COPYRIGHT NOTICE:

Ariel Rubinstein: Lecture Notes in Microeconomic Theory

is published by Princeton University Press and copyrighted, c 2006, by Princeton
University Press. All rights reserved. No part of this book may be reproduced in any form
by any electronic or mechanical means (including photocopying, recording, or information
storage and retrieval) without permission in writing from the publisher, except for reading
and browsing via the World Wide Web. Users are not permitted to mount this file on any
network servers.
Follow links for Class Use and other Permissions. For more information send email to:
permissions@pupress.princeton.edu

September 29, 2005 12:58

master

Sheet number 68 Page number 52

LECTURE 5

Demand: Consumer Choice

The Rational Consumers Choice from a Budget Set

In Lecture 4 we discussed the consumers preferences. In this lecture we adopt the rational man paradigm in discussing consumer
choice.
Given a consumers preference relation  on X = K+ , we can talk
about his choice from any set of bundles. However, since we are
laying the foundation for price models, we are interested in the
consumers choice in a particular class of choice problems called
budget sets. A budget set is a set of bundles that can be represented
as B(p, w) = {x X |px w}, where p is a vector of positive numbers
(interpreted as prices) and w is a positive number (interpreted as the
consumers wealth).
Obviously, any set B(p, w) is compact (it is closed since it is dened by weak inequalities, and bounded since for any x B(p, w)
and for all k, 0 xk w/pk ). It is also convex since if x, y B(p, w),
then px w , py w, xk 0, and yk 0 for all k. Thus, for all
[0, 1], p[x + (1 )y] = px + (1 )py w and xk + (1 )yk
0 for all k, that is, x + (1 )y B(p, w).
We will refer to the problem of nding the -best bundle in B(p, w)
as the consumers problem.

Claim:
If  is a continuous relation, then all consumer problems have a
solution.

Proof:
If  is continuous, then it can be represented by a continuous utility
function u. By the denition of the term utility representation,

September 29, 2005 12:58

master

Sheet number 69 Page number 53

Demand: Consumer Choice

53

nding an  optimal bundle is equivalent to solving the problem

maxxB(p,w) u(x). Since the budget set is compact and u is continuous,
the problem has a solution.
To emphasize that a utility representation is not necessary for the
current analysis, let us study a direct proof of the previous claim,
avoiding the notion of utility.

Direct Proof:
For any x B(p, w) dene the set Inferior(x) = {y B(p, w)|x  y}. By
the continuity of the preferences, every such set is open. Assume
there is no solution to the consumers problem of maximizing  on
B(p, w). Then, every z B(p, w) is a member of some set Inferior(x),
that is, the collection of sets {Inferior(x)| x X} covers B(p, w). A collection of open sets that covers a compact set has a nite subset of
sets that covers it. Thus, there is a nite collection Inferior(x1 ), . . . ,
Inferior(xn ) that covers B(p, w). Letting xj be the optimal bundle
within the nite set {x1 , . . . , xn }, we obtain that xj is an optimal bundle in B(p, w), a contradiction.

Claim:
If  is convex, then the set of solutions for a choice from B(p, w) (or
any other convex set) is convex.

Proof:
If both x and y maximize  given B(p, w), then x + (1 )y B(p, w)
and, by the convexity of the preferences, x + (1 )y  x  z for
all z B(p, w). Thus, x + (1 )y is also a solution to the consumers problem.

Claim:
If  is strictly convex, then every consumers problem has at most
one solution.

September 29, 2005 12:58

54

master

Sheet number 70 Page number 54

Lecture Five

Proof:
Assume that both x and y (where x  = y) are solutions to the consumers problem B(p, w). Then x y (both are solutions to the same
maximization problem) and x + (1 )y B(p, w) (the budget set
is convex). By the strict convexity of , x + (1 )y  x, which is
a contradiction of x being a maximal bundle in B(p, w).

The Consumers Problem with Differentiable Preferences

When the preferences are differentiable, we are provided with a useful condition for characterizing the optimal solution.

