm. [Hint: separately analyze the case in which + < 1, + = 1,
and + > 1, and relate them with the presence of increasing, constant, or
decreasing returns to scale.]
The pro…t-maximization problem of the merged …rm is: maxq pq C(q):]

22
Substituting the cost function C(q) from that obtained in the previous exer-
cise for cases (1)-(3), we obtain:
+
max pq q in case (1), < 1.and (2), i.e., = 1 and
q

max pq N1 q +
in case (3), i.e., >1
q

Existence of solutions to the PMP: Note that the (physical) average cost of
the merged …rm is

q + 1
in cases 1,2,
C A (q) =
N1 q + 1
in case 3,

and the (physical) marginal cost of the merged …rm is

( + )q + 1
in cases 1,2,
C M (q) =
( + )N 1 q + 1
in case 3,

If + < 1, then the average cost of the merged …rm is decreasing, and its
technology displays increasing returns to scale, which are incompatible with
pro…t-maximizing, price-taking behavior. Thus, no solution to the pro…t
maximization problem of the merged …rm exists in this case.
If + = 1, then the average cost of the merged …rm is constant, and its tech-
nology displays constant returns to scale. Pricing-taking pro…t maximization
then requires p C A (q); i.e., either p < C A (q) and q = 0 in the case of corner
solutions, or p = C A (q) and q 2 R for interior solutions.
If + > 1 , then the average cost of the merged …rm is increasing, and
its technology displays decreasing returns to scale. Because, in this case, the
marginal cost is strictly increasing and satis…es C M (0) = 0 and lim C M (q) =
q!1
1;the equation “pricing=marginal cost”has a unique solution. In cases 1 and
2, the equation is with solution whereas in Case 3 the equation is p ( +
1
p + 1
+ 1 M
)q , with solution q = +
, whereas in Case 3 the equation is

p = ( + )N 1 q + 1
; (2.4)
1
p + 1
M 1
with solution q = N +
:
See Figure which, for parameters + > 1 and > 1separately depicts the
left-hand side of equation (2.4), p, and its right-hand side, illustrating the

23
 pro…t-maximizing output level that solves (2.4).

(d) Suppose now that each of the N …rms in an independent, price-taking, pro…t-
maximizing decision maker. Here and in what follows, we assume that parameters
satisfy + > 1, and that > 1. Write …rm i’s P M P [1; p], and its …rst-order
conditions. Rewrite the …rst order equality, as a function of aggregate output q,
under the symmetry assumption that qi = 1; :::; N , and solve it for the aggregate
output, denoting its solution by qb.
The pro…t-maximization Problem of the i …rm is

maxqi pqi -(qi + q i ) qi ;

with …rst-order equality

1 1
p = (qi + q i ) qi + (qi + q i ) qi ;

which, under symmetry becomes

1 q q 1
+ 1
p= q +q =q ( N + N1 )
N N
which simpli…es to

p= + N1 q + 1
: (2.5)
N
1
+ 1
Notice that equation (2.5), with solution q^ = N 1 p+ , looks like
N
equation (2.4) except that N substitutes in the right-hand side of the ex-
pression.
(e) Compare the supply of output in exercises (c) and (d) for various values of the
parameter , and discuss the dependence of aggregate supply on the organization
of production in one or more …rms.[Hint: use the …rst-order conditions you found
in each exercise, and separately analyze the case in which < 0, and > 0.]

24
Comparing expressions (2.4) and (2.5), we see that the output of the merged
…rm equals the aggregate output of the N independent …rms if and only if =
0 . This is the externality-free case, where the “decentralization theorem”
holds: the organization of production in either private …rms (or a merged
…rm, acting as a social planner) does not a¤ect aggregate supply. However,
when 6= 0 , this decentralization theorem does not hold. In particular:
1. If > 0 (negative externality), then the right-hand side of (2.5) is below
that of (2.4) (see the …gure below), thus implying that q^ > q M . As a
consequence, the independent …rms overproduce relative to the merged
…rm.

2. If, on the contrary, < 0 (positive externality), then the right-hand side
of (2.5) is above that of (2.4) (see the …gure below), thus implying that
q^ < q M . As a consequence, the independent …rms underproduce relative
to the merged …rm.

25
 Chapter 6 - Competitive Markets
End-of-Chapter Exercises - Answer key

1. [Equilibrium in competitive markets]. A perfectly competitive industry consists

of two types of …rms: 100 …rms of type A and 30 …rms of type B. Each type A …rm has
a short-run supply curve sA (p) = 2p. Each type B …rm has a short-run supply curve
sB (p) = 10p. The Marshallian market demand curve is x(p) = 5000 500p.

(a) What is the short-run equilibrium price in this market?

Summing the individual supply curves of the 100 type-A …rms and the 30
type-B …rms, we have an aggregate supply curve of S(P ) = 200p + 300p =
500p. The short-run equilibrium occurs at the price at which quantity sup-
plied equals quantity demanded,

5000 500p = 500p, or p = 5

(b) At this price, how much does each type A …rm produce, and how much does each
type B …rm produce?
At this price, each type-A …rm supplies sA (p) = 2p = 2 5 = 10 units, and
every type-B …rm supplies sB (p) = 10p = 10 5 = 50 units.

2. [Firm supply curve]. A competitive …rm has the following short run total cost
function:
C(q) = q 3 8q 2 + 30q + 5

(a) Find the expression for the marginal cost (MC), average total cost (AC) and the
average variable cost (AVC). Sketch MC, AC and AVC on a graph (a graph that
approximates the pattern of these curves is enough).
Marginal costs, M C (q) = 3q 2 16q + 30.
Average total costs, AC (q) = q 2 8q + 30 + 5q .
Average variable costs, AV C (q) = q 2 8q + 30

1
 (b) At what range of prices will the …rm supply zero output?
In the short-run, the …rm will …nd pro…table to produce at any price p
AV C (q). Hence, we need to …nd the minimum of the AV C(q) function.
A useful trick to …nd such minimum point is remembering that such point
occurs at the output level for which AV C(q) = M C(q),

q2 8q + 30 = 3q 2 16q + 30, or q = 4 units

And when the …rm produces q = 4 units, its average variable cost is at the
minimum, with a value of

AV C(8) = 42 (8 4) + 30 = 14

Hence, the …rm supplies zero output if p < $14.

(c) Identify the …rm’s supply curve on your graph.
The …rm supply curve is the marginal cost curve above the point where M C =
AV C. The …rm will produce at the point where p = M C as long as M C
AV C:
(d) At what price would the …rm supply exactly 6 units of output?
The …rm maximizes pro…t by choosing the level of output at which p = M C.
To …nd the price at which the …rm will supply exactly q = 6 units of output,
we then need

p = M C(6), or
p = 3 62 16 6 + 30 = $42

3. [Entry and exit in perfectly competitive markets]. Suppose you are given the
following information about a particular industry

QD = 6500 100p Market demand

QS = 1200p Market supply
q2
C(q) = 722 + 200 Firm total cost function

Assume that all …rms are identical, and that the market is characterized by pure
competition.

(a) Find the equilibrium price and the equilibrium quantity.

Setting market demand and supply equal to each other,

6500 100p = 1200p

we …nd that the equilibrium price is p = $5, and substituting it back into
either market demand or market supply, we …nd that the equilibrium quantity
is Q = 6; 000 units.
(b) Find the output supplied by the …rm, and the pro…t of the …rm.

2
 Setting the market equilibrium price equal to marginal cost, p = M C , for
2q
every …rm, 5 = 200 , or q = 500 units for every …rm. Pro…ts of each …rm are
then
5002
(500) = (5 500) 722 = $528
200
Notice that market output is Q = 6; 000 units and the output of every …rm
is q = 500. Therefore, there must be n = 6;000
500
= 12 …rms.
(c) Would you expect to see entry into or exist from the industry in the long-run?
Explain.
Entry, because the …rms in the industry are making positive pro…ts.
(d) What e¤ect will entry or exit have on market equilibrium?
As …rms enter, the supply curve of the industry will shift down and to the
right, and the new equilibrium price will fall. This would reduce each …rm’s
pro…t down to zero until there is no incentive for further entry.
(e) [4 points] What is the lowest price at which each …rm would sell its output in the
long run?
In the long-run, every …rm would sell its output as long as p AV (q). To
…nd the minimum of the AC(q), set M C = AV ,
2q 722 q
= +
200 q 200

and solving for q, we …nd q = 380; which implies an average cost of AC (380) =
3:8
(f) Is pro…t positive, negative, or zero if the …rm sells its products for a price below
the one you …nd? Explain
The …rm will not sell for any price p < 3:8 in the long run. At any price below
the minimum of the average cost function, pro…ts are negative, and the …rm
is better o¤ selling its …xed resources and producing zero output.
(g) What is the lowest price at which each …rm would sell its output in the short run?
q q
The …rm will sell for any positive price, since M C = 100
, AV C = 200
, at any
price p > 0, we have that M C > AV C.

4. [Per unit taxes vs. Ad valorem taxes] A tax is to be levied on a commodity

bought and sold in a competitive market. Two possible forms of tax may be used: In
one case, a speci…c tax is levied, where an amount t is paid per unit bought or sold
(this is the case considered in the text). In the other case, an ad valorem tax is levied,
where the government collects a tax equal to times the amount the seller receives
from the buyer. Assume that a partial equilibrium approach is valid.

(a) Show that, with a speci…c tax, the ultimate cost of the good to consumers and the
amounts purchased are independent of whether the consumers or the producers
pay the tax. As a guidance, you can use the following steps:

3
 1. Consumers: Let pC be the competitive equilibrium price when the consumer
pays the tax. Note that when the consumer pays the tax, he pays pC + t
whereas the producer receives pC . State the equality of the (generic) demand
and supply functions in the equilibrium of this competitive market when the
consumer pays the tax.
If the speci…c tax t is levied on the consumer, then he pays p + t for every
unit of the good, and the demand at market price p becomes x (p + t).
The equilibrium market price pc is determinded from equalizing demand
and supply:
x (pc + t) = q (pc ) .
2. Producers: Let pP be the competitive equilibrium price when the producer
pays the tax. Note that when the producer pays the tax, he receives pP t
whereas the consumer pays pP . State the equality of the (generic) demand
and supply functions in the equilibrium of this competitive market when the
producer pays the tax.
On the other hand, if the speci…c tax t is levied on the producer, then he
collects p t from every unit of the good sold, and the supply at market
price p becomes q (p t). The equilibrium market price pp is determined
from equalizing demand and supply:
x (pp ) = q (pp t) .
3. Show that if an equilibrium price p solves your equality in (i), then p + t
solves the equality in (ii). Show that, as a consequence, equilibrium amounts
are independent of whether consumers of producers pay the tax.
It is easy to see that p solves the …rst equation if and only if p + t solves
the second one. Therefore, pp = pc + t, which is the ultimate cost of the
good to consumers in both cases. The amount purchased in both cases
is
x (pp ) = x (pc + t) .
(b) Show that this is not generally true with an ad valorem tax. In this case, which
collection method leads to a higher cost to consumers? [Hint: Use the same steps
as above, …rst for the consumer and then for the producer, but taking into account
that now the tax increases the price to (1 + )p. Then, construct the excess
demand function for the case of the consumer and the producer. You should …nd
that z (1 + ) pC > 0 and z pP = 0. Since, in addition, z( ) is non-increasing,
then (1 + ) pC pP , implying a lower cost of the ad valorem tax for consumers
than for producers.]
If the ad valorem tax is levied on the consumer, then he pays (1 + ) p
for every unit of the good, and the demand at market price p becomes
x ((1 + ) p). The equilibrium market price pc is determined from equalizing
demand and supply:
x ((1 + ) pc ) = q (pc ) .
On the other hand, if the ad valorem tax is levied on the producer, collects
(1 ) p from then he pays (1 + ) p for every unit of the good sold, and the

4
 supply at market price p becomes q ((1 ) p). The equilibrium market price
pp is determined from equalizing demand and supply:

x (pp ) = q ((1 ) pp ) .

Consider the excess demand function for this case:

z (p) = x (p) q ((1 ) p) (1)

Since x ( ) is non-increasing and q ( ) is non-decreasing, z (p) must be non-

increasing. From (1) we have

z ((1 + ) pc ) = x ((1 + ) pc ) q ((1 ) [(1 + ) pc ]) =

= x ((1 + ) pc ) q ((1 2
) pc )
x ((1 + ) pc ) q (pc ) = 0,

taking into account that q ( ) is non-decreasing and using (1).

Therefore, z ((1 + ) pc ) 0 and z (pp ) = 0. Since z ( ) is non-increasing,
this implies that (1 + ) pc pp . In words, levying the ad valorem tax on
consumers leads to a lower cost on consumers than levying the same tax on
producers. (In the same way it can be shown that levying the ad valorem
tax on consumers leads to a higher price for producers than levying the same
tax tax to producers).
(c) Are there special cases in which the collection method is irrelevant with an ad
valorem tax? [Hint: Think about cases in which introduces the same wedge on
consumers and producers (inelasticity). Then prove your statement by using the
above argument on excess demand functions].
If q ( ) is strictly increasing, the argument can be strengthened to obtain the
strict inequality: (1 + ) pc < pp . On the other hand, when the supply is
perfectly inelastic, i.e., q (p) = q = const, then (1) and (2) combined yield

x ((1 + ) pc ) = q = x (pp ) ,

and therefore pp = (1 + ) pc . Here both taxes result in the same cost to

consumers. However, the producers still bear a higher burden when the tax
is levied directly on them:

(1 ) pp = (1 ) (1 + ) pc < pc .

Therefore, the two taxes are still not fully equivalent. (As depicted in the

5
 …gure )

The intuition behind these results is simple: with a tax, there is always a
wedge between the "consumer price" and the "producer price". Levying an
ad valorem tax on the producer price, therefore, results in a higher tax burden
(and a higher tax revenue) than levying the same percentage tax on the lower
consumer price.

5. [Distribution of tax burden]. An ad valorem tax of is to be levied on consumers

in a competitive market with aggregate demand curve x(p) = Ap" , where A > 0 and
" < 0, and aggregate supply curve q(p) = p , where > 0 and > 0. Calculate
the percentage change in consumer cost and producer receipts per unit sold for a small
(marginal) tax. Denote = (1 + ). Assume that a partial equilibrium analysis is
valid.

(a) Compute the elasticity of the equilibrium price with respect to . [Hint: To
compute the price received by producers, use expression (10.C.8) from MWG,
note that the tax t is marginal, so assume that t = 0, and use the expressions of
x(p) and q(p). Then multiply the numerator and denominator by p , and use that
in equilibrium x(p) = q(p). Regarding consumers, note that they will pay p + t,
and its derivative with respect to t at t = 0 is simply p0 (0) + 1 (the expression
you found for producers plus one)].
To compute the price received by producers, we can use equation (10.C.8) in
the textbook:
0 x0 (p ) A"p" 1
A"p"
p (0) = = 1 = =
x0 (p ) q 0 (p ) A"p" 1 p A"p" p
"x (p ) "
= = .
"x (p ) q (p ) "

(We have multiplied both the numerator and the denominator by p and used
the fact that p is an equilibrium price, therefore x (p ) = q (p ).) The price
paid by consumers is (p ) + t, and its derivative with respect to t at t = 0 is
"
p0 (0) + 1 = +1= .
" "

6
 From these expressions we can see that when = 0 (supply is perfectly in-
elastic) or " ! 1 (demand is perfectly elastic), the price paid by consumers
is unchanged, and the price received by producers decreases by the amount
of the tax. On the other hand, when " = 0 (demand is perfectly inelastic)
or ! 1 (supply is perfectly elastic), the price received by producers is
unchanged and the price paid by consumers increases by the amount of the
tax.
(b) Argue that when = 0 producers bear the full e¤ect of the tax while consumers’
total costs of purchase are una¤ected.
From the above expression,
"
p0 (0) + 1 = +1= .
" "

we can see that when = 0 (supply is perfectly inelastic) or " ! 1

(demand is perfectly elastic), the price paid by consumers is unchanged, and
the price received by producers decreases by the amount of the tax.
(c) Argue that when " = 0 consumers bear the full burden of the tax.
On the other hand, when " = 0 (demand is perfectly inelastic) or ! 1
(supply is perfectly elastic), the price received by producers is unchanged and
the price paid by consumers increases by the amount of the tax.
(d) What happens when each of these elastiticities approaches 1 in absolute value?
As suggested above, when " ! 1 (demand is perfectly elastic), the price
paid by consumers is unchanged, and the price received by producers de-
creases by the amount of the tax. In contrast, when ! 1 (supply is
perfectly elastic), the price received by producers is unchanged and the price
paid by consumers increases by the amount of the tax.