Claim:
If x is an optimal bundle in the consumer problem and k is a consumed commodity (i.e., xk > 0), then it must be that vk (x )/pk
vj (x )/pj for all other j, where vk (x ) are the subjective value numbers (see the denition of differentiable preferences in Lecture 4).
For the case in which the preferences are represented by a utility function u, we have vk (x ) = u/xk (x ). In other words, the
value per dollar at the point x of the k-th commodity (which
is consumed) must be as large as the value per dollar of any other
commodity.

Proof:
Assume that x is a solution to the consumers problem B(p, w) and
that xk > 0 and vk (x )/pk < vj (x )/pj (see g. 5.1). A move in the
direction of reducing the consumption of the k-th commodity by
1 and increasing the consumption of the j-th commodity by pk /pj
is an improvement since vj (x )pk /pj vk (x ) > 0. As xk > 0, we can
nd > 0 small enough such that decreasing ks quantity by and
increasing js quantity by pk /pj is feasible. This brings the consumer
to a strictly better bundle, contradicting the assumption that x is a
solution to the consumers problem.

September 29, 2005 12:58

master

Sheet number 71 Page number 55

Demand: Consumer Choice

Figure 5.1
(a) x is a solution to the consumer problem B(p, w).

55

(b) x is not a solution to the

consumer problem B(p, w).

From the above we can derive the classic necessary conditions

on the consumers maximization:

Claim:
If x is a solution to the consumers problem B(p, w) and both xk > 0
and xj > 0, then the ratio vk (x )/vj (x ) must be equal to the price
ratio pk /pj .
In order to establish sufcient conditions for maximization, we
require also that the preferences be convex.

Claim:
If  is monotonic, convex, continuous, and differentiable, and if
at x
px = w,
for all k such that xk > 0, and for any commodity l, vk (x )/pk
vl (x )/pl ,
then x is a solution to the consumers problem.

September 29, 2005 12:58

56

master

Sheet number 72 Page number 56

Lecture Five

Proof:
If x is not a solution, then there is a bundle z such that pz px
and z  x . By continuity and monotonicity, there is a bundle y  =
z, with yk zk such that y  x and py < pz px . By convexity,
any small move in the direction (y x ) is an improvement and by
differentiability, v(x )(y x ) > 0.
Let = vk (x )/pk for all k with xk > 0. Now,


0 > p(y x ) =
pk (yk xk )
vk (x )(yk xk )/
(since for a good with xk > 0 we have pk = vk (x )/, and for a good k
with xk = 0, (yk xk ) 0 and pk vk (x )/.) Thus, 0 v(x )(y x ),
a contradiction.

The Demand Function

We have arrived at an important stage on the way to developing
a market model in which we derive demand from preferences. Assume that the consumers preferences are such that for any B(p, w),
the consumers problem has a unique solution. Let us denote this
solution by x(p, w). The function x(p, w) is called the demand function. The domain of the demand function is K+1
++ whereas its range
is K+ .

Example:
Consider a consumer in a world with two commodities having the
following lexicographic preference relation, attaching the rst priority to the sum of the quantities of the goods and the second priority
to the quantity of commodity 1:
x  y if x1 + x2 > y1 + y2 or both x1 + x2 = y1 + y2 and x1 y1 .
This preference relation is strictly convex but not continuous. It
induces the following noncontinuous demand function:

(0, w/p2 ) if p2 < p1
x((p1 , p2 ), w) =
.
(w/p1 , 0) if p2 p1
We now turn to studying some properties of the demand function.

September 29, 2005 12:58

master

Sheet number 73 Page number 57

Demand: Consumer Choice

57

Claim:
x(p, w) = x(p, w) (i.e., the demand function is homogeneous of degree zero).

Proof:
This follows (with no assumptions about the preference relations)
from the basic equality B(p, w) = B(p, w) and the assumption that
the behavior of the consumer is a choice from a set.
Note that this claim is sometimes interpreted as implying that
uniform ination does not matter. This is an incorrect interpretation. We assumed, rather than concluded, that choice is made
from a set independently of the way that the choice set is framed.
Ination can affect choice since behavior may be sensitive to the
nominal prices and wealth even if the budget set is unchanged.

Claim (Walrass law):

If the preferences are monotonic, then any solution x to the consumers problem B(p, w) is located on its budget curve (and thus,
px(p, w) = w).

Proof:
If not, then px < w. There is an > 0 such that p(x1 + , . . . , xK + ) <
w. By monotonicity, (x1 + , . . . , xK + )  x, thus contradicting the
assumption that x is optimal in B(p, w).

Claim:
If  is a continuous preference, then the demand function is continuous in prices (and also in w, see problem set).

September 29, 2005 12:58

58

master

Sheet number 74 Page number 58

Lecture Five

Proof:
Once again, we could use the fact that the preferences have a continuous utility representation and apply a standard maximum theorem. (If the function f (x, a) is continuous, then the function h(a) =
argmaxx f (x, a) is continuous.) However, I prefer to present a proof
that does not use the notion of a utility function:
Assume not. Then, there is a sequence of price vectors pn converging to p such that x(p , w) = x , and x(pn , w) does not converge
to x . Thus, we can assume that (pn ) is a sequence converging to p
such that for all n distance (x(pn , w), x ) > for some positive .
All numbers pkn are greater than some positive number m. Therefore, all vectors x(pn , w) belong to some compact set (the hypercube
of bundles with no quantity above w/m) and thus, without loss of
generality, we can assume that x(pn , w) y for some y  = x .
Since pn x(pn , w) w for all n, it must be that p y w, that is, y
B(p , w). Since x is the unique solution for B(p , w), we have x  y .
By the continuity of the preferences, there are neighborhoods of x
and y in which the strict preference is preserved. For sufciently
large n, x(pn , w) is in that neighborhood of y . Choose a bundle z
in the neighborhood of x so that p z < w. For all sufciently large
n, pn z < w; however, z  x(pn , w), which is a contradiction.

Rationalizable Demand Functions

As in the general discussion of choice, we will now examine whether
choice procedures are consistent with the rational man model. We
can think of various possible denitions of rationalization.
One approach is to look for a preference relation (without imposing any restrictions that t the context of the consumer) such
that the chosen element from any budget set is the unique bundle
maximizing the preference relation in that budget set. Thus, we say
that the preferences  fully rationalize the demand function x if for
any (p, w) the bundle x(p, w) is the unique  maximal bundle within
B(p, w).
Alternatively, we could say that being rationalizable means that
there are preferences such that the consumers behavior is consistent
with maximizing those preferences, that is, for any (p, w) the bundle x(p, w) is a  maximal bundle (not necessarily unique) within

September 29, 2005 12:58

master

Sheet number 75 Page number 59

Demand: Consumer Choice

59

B(p, w). This denition is empty since any demand function is

consistent with maximizing the total indifference preference. This
is why we usually say that the preferences  rationalize the demand
function x if they are monotonic and for any (p, w), the bundle x(p, w)
is a  maximal bundle within B(p, w).
Of course, if behavior satises homogeneity of degree zero and
Walrass law, it is still not necessarily rationalizable in any of those
senses:

Example 1:
Consider the demand function of a consumer who spends all his
wealth on the more expensive good:

(0, w/p2 ) if p2 p1
x((p1 , p2 ), w) =
.
(w/p1 , 0) if p2 < p1
This demand function is not entirely inconceivable, and yet it
is not rationalizable. To see this, assume that it is fully rationalizable or rationalizable by . Consider the two budget sets B((1, 2), 1)
and B((2, 1), 1). Since x((1, 2), 1) = (0, 1/2) and (1/2, 0) is an internal
bundle in B((1, 2), 1), by any of the two denitions of rationalizability, it must be that (0, 1/2)  (1/2, 0). Similarly, x((2, 1), 1) = (1/2, 0)
and (0, 1/2) is an internal bundle in B((2, 1), 1). Thus, (0, 1/2)
(1/2, 0), a contradiction.

Example 2:
A consumer chooses a bundle (z, z, . . . , z), where z satises zpk = w.
This behavior is fully rationalized by any preferences according
to which the consumer strictly prefers any bundle on the main diagonal over any bundle that is not (because, for example, he cares
primarily about purchasing equal quantities from all sellers of the K
goods), while on the main diagonal his preferences are according to
the more the better. These preferences rationalize his behavior in
the rst sense but are not monotonic.
This demand function is also fully rationalized by the monotonic preferences represented by the utility function u(x1 , . . . , xK ) =
min{x1 , . . . , xK }.