6. [Ad valorem taxes] Throughout this chapter’s analysis of taxes we have used per-
unit taxes–that is, a tax of a …xed amount for each unit traded in market. A similar
analysis would hold for ad valorem taxes–that is, taxes on the value of the transaction (
or, what amounts to the same thing, proportional taxes on price). Given an ad valorem
tax rate of t (t=0.05 for a 5 percent tax), the gap between the price demanders pay
and what suppliers receive is given by PD = (1 + t)PS .

(a) Show that, for an ad valorem tax,

d ln PD eS d ln PS eD
= : and =
dt eS eD dt eS eD
Answer: Market equilibrium requires quantity demanded (evaluated at the
price consumers pay) to equal quantity supplied (evaluated at the price pro-
ducers receive):
QD (PD ) = QS (PS )

7
Substituting the relationship between consumer and producer prices PD =
(1 + t)PS ;we have
QD ((1 + t)PS ) = QS (PS ):
Rearranging,
0 = QS (PS ) QD ((1 + t)PS ):
Totally di¤erentiating this identity with respect to t,

dPS dPS
0 = Q0 S Q0 D PS + (1 + t) :
dt dt

The formulae we are asked to verify are only exact for an in…nitesimal tax
increase above an initial tax oft = 0, So throughout the remainder of the
analysis, we will use the approximation t 0. At these small tax levels, it
is approximately true that PS PD and QS QD :Given t 0 the previous
total derivative becomes
dPS dPS
0 = Q0 S Q0 D PS + :
dt dt

Solving for dPS =dt ;

dPS Q0 PS
= 0 D 0 :
dt QS QD
This implies
d ln PS dPS 1
=
dt dt PS
Q0 D
=
Q0 S Q0 D
Q0 D (PD =QD )
=
Q0 S (PS =QS ) Q0 D (PD =QD )
eD
=
eS eD
The second to last line uses the approximations PS PD and QS QD :We
can use a shortcut to …nd d ln PD =dt :Totally di¤erentiating the identity PD =
(1 + t)PS ;
dPD dPS
= PS + (1 + t)
dt dt
dPS
PD + ;
dt
where the last line uses our approximations t 0 and PS PD for tiny tax
rates. Hence
d ln PD dPD 1
=
dt dt PD
dPS 1
1+
dt PS

8
 d ln PS
=1+
dt
eD
=1+
eS eD
eS
= :
eS eD
(b) Show that the excess burden of a small tax is
eD eS 2
DW = 0:5 t P0 Q0 ;
eS eD
Answer: DW is given as the area of the shaded region in the graph below.
For a small tax increase starting from t = 0, DW can be approximated using
the formula for the area of a triangle. (This is only an approximation because
the supply and demand curves may not be straight lines.). Thus
1
DW = [PS (1 + t) PS ] (QS Q0 )
2
tPS
= (QS Q0 )
2
tPS dPS
Q0 S ( t)
2 dt
or, rearranging,
t2 0 dPS
DW = (Q S PS ) :
2 dt
As shown in part a,
dPS 1 d ln PS eD
= =
dt PS dt eS eD
dPS eD eD
) = PS P0 :
dt eS eD eS eD
The last approximation is good for a small tax increase above 0, implying
PS P0 :Further, manipulating the expression to have an elasticity show up,

QS
Q0 S PS = Q0 S PS
QS

= eS QS
eS Q0 ;
where the approximation QS Q0 is again good for a small tax increase
above 0. Substituting these results into the expression for DW ,

t2 eD eS
DW P0 Q0 ;
2 eS eD

9
 as was to be shown.

(c) Compare these result to those derived in this chapter for a unit tax.
Answer: The unit tax described in this chapter is equivalent to the value of
the ad-valorem tax. In other words, the unit tax is equal to the ad-valorem
tax multiplied by Ps . Therefore, the results obtained using the ad-valorem
tax are equivalent to the ones obtained using the unit tax.

10
 Chapter 7 - Choice under Uncertainty
End-of-Chapter Exercises - Answer key

1. [Independence axiom] State the independence axiom.

A preference relation % over the space of simple lotteries L satis…es the indepen-
dence axiom if for all lotteries L; L0 ; L00 2 L and 2 [0; 1], we have

L % L0 if and only if L + (1 ) L0 % L00 + (1 ) L00

That is, if we mix each of two lotteries with a third one, then the preference
ordering of the two resulting lotteries does not depend on (is independent of) the
particular third lottery used.

(a) Show that if indi¤erence curves in the Machina triangle are not parallel straight
lines, then the independence axiom is violated. (You can help your discussion
with a …gure).
In order to show that when indi¤erence curves are not parallel straight lines
(as we can see in …gure below) the independence axiom is violated, we need
to show that

L % L0 does not imply L + (1 ) L00 % L0 + (1 ) L00

This is easy to prove for the case in which the individual is indi¤erent between
lotteries L and L0 , L L0 That is, we must show

L L0 does not imply L + (1 ) L00 L0 + (1 ) L00

Take a case in which L L0 , as those lotteries on the indi¤erence curve of

the …gure. As we can see, the mix of each of these two lotteries with a third
lottery L00 leads to compound lotteries 31 L + 32 L00 and 31 L0 + 23 L00 , respectively.

1
 The possibility of indi¤erence curves which are not parallel straight lines leads
to situations like that in the …gure, where
1 2 1 0 2 00
L + L00 L + L
3 3 3 3
since the decision maker prefers the second compound lottery to the …rst.
Hence, we cannot guarantee that
L + (1 ) L00 L0 + (1 ) L00
and as a consequence, the independence axiom does not hold for indi¤erence
curves which are not parallel straight lines.

2. [Independence axiom and convexity].Show that the independence axiom implies

convexity, i.e., for three di¤erent lotteries L, L0 and L00 , if L % L0 and L % L00 , then
L % L0 + (1 ) L00 :

From L % L0 we can apply the independence axiom, and obtain

L + (1 )L L0 + (1 )L
Similarly, from L % L00 we can apply the independence axiom, and obtain
(1 )L + L0 (1 )L00 + L0
And by transitivity (from the two previous expressions), we have
L + (1 )L (1 )L00 + L0
and rearranging
L L0 + (1 )L00

(a) Additionally, argue that a decision maker whose preference violate convexity can
be o¤ered a sequence of choices that lead to a sure loss of money (Dutch books).
If a decision maker’s preferences over lotteries violate convexity, then we must
have that for three di¤erent lotteries L, L0 and L00 , where L % L0 and L % L00 ,
we obtain the opposite result than above; that is
L0 + (1 ) L00 L
Note that, if the decision maker initially owns the right to participate in
lottery L, he will be willing to pay an amount $X in order to switch to the
compound lottery L0 + (1 ) L00 given that L0 + (1 ) L00 L. Now he
owns the compound lottery L0 + (1 ) L00 , and either lottery L0 or lottery
L00 are realized. But we know that the decision maker prefers lottery L to
either of these lotteries since
L % L0 and L % L00
was an initial assumption of this decision maker’s preferences over lotteries.
Therefore, he would be willing to pay again $Y in order to get L. Hence,
the decision maker is exactly as at the beginning (lottery L) and has lost
$X + $Y . We can then start the process again, and make this individual pay
$X + $Y dollars, keeping him exactly where he started!

2
3. Let G be the set of compound gambles over a …nite set of deterministic payo¤s
fa1 ; a2 ; :::an g R+ . A decision maker’s preference relation % over G can be rep-
resented by the following utility function v : G ! R. Let g 2 G, and let probability
pi be associated to the corresponding payo¤ ai . Finally, consider that the decision
maker’s utility function v( ) is given by

p1 p2 pn
Y
n
v(g) = (1 + a1 ) (1 + a1 ) ::: (1 + an ) = (1 + ai )pi
i=1

(a) Show that this is not a von Neumann-Morgenstern utility function.

Since v(g) is not linear in the probabilities, then v(g) cannot be a vNM
expected utility function, with general form
X
N
v(g) = pi u(ai )
i=1

(b) Show that the decision maker has the same preference relation as an expected
Xn util-
ity maximizer with von-Neumann Morgenstern utility function u(g) = pi ln (1 + ai ).
i=1
Since ln( ) is a monotonic transformation of v( ), so they have the same
preference relation. Applying the monotonic transformation u(g) = ln [v(g)]
to the original function v(g), we obtain
!
Yn X
n
ln (1 + ai )pi = pi ln (1 + ai )
i=1 i=1

which represents the initial preference relation over lotteries, and it is linear
in the probabilities. Hence, it is a vNM utility function.
(c) Assume now that the expected utility maximizer in (b) has utility function u(w) =
ln (1 + w) over wealth w > 0. Evaluate his risk attitude (concavity in his utility
function). Additionally, …nd the Arrow-Pratt coe¢ cient of absolute risk aversion.
How does it vary in wealth?
Given that u(w) = ln(1 + w), where w > 0, then
1 1
u0 (w) = > 0 and u00 (w) = <0
1+w (1 + w)2
Which implies that the utility function is concave, and that the decision maker
is risk-averse.
Let us now …gure out the Arrow-Pratt coe¢ cient of absolute risk-aversion,
1
u00 (w) (1+w)2 1
rA (w; u) = = 1 =
u0 (w) 1+w
1+w
Finally, we want to know how this coe¢ cient of absolute risk aversion varies
with wealth,
@rA (w; u) 1
= < 0 for all w > 0
@w (1 + w)2
Hence, the agent becomes less risk-averse as he becomes more wealthy.

3
4. [Some Regret theory] Let G be the set of compound gambles over a …nite set
of deterministic payo¤s fa1 ; a2 ; :::an g R+ . In regret-based decision making, one
sometimes considers the utility function v : G ! R de…ned as follows. First, de…ne the
highest deterministic payo¤ that could be reached in gamble g by using the following
function h(g) = max fak : k 2 f1; 2; :::; ng and pk > 0g. Substracting h(g) from all
deterministic outcomes and computing expected value yields:
X
n X
n
v(g) = pi (ai h(g)) = pi ai h(g)
i=1 i=1

Assume that this individual’s preference relation is represented by v(g).

Note that the individual compares the monetary payo¤ resulting from each of the
possible events, ai , with respect to the highest possible payo¤ of the lottery, h(g).
For this reason, utility functions of this type re‡ect what is usually denoted as
“regret theory”, as if individuals experience a disutility from not receiving the
highest possible monetary payo¤ in the lottery (regret).

(a) Compute the expected value of the following two gambles:

1 1 1 1 1 1
g1 = 0; 1; 2; ; ; and g 2 = 1; 4; 5; ; ;
3 3 3 2 3 6

First, note that h (g) = max f0; 1; 2g since all these events can occur with
strictly positive probability. Then, h (g) = 2, and therefore the individual’s
expected utility from playing the …rst gamble, g1 ; is
1 1 1
v(g1 ) = (0 2) + (1 2) + (2 2) = 1
3 3 3
Similarly, we can …nd the expected utility from playing the second gamble,
g2 . Importantly, note that in this case h (g) = max f1; 4; 5g = 5:
1 1 1 7
v(g2 ) = (1 5) + (4 5) + (5 5) =
2 3 6 3
Note that the individual experiences a lower expected utility from playing the
second lotter than the …rst lottery. Intuitively, this happens because: (1) the
second lottery’s distribution of payo¤s is more spread than the …rst lottery,
and this makes the lower payo¤s on the second gamble to be compared to
a higher h(g); and (2) because the lowest payo¤s on the second gamble are
more likely than in the …rst, and as a consequence the individual assigns a
higher weight in the expected utility calculation to those monetary payo¤s in
which he is experiencing the biggest regret.
(b) Show that all deterministic outcomes (the outcomes with probability of 1) have
the same utility level. That is, v(a1 ) = v(a2 ) = ::: = v(an )

4
 Let us represent by v(ai ) the individual’s utility level from a certain deter-
ministic outcome ai , i.e., pai = 1. But if outcome ai always occurs, then there
is not potential regret since function h(g) can only …nd the max among all
outcomes of the lottery whose probability is strictly greater than zero. Since
pai = 1, then all other outcomes of the lottery receive probability zero, and
hence

h(g) = max fak : k 2 f1; 2; :::; ng and pk > 0g

= max fai g = ai

Therefore, the individual’s expected utility is

X
n
v(ai ) = pi (ai h(g)) = 1 (ai ai ) = 0
i=1

Thus, v(a1 ) = v(a2 ) = ::: = v(an ) = 0, regardless of the monetary payo¤

associated to outcome ai . If there is just one event to be regretful about, my
expected utility is zero!
(c) Show that the preference relation does not satisfy monotonicity if outcomes are
deterministic.
From the de…nition of monotonicity, we have that

(a1 ; an ; ; 1 ) % (a1 ; an ; ; 1 )

for all ; 2 [0; 1] if and only if > . So if we make = 1 and = 0, then

the above condition on monotonicity becomes

(a1 ; an ; 1; 0) % (a1 ; an ; 01)

Then, clearly a1 % an and a1 an , what implies that a1 an strictly.

However, in part (b) we have shown that the individual’s utility is the same
(and equal to zero) when outcomes are deterministic. In other words, he is
indi¤erent between gambles whose outcomes are deterministic, i.e., a1 a2
::: an . But this contradicts that a1 an strictly. Therefore, this “regretful”
preference relations cannot satisfy monotonicity.

5. [An introductory example p on risk aversion] Let us assume that your utility
function is given by U (I) = I. You have been o¤ered two wage o¤ers.

In the …rst one you will receive a …xed salary of $54,000.

In the second one, you will only receive $4,000 as a …xed payment, plus a bonus of
$100,000 if the …rm goes pro…table. The probability that the …rm goes pro…table
(and you get a total of $104,000 that year) is 0.5, while the probability that the
…rm does not make enough pro…ts is 0.5.

(a) Find the expected value of the lottery induced by accepting the second wage o¤er.

5
 The expected value of accepting the second wage o¤er is:

EVSecond = 0:5($4; 000) + 0:5($104; 000) = 2; 000 + 52; 000 = 54; 000

(b) Find the expected utility associated with the second o¤er.
The expected utility is
p p
EUSecond = 0:5 4; 000 + 0:5 104; 000 = 192:87

(c) Draw an approximate …gure where the following elements are clearly illustrated:
1. Utility function (either concave, linear or convex);
2. Utility level (in the vertical axis) from the …rst wage o¤er;
3. Utility level (in the vertical axis) from each of the two possible outcomes of
the second wage o¤er.
4. Expected utility level from the second wage o¤er. [Hint: notice that you just
have to connect with a line the two points you identi…ed in the previous bullet
and …nd the midpoint of this line].

U
U($104,000)
=322.5
U($EV)=
232.38
EU

U($4,000)
=63.25

$4,000 $54,000 $104,000 Income

EV= First wage offer

p
Note that the utility of …rst wage o¤er is U (54; 000) = 54000 = 232:38
(d) Using your answers from parts (a) and (b), …nd the risk premium associated with
the second o¤er.
We know that the general expression of the risk premium (RP) of a lottery is

pU (I1 ) + (1 p)U (I2 ) = U (EV RP )

Since the left hand side is just the expected utility from the lottery, EU , we
have
EU = U (EV RP )

6
 Given that we know EU = 192:87 from part (b) and EV = 54; 000 from part
(a), p
192:87 = 54; 000 RP
Squaring both sides of this equation and rounding to the nearest integer,

37; 199 = 54; 000 RP () RP = 16; 801

(e) What amount of money should the …rst wage o¤er propose you in order to make
you exactly indi¤erent between accepting the …rst and the second wage o¤ers?
The certain amount of money that would make you exactly indi¤erent be-
tween the utility form this certain payment and the utility from accepting the
second (uncertain) wage o¤er is the so-called Certainty Equivalent. As can
be seen from the above …gure, the certain equivalent is just obtained from
subtracting the risk premium to the original certain amount (…rst wage o¤er),

Certainty Equivalent = $54; 000 $16; 801 = $37; 199

(f) In your …gure from part (c) include the risk premium and the certainty equivalent
of the second wage o¤er.