September 29, 2005 12:58

60

master

Sheet number 76 Page number 60

Lecture Five

Example 3:
Consider a consumer who spends k of his wealth on commodity k
K
(where k 0 and k=1
k = 1). This rule of behavior is not formulated as a maximization of some preference relation. It can however be fully rationalized by the preference relation represented by
the Cobb-Douglas utility function u(x) = Kk=1 xk k . A solution x to
the consumers problem B(p, w) must satisfy xk > 0 for all k (notice
that u(x) = 0 when xk = 0 for some k). Given the differentiability
of the preferences, a necessary condition for the optimality of x is
that vk (x )/pk = vl (x )/pl for all k and l where vk (x ) = du/dxk (x ) =
k u(x )/xk for all k. It follows that pk xk /pl xl = k /l for all k and l
and thus xk = k w/pk for all k.
Example 4:
Let K = 2. Consider the behavior of a consumer who allocates his
wealth between commodities 1 and 2 in the proportion p2 /p1 (the
cheaper the good, the higher the share of the wealth devoted to
it). Thus, x1 p1 /x2 p2 = p2 /p1 and xi (p, w) = (pj /(pi + pj ))w/pi . This
demand function satises Walrass law as well as homogeneity of
degree zero.
To see that this demand function is fully rationalizable, note that

xi /xj = pj2 /pi2 (for all i and j) and thus p1 /p2 = x2 / x1 . The quasi

concave function x1 + x2 satises the condition that the ratio of

its partial derivatives is equal to x2 / x1 . Thus, for any (p, w), the

bundle x(p, w) is the solution to the maximization of x1 + x2 in

B(p, w).

The Weak and Strong Axioms of Revealed Preferences

We now look for general conditions that will guarantee that a demand function x(p, w) can be fully rationalized (a similar discussion
would apply to the other denition of rationalizability that requires
that x(p, w) maximizes a monotonic preference relation). Of course,
one does not necessarily need these general conditions to determine
whether a demand function is rationalizable. Guessing is often an
excellent strategy.

September 29, 2005 12:58

master

Sheet number 77 Page number 61

Demand: Consumer Choice

Figure 5.2
(a) Satises the weak axiom.

61

(b) Does not satisy the weak axiom.

In the general discussion of choice functions, we saw that the

weak axiom (WA) was a necessary and sufcient condition for a
choice function to be derived from some preference relation. In the
proof, we constructed a preference relation out of the choices of the
decision maker from sets containing two elements. We showed (by
looking into his behavior at the choice set {a, b, c}) that WA implies
that it is impossible for a to be revealed as better than b, b revealed
as better than c, and c revealed as better than a. However, in the
context of a consumer, nite sets are not within the scope of the
choice function.
In the same spirit, adjusting to the context of the consumer, we
might try to dene x  y if there is (p, w) so that both x and y are in
B(p, w) and x = x(p, w). In the context of the consumer model the
Weak Axiom is read: if px(p
, w
) w and x(p, w)  = x(p
, w
), then
p
x(p, w) > w
. WA guarantees that it is impossible that both x  y
and y  x. However, the dened binary relation is not necessarily
complete: there can be two bundles x and y such that for any B(p, w)
containing both bundles, x(p, w) is neither x nor y. Furthermore, in
the general discussion, we guaranteed transitivity by looking at the
union of a set in which a was revealed to be better than b and a set in
which b was revealed to be as good as c. However, when the sets are
budget sets, their union is not necessarily a budget set. (See g. 5.2.)

September 29, 2005 12:58

62

master

Sheet number 78 Page number 62

Lecture Five

Apparently WA is not a sufcient condition for extending the binary relation , as dened above, into a complete and transitive
relation (an example with three goods from Hicks 1956 is discussed
in Mas-Colell et al. 1995). A necessary and sufcient condition for
a demand function x satisfying Walrass law and homogeneity of
degree zero to be rationalized is the following:

Strong Axiom of Revealed Preference (SA):

If (xn )n=1,...,N is a sequence of bundles and (B(pn , wn ))n=1,...,N is a sequence of budget sets so that for all n N 1, xn  = xn+1 and xn is
chosen from B(pn , wn ) which also contains xn+1 , then x1
/ B(pN , wN ).
The Strong Axiom is basically equivalent to the assumption that
the relation  derived from revealed behavior is transitive. But 
is not necessarily a complete relation, and thus we are left with the
question of whether  can be extended into preferences. Proving
that this is possible is beyond the scope of this course. In any case,
the SA is cumbersome, and using it to determine whether a certain
demand function is rationalizable may not be a trivial task.