U
U($104,000)
=322.5
U($EV)=
232.38
EU

U($4,000)
=63.25

Risk
premium

$4,000 $37,199 $54,000 $104,000 Income

EV= First wage offer

Certain Equivalent

6. [Does risk-aversion imply variance aversion?] Consider an individual with the

following utility function

2x if x 25
u(x) = 5
2
+ x if x > 52

(a) Draw it. Show that this individual is (weakly) risk averse. [Hint: Note that the
above utility function can be expressed as the minimum of two expressions].
This individual’s utility can be expressed as the minimum of 2x and 52 + x,
5
since 2x is the minimum of these linear functions for x 2
, and 52 + x is

7
 the minimum of both linear functions for x > 25 . Hence, the function can
be represented u(x) = min 2x; 52 + x . And clearly, the function u(x) =
min 2x; 25 + x is a (weakly) concave function.

u(x) = min 2x; 52 + x

(b) Suppose that there are three states of the world, each equally likely. There are
two assets, x and y. The asset x is the random variable (1; 5; 9) and the asset y is
the random variable (2; 3; 10). Calculate the expected utility of asset x and asset
y. Which asset, hence, would be preferred by this individual, if both of them were
o¤ered at the same price?
Let us …rst …nd the expected utility of asset x,
1 5 1 5
EU (x) = min 2 1; + 1 + min 2 5; + 5
3 2 3 2
1 5
+ min 9 1; + 9
3 2
1 1 15 1 23 21
= 2+ + =
3 3 2 3 2 3
And similarly for the expected utility of asset y,
1 5 1 5
EU (y) = min 2 2; + 2 + min 2 3; + 3
3 2 3 2
1 5
+ min 9 10; + 10
3 2
1 1 11 1 25 22
= 4+ + =
3 3 2 3 2 3
Therefore, EU (x) < EU (y), making asset y to be preferred by the individual,
if both assets were o¤ered at the same price.
(c) Calculate the expected value of each asset (you previously found the expected
utility). Calculate the variance of both assets. Which asset would be chosen by
this individual if he were variance averse?

8
 The expected value of each asset is
1 1 1
E(x) = 1+ 5+ 9=5
3 3 3
1 1 1
E(y) = 2+ 3+ 10 = 5
3 3 3
Hence, both assets have the same expected value. Let us now …nd their
variance. For convenience, we use the formula V ar(x) = E(x2 ) E(x)2 . We
already know E(x), let us then …nd E(x2 ) for each asset,
1 1 1 107
E(x2 ) = (1)2 + (5)2 + (9)2 =
3 3 3 3
1 1 1 114
2
E(y ) = (2)2 + (3)2 + 2
(10) =
3 3 3 3
Therefore, we can use this information, together with the expected values of
the assets, E(x) and E(y), found above, and compute the variance of every
asset,
107 32
V ar(x) = E(x2 ) E(x)2 = 25 =
3 3
114 38
V ar(y) = E(y 2 ) 2
E(y) = 25 =
3 3
Hence, V ar(y) > V ar(x).
(d) From your previous answers, comment on the validity of the following statement:
“Every risk-averse individual is also variance averse”.
First, note that this individual is risk-averse since in part (a) we showed that
his utility function is weakly concave. Moreover, if he were variance-averse,
he would prefer the asset (or the lottery) with the lowest possible variance
(asset x). However, we showed in part (b) that he chooses asset y. Hence,
risk-aversion does not necessarily imply variance-aversion.

7. [Investing in two di¤erent assets]. An investor has the von Neumann Morgenstern
utility function u(c) = e c , where c is consumption, and where > 0. There are
two states of the world, labelled 1 and 2, which are equally likely. There two types of
assets:

Asset 1 yields one unit of consumption in state 1, but nothing in state 2.

Asset 2 yields nothing in state 1, but one unit of consumption in state 2.
The price of the …rst asset is 1 , while the price of the second asset is 2 , where
for simplicity 1 + 2 = 1. The investor starts with an endowment of w units of
both assets, but seeks to balance her portfolio so as to maximize expected utility.
Denote by x1 the number of units of the …rst asset, and by x2 the number of units
of the second asset.

(a) Formulate the investor’s maximization problem.

9
 First, note that in state 1, the investor gets one unit of consumption for every
unit of consumption of asset x1 the investor purchased, and no units of con-
sumptions for every unit of consumption of asset 2, i.e., u(c) = e c =
e x1 . In state 2 the opposite happens, leading to a utility of u(c) =
e c = e x2 . We can then express the investor’s maximization prob-
lem as
1 1
max e x1 + e x2
x1 ;x2 0 2 2
s.t. 1 x1 + 2 x2 w
(b) By solving the investor’s maximization problem, show that
2
x1 = w + [log 2 log 1]

1
x2 = w + [log 1 log 2]

Setting up the Lagrangian associated to this maximization problem,

1 x1 1 x2
L (x1 ; x2 ; ) = e + e + [w 1 x1 2 x2 ]
2 2
and solving for the …rst order conditions,
@L x1
= e 1 =0
@x1 2
@L x2
= e 2 =0
@x1 2
@L
= w 1 x1 2 x2 =0
@
Solving for in the …rst two expressions, and setting them equal to each
other, we obtain
x1 x2
e e x1 + x2 1
= () e =
1 2 2

And applying logs,

x1 + x2 = ln 1 ln 2

Solving for x1 on the binding constraint, and plugging the result into the
above expression,
w 2
x2 + x2 = ln 1 ln 2
1 1

Multiplying both sides by 1

,
1
x2 [ 1 + 2] =w+ [ln 1 ln 2]

and using the property that 1 + 2 = 1, then we obtain the optimal amount
of asset x2 that the investor demands,
1
x2 = w + [ln 1 ln 2]

10
 Operating similarly, we can …nd the optimal amount of asset x1 ,
2
x1 = w + [ln 2 ln 1]

(c) How does the holding of assets vary with ? Give an explanation.
The demand for assets 1 and 2 found above, x1 and x2 , is decreasing in . It
is easy to show that the Arrow-Pratt coe¢ cient of absolute risk-aversion of
this investor is
u00 (x)
rA (x; u) = = ;
u0 (x)
which is constant for all wealth levels. Hence, an increase in his risk aversion
(measured by ) reduces his demand for any of these two risky assets.
(d) How does the holding of assets vary with w? How sensitive is this result to the
speci…cation of the utility function?
As the previous section pointed out, the investor’s demand for risky assets is
indepent on his wealth level. This result, however, depends on the particular
speci…cation of the investor’s utility function. In this case, his utility function
has an Arrow-Pratt coe¢ cient of absolute risk-aversion, rA (x; u) = , which
is constant for all wealth levels. Many other utility function can be used
which do not satisfy this property, and where the investor would vary his
holdings of risky assets when he becomes more risk-averse.

8. [Second order stochastic dominance]. Assume that distribution G(x) is a elemen-

tary increase in risk from distribution F (x). Show that F (x) second order stochastically
dominates G(x). [Hint: Use the de…nition of the elementary increase in risk. Recall
that then distribution G(x) is generated from F (x) by taking all the weight that F (x)
assigns to an interval [x0 ; x00 ] and transfering it to the endpoints x0 and x00 , such that the
same mean is preserved. Of course the variance of distribution G(x) does not coincide
with that of distribution F (x)].

Distribution G(x) is considered an elementary increase in risk from F (x) if dis-

tribution G(x) is generated from F (x) by taking all the weight that F (x) assigs
to an interval [x0 ; x00 ] and transferring it to the endpoints x0 and x00 , such that
the same mean is preserved. Hence, an elementary increase in risk is a mena
preserving spread. Therefore, since G(x) is a mean preserving spread of F (x),
then

E [F (x)] = E [G (x)] , but

V ar [F (x)] V ar [G(x)]

And as a consequence, F (x) second order stochastically dominates G(x). [See

11
 Figure below, for a graphical illustration of a mean preserving spread.]

Mean preserving spread

9. [What is the relationship between FOSD and dominance in means?] Prove

that if the cummulative distribution function
Z F (x) …rst-order stochastically dominates
Z
G(x), then the mean of x under G(x), x dG(x), cannot exceed that under F (x), x
dF (x), i.e., Z Z
xdF (x) > xdG(x)

We know that distribution function F (x) …rst-order stochastically dominates G(x)

if Z Z
u (x) dF (x) > u (x) dG(x)

Using the fact that the utility function is weakly increasing, and using u(x) = x,
we have Z Z
xdF (x) > xdG(x)

Z Z
(a) Provide now an example where xdF (x) > xdG(x) is satis…ed, but F (x)
does not …rst order stochastically
Z dominates
Z G(x). [Note that these two proofs
show that FOSD implies xdF (x) > xdG(x), but the latter does not imply
FOSD.
Z In other
Z words, FOSD is a su¢ cient (but not necessary) condition for
xdF (x) > xdG(x) ].

Consider the example of a mean-preserving spread discussed in class. The

mean of both distribution functions F (x) and G(x) was 52 . However, neither
F (x) FOSD G(x), nor G(x) FOSD F (x), for all x.

12
10. [Coe¢ cient of risk aversion] Show that the coe¢ cient of absolute risk-aversion is
invariant to linear transformations of the utility function u( ). [Hint: Use a linear
transformation like v = a + bu. Then, simply …nd the coe¢ cient of absolute risk-
aversion for v, for u; you should obtain the same expression].

First note that

v 0 = bu0 and v 00 = bu00
Then, the Arrow-Pratt coe¢ cient of absolute risk aversion for the linear transfor-
mation v is
v 00 (x) bu00 u00
rA (x; v) = = =
v 0 (x) bu0 u0
which exactly coincides with the Arrow-Pratt coe¢ cient of absolute risk aversion
for utility function u(x).

11. [Concavity and Coe¢ cient of risk aversion] Let u and v be two utility functions,
where v(W ) = f (u(W )), and f ( ) is a concave function, i.e. v is more concave than u.

(a) Find the coe¢ cient of absolute risk-aversion for v.

First, note that v 0 = f 0 u0 and v 00 = f 00 [u0 ]2 +f 0 u00 . Therefore, the Arrow-Pratt
coe¢ cient of absolute risk aversion of v is
v 00 (x) f 00 [u0 ]2 + f 0 u00
rA (x; v) = =
v 0 (x) f 0 u0
u00 f 00 u0
=
u0 f0
(b) Prove that the coe¢ cient of absolute risk-aversion for v is greater than for u.
We want to compare the Arrow-Pratt coe¢ cient of absolute risk aversion of v,
u00 f 00 u0 00
u0 f0
, with respect to that of u, uu0 . Hence, we want to know the sign
00 0
of f f u0 , First, note that u ( ) is increasing in wealth, and then u0 > 0. Second,
note that function f ( ) is a concave, i.e., f 0 > 0 and f 00 < 0. Summarizing,
00 0
the ratio f f u0 is negative, which implies
u00 f 00 u0 u00
>
u0 f0 u0
which implies that rA (x; v) > rA (x; u). Intuitively, the coe¢ cient of absolute
risk aversion is higher than the more concave is the utility function.

12. [Nonconstant coe¢ cient of absolute risk aversion] Suppose that the utility func-
tion is given by u(W ) = aW bW 2 , where a; b > 0. First, …nd the coe¢ cient of absolute
risk-aversion. Does it increases or decreases in wealth? Interpret.

First, note that u0 = a 2bW and u00 = 2b. Hence, the coe¢ cient of absolute
risk-aversion is
u00 (x) 2b
rA (x; u) = 0
=
u (x) a 2bW
Note that as W rises, the denominator decreases, and as a consequence rA (x; u)
rises, i.e., the decision maker becomes more risk averse as he wealth increases.

13
 (a) Let us now consider that this decision maker is deciding how much to invest in
a risky asset. This risky asset is a random variable R, with mean R > 0 and
variance 2R . Assuming that his initial wealth is W , state the decision maker’s
(expected) utility maximization problem, and …nd …rst order conditions. [Hint:
First, note that the decision maker’s wealth (W in his utility function) is now a
random variable W + xR, where x is the amount of risky asset that he acquires.
You must insert this expression in the decision maker’s utility function, for every
W . Then, we must take expectations over the entire expression, since the risky
asset is a random variable.]
The choice problem of the decision maker is

max E a (W + xR) b (W + xR)2

x

And the associated …rst order condition is

E [aR 2bR (W + x R)] = 0

(b) Simplify the …rst order condition you found before. [Hint: Note that you must
2
use the property that E[R2 ] = R + 2R ].
Simplifying the above …rst order condition,

E [aR 2bR (W + x R)] = aR 2bRW E 2bR2 x =

2 2
= aR 2bRW 2bx R + R =0

(c) [6 points] What is the optimal amount of investment in risky assets?

Solving for x in the above expression,
(a 2bW ) R
x = 2 2
2b R + R

(d) Show that the optimal amount of investment in risky assets (the expression you
found in the previous part) is a decreasing function in wealth. Interpret.
Di¤erentiating x with respect to wealth,
@x R
= 2
@W R + 2
R

which is negative, since R; 2R > 0. Intuitively, the larger the decision maker’s
wealth, the lower is the amount of risky assets he wants to hold. This expla-
nation is consistent with his coe¢ cient of absolute risk aversion found at the
beginning of the exercise, where we showed that the individual becomes more
risk averse as his wealth increases.

13. [Purchasing insurance] Consider the insurance problem studied in class, but assume
now that insurance is not actuarilly fair, i.e., q > , and the cost of every unit of
insurance is higher than the probability of a loss.

14
 (a) State the individual’s utility maximization problem, specifying his choice variable.
Let us …rst recall the notation in MWG. The probability of a loss is denoted
by ; the amount of money that can be lost if the bad event (loss) occurs
is $D; buying one unit of insurance costs $q dollars, and pays $1 if the loss
occurs (every unit of insurance pays one dollar); and the control variable of
the individual is the number of units of insurance he buys, which we will
denote by . The wealth of the individual is given by

w q if the loss does not occur,

w q D + if the loss occurs

Therefore, his maximization problem is

max (1 ) u (w q) + u (w q D+ )
0

(b) Find the …rst order conditions of the above problem.

The …rst order conditions are given by

q (1 ) u0 (w q) + (1 q)u0 (w q (1 q)) 0

(c) Show that in this case the individual will not insure completely [Hint: You must
show that the …rst order condition of the individual’s maximization problem is
not satis…ed at the point where the amount of insurance is total, i.e., insurance
covers all the potential loss, = D].
Let us prove it by contradiction, so let us assume that = D. Then, the
above …rst order condition becomes

q (1 ) u0 (w Dq) + (1 q)u0 (w D) 0

() u0 (w D) [ q] 0
On the one hand, we know that the utility function is strictly increasing, and
hence u0 (w D) > 0. On the other hand, we know that insurance is not
actuarilly fair, i.e., q > , meaning that the cost of every unit of insurance is
higher than the probability of a loss. As a consequence, the above …rst order
condition cannot be satis…ed at the point of complete insurance = D, and
therefore the individual does not insure completely.

14. [Purchasing health insurance] Consider an individual with the following Bernouilli
utility function
u(C; H) = ln C
H
where C is his expenditure in consumption goods and H is his expenditure on health
insurance. Parameter denotes his monetary loss if he becomes sick, where for sim-
plicity
1 if he is sick, and
=
0 if he is healthy

15
Note that this utility function implies that, when getting sick, this individual’s disu-
tility is decreasing in the amount of health insurance that he purchased (e.g., he can
have access to better doctors and care facilities, and the negative e¤ects of the illness
are reduced). Finally, the probability of getting sick is given by 2 [0; 1], and this
individual’s wealth is given by m > 0, where m = C + H.

(a) What is this individual utility maximization problem? [Hint: it is easier to choose
C as your choice variable. You can …nd the optimal amount of H later on]

1
max (1 ) ln C + ln C
C;H H
where we subsitute = 0 when the individual is healthy (which occurs with
probability 1 ), and = 1 when the individual is sick (which occurs with
probability ). And since m = C + H, then H = m C, hence,
1
max (1 ) ln C + ln C
C m C
which reduces the choice variables of this maximization problem to only one:
C.
(b) Find the …rst order conditions associated to the previous maximization problem.
1
=0
C (m C)2
(c) Determine the optimal amount of consumption goods, C , and health insurance,
H .
Rearranging,
C2 (2m + )C + m2 = 0
with solutions
p p
2m + + 2 + 4m 2m + 2 + 4m
C= and C =
2 2
but given that the amount spent on consumption cannot exceed p the indi-
2m+ 2 +4m
vidual’s wealth, C m, the only feasible solution is C = 2
.
Therefore, the optimal amount of health insurance that this individual buys
is p p
2m + 2 + 4m 2 + 4m
H =m C =m =
2 2
(d) Determine if the optimal amount of health insurance, H , is increasing, decreasing,
or constant in m. Interpret.
Di¤erentiating H with respect to m,
@H
=p > 0 for all parameter values
@m + 4m
That is, the optimal amount of health insurance, H , is increasing in the
individual’s wealth level, m.

16
15. [Recovering the utility function from rA (x; u)] Show that the utility function
u( ) can be recovered from the Arrow-Pratt index of absolute risk aversion up to two
constants.