Decreasing Demand
The consumer model discussed so far constitutes the standard framework for deriving demand. Our intuition tells us that demand for
a good falls when its price increases. However, this does not follow
from the standard assumptions about the rational consumers behavior which we have discussed so far. The following is an example
of a preference relation that induces demand that is nondecreasing
in the price of one of the commodities:

An Example in Which Demand for a Good May Increase with Price

Consider the preferences represented by the following utility function:

x1 + x2 if x1 + x2 < 1
u(x1 , x2 ) =
.
x1 + 4x2 if x1 + x2 1

September 29, 2005 12:58

master

Sheet number 79 Page number 63

Demand: Consumer Choice

63

Figure 5.3

These preferences might reect reasoning of the following type:

In the bundle x there are x1 + x2 units of vitamin A and x1 + 4x2
units of vitamin B. My rst priority is to get enough vitamin A.
However, once I satisfy my need for 1 unit of vitamin A, I move on
to my second priority, which is to consume as much as possible of
vitamin B. (See g 5.3.)
Consider x((p1 , 2), 1). Changing p1 is like rotating the budget lines
around the pivot bundle (0, 1/2). At a high price p1 (as long as p1 >
2), the consumer demands (0, 1/2). If the price is reduced to within
the range 2 > p1 > 1, the consumer chooses the bundle (1/p1 , 0).
So far, the demand for the rst commodity indeed increased when
its price fell. However, in the range 1 > p1 > 1/2 we encounter an
anomaly: the consumer buys as much as possible from the second
good subject to the constraint that the sum of the goods is at least
1, i.e., x((p1 , 2), 1) = (1/(2 p1 ), (1 p1 )/(2 p1 )).
The above preference relation is monotonic but not continuous.
However, we can construct a close continuous preference that leads
to demand that is increasing in p1 in a similar domain. Let (t) be
a continuous and increasing function on [1 , 1 + ] where > 0,
so that (t) = 0 for all t 1 and (t) = 1 for all t 1 + . The
utility function
u (x) = ( (x1 + x2 )(x1 + 4x2 )) + (1 (x1 + x2 )(x1 + x2 ))
is continuous and monotonic. For close to 0, the function u = u
except in a narrow area around the set of bundles for which
x1 + x2 = 1.

September 29, 2005 12:58

64

master

Sheet number 80 Page number 64

Lecture Five

Now, take two prices, H > 1 and L < 1, such that a consumer
with utility function u consumes more of the rst commodity when
facing the budget set ((H, 2), 1) than when facing the budget set
((L, 2), 1) (that is, 1/H > 1/(2 L)). When is close enough to 0,
the demand induced from u at B((H, 2), 1) is (1/H, 0). Choose
such that 1/(2 L) + < 1/H. For close enough to 0, the bundle
in the budget set of B((L, 2), 1) with x1 = 1/(2 L) + is preferred
(according to u ) over any other bundle in B((L, 2), 1) with a higher
quantity of x1 . Thus, for small enough , the induced demand for
the rst commodity at the lower price is at most 1/(2 L) + , and
is thus lower than the demand at the higher price.

The Law of Demand

We are interested in comparing demand in different environments.
We have just seen that the classic assumptions about the consumer
do not allow us to draw a clear conclusion regarding the relation
between a consumers demand when facing B(p, w) and his demand
when facing B(p + (0, . . . , , . . . , 0), w).
A clear conclusion can be drawn when we compare the consumers
demand when he faces the budget set B(p, w) to his demand when
facing B(p
, x(p, w)p
). In this comparison we imagine the price vector changing from p to an arbitrary p
and wealth changing in such
a way that the consumer has exactly the resources allowing him to
consume the same bundle he consumed at (p, w). (See g. 5.4.)