From the formula of the Arrow-Pratt index of absolute risk aversion,

u00 (x)
rA (x; u) =
u0 (x)
For convenience, let us assume rA (x; u) = a. That is,
u00 (x)
= a () u00 (x) + au0 (x) = 0
u0 (x)
The above expression is a second-order di¤erential equation, whose characteristic
polynomial is given by
r2 + ar = 0 with roots r = f a; 0g
Therefore, the general solution of this second-order di¤erential equation is,
u(x) = e0x + e ax
= + e ax

Finally, note that the condition that u0 (x) > 0 (utility is nondecreasing in wealth),
implies that parameter > 0.
Intuition: That is, we recovered the utility function u(x) from the Arrow-Pratt
coe¢ cient of absolute risk aversion, up to two constants ( and ). In other
words, the “recovered”u(x) is the family of a¢ ne transformations of e ax (roughly
speaking: all combinations of e ax times a constant, and plus another constant).
Note: for more on …rst- and second-order di¤erential equations, see Simon and
Blume’s Mathematics for Economists, sections 24.2 and 24.3 (in fact, this exercise
is an example of these sections).

16. Prove that risk aversion is equivalent to the following two points.

(a) u( ) is concave;
By de…nition, a decision maker is risk averter if his expected utility from
playing a lottery is weakly lower than his utility from the expected value of
the lottery. That is, if preferences admit Bernouilli function representation,
risk aversion is captured by Jensen’s inequality:
Z Z
u(x)dF (x) u xdF (x)

And in the case in a …nite number of states, x1 ; x2 ; :::; xN , with associated

X
N
probabilities 1 ; 2 ; :::; N respectively, where i = 1this inequality can
i=1
be rewritten as
u( 1 x1 + 2 x2 + ::: + N xN ) 1 u (x1 ) + 2 u (x2 ) + ::: + N u (xN )

and this condition (together with Jensen’s inequality) state the concavity of
function u( ).

17
 Z
(b) c(F; u) x dF (x) for all F (x);

By the de…nition of the Certain Equivalent of F ( ), c (F; u),

Z
u (c (F; u)) = u(x)dF (x)

Recall that the Certain Equivalent of a lottery de…nes the amount of money
for which the decision maker is indi¤erent between playing the gamble (left-
hand side of the above expression) and accepting such amount of money
c (F; u) (right-hand side).
Using Jensen’s inequality
Z Z
u(x)dF (x) u xdF (x)

we can substitute the …rst term by u (c (F; u)) from the de…nition of the
Certain Equivalent.
Z
u (c (F; u)) u xdF (x)

and since u( ) is nondecreasing,

Z
c (F; u) xdF (x)

which is exactly what we had to show: the Certain Equivalent must be o¤ered
in order for him to give up the lottery is weakly lower than the expected value
of the lottery.

17. [Uncertainty about being audited by the IRS] Consider a taxpayer with exoge-
nous income y > 0 who faces a tax rate t, where 0 < t < 1. She is asked to report
a number x to the government and pays taxes tx. If the taxpayer is honest she will
report x = y, but she may cheat by reporting a lower income 0 x < y. Let z = y x
represent the amount by which income is understated. The government does not know
the true income y and must enforce compliance through a system of audits and penal-
ties. Assume that the enforcement policy, known by the taxpayer, is to audit reports
with probability p, where 0 < p < 1. Assume that p is constant and hence independent
of x. If there is an audit, we assume that the government always learns the true income
y. If the taxpayer is caught cheating she must pay a penalty on each dollar of income
evaded, z, in addition to the evaded tax. Assume that the taxpayer is risk averse and
maximizes expected utility.

(a) For any z, where 0 z y, write the income the consumer will enjoy in each one
of the two possible situations, i.e., if there is an audit and if there is not an audit
(notice that the choice variable for the consumer is z).
With probability 1 p, no audit occurs and then she gets an amount of money
y tx = y(1 t) + tz.

18
 With probability p there is an audit, and she gets an amount of money y
tx ( + t)z, or simplifying, y(1 t) z.
(b) Calculate the minimum value of t such that she will choose to cheat [Hint: You
just have to provide a condition under which z > 0).
The problem of the individual is
max U (z) = (1 p) u(y(1 t) + tz) + p u(y(1 t) z)
0 z y

where note that the only choice variable of this individual is the amount of
money he reports, x, or equivalently, the amount of money he understates,
z = y x. By risk aversion we know that this function is concave. Computing
the …rst derivative with respect to z:
U 0 (z) = t(1 p) u0 (y(1 t) + tz) p u0 (y(1 t) z):
Evaluating it at z = 0 we get:
U 0 (0) = [t(1 p) p] u0 (y(1 t)):
This expression will be positive if and only if:
p
t>
1 p
Intuitively, this implies that the marginal utility the individual obtains from
marginally increasing z (away from z = 0) is strictly positive, i.e., U 0 (0) > 0,
when the income tax rate satis…es t > 1 p p . Under this condition, he selects
z > 0 (underreporting some of his income), rather than z = 0 (fully reporting
all his income).
(c) Assume that the consumer chooses z > 0. Prove that the optimal value of z
decreases in the probability of being audited p and in the …ne . [Hint: use the
implicit function theorem].
The optimal value of z > 0 must satisfy the …rst order condition:
U 0 (z ) = t(1 p) u0 (y(1 t) + tz ) p u0 (y(1 t) z ) = 0:
To see the e¤ect of changes in the parameters, we use the Implicit Function
Theorem.
@U 0 (z )
@z @p
= @U 0 (z )
@p
@z
00 @U 0 (z )
As the function is concave, U (z ) = @z
< 0, and therefore the sign of
@z @U 0 (z )
@p
will be the sign of @p
. We have:

@U 0 (z )
= tu0 (y(1 t) + tz ) u0 (y(1 t) z )<0
@p
Therefore, @z
@p
< 0 implying that the individual reduces the amount of money
he underreports to the IRS, z , if the probability of being audited, p, increases.

19
 In the same way we have that using the Implicit Function Theorem,
@U 0 (z )
@z @
= @U 0 (z )
@
@z
@z @U 0 (z )
and by concavity, the sign of @
coincides with the sign of @
. Hence,
@U 0 (z )
= pu0 (y(1 t) + tz ) z u00 (y(1 t) z )<0
@
Therefore, @z
@
< 0 implying that the individual reduces the amount of money
he underreports to the IRS, z , if the …ne on each dollar of income evaded, ,
increases.
(d) Can you prove that z changes monotonically with the tax rate t? If you cannot
provide an unambiguous answer, use the taxpayer’s expected utility maximization
problem from part (a) to provide an interpretation about how an increase in t on
one hand raises the incentives to cheat, and on the other hand decreases those
incentives.
@z @U 0 (z )
The sign of @t
, by concavity, coincides with the sign of @t
. In particular,
@U 0 (z )
= (1 p)u0 (y(1 t)+tz )+(z y)t(1 p)u00 (y(1 t)+tz )+y pu00 (y(1 t)+tz )
@t
The …rst two terms are positive, and the third term is negative. The e¤ect
is ambiguous. Maybe it is better to have a look at this individual’s expected
utility maximization problem. Raising t has two e¤ects.
– First, it lowers y(1 t). This is the after-tax income of full compliance
(no cheating). If absolute risk aversion is decreasing (this is the standard
assumption), this change in the tax rate would make people more risk
averse for all z and consequently less likely to cheat.
– Second, the return to cheating goes up (tz), while the penalty of being
caught ( ) stays unchanged. This second e¤ect encourages cheating.
If the …rst e¤ect dominates the second, then cheating decreases in the tax
rate, otherwise it increases.

18. [Uncertainty about the future] Mr. Pullman lives for exactly two periods, t = 0; 1.
Let ct 2 R denote his consumption in period t. Mr Pullman’s preferences (evaluated
at t = 0) over two-period consumption streams are represented by the function
U (c0 ; c1 ) = u(c0 ) + Eu(c1 )

where is a discount factor, u( ) is an increasing and strictly concave utility function,

and the E operator denotes his expectation (at t = 0) about events in period t = 1.
Initially, suppose that there is no uncertainty. Let w0 > 0 be Mr. Pullman’s income
in period 0 and let w1 0 denote his income in period 1. Mr. Pullman can save or
borrow. Let s 2 R denote his saving (notice that s could be negative if he borrows)
and let denote the gross return on saving (i.e., = 1 + r where r is the interest rate).
Thus, his consumption in period 0 is w0 s and his consumption in period 1 is w1 + s.
Assume interior solutions throughout the exercise.

20
(a) Write down necessary and su¢ cient conditions for Mr. Pullman’s choice of saving,
s , to be greater than zero.
Mr. Pullman’s utility maximization problem is

max u(c0 ) + u(c1 ) subject to c0 = w0 s and c1 = w1 + s

c0 ;c1

or alternatively,
max u(w0 s) + u(w1 + s)
s

Taking FOCs with respect to s, we obtain

u0 (w0 s )+ u0 (w1 + s ) = 0

Hence, for s > 0, we require that u0 (w0 ) + u0 (w1 ) > 0, as depicted in the
…gure below. And in addition, the expression u0 (w0 s ) + u0 (w1 + s )
decreases in s, which is guaranteed by the strict concavity of u( )

Positive savings, s > 0.

u0 (w0 )
and u0 (w0 ) + u0 (w1 ) > 0 implies > u0 (w1 )
.
Intuition: the gross return on saving (the rate at which Mr. Pullman
can transfer consumption from today to tomorrow) must be greater than
his marginal rate of substitution evaluated at the endowment (w0 ; w1 ) (that
is, the rate at which he is willing to transfer consumption from today to
consumption tomorrow).
(b) Suppose that w1 = 0 and that the conditions you found in part (a) hold. Find a
condition on Mr. Pullman’s coe¢ cient of relative risk aversion that is necessary
and su¢ cient for s to be (locally) increasing in .

21
 Evaluating the FOC we found in (a) at w1 = 0, we have

u0 ( s ) = u0 (w0 s)

Di¤erentiating this FOC with respect to we obtain

2 @s @s
u0 ( s ) + u00 ( s ) s + = u00 (w0 s)
@ @

and rearranging,
@s
u00 ( s ) 2
u00 (w0 s) = [u00 ( s )( s ) + u0 ( s )]
@

and dividing both sides of the equality by u0 (w0 s ) yields

u00 ( s ) 2 u00 (w0 s ) @s u00 ( s )( s ) u0 ( s )

= + 0
u0 (w0 s ) u0 (w0 s) @ u0 (w0 s ) u (w0 s )

Furthermore, since from the FOC we know that u0 (w0 s ) = u0 ( s ) > 0,

it follows from the above expression can be rewritten as:

u00 ( s ) 2 u00 (w0 s ) @s u00 ( s )( s ) u0 ( s )

= +
u0 ( s ) u0 (w0 s) @ u0 ( s ) u0 ( s )
1 u00 ( s )( s )
= +1
u0 ( s )
u00 ( s ) 2
In addition, since the term on the left-hand side is positive, i.e., u0 ( s )
u00 (w0 s )
u0 (w0 s )
> 0, given that u0 > 0 and u00 < 0, a necessary and su¢ cient
condition for @s
@
> 0 is that the right-hand side of the equality is positive as
well, that is

u00 ( s )( s ) u00 ( s )( s )
+ 1 > 0 () <1
u0 ( s ) u0 ( s )

Interpretation: Hence, the Arrow-Pratt’s coe¢ cient of relative risk aversion,

u00 (x)x
rR (x) = u0 (x) , must be lower than one. This property holds for most of the
p
concave utility functions described in class, e.g., for u(x) = x the coe¢ cient
rR (x) becomes rR (x) = 12 < 1. Intuitively, this condition states that for Mr.
Pullman to increase his savings as a response of a larger return , his utility
function must be su¢ ciently concave.
(c) Now suppose that Mr. Pullman faces uncertainty over his period 1 income. Specif-
e where w1
ically, suppose that his period 1 income is given by w1 + x 0 and
E(ex) = 0. Let s denote Mr. Pullman’s new optimal saving in this context.
Show that s > s . [Hint: Suppose that s = s and compare the …rst order
conditions using Jensen’s inequality.]

22
Under uncertainty, the FOC becomes

u0 (w0 s )+ E[u0 (w1 + x

e + s )] = 0

By Jensen’s inequality, the strict convexity of u0 (i.e., u000 > 0) implies

u0 (w0 s )+ E[u0 (w1 + x

e + s )] > u0 (w0 s )+ u0 (w1 + E(e
x) + s )]

and, since E(e

x) = 0, the right-hand side of the inequality can be more com-
pactly expressed as

u0 (w0 s )+ u0 (w1 + s )] = 0

which exactly coincides with the …rst-order condition of exercise (a) and,
hence, it equal to zero. Hence, the above inequality becomes

u0 (w0 s )+ E[u0 (w1 + x

e + s )] > 0

indicating that, at a level of savings s , the curve u0 (w0 s ) + E[u0 (w1 +

e + s )] is still positive, i.e., it has not crossed the horizontal axis yet.
x
Under uncertainty, the FOC becomes

u0 (w0 s )+ E[u0 (w1 + x

e + s )] = 0

By Jensen’s inequality, the strict convexity of u0 (i.e., u000 > 0) implies

u0 (w0 s )+ E[u0 (w1 + x

e + s )] > u0 (w0 s )+ u0 (w1 + E(e
x) + s )]

and, since E(e

x) = 0, the right-hand side of the inequality can be more com-
pactly expressed as

u0 (w0 s )+ u0 (w1 + s )] = 0

which exactly coincides with the …rst-order condition of exercise (a) and,
hence, it equal to zero. Hence, the above inequality becomes

u0 (w0 s )+ E[u0 (w1 + x

e + s )] > 0

indicating that, at a level of savings s , the curve u0 (w0 s ) + E[u0 (w1 +

e + s )] is still positive, i.e., it has not crossed the horizontal axis yet.
x

23
 The following …gure illustrates this comparison.

Comparing s and s

In particular, the …gure depicts curve u0 (w0 s ) + u0 (w1 + s )] that we

used in part (a) to identify the level of savings s , and curve u0 (w0 s ) +
[u0 (w1 + E(e
x) + s )] that we used in part (c) to identify the level fo savings
s . Graphically, when evaluating both curves at the same level of savings s ,
we obtain that the latter is above the former. Hence, s > s .

19. [Uncertainty in agricultural markets] A farmer uses his own labor, x, to produce
wheat, q, with the production function q = f (x). At the end of the harvesting season,
the farmer sells its production at the given world price of wheat. The price he receives
is a random variable, in particular
pL with probability , and
p=
pH with probability 1
where pH denotes a high price, whereas pL represents a low price of wheat, and pH > pL .

The farmer’s preferences are given by utility function u(wk ) f (x), where wk
represents the farmer’s income when the realization of the random variable is
k, where k = fH; Lg. Consider that u( ) is increasing and concave in wk , i.e.,
@u @u2
@wk
> 0 and @w 2 < 0. Assume that the cost of e¤ort is increasing and convex,
k
@f @f 2
i.e., @x
> 0 and @x2
> 0, and denote by w0 the farmer’s initial income
The government has just created a price-guarantee program in which the farmer is
guaranteed a price pG , where pH > pG > pL , for the proportion of his production
that he includes in this program at the beginning of the season, that is, before
knowing the particular world price for that year. Let 2 [0; 1] denote the propor-
tion of the farmer’s production that he chooses to include in the price-guarantee
program.

24
(a) Describe the farmer’s income level when world prices are high and low. Set up
the farmer’s expected utility maximization problem.
When world prices are low, the farmer’s income level becomes

w0 + pL (1 )x + pG x;

where the third component represents the revenue he obtains from the price-
guarantee program, whereas the second component represents the revenue
he obtains from selling the rest of his production at the current world price
pL . Similarly, when world prices are high, the farmer’s income becomes w0 +
pH (1 )x + pG x.
Therefore, the farmer’s expected utility maximization problem is

max u w0 + pL (1 )x + pG x +(1 ) u w0 + pH (1 )x + pG x f (x)

x;

(b) Take …rst-order conditions with respect to x and , and characterize the (implicit)
solution. For simplicity, assume interior solutions.
Taking FOCs with respect to x, we obtain

u0 w0 + pL (1 )x + pG x pL (1 ) + pG
+(1 ) u0 w0 + pH (1 )x + pG x pH (1 ) + pG f 0 (x)(1)

and taking FOCs with respect to , we have

u0 w0 + pL (1 )x + pG x pL x + pG x
+(1 ) u0 w0 + pH (1 )x + pG x pH x + pG x (2)

Combining (1) and (2), we can rewrite them as the following implicit equation

u0 w0 + pL (1 )x + pG x 1 pH pG
=
u0 (w0 + pH (1 G
)x + p x) pG pL

(c) Consider now that the price-guarantee program produces zero expected pro…ts
for the government. Explain how your implicit solution for x and in question
(b) is a¤ected.
When world prices are low, the government must pay a price pG for an amount
of output x, and sells that output later to the open market at a (lower) price
of pL , making a loss of pL pG x.
When world prices are high, the government must pay a price pG for an
amount of output x, and sells that output later to the open market at a
(higher) price of pH , making a pro…t of pH pG x.
Hence, the government makes no expected pro…t from the price-guarantee
program if

pL pG x + (1 ) pH pG x=0

25
 we can rearrange this expression as
(1 ) pH pG x= pG pL x
and dividing both sides by pG pL x, we obtain
1 pH pG
=1
pG pL
Using this result into the implicit condition found in question (b), we have
u0 w0 + pL (1 )x + pG x
=1
u0 (w0 + pH (1 )x + pG x)
which implies that pL (1 )x = pH (1 )x. However, given that pH > pL ,
this condition implies that = 1, i.e., the farmer includes all his production
in the price-guarantee program.
On the other hand, if we plug = 1 into the FOC we obtained in question
(b) for x, we obtain
u0 w0 + pG x pG + (1 ) u0 w0 + pG x pG f 0 (x) = 0
that is, u0 w0 + pG x = f 0 (x).

20. [Uncertainty in production decisions] Consider a …rm with initial pro…ts given by
2
0 = 15, and cost function c(q) = 10 + q . The price at which this …rm sells its pro-
duction is stochastically distributed according to the following probability distribution
pH = $8 with probability 12 , and
pe =
pL = $2 with probability 12
where the tilde sign (e) emphasizes that the price level is a random variable.

(a) If the …rm manager’s utility function is given by

1
u(e) = E[e]V ar(e)
9
determine the …rm’s optimal production level, and the associated equilibrium
pro…ts from producing such an output level.
We …rst separately …nd E[e] and V ar(e). On one hand, E(e) is given by
E[e] = E[ + peq
0 10 q 2 ]
1 1
= 15 + 8+ 2 q 10 q2
2 2
= 5 + 5q q 2 .
On the other hand, V ar(e) is given by
V ar(e) = V ar[ 0 + peq 10 q 2 ]
= V ar[epq]
1 1
= (8q 5q)2 + (2q 5q)2 q
2 2
2
= 9q

26
 Therefore, the …rm manager’s utility function can be expressed as
1
u(e) = E[e] V ar(e)
9
1 2
= 5 + 5q q2 9q
9
= 5 + 5q 2q 2

Hence, the …rm manager’s maximization problem becomes

max u(e) = 5 + 5q 2q 2
q 0

Taking FOCs with respect to q yields

@u(e)
=5 4q 0
@q

implying that q = 54 at the interior solution. Plugging this result into the
objective function u(e) entails equilibrium pro…ts of
1 155 1 225 585
u(e) = E[e] V ar(e) = =
9 16 9 16 72
(b) Consider now that the …rm manager’s utility function is described by u( ) = E[ ].
What is now the optimal production level and the associated equilibrium pro…ts?
In this case, the …rm manager is risk neutral. Hence, when selecting the op-
timal production level, he only considers expected pro…ts. His maximization
problem becomes
max u(e) = 5 + 5q q 2
q 0 | {z }
E[e]

Taking FOCs with respect to q yields

@u(e)
=5 2q 0
@q

implying that q = 52 at the interior solution, i.e., he increases his production

decision relative to part (a) where he is risk averse. Plugging this result into
the objective function u(e) entails equilibrium pro…ts of
45
u(e) = E[e] =
4

27
 Chapter 8 - Monopoly
End-of-Chapter Exercises - Answer key

1. [Lerner Index]. The expression

pm c0 (q m ) 1
=
pm "d (pm )

where pm and q m .are the monopolist’s price and output level, respectively, is known
as the monopolist’s price-cost margin (or as the Lerner index of monopoly power).
It measures the distortion of the monopolist price, pm , above its marginal cost (as a
proportion of its price). Interestingly, it can be shown that this price-cost margin is
always equal to 1="d (pm ), i.e., the inverse of the price elasticity of demand at price pm .
Assume c0 ( ) > 0 for all output levels q > 0.

(a) Prove the above equality. [Hint: start by di¤erentiating total revenue].
The monopolists maximizes x (p) p c (x (p)) ;which leads to the following
…rst order condition:

x0 (pm ) pm + x (pm ) = c0 (x (pm )) x0 (pm ) .

Rearranging this we get

[pm c0 (x (pm ))] 1

m
= ;
p " (pm )

where the elasticity of demand is

pm
" (pm ) = x0 (pm ) :
x (pm )

(b) Discuss the relationship between the Lerner’s index, and elasticity of demand.
Since c0 (x (pm )) > 0 always, then

[pm c0 (x (pm ))] [pm 0]

< = 1.
pm pm
Therefore,
1
< 1; or " (pm ) > 1.
" (pm )
Which implies that if the marginal cost is positive, necessarily, the elasticity
of demand will always be larger than 1, i.e., the monopolist produces in the
elastic section of the demand curve.
(c) Show that, at the monopolistic’s optimal price, demand must be elastic (i.e., price
elasticity is greater than one).

1
 From above we know that
1
< 1; or " (pm ) > 1.
" (pm )

so the monopolist operates in the elastic section of the demand curve.

2. Suppose that the government can tax or subsidize a monopolist who faces inverse
demand function p(q) and has cost function c(q)[assume both are di¤erentiable and
that p(q)q-c(q) is concave in q]. What tax or subsidy per unit of output would lead
the monopolist to act e¢ ciently?

Let t be the tax/subsidy per unit of output. Then the monopolist maximizes

max p(q)q c(q) tq

q 0

and the FOC is

p0 (q)q + p(q) c0 (q) t 0, with equality if q > 0

Solving for t (and assuming an interior solution) we obtain

t = p0 (q)q + p(q) c0 (q)

Since we want an e¢ cient outcome, the monopolist must choose a level of output
q that satis…es p(q) = c0 (q). This implies that the tax must be t = p0 (q)q, so that
the above condition becomes

p0 (q)q = p0 (q)q + p(q) c0 (q) =) c0 (q) = p(q)

However, we know that t = p0 (q)q < 0 since p0 (q) < 0. Therefore, the tax that the
government imposes on the monopoly is actually a "negative tax", i.e., a subsidy.
Intuitively, the government must subsidize the monopolist in order to induce a
larger level of production.

3. [Monopolist with interdependent demands] Consider a monopolist selling two

di¤erent goods, q1 and q2 , whose demands are interdependent and given by

qi = a bpi + gpj for i = f1; 2g and j 6= i

where b > 0, which guarantees that the demand for a particular good decreases in the
price of that same good. In addition, note that if g > 0 then goods are substitutes while
if g < 0 they are complements (and if g = 0 the products are independent of each other).
Assume also that jgj < b, which guarantees that the own-price e¤ect on the demand of
a product i is larger than the cross-price e¤ect. Finally, consider that a > c(b g) in
order to guarantee that output is strictly positive in equilibrium. In order to focus on
the interdependence of demands, let us assume that the marginal costs of production
coincide across both products. That is, total costs are T C(q1 ; q2 ) = c(q1 + q2 ).

2
(a) Find the pro…t-maximizing price for good 1, p1 , and for good 2, p2 , for this monop-
olist. [Hint: since we have the direct demand function, this exercise is probably
easier if you …nd pro…t-maximizing prices than pro…t-maximizing output.]
The monopolist PMP is
max = (a bp1 + gp2 )(p1 c) + (a bp2 + gp1 )(p2 c)
p1 ;p2

and FOCs are

@
= a 2bpi + 2gpj + c(b g) = 0, for every product i; j = f1; 2g and i 6= j
@pi
and at the symmetric solution p1 = p2 = pm , which implies
a 2bpm + 2gpm + c(b g) = 0,
and solving for pm , we obtain
a + c(b g)
pm = ,
2(b g)
where clearly pm > c.
(b) Do prices increase or decrease in parameter g?
Taking the …rst order derivative of pm with respect to g,
@pm a
= >0
@g 2(b g)2
implying that as g increases in the interval ( b; b), the price charged by the
monopolist on both products also increases. [It is easy to check that pm is
concave in g, with its lowest possible value being a+2bc
4b
, which occurs when
g ! b, and an asymptote as g ! b. ]
For illustration purposes, the next …gure assumes parameters a = 1, b = 1
and c = 0, implying a price of pm = 2(11 g) , and allows for g to take values in
the interval ( b; b), i.e., ( 1; 1).

3
 (c) Compare prices if g > 0 and if g < 0 with those where g = 0. Explain
Relative to the benchmark case where the two products are independent
(g = 0), this implies that the monopolist reduces the price of its product when
they are complements (g < 0) and it increases it when they are substitutes
(g > 0). The intuition for this result is straightforward. When products are
complements, they exercise a positive e¤ect on each other and the monopolist
internalizes such e¤ect by decreasing its prices (i.e., a lower price of good 1
stimulates sales of good 2, and vice versa). In other words, if products 1 and
2 were sold by two distinct monopolists, consumer would pay more for them
than when they are sold by the same …rm. [This is a standard result in the
literature on vertically integrated …rms studied in industrial organization].
When instead, products are substitutes (g > 0), the external e¤ect they
exercise on each others’demands is negative, and the monopolist controls it
by raising prices (a lower price of good 1 crowds out sales of good 2 and vice
versa). If products 1 and 2 were sold by two distinct …rms, consumers would
pay less than when they are sold by the same …rm.
Remark: note that the insights obtained above carry over to the case where
the monopolist sells the same product in sequential markets, i.e., to a group of
customers in the …rst period and to a di¤erent set of customers in the second
period. In particular, we can re-interpret the demand relationship as one of
intertemporal substitutability and complementarity, i.e., demand in the …rst
period stimulates (reduces) demand in the second period when g > 0 (g < 0,
respectively).

4. [Monopoly: leasing vs. selling] Consider a two-period game where a monopolistic

…rm wants to sell its durable good. The durable good will last only two periods, and
after that it will become obsolete. There is no depreciation of the good between the
two periods. The discount factor is identical for all consumers and the …rm. Demand
for the utilization of the good is given by p = 1 Q, where Q denotes the aggregate
quantity. Production is assumed to be costless. A resale market exists, since consumers
who buy the good in one period might want to re-sell (or lease) it in the second period.

(a) Consider …rst the case where the …rm sells in each period.
1. Starting from the second period, set up the pro…t-maximization problem for
the monopolist, where it selects a production level q2 given a demand function
p2 = 1 q1 q2 . Determine q2 , p2 , and pro…ts 2 during this second period.
If the monopolist decides to sell, the quantity sold during period 1 is
re-o¤ered in period 2 (remember that a resale market exists). Hence, a
monopolist who has produced q1 units in period 1 will sell in period 2 the
quantity which maximizes its pro…t

max (1 q1 q2 )q2 .
q2

that is q2 = 1 2q1 . Notice that the price of the good at period 2, p2 , is

the one for which the total quantity produced in both periods (q1 + q2 ) is

4
 equal to the quantity demanded, i.e., p2 = 1 q1 q2 . Hence,
1 q1 1 q1
p2 = 1 q1 =
2 2
2
yielding second-period pro…ts of 2 = (1 4q1 ) .
2. Given the equilibrium price you found for the second-period monopoly, p2 , the
…rst-period demand is p1 = (1 q1 )+ p2 , which intuitively represents that the
willingness to pay for the good in the …rst period is given by the current value
that the consumer assigns to this good (given by the demand function, 1 q1 ),
plus the discounted value of the good tomorrow (which arises if the current
consumer leases the good in the second period at a price p2 ). Given this
…rst-period demand, set up the monopolist’s pro…t-maximization problem,
where its choice variable is now q1 , and its objective function considers not
only …rst-period but also the discounted value of second-period pro…ts, 2 .
Determine q1 , p1 , and overall pro…ts across both periods.
At stage 1, the price consumers will be willing to pay for the good will ob-
viously depend on their expectations about p2 . Assume that consumers
are rational, their expectations must satisfy E(p2 ) = p2 . Hence, con-
sumers’willingness to pay, p1 , is given by the current value of the good,
1 q1 , plus the discounted value of the good tomorrow (which arises if
the current consumer leases it tomorrow at p2 ),
1 q1
p1 = (1 q1 ) + p2 = (1 q1 ) + = (1 q1 ) 1 + ;
2 2
where notice that p1 > p2 . When the monopolist sells in each period,
price fall over time. The monopolist then chooses q1 as to maximize its
present value of the pro…ts from selling the good, S , given by

(1 q1 )2
max (1 q1 ) 1 + q1 +
q1 2 4
Taking …rst-order conditions with respect to q1 , and solving for q1 entails
q1 = 4+2 . Therefore, the …rst-period price is

2 (2 + )2
p1 = 1 1+ =
4+ 2 2(4 + )
2
(2+ )2 2 (1 4+2 ) (2+ )2
yielding pro…ts of S = 2(4+ ) 4+
+ 4
= 4(4+ )
.
3. Show that equilibrium prices decline over time, i.e., p1 > p2 .
p1 > p2 , since (1 q1 ) 1 + 2 > 1 2q1 holds for any discount factor 2
(0; 1).
(b) Consider now that the monopolist leases (i.e., rents) the good in each period, and
…nd equilibrium prices and output. [Hint: when leasing its goods, the monopolist’s
pro…ts on a given time period t become independent on other period prices, pk ,
where k 6= t]

5
 If, instead, the monopolist decides to lease, then at each period of time t =
f1; 2g, it sets a price such that

max pt (1 pt ).
pt

This implies that p1 = p2 = 21 , since the pro…ts of a given time period are
independent on other period’s prices. Then, it produces q1 = 1 p1 = 1 12 =
1
2
and

1 1
p2 = 1 q1 q2 =) q2 = (1 q1 ) p2 = 1 =0
2 2
since it was assumed that there is no depreciation.
(c) Find the monopolist’s equlibrium pro…ts from leasing. Are they higher or lower
than the pro…t the monopolist makes from selling the good, i.e., your result from
part a(3)?
The present value of the pro…ts from leasing is then given by
11 11 1+
L = + = .
22 22 4
As can be easily checked, L > S . In particular, solving for in the inequal-
(2+ )2
ity 1+4 > 4(4+ )
we obtain > 0, which is true by de…nition.

5. [Discrimination or Nondiscrimination?]. Ann’s total demand for good x is given

by xA (p) = a A p, and Bob’s total demand is xB (p) = a B p, where B < A . The
cost of production is c per unit.

(a) Suppose that the market for good x is competitive. Find the equilibrium quantity
and price.
In a competitive equilibrium we must have pB = pA = c, and therefore
xB = a B c, and xA = a A c:

(b) Suppose, instead, that …rm F is a monopolist. If …rm F is prohibited from “dis-
criminating” (i.e., charging di¤erent prices to Ann and Bob), what is its pro…t
maximizing price? Under which conditions do Ann and Bob consume a positive
amount of good x in this solution?
If the monopolist will serve both markets it solves,

max (a B p) (p c) + (a A p) (p c) .
p

The FOC yields

a c
p = + .
B + A 2
However, since the women are willing to pay higher prices than men for the
same quantities ( B > A ) then it may be better to charge a price above
a
B
. This cannot be captured in the program above since it causes negative

6
 quantities in Bob’s demand, so we have to compare the solution p above to
the solution of max (a A p) (p c) which yields
a c
pb = + .
2 A 2
The following …gure shows the two cases.

The aggregate demand curve must have a kink: for prices above the kink, only
women will buy positive units of the good, implying that aggregate demand
coincides with women’s demand. In contrast, for prices below the kink both
Bob and Ann buy positive units of the good, implying that aggregate demand
is equal to the sum of the demand of Bob and the demand of Ann.
Due to the kink in the demand curve, the marginal revenue curve M R has a
jump at that point (the dashed line). When equating the marginal cost c,
to M R, we could have one or two solutions:

(a) 1. When we have one solution, it means that c cuts M R only to the left of
the discontinuity, or only to the right. If c cuts M R only to the left, then
the optimal price is above aB and we only serve Ann’s markets.
2. When we have two solutions, then c cuts M R both to the left and to the
right of the discontinuity, and this is shown in the …gure by points x and
y. When the monopolist moves from point x to y (i.e., from pb to p ) it
loses pro…ts equal to the triangle A, and gains pro…ts equal to the triangle
B. The costs and bene…ts of such a move will determine if pb is optimal
(only Ann is served) or if p is (both markets are served).