Claim:
Let x be a demand function satisfying Walrass law and WA. If
w
= p
x(p, w), then either x(p
, w
) = x(p, w) or [p
p][x(p
, w
)
x(p, w)] < 0.

Proof:
Assume that x(p
, w
)  = x(p, w). Then,
[p
p][x(p
, w
) x(p, w)]
= p
x(p
, w
) p
x(p, w) px(p
, w
) + px(p, w)
= w
w
px(p
, w
) + w = w px(p
, w
)

September 29, 2005 12:58

master

Sheet number 81 Page number 65

Demand: Consumer Choice

65

Figure 5.4
A compensated price change from (p, w) to (p
, w
).

(by Walrass law and the assumption that w
= p
x(p, w)), and by WA
the right-hand side of the equation is less than 0.

Bibliographic Notes

Recommended readings: Kreps 1990, 3745, Mas-Colell et al. 1995,

Chapter 2, AD, 3, D,J.
The material in this lecture, up to the discussion of differentiability, is fairly standard and closely parallels that found in Arrow and
Hahn (1971) and Varian (1984).

September 29, 2005 12:58

master

Sheet number 82 Page number 66

Problem Set 5

Problem 1. (Easy)
Calculate the demand function for a consumer with the utility function

k k ln(xk ).
Problem 2. (Easy)
Verify that when preferences are continuous, the demand function x(p, w)
is continuous in prices and in wealth (and not only in p).
Problem 3. (Easy)
Show that if a consumer has a homothetic preference relation, then his
demand function is homogeneous of degree one in w.
Problem 4. (Easy)
Consider a consumer in a world with K = 2, who has a preference relation
that is quasi-linear in the rst commodity. How does the demand for the
rst commodity change with w?
Problem 5. (Moderately Difcult)
Let
be a continuous preference relation (not necessarily strictly convex)
and w a number. Consider the set G = {(p, x) K K | x is optimal in
B(p, w)}. (For some price vectors there could be more than one (p, x) G.)
Calculate G for the case of K = 2 and preferences represented by x1 + x2 .
Show that (in general) G is a closed set.
Problem 6. (Moderately difcult)
Determine whether the following behavior patterns are consistent with the
consumer model (assume K = 2):
a. The consumers demand function is x(p, w) = (2w/(2p1 + p2 ),
w/(2p1 + p2 )).
b. He consumes up to the quantity 1 of commodity 1 and spends his
excess wealth on commodity 2.

September 29, 2005 12:58

master

Sheet number 83 Page number 67

Demand: Consumer Choice

67

c. The consumer chooses the bundle (x1 , x2 ) which satises x1 /x2 = p1 /p2
and costs w. (Does the utility function u(x) = x21 + x22 rationalize the
consumers behavior?)

Problem 7. (Moderately difcult)

In this question, we consider a consumer who behaves differently from the
classic consumer we talked about in the lecture. Once again we consider a
world with K commodities. The consumers choice will be from budget sets.
The consumer has in mind a preference relation that satises continuity,
monotonicity, and strict convexity; for simplicity, assume it is represented
by a utility function u.
The consumer maximizes utility up to utility level u0 . If the budget set
allows him to obtain this level of utility, he chooses the bundle in the budget
set with the highest quantity of commodity 1 subject to the constraint that
his utility is at least u0 .
a.
b.
c.
d.
e.

Formulate the consumers problem.

Show that the consumers procedure yields a unique bundle.
Is this demand procedure rationalizable?
Does the demand function satisfy Walrass law?
Show that in the domain of (p, w) for which there is a feasible bundle yielding utility of at least u0 the consumers demand function for
commodity 1 is decreasing in p1 and increasing in w.
f. Is the demand function continuous?

Problem 8. (Moderately difcult)

A common practice in economics is to view aggregate demand as being
derived from the behavior of a representative consumer. Give two examples of well-behaved consumer preference relations that can induce
average behavior that is not consistent with maximization by a representative consumer. (That is, construct two consumers, 1 and 2, who choose
the bundles x1 and x2 out of the budget set A and the bundles y 1 and y 2 out
of the budget set B so that the choice of the bundle (x1 + x2 )/2 from A and of
the bundle (y 1 + y 2 )/2 from B is inconsistent with the model of the rational
consumer.)