(a) c. If …rm F (still monopolist) has produced some total level of output X, what is
the welfare-maximizing way to distribute it between Ann and Bob? [Assume here
and below that Marshallian aggregate surplus is a valid measure of welfare].
Given quantity X, we maximize consumer surplus by solving,
Z qB Z qA
max pB (x) dx + pA (x) dx
X 0 0

7
 subject to qB + qA = X
and letting denote the Lagrange multiplier we get the FOCs pB (qB ) =
and pA (qA ) = ;which yield pB (qB ) = pA (qA ), or

(a qB ) (a qA )
= .
B A

Together with the constraint we solve for qB and qA , i.e., qA + qB = X

(b) d. Suppose that …rm F (still a monopolist) is allowed to discriminate. What
prices does it charge?
The discriminatory monopolist solves

max (a B pB ) (pB c) + (a A pA ) (pA c)

pB ;pA

which FOC’s yield

a+c B a+c A
pB = , and pA =
2 B 2 A
These prices yield quantities of

(a c B) (a c A)
qB = , and qA =
2 2
Note that aggregate output under price discrimination is equal to the aggre-
gate output without price discrimination if both markets were served

( B + A) c
Q=a
2
(c) e. In the case where the nondiscriminatory solution in (b) has positive consump-
tion of good x by both Ann and Bob, does aggregate welfare as measured by
the Marshallian aggregate surplus rise or fall, relative to when discrimination is
allowed? Relate your conclusion to your answer in (c).
From part (c), we know that the welfare maximizing distribution for a given
output is to set pB q B = pA q A , which is the case without price discrimi-
nation when both markets are served.
Notice that under price discrimination pB q B 6= pA q A if B 6= A , which
implies that welfare is lower under price discrimination.
(d) f. What if the nondiscriminatory solution in (b) has only one type of consumers
being served?
If, however, without price discrimination only Ann was served, then by al-
lowing the monopolist to discriminate it will open Bob’s market (assuming
that c < aB ) without changing the price in the women’s market. This means
that there is added surplus from Bob’s market and welfare under price dis-
crimination is higher than welfare when price discrimination is forbidden.

8
6. [Multiplant Monopoly]. Imagine that Gillette has a monopoly in the market for
razor blades in Mexico. The market demand curve for blades in Mexico is p (Q) =
968 20Q, where p is the price of blades and Q is annual demand for blades. Gillette
has two plants in which it can produce blades for the Mexican market: one in Los
Angeles and one in Mexico City. In its L.A. plant, Gillette can produce any quantity
of blades it wants at a marginal cost of 8 cents per blade. Letting and denote the
output and marginal cost at the L.A. plant, we have M C1 (Q1 ) = 8. The Mexican
plant has a marginal cost function given by M C2 (Q2 ) = 1 + 0:5Q2 .

(a) Find Gillette’s pro…t maximizing total quantity of output (and denote it by QT )
and price, for the Mexican market overall.
Since the pro…t-maximizing …rm will always allocate output among plants so
as to keep marginal costs equal, and since the …rst plant in this example has a
constant marginal cost of 8, the pro…t-maximizing solution will have M C = 8
at both plants. To compute total output, set M R = M C. With a demand
function of p (Q) = 968 20Q, marginal revenue is M R = 968 40Q. This
implies,
968 40Q = 8 () Q = 24 units
At this quantity, the …rm will charge a price of P = 968 20(24) = 488.
Therefore, the price for one razor will be $488 and Gillette will supply 24
million blades.
(b) How will Gillette allocate production between its Mexican plant and its L.A.
plant? That is, what part of QT is due to Q1 , and what part is due to Q2 ?
The allocation between plants will require M C = 8 at both plants. At plant
2, M C2 (Q2 ) = 1 + 0:5Q2 . Setting this equal to 8 implies Q2 = 14. Then,
since total output is 24, the …rm will produce 10 at plant 1, Q1 = 10:
(c) Suppose Gillette’s plant in L.A. had a marginal cost of 10 cents rather than 8
cents. That is, suppose now that M C1 (Q1 ) = 10 while M C2 (Q2 ) is unchanged.
How would your answer in parts a) and b) change?
If at plant 1, the setting M R = M C implies

968 40Q = 10 () Q = 23:95

At this quantity, price will be p = $489. The allocation between plants will
require M C = 10 at both plants. Setting M C = 10 at plant 2 implies
10 = 1 + 0:5Q2 implying Q2 = 18. Since total output is 23.95, the …rm will
produce the remaining Q1 = 5:95 at plant 1.

7. [Cost reducing investment]. Consider a situation in which there is a monopolist

in a market with inverse demand function p(q). The monopolist makes two choices:
How much to invest in cost reduction, I, and how much to sell, q. If the monopolist
invests I units in cost reduction, his (constant) per-unit cost of production is c(I).
Asume that c0 (I) < 0 and that c00 (I) > 0, i.e., investing in cost reduction reduces the
monopolist’s per-unit cost of production, but at a decreasing rate. Assume throughout
that the monopolist’s objective function is concave in q and I.

9
(a) Derive the …rst-order conditions for the monopolist’s choices.
The monopolist will solve

max p (q) q c (I) q I

q;I

which yields the FOCs are

(i) p0 (q m ) q m + p (q m ) = c (I m ) ,
(ii) c0 (I m ) q m = 1.

(b) Compare the monopolist’s choices with those of a benevolent social planner who
can control both q and I (a “…rst-best”comparison). Interpret your results.
The social planner will maximize total surplus,
Z q
max p (x) dx c (I) q I,
q;I 0

and the FOCs are,

(iii) p (q ) = c (I ) ,
(iv) c0 (I ) q = 1.
The monopolist produces less output than is socially optimal, q m < q , and
price is above marginal cost. Given this, equations (ii) and (iv) imply that
c0 (I ) < c0 (I m ) =) c0 (I ) > c0 (I m ), in addition, since c00 (I) > 0 by
assumption, then I > I m That is, the monopolist invests less in cost-reducing
technologies than the social planner would.
(c) Compare the monopolist’s choices with those of a benevolent social planner who
can control for I but not for q (a “second-best” comparison). In particular,
suppose that the social planner chooses I and then the monopolist chooses q.
Given a level Ib set by the government, the monopolist will set q to maximize
its pro…ts, i.e., it will set q to equate M R = M C. Therefore, the governments
problem is to maximize social surplus subject to the monopolists’s behavior.
That is, Z q
max p (x) dx c (I) q I
q;I 0

subject to p0 (q) q + p (q) = c (I)

The Lagrangian is
Z q
L= p (x) dx c (I) q I [p0 (q) q + p (q) c (I)] ,
0

which yields the FOCs,

(v) p (b
q) c Ib [p00 (b
q ) qb + 2p0 (b
q )] = 0,
(vi) c0 Ib (b
q ) = 1.

10
 When comparing (ii) and (vi) we can see that

c0 (I m ) q m = c0 Ib (b
q ) , where >0

Hence, the social planner’s investment, Ib is greater than the optimal invest-
ment level I , found in the previous exercise The intuition is that in this
second-best environment, the social planner chooses an investment level Ib
larger than optimal for the given level of output in order to induce the mo-
nopolist to produce more.

8. [Intertemporal pricing] Consider the following two-period monopoly model: A …rm

is a monopolist in a market with an inverse demand function (in each period) of
p(q) = a bq. The cost per unit in period 1 is c1 . In period 2, however, the monopolist
has “learned by doing,” and so its marginal costs decrease to c2 = c1 mq1 , where q1
is the monopolist’s period 1 output level. Assume that a > c > 0 and b > m > 0. Also
assume that the monopolist does not discount future earnings, i.e., the discount factor
= 1.

(a) What is the monopolist’s output level in each of the periods, q1 and q2 ? Denote
them by q1M and q2M .
The monopolists intertemporal problem is
max (a bq1 c) q1 + (a bq2 (c mq1 )) q2 ,
q1 ;q2

and the FOCs yield

a c
q1m = q2m = >0
2b m
(which is positive by the exercise’s assumptions a > c > 0 and b > m > 0).
(b) Consider a benevolent social planner with social welfare function W given by
W = (CS1 + 1) + (CS2 + 2)

where CSt and t represent respectively consumer surplus and pro…ts during
a given period t = f1; 2g. What output levels would be implemented by the
benevolent social planner? Denote them by q1SP and q2SP .
A benevolent social planner maximizes total surplus (assuming no discounting
of the future, we just add up both periods’consumer surplus, to both periods’
…rm’s pro…ts, and subtract both periods’costs),
Z q1 Z q2
max SW = p (x) dx + p (x) dx cq1 (c mq1 ) q2
q1 ;q2 0 0
1
= a (q1 + q2 ) b q12 + q22 cq1 (c mq1 ) q2
2
and the FOCs are,
(i) (a bq1 ) + mq2 = c
(ii) (a bq2 ) = c mq1
which yield
a c
q1SP = q2SP = > 0.
b m

11
 (c) Can you interpret the choice of q1SP as being selected according to the “price equal
to marginal cost”rule?
Comparing this with the monopoly quantities we see that qim < qiSP . The
way we wrote down the FOCs show that in fact there is a sense of "price
equals marginal cost". Recall that price is marginal surplus, and the left
hand side of both FOCs is exactly the e¤ective marginal surplus from each
period’s good: In the …rst period, aside from marginal consumer surplus,
given q2 , any additional unit of q1 reduces marginal cost next period by mq2 .
The right hand side is the e¤ective marginal cost in each period.
(d) Given that the monopolist will be selecting the period 2 output level, q2M , would
the social planner like the monopolist to slightly increase the level of period 1
output above that identi…ed in part (a), q1M ? Can you given any intuition for
this?
As we have seen, the social planner would want to produce more in every
period. By increasing the output in the …rst period above am i , welfare in the
…rst period will be higher, and this will lead to a lower second period marginal
cost. This lower second period marginal cost will induce the monopolist to
produce more in the second period and will therefore further increase welfare.

9. [Regulating a natural monopoly] A water supply company provides water to Pull-

man. The demand for water in Pullman is p(q) = 10 q, and this company’s costs are
c(q) = 1 + 2q.

(a) Depict in a …gure: the demand curve p(q), the associated marginal revenue M R(q),
the marginal cost of production M C(q) and the average cost of production AC(q).
Discuss why this situation illustrates a “natural monopoly.”
See the …gure below.

(b) Determine the amount of water that this …rm will produce if left unregulated as
a monopolist. Determine the corresponding prices and pro…ts for the …rm.
The monopolist maximizes
max (10 q)q (1 + 2q)
q

12
 Taking FOCs and solving for q we obtain q = 4 units, sold at a price p = $6
, with monopoly pro…ts = $11:
(c) Determine the amount of water that this …rm will produce if a regulatory agency
in Pullman forces the …rm to produce an amount of output q that solves p(q ) =
M C(q ). Determine the corresponding prices and pro…ts for the …rm.
In that case, the monopolist sets 10 q = 2, i.e., q = 8 units, at a price
p = $2, with corresponding losses of = (10 8)8 (1 + 2 8) = 1. (This
result arises because the presence of decreasing average costs).
(d) Consider now that the regulatory agency allows the monopoly to charge two
di¤erent prices: a high price p1 for the …rst q1 units, and a low price p(q ) for the
remaining (q q1 ), i.e., the units from q1 until the output level you found in part
(c), q . In addition, the regulatory agency imposes the condition that the …rm
cannot make any pro…ts, = 0, when charging these two prices.
1. Find the value of q1 , and the associated value of p(q1 ).
First, note that the value of q1 must satisfy the “no pro…ts” condition,
that is

= [p(q1 ) AC(q1 )] q1 + [p(q ) AC(q )] (q q1 ) = 0

and since we know from part (c) that q = 8 units, and that p(q) = 10 q
and AC(q) = 1+2q
q
= 1q + 2, we can rewrite the above condition as

1 1
= (10 q1 ) +2 q1 + (10 8) +2 (8 q1 ) = 0
q1 8

and solving for q1 we obtain two solutions:

First solution: q1 = 0:82 with a corresponding price of p1 = $9:18. This
means that the …rst 0.82 units are sold at $9.18 each, while the remaining
7.18 (up to 8 units) are sold at a price of $2.
Second solution: q1 = 0:30 with a corresponding price of p1 = $9:7. This
means that the …rst 0.30 units are sold at $9.7 each, while the remaining
9.7 (up to 8 units) are sold at a price of $2.
2. Depict these two prices and quantities in a …gure, and shade the area of
bene…ts and losses for the …rm.

13
See the …gure below.

14
 Chapter 9 - Externalities and Public Goods
End-of-Chapter Exercises - Answer key

1. [Externalities] Consider an economy with two individuals i = f1; 2g with the follow-
ing quasi-linear utility function

ui (si ; q i ) = v i (si ) + q i

where si denotes the speed at which individual i drives his car, q i is money, and > 0.
The utility that individual i obtains from driving fast cars is v i (si ), which is increasing
i i 2 v i (si )
but concave in speed, i.e., @v@s(si ) > 0 and @(@s i )2
< 0. Driving fast, however, increases
the probability of su¤ering a car accident, represented by (si ; sj ). This probability is
increasing both in the speed at which individual i drives, si , and the speed at which
the other individual drives, sj where j 6= i. Hence, the speed of other individuals
imposes a negative externality on driver i, since it increases the risk of him su¤ering
a car accident. If individual i su¤ers an accident, he bears a cost of ci > 0, which
intuitively embodies the cost of …xing his car, health-care expenses, etc.

(a) Set up individual i’s expected utility maximization problem. Take …rst-order
conditions with respect to si . Denote the (implicit) solution to this …rst-order
condition as sbi .
Every individual i maximizes his expected utility as follows

max
i
v i (si ) + q i (si ; sj ) ci
s

Taking …rst-order conditions with respect to si we obtain

@v i (si ) @ i
c =0 (1)
@si @si
i i
Hence, the speed individually selected by driver i, sbi , solves @v@s(si ) = @s
@ i
ic .

Intuitively, driver i increases his speed si until the point where the additional
utility from marginally increasing si coincides with its associated expected
individual cost from speed.
(b) Set up the social planner’s expected welfare maximization problem. Take …rst-
order conditions with respect to s1 and s2 . Denote the (implicit) solution to this
…rst-order condition as si . Show that drivers have individual incentives to drive
too fast, relative to the socially optimal speed, i.e., show that sbi > si .
The social planner solves the expected welfare maximization problem

max
1 2
v 1 (s1 ) + q 1 + v 2 (s2 ) + q 2 (s1 ; s2 ) c1 + c2
s ;s

Taking …rst-order conditions with respect to s1 , we obtain that s1 solves

@v 1 (s1 ) @
1
= 1 c1 + c2 (2)
@s @s

1
 and similarly with respect to s2 , we obtain that s2 solves
@v 2 (s2 ) @
= c1 + c2 (3)
@s2 @s2
Intuitively, at the social optimum every driver i increases his speed si until
the point where the additional utility from marginally increasing si coincides
with its associated expected social cost from speed.
Comparing (1) and (3), we have
@v 1 (^
s1 ) @v 1 (s1 )
<
@^s1 @s1
2 i i
Since @(@s
v (s )
i )2
< 0, we can con…rm that sb1 > s1 . Similarly, comparing (1) and
(4), we have that sb2 > s2 .
Here is a graphical representation of the marginal utility and marginal ex-
pected costs (individual and social) to support their explanation.

– Note that for simplicity, we consider that the marginal utility decreases in
3 v i (si )
si at a constant rate, i.e., @(@s i )3
= 0. In addition, we also assumed that
for the increases in speed si impose a constant increase in the probability
@2 @3
of an accident, i.e., (@s i )2 > 0, and (@si )3 = 0, which also implies that the

marginal cost curve is straight.

(c) Let us evaluate the e¤ect of speeding tickets (…nes) to individuals driving too fast,
sbi > si . What is the dollar amount of the …ne mi that induces every individual i
to fully internalize the externality he imposes onto others?
Comparing (1) and (3) for driver 1, we must impose a …ne of m1 = c2 in order
to guarantee that (1) coincides with (3). Intuitively, driver 1 internalizes the
negative externality (cost) that he imposes on driver 2. Similarly comparing
(1) and (4) for driver 2, we must impose a …ne of m2 = c1 in order to guarantee
that (1) coincides with (3).
(d) Let us now consider that the individuals obtain utility from driving fast, v i (si ),
only in the case that no accident occurs. Repeat steps (a)-(c), …nding the optimal
…ne mi that induces individuals to fully internalize the externality.

2
 In this section of the exercise, driver i only obtains utility from driving fast,
v i (si ), when no accident occurs. Given that the probability that an accident
does not occur is 1 (s1 ; s2 ), the utility of driver i is

1 (s1 ; s2 ) v i (si ) + q i (si ; sj ) ci

which can be rearranged as

v i (si ) + q i (si ; sj ) ci + v i (si )

Taking FOC with respect to si , we obtain that the individually selected speed
si solves
@v i (si ) @ i i i
i i
i j @v (s )
= c + v (s ) + (s ; s ) (4)
@si @si @si
The social planner’s maximization problem in this case becomes

max
1 2
v 1 (s1 )+ q 1 (s1 ; s2 ) c1 + v 1 (s1 ) +v 2 (s2 )+ q 2 (s1 ; s2 ) c2 + v 2 (s2 )
s ;s

Taking FOC with respect to si , we obtain that sbi solves

@v i (si ) @ i i i
i i
i j @v (s ) @
i
= i
c + v (s ) + (s ; s ) i
+ i cj + v j (sj ) (5)
@s @s @s @s
Comparing (4) and (5), we obtain that the …ne mi that induces every in-
dividual i to internalize the externality that his driving imposes on others
is
mi = cj + v j (sj )

2. [Pollution and Income Level] It is often thought that when the level of income in
the economy increases, consumers are more willing to exchange consumption against
investment in abatement techniques for the negative externalities associated with pro-
duction, so that “pollution” decreases, at least in relative terms. Let us study the
predictions of a very simple model. Suppose the economy has two goods, good 2 be-
ing produced from good 1 with the production function y2 = f (z1 ), where z1 is the
quantity of good 1 used as input, y2 is the quantity of good 2 produced and f ( ) is a
di¤erentiable increasing concave production function. The production of good 2 creates
harmful emission e, which can be reduced by using additional good 1 in a pollution
abatement technology. Thus the level of emissions is described by e = g(y2 ; ze ), where
ze is the quantity of good 1 used to reduce pollution, and g( ) is a convex and di¤eren-
tiable function, being increasing in the …rst component and decreasing in the second
component. The consumer sector of the economy is represented by a consumer with
utility function u(x1 ; x2 ; e), which is di¤erentiable and concave, and increasing in the
…rst two components and decreasing in the third one. The consumer is endowed with
w units of good 1, where w representing the wealth of the economy.

(a) Derive the …rst-order conditions which must be satis…ed at an interior Pareto
optimal allocation.

3
 [Hint: Be sure to answer this question correctly: all the following questions
depend on it].
Answer: An interior Pareto optimal allocation is solution to the maximization
problem:
max u(x1 ; x2 ; e)
x2 ;y2 ;e

subject to:
x2 y2
y2 f (z1 )
e g(y2 ; ze )
x1 + z1 + ze w
where all inequalities bind. This maximization problem can be simpli…ed to

maxu(w z1 ze ; f (z1 ); g(f (z1 ); ze ))

z1 ;ze

which an unconstrained maximization problem. The …rst order conditions

with respect to z1 and ze are
@u @u @u @g
()+ ( )f 0 (z1 ) + () ( )f 0 (z1 ) = 0 (FOC1 )
@x1 @x2 @e @y2
@u @u @g
()+ () ()=0 (FOC2 )
@x1 @e @ze
where, for compactness,( ) means that the function is calculated at the value
of the argument at the Pareto optimal allocation.
(b) Interpret the FOCs in terms of marginal cost and marginal bene…t.
Interpretation of FOC1 : A marginal increase in z1 entails, on the one
hand, a reduction in the quantity of good 1 that the consumer enjoys as a
@u
consumption good, measured by @x 1
( ) . This additional unit of z1 , however,
increases the quantity of good 2 available by4y2 = f 0 (z1 ) , leading to an
@u
increase in utility of @x2
( )f 0 (z1 ) . Furthermore, the additional production
@g
creates additional externality, which has a utility cost of @u @e
( ) @y2
( )f 0 (z1 ) .
Thus, the increase in z1 increases the amount of good 2, ultimately entailing
a net marginal bene…t of
@u @u @g
( )f 0 (z1 ) + () ( )f 0 (z1 )
@x2 @e @y2
Therefore, expression FOC1 explains that marginal bene…t of increasing z1
@u
by one unit, @x 1
( ) , must be equal to the net marginal bene…t that the ad-
ditional unit of z2 entails in terms of more production of good 2 (as measured
in the net marginal bene…t expression found above).
Interpretation of FOC2 : Similarly, expression FOC2 explains that a mar-
ginal increase in the amount of good 1 that is taken away from consumption
and used into pollution abatement yields:

4
 @u
– 1. a marginal cost in terms of lower utility, measured by @x 1
( ) ; but
also
– 2. a marginal bene…t, since the consumer su¤ers a lower level of pollution,
@g
measured by @u
@e
( ) @ze
( ).
the marginal cost of taking away one unit of consumption in good 1 is equal
to the net marginal bene…t of the increased production in good 2.
(c) Let us next develop a parametric example of the above exercise. Assume that

u(x1 ; x2 ; e) = ln(x1 ) + ln(x2 ) ln(e); where < 1, and w > 2,

f (z1 ) = z1 , and
ky2
g(y2 ; ze ) = , where k > 0
1 + ze
Find the Pareto optimal allocation of this economy.
Replacing the derivative by their values in FOC1 and FOC2 ,
1 1 k
+ kz1
=0
w z1 ze z1 1+ze
1 + ze

1 kz1
=0
w1 z1 ze kz1
1+ze (1 + ze )2
At this point, we need to solve for z1 and ze . Simplifying these two expressions,

z1 ( 11 ) + ze = w
+1
z1 + ze = w 1

Solving for ze in the …rst expression,

2 2
z1 + ze = w ) ze = w z1
1 1
+1 1
and plugging it into the second equation, z1 + ze = w , we obtain

+1 +1 2 1
z1 + ze = z1 + w z1 =w
1
and further simplifying
(2 )( + 1) + (1 ) w+1
z1 =
(1 )
w+1 (1 ) (w + 1)(1 )
z1 = =
(2 )( + 1) + (1 ) 2
Thus, z1 is
(w + 1)(1 )
z1 =
2

5
 2
Plugging this value into ze = w z1 1
, we obtain

(w + 1)(1 ) 2 (w + 1) 2
ze = w =
2 1 2

The externality level e is, hence,

ky2 k 1 2 (w + 1) 1
e = g(y2 ; ze ) = = = k
1 + ze 1 + (1+w)
2
1

(d) Discuss how this Pareto optimal allocation depends on w. How does the exter-
nality level change with wealth: in absolute or relative terms?
Answer:
Input z1. We can …rst examine the comparative statics of z1 with respect to
the income level, w, obtaining @z
@w
1
= 1 2 , which is positive for all parameter
values; thus indicating that, as consumers become richer they employ larger
amounts of good 1 into the production of good 2.
Abatement ze . We can operate similarly for the amount of resources spent on
pollution abatement, …nding that @z @w
e
= 2 which is also positive for all para-
meter values. Hence, consumers spend more resources on pollution abatement
as they become wealthier.
Emission level, e. Finally, in order to examine the comparative statics
of the emission level, we …rst need to …nd the optimal level of emissions
(which we didn’t in previous exercises). As we described at the end of part
(c), emissions are given by e = k 1 , implying that emissions are constant
in the income level, w. In particular, when w increases, both the input
used to produce good 2 and the input used to reduce pollution increases
proportionally to w, ultimately implying that the pollution level is constant
in w. If, instead, we measure emissions as a proportion of consumption, i.e.,
e
y2
, we obtain
e k1 2k
= (w+1)(1 ) =
y2 + w
2
which decreases in income, w. Hence, while the absolute level of emissions
is constant in income, w, its relative size is decreasing as consumers become
richer.

3. [Regulating externalities under incomplete information.] Consider the setting

studied in class where a regulator observes neither the type of the …rm emitting pollu-
tion (i.e., the realization of parameter ) nor the type of the consumers being a¤ected
by such pollution (the realization of parameter ). Suppose that the …rm’s marginal
bene…t from an addition unit of pollution is

@ (h; )
= bh + ,
@h

6
and that the marginal utility from an additional unit of pollution for the consumer is
@ (h; )
= ch + ,
@
where and are random variables with expectation E[ ] = E[ ] = E[ ] = 0, and
all take strictly positive values, i.e., ; > 0. Parameters b; c and are also strictly
positive by de…nition, i.e., b; c; > 0. Finally, denote E[ 2 ] = 2 and E[ 2 ] = 2 .

(a) Identify the best quota b

h that a social planner selects to maximize the expected
value of aggregate surplus. (Assume that the …rm must produce an amount
exactly equal to the quota.)
Answer: Firm must produce exactly equal to the quota c. The social planner
determines the optimal quantity h^ by choosing the value of h that maximizes
expected value of aggregate surplus (since the social planner does not know
the precise realization of parameters and ),
max E [ (h; )] + E [ (h; )]
^
h

And taking FOC with respect to h, we obtain

" # " #
^ ; )
@ (h ^ ; )
@ (h
E +E 0
@h @h

We can now substitute the functional forms for the marginal bene…t for con-
sumers, @ @h
(h; )
, and the marginal pro…ts for the …rm, @ @h
(h; )
, we obtain,
^ + E[ ] +
ch ^ + E[ ]
bh 0
from which we can solve form h^ to have
^ + ^
h , with equality for h 0
c+b
(b) Identify the best tax t for this same planner.
Answer: Given a tax t , the …rm will maximize pro…ts and will chooseh
that maximizes its pro…ts (reduced in tax payments), that is
max (h; ) th
h

The …rm hence takes FOC with respect to h, obtaining

@ (h; )
t=0
@h
And since we know that @ @h(h; )
= bh + by de…nition, the above FOC
becomes bh + t = 0. Solving for h, we obtain the …rm’s pro…t-
maximizing externality h(t; ), as a function of the tax ratet and its “type” ,
as follows
+ +t
h(t; ) =
b

7
 Importantly, note that describes the …rm’s “reaction function”(or “best re-
sponse function”) after observing that the regulator imposes a particular tax
rate t: Provided this best response function, we can now …nd what is the opti-
mal tax that the social planner imposes, anticipating the …rm’s best response
function, as follows

max E[ (h(t; ); )] + E[ (h(t; ); )]

t

(where note that, rather than writing any general level of h, we wrote the
level of h that the …rm optimally chooses in the second stage, after observing
the tax rate t imposed by the regulator in the …rst stage). Taking FOC with
respect to h, we obtain

@ (h(t; ); ) @h(t; ) @ (h(t; ); ) @h(t; )

E +E =0
@h @t @h @t

(note that we needed to use the chain rule in this FOC).

And since h(t; ) = +b +t then the derivative @h(t;
@t
)
= 1b is a constant, that
can be taken out of the expectation operator. Therefore, we can cancel out
the @h(t;
@t
)
from the FOC. Substituting the functional form of our marginal
bene…t and marginal pro…t functions, the above FOC becomes:

E[ ] + t E[ ] + t
c + E[ ] + b + E[ ] = 0
b b
Rearranging, and solving for t, we obtain

c(E[ ] + t) b(E[ ] + t) = b(E[ ] + + E[ ] )

t(c + b) = bE[ ] + b + bE[ ] b + cE[ ] + c + bE[ ] + b

Recalling that E[ ] = 0 and E[ ] = 0, the above expression simpli…ed to

t(c + b) = b b +c +b

and solving for t, we …nd

c b
t =
(c + b)
(c) Compare the two instruments in terms of their associated deadweight loss. Two
…gures are enough: one where the quota performs better and another where the
tax performs better.
Answer: We need to compare the expected di¤erence in losses in order to
determine when a tax or a quota instrument is better. The …gure below

8
illustrates the choices of h and t

In the above …gure the intersection of the expected marginal pro…ts and
marginal utility curves determine h and t . Consider a realization of and
that results in the curves intersecting at point x. The optimal level of the
externality would then be h . If we use a quota instrument h the loss is the
shaded triangle xuv. If we use a tax instrument then the …rm will choose
h(t ; ) and the loss is the shaded triangle xyz. Thus in the case pictured the
tax instrument is better. (This is, of course, not a proof, but an introduction
to the proof). Let’s consider when each instrument will be best.
First, we must introduce a non-standard way of calculating the area of a
triangle. Area is normally calculated as: Area = 21 base height. In the
…gure below this would be A = 12 ed .

We can divide the edge e into e1 and e2 , and we can then write d = eb1 , where
b
b is the slope of the top edge of the triangle. We can also write e1 = b+c e.

9
 e
Combining d = b+c . Next we plug d back into our normal area calculation to
1 e2
get: A = 2 b+c .
We apply this non-standard area calculation to determine the area of triangle
xuv. In words, the calculation is the edge of uv squared, divided by twice the
sum of the slopes of both marginal curves. The height of the edge uv is
@ (h ; ) @ (h ; )
=
@h @h
+ +
b + + c + = +
c+b c+b
We may therefore calculate the loss from quantity regulation (the loss from
+ )2
setting a quota h) as Lh = (2(b+c) .
Next, we calculate the area of the loss from taxation, which we denote as Lt .
First, note that the height of the edge yz is
@ (h(t ; ); ) @ (h(t ; ); )
=
@h @h
+ t + t
c + b + =
b b
c
+
( + t) t =
b
and since we know that t = c h , the above expression becomes

c(h(t ; ) h)
c
Next, calculate h(t ; ) h = b and plug back in b
. We may therefore
calculate the loss from taxation, Lt , as
c 2
b
Lt =
2(b + c)
Comparison of losses. Last, we calculate the expected di¤erences in losses
" #
2
1 2 c
E(Lh Lt ) = E( + )
2(b + c) b

2
(b c)
text =
2b2
To conclude, we have just found that the optimal choice of quantity or tax
instrument depends on the sign of (b c):
– 1. When this term is positive, (b > c) , the loss from the quota system is
greater so the tax instrument is preferred.
– 2. When instead c > b , the reverse is true and the quota instrument is
preferred.

10
 Recall that under the tax system the level of the externality is changed de-
pending on the …rm’s realized marginal pro…ts.

4. [The problem of the commons] Lake Washington can be freely accessed by …sh-
ermen. The cost of sending a boat out on the lake is r > 0. When b boats are sent
out onto the lake, f (b) …sh are caught in total (so each boat catches f (b)
b
…sh), where
0 00
f (b) > 0 and f (b) < 0 at all b 0. The price of …sh is p 0, which is una¤ected by
the level of catch from Lake Washington.

(a) Characterize the equilibrium number of boats that are sent out on the lake.
Answer: This is a model of free entry so …shermen will send out boats as
long as there are a positive pro…ts from doing so. Therefore, the equalibrium
number of boats, b , will be reached when p f (b
b
)
r = 0, or, f (b
b
)
= pr . This
condition is that average revenue equals to average cost. (We ignore integer
problems, but if we are to give the interger equilibrium number then it is b
such that p f (b
b
)
r 0 and p f (b +1)
b +1
r < 0.)
(b) Characterize the optimal number of boats that should be sent out on the lake.
Answer: To characterize the optimal number of boats we must solve for
maximum total surplus, i.e., maxb pf (b) rb, the FOC is pf 0 (b0 ) r 0, which
is necessary and su¢ cient since the SOC, pf 00 (b) < 0, is satis…ed. Therefore,
the condition for the optimal number of boats is f 0 (b0 ) = pr , i.e., that marginal
revenue equals marginal (and in this case average) cost. Assuming that f (0) =
0,ensures that b0 b (equality only at 0).
(c) Compare your answers in parts (a) and (b). Explain.
Answer: To restore e¢ ciency we need the equilibrium condition satis…ed at
b0 , i.e., we need the tax level to satisfy f (b
b
)
= r+t
p
, or t = p f (b
b
)
r.
(d) Suppose that the lake is instead owned by a single individual who can choose how
many boats to send out. What number of boats would this owner choose?
Answer: Clearly, if owned by a single individual, the problem to be solved
is that solved in part (b) above, which result in b0 .

5. [Private contributions to a public good] Take an economy with 2 consumers

i = A; B, 1 private good x, and 1 public good G. Let each consumer have an income
of M . The prices of public and private good are both 1. Let the consumers have a
utility functions:

U A = log(xA ) + log(G), for individual A, and

U B = log(xB ) + log(G), for individual B

Assume that the public good is privately provided, so total contributions to the public
good are G = g A + g B . Note that you can eliminate xi from the utility function using
the budget constraint M = xi + g i .

11
(a) Consider individual A choosing his contribution to the public good g A to maximize
utility. Show that the optimal choice satis…es:

M gB
gA =
2 2
Answer: The utility maximization decision of A is:

max U A = log(M g A ) + log(g A + g B )

fg A g

which has a necessary condition:

1 1
+ =0
M gA gA + gB

solving for g A we have:

M gB
gA =
2 2
which represents individual A’s best response function (see …gure at below).

(b) Repeat part (a) for consumer B.

Answer: The utility maximization decision of B is:

max U B = log(M g B ) + log(g A + g B )

fg B g

which has a necessary condition:

1 1
+ =0
M gB gA + gB

solving for g B we have:

M gA
gB =
2 2

12
 which represents individual B’s best response function (see …gure below). Note
that we use the same axes as in the best response function of individual A, so we
can afterwards superimpose both best response functions in the same …gure to
…nd their crossing point.

(c) Find the competitive (Nash) equililibrium contributions to the public good by
consumer A and B.
Answer: Plugging individual B’s best response function into individual A’s
best response function,
M gA
M 2 2
gA =
2 2
A
and solving for g , we obtain
M
gA =
3
which identi…es the crossing point between both individuals’ best response
function, as depicted in the …gure below. (A similar equilibrium contribution
arises for individual B.)

Thus implying that aggregate contributions in the Nash equilibrium are g A +

gB = G^ = 2M .
3

13
(d) Find the e¢ cient (socially optimal) contribution to the public good by consumers
A and B.
Answer: The e¢ cient level of provision will have the cost equally allocated
between the consumers. Recall W = U A + U B , it therefore solves:

max U A + U B = log M g A + log(g A + g B ) + log M g B + log(g A + g B )

fg A ;g B g

and given that individuals are symmetric, their optimal contributions must
coincide, i.e., g A = g B = g, implying that we can simplify the above problem
to

maxU A + U B = log (M g) + log(g + g) + log (M g) + log(g + g)

fgg

or
maxU A + U B = 2log (M g) + 2log(2g)
fGg

and since g + g = G, we can further simplify the above problem to

G
maxU A + U B = 2log M + 2log(G)
fGg 2
1 2
Taking FOCs with respect to G, we obtain + G2 = 0, and solving for
2 M G
2

G we have that the optimal aggregate contribution is Ge = M.

As we can see comparing G e = M with G ^ = 2M shows that provision at the
3
e
^ < G.
Nash equilibrium is below what is optimal, i.e., G
(e) Show that along an indi¤erence curve the following property must be satis…ed:

1 1 1
dg A + dg B =0
gA + gB M gA gA + gB

and hence that:

dg B gA + gB
= 1:
dg A M gA
Answer: The utility of consumer A is given by U A = log(xA )+log(G), which
can be written as U A = log(M g A ) + log(g A + g B ). Totally di¤erentiating
gives:
1 1 1
dU A = A
+ A B
dg A + dg B
M g g +g g + gB
A

In addition, since we know that along the indi¤erence curve the chance in
utility is zero, then, dU A = 0, then

1 1 1
dg A + dg B = 0 (First result)
gA + gB M gA gA + gB

14
 Rearranging, we obtain

1 1 1
dg B = dg A
gA + gB gA + gB M gA

1 1
dg B g A +g B M gA
= (Second result)
dg A 1
g A +g B

gA + gB
= 1
M gA

(f) Solve the last equation to …nd the locus of points along which the indi¤erence
curve of individual A is horizontal and use this to sketch the indi¤erence curves
of the individual A.
B
Answer: The indi¤erence curve of A is horizontal when dg dg A
= 0. Hence
the locus of points where the indi¤erence curves of A are horizontal is the
solution to:
gA + gB gA + gB
0= 1 or =1
M gA M gA
Rearranging we obtain g A + g B = M g A , or 2g A = M g B . Finally, solving
for g A , we have
M gB M gB
gA = =
2 2 2
which exactly coincides with individual A’s best response function.
The following …gure illustrates this property.

(g) Use a …gure to contrast the Pareto e¢ cient level of private provision and the Nash
equilibrium level of provision.
Answer: The following …gure compares individual contributions in the Nash
equilibrium, g^ = M3 , and socially optimal (Pareto e¢ cient) contributions, M2 ,
which lie on the middle of the lens of Pareto improving allocations. The …gure

15
 also depicts aggregate contributions in the Nash equilibrium and at the socially
optimal allocation.

6. [Production and Externalities] A …rm’s production of bacon generates a smelly

gas as an unpleasant side product. Let c(y; m; w) denote the (minimum) input cost of
producing y tons of bacon and m cubic meters of gas when input prices are given by
@c
the vector w >> 0. Let p > 0 denote the market price of bacon. Assume that @y > 0,
@c
@m
< 0 and that c(y; m; w) is strictly convex in y and m. Let stars denote solutions
and assume throughout that y > 0.

(a) Show that c(y; m; w) is concave in input prices, w.

Answer: Fix two input price vectors w and w0 and consider w00 = w + (1
)w00 , for some 2 (0; 1). Let x (respectively, x0 and x00 ) be the minimum cost
bundle for input prices w (respectively, w0 and w00 ). By cost minimization
we have

c(y; m; w00 ) = wx00 + (1 )w0 x00

wx + (1 )w0 x0
= c(y; m; w) + (1 )c(y; m; w00 )

So c(y; m; w) is concave in input prices w.

(b) Suppose that the government imposes a ceiling on gas emissions such that m m.
Assuming that this constraint binds,
1. write down the …rm’s pro…t maximization problem with respect to y, and …nd
necessary and su¢ cient conditions for the …rm’s cost-minimizing production,
y ,
2. under which condition on c(y; m; w) can we guarantee that to increase in the
ceiling on gas emissions, m, produces a raise in the …rm’s cost-minimizing
production, y , i.e., @y
@m
> 0? [Hint: use the implicit function theorem.]

16
 Answer: The pro…t maximization problem for the …rm is

max py c(y; m; w)
y;m m

If the constraint binds then the FOC with respect to output y is

@c(y ; m; w)
p=
@y
that is, price equals marginal cost at the optimum. Di¤erentiating this
expression again with respect to m, we obtain
@ 2 c(y ; m; w) @y @ 2 c(y ; m; w)
0= +
@y 2 @m @m@y
and rearranging we obtain the usual expression of the implicit function
theorem,
@ 2 c(y ;m;w)
@y @m@y
= @ 2 c(y ;m;w)
@m 2
@y

Since c( ) strictly convex in output y the denominator is positive, so a

@ 2 c(y ;m;w)
necessary and su¢ cient condition for @y
@m
> 0 is that @m@y
< 0, i.e.,
an increase in pollution, m, reduces the marginal cost of production.
(c) Suppose now that the government abandons its emissions ceiling and replaces it
with a tax t > 0 on gas emissions. Thus, the new cost of producing (y; m) is given
by c(y; m; w) + tm.
1. Write down the …rm’s pro…t maximization problem with respect to y and m:
Answer: The pro…t maximization problem for the …rm is

M ax py c(y; m; w) tm
y;m

2. Show that maximized pro…ts are convex in t, and that the …rm’s choice of
pollution decreases in the pollution tax, i.e., @m @t
0.
Answer: Suppose (y; m), (y 0 ; m0 ) and (y 00 ; m00 ) maximize pro…ts for tax
levels t, t0 and t00 , respectively, where t00 = t + (1 )t0 and 2 (0; 1).
By pro…t maximization it follows that

(p; w; t) = py c(y; m; w) tm py 00 c(y 00 ; m00 ; w) tm00 and

(p; w; t0 ) = py 0 c(y 0 ; m0 ; w) t0 m0 py 00 c(y 00 ; m00 ; w) t0 m00

Hence,

(p; w; t) + (1 ) (p; w; t0 )
py 00 c(y 00 ; m00 ; w) [ t + (1 )t0 ]m00
= py 00 c(y 00 ; m00 ; w) t00 m00 = (p; w; t00 )

Then the pro…t function (p; w; t) is convex in the tax level t.

17
 Let x and x0 be the input vectors for the pro…t maximizing plans y and
y 0 associated with t and t0 , respectively. By pro…t maximization,
(p; w; t) = py wx tm py 0 wx0 tm0
and rearranging,
p(y 0 y) + w(x0 x) + t(m0 m) 0 (1)
Similarly for tax level t0 ,
(p; w; t0 ) = py 0 wx0 t0 m0 py wx t0 m
and rearranging,
p(y 0 y) w(x0 x) t0 (m0 m) 0 (2)
So by adding inequalities (1) and (2), we obtain
(t0 t)(m0 m) 0
=) (t0 t)(m0 m) 0
it means m decreases as t increases, or in di¤erential terms
@m
0:
@t
7. [Positive and Negative Externalities] Consider an economy with two …rms which
produce an homogeneous good. Firm 1 produces q1 units of the good, and its cost
function is c1 (q1 ; q2 ) = 2q12 + 5q1 + q2 , while …rm 2 produces q2 units of the same good
and its cost function is c2 (q2 ; q1 ) = q22 + 3q2 4q1 . Finally, market demand is given by
p(Q) = 34 Q, where Q = q1 + q2 . Therefore, the pro…ts of …rm i = f1; 2g are given
by (34 qi qj ) qi ci (qi ; qj ), where j 6= i.

(a) Considering that every …rm independently and simultaneously selects is produc-
tion level, determine equilibrium output q1 and q2 . Which are the associated
pro…ts for each …rm? Measure consumer surplus, pro…ts and social welfare.
Answer: Since @c1@q
(q1 ;q2 )
2
= 1 > 0 and @c2@q
(q2 ;q1 )
1
= 4 < 0, …rm 2 generates a
negative externality on …rm 1, while …rm 1 produces a positive externality
on …rm 2. In order to determine the equilibrium level of q1 and q2 , we need
to separately consider each …rm’s pro…t-maximization problem. First, …rm 1
solves
max (34 q1 q2 ) q1 2q12 + 5q1 + q2
q1 0

Taking FOC with respect to q1 , we obtain 29 6q1 q2 = 0, and …rm 1’s best
response function is q1 (q2 ) = 29 6 q2 . Similarly, …rm 2 solves
max (34 q1 q2 ) q2 q22 + 3q2 4q1
q2 0

and taking FOC with respect to q2 , we have 31 q1 4q2 = 0, and …rm

2’s best response function is q2 (q1 ) = 31 4 q1 . Plugging q2 (q1 ) into q1 (q2 ), we
obtain q1 = 85
23
' 3:69 and q2 = 157
23
' 6:82. Therefore, the aggregate supply is
QS (p) = q1 + q2 = 242
23
' 10:52, with an equilibrium price of p = 540 23
' $23:47.

18
 Equilibrium pro…ts are therefore 1 = 34:14 and 2 = 107:97 for …rm 1 and
2, respectively, and aggregate pro…ts are = 142:11.
Consumer surplus is hence
Z 23:47
CS = (34 Q)dQ (23:47 10:52) = 275:65
0

Thus, social welfare is W = CS + = 417:76.

Finally, notice that this output allocation is ine¢ cient: …rm 1 (the agent who
generates the positive externality) produces too little, whereas …rm 2 (the
agent who causes the negative externality) produces too much.
(b) Assume that the government is aware of these mutual externalities between …rm
1 and 2, but does not want to directly regulate their production by the imposition
of quotas or fees. Instead, the regulator allows both …rms to merge. Determine
the equilibrium level of q1 and q2 that the newly merged …rm will choose, and
check if …rm 1 and 2 have incentives to merge.
Answer: This merge is equivalent to a horizontal integration, whereby …rms
choose the level of q1 and q2 in order to maximize their joint pro…ts, as follows

max (34 q1 q2 ) (q1 + q2 ) 2q12 + 5q1 + q2 q22 + 3q2 4q1

q1 0;q2 0

and taking FOCs with respect to q1 and q2 , we obtain

33 6q1 2q2 = 0, and

30 2q1 4q2 = 0

where we can simultaneously solve for q1 and q2 to …nd q1 = 18 5

= 3:6 and
q2 = 57
10
= 5:7. Aggregate supply is hence QS
(p) = 93
10
= 9:3 units, with an
equilibrium price of p = 34 9:3 = 24:7.
Equilibrium pro…ts are therefore 1 = 39:3 and 2 = 105:6 for …rm 1 and 2,
respectively, and aggregate pro…ts are = 144:9. Aggregate pro…ts increase
as a result of the merger, and hence …rms have incentives to merge.
(c) Compare consumer surplus, pro…ts and welfare after the merger (as you found in
part b) and before the merger (as found in part a). Does the merger ameliorate
the negative externality that the production of …rm 2 generates? Does social
welfare increase as a result of the merger?
Answer: Consumer surplus is hence
Z 24:7
CS = (34 Q)dQ (24:7 9:3) = 305:045
0

Thus, social welfare is W = CS + = 449:94.

Comparing our results in parts (a) and (b), we can summarize that, as a
result of the merger:
– The production of …rm 1 (the agent who generates the positive external-
ity) increases; whereas

19
 – The production of …rm 2 (the agent who causes the negative externality)
decreases.
– Aggregate output and consumer surplus increase.
– Social welfare increases.

8. [Externalities] Consider two consumers with utility functions

1
uA = log(xA A
1 ) + x2 log(xB
1 ) for consumer A, and
2
1
uB = log(xB B
1 ) + x2 log(xA
1 ) for consumer B.
2
where the consumption of good 1 by individual i = fA; Bg creates a negative external-
ity on individual j 6= i. For simplicity, consider that both individuals have the same
wealth, m, and that the price for both goods is 1.

(a) Equilibrium. Set up consumer A’s utility maximization problem, and determine
his demand for goods 1 and 2, i.e., xA A
1 and x2 . Then operate similarly to …nd
consumer B’s demand for good 1 and 2, i.e., x1 and xB
B
2.

Answer: Consumer A solves

max U A subject to xA A
1 + x2 = M
(xA A
1 ;x2 )

The Lagrangian for this optimization problem is

1 A
L = log(xA A
1 ) + x2 log(xB
1)+ (M xA
1 xA
2)
2
and from the forst-order conditions
@L 1 A
A
= A =0
@x1 x1

@L A
=1 =0
@xA
2
@L
=M xA
1 xA
2 = 0
@
We …nd his optimal consumption

xA A
1 = 1 and x2 = M 1

Similarly,
xB B
1 = 1 and x2 = M 1
(b) Social optimum. Calculate the social optimum amounts of xA A B B
1 , x2 , x1 and x2 ,
considering that the social planner maximizes a utilitarian social welfare function,
i.e., W = UA + UB .

20
 A utilitarian welfare function is de…ned as sum of the individual utility func-
tions, in general, with individual weights. Here we assume a symmetric
utilitarian welfare function, meaning W = UA + UB . The socially optimal
consumption in this case solves
max U A + U B subject to xA;B
1 + xA;B
2 =M
(xA A
1 ;x2 )

The Lagrangian for this social planner’s problem is

1 1 A B
L = log(xA 1 )+ log(xB A B
1 ) + x2 + x2 + (M xA 1 xA
2 )+ (M xB1 xB2)
2 2
and from the forst-order conditionsWe …nd socially optimal consumption pro-
…le:
1 1
xA
1 = and xA
2 = M
2 2
1 1
xB1 = and xB
2 = M
2 2
(c) Restoring e¢ ciency. Show that the social optimum you found in (b) can be
sustained by a tax on good 1 (so the after-tax price becomes 1 + t) with the
revenue returned equally to both consumers in a lump-sum transfer.
With tax tA placed on good 1 and with lump-sum transfer T A , consumer A
solves
max U A subject to (1 + tA )xA A
1 + x2 = M + T
A
(xA A
1 ;x2 )

The Lagrangian for this optimization problem is

1 A
L = log(xA A
1 ) + x2 log(xB
1)+ (M + T A (1 + tA )xA
1 xA
2)
2
and from the forst-order conditions
@L 1 A
A
= A (1 + tA ) = 0
@x1 x1
@L A
=1 =0
@xA
2
@L
= M + T A (1 + tA )xA 1 xA
2 = 0
@
We …nd his optimal consumption
1 1
xA
1 = and x A
2 = M T A
1 + tA 1 + tA
Similarly we …nd the optimal consumption of consume B who pays tax tB on
good 1 and receives T B as a lump-sum transfer:
1 1
xB
1 = B
and xB2 = M TB
1+t 1 + tB
Comparing the optimum consumtion levels with the parket outcome,we …nd
the set of taxes and transfers that decentralizes the socially optimal outcome,
we …nd the set of taxes and transfers that decentralizes the socially optimal
outcome.

21