ApEc 8001

Applied Microeconomic Analysis: Demand Theory

Lecture 2: Consumer Choice (MWG, Ch. 2, pp.17-28)
I. Introduction
The discussion of preference and choice in Lecture 1 was
rather abstract. In particular the objects of choice could
be almost anything, and the decision maker could be
almost any kind of person.
In this lecture we narrow this down. The objects of
choice are bundles of goods and services, and the
decision maker is a consumer. The consumer has an
amount of wealth to spend, and all items in the bundles
have a price.
We begin by describing the commodities, and then move
on to the consumption set (all feasible choices within the
budget constraint), followed by a discussion of demand
functions.
Notice that we do not yet introduce the concept of a
utility function! That will be done next week.

II. Commodities
Commodities are simply goods and services that the
consumer values. We assume that there a finite number
of possible commodities, denoted by L. We use to
index these commodities, so that = 1, 2, L.
A commodity vector, or commodity bundle, is denoted
by x. It is a list of the individual commodities in a
specific bundle. It is complete in that it indicates how
much of all L commodities are in the bundle. If some
commodity is not in the bundle then we assign it a value
of zero. Thus the vector x can be explicitly described as
follows:
x1
x
x = 2

x
L

Although we will usually think of x as amounts consumed

during a specific time period, it could be made more
general to represent consumption over several time
periods. For example, for two time periods we could have
consumption of good 1 during time period 1 as on
element of the vector x and consumption of good 1 during
time period 2 as another element of the vector x.
2

III. The Consumption Set

Before talking about budget constraints, it is useful to
point out that there are physical limits to the set of all
possible vectors (all possible values of the vector x). For
example, for most commodities it is not possible to
consume a negative amount.
Formally, we can define the consumption set as a subset
of the commodity space, denoted by RL, whose
elements are the consumption bundles that the consumer
could possibly consume, subject to physical constraints.
Mas Colell et al. give four examples of physical
constraints on pp.18-19:
1. It is not possible to consume more than 24 hours of
leisure time in a given day.
2. For some goods, it is not possible to purchase parts
of a good. That is the number of goods must be an
integer. What examples can you think of ?
3. For some pairs of goods, you can only purchase one
of the pair, not both. Their example is eating bread in
New York City at a given point in time and eating
bread in Washington D.C. at the same point in time.
3

4. For some necessities, such as food, you cannot

survive if you eat a total amount that provides
insufficient calories for survival.
The consumption sets for these 4 examples are shown in
diagrams on p.19 of Mas Colell et al.
These examples are physical constraints, but institutional
constraints are also possible. For example a law could be
passed saying that it is illegal for someone to work more
than 16 hours per day. This means that one is forced to
consume at least 8 hours of leisure every day (assuming
any time that is not work time is leisure time). A diagram
shows this on p.20 of Mas Colell et al.
For most economic models, it is reasonable to assume that
the consumption set, which we can denote by X, consists
of collections of commodity vectors x for which all
elements are nonnegative:
X = R L+ = { x RL: x 0 for = 1, 2, , L}
One feature of the consumption set X = R L+ is that it is
convex. Intuitively, a set is convex if, for any two
elements in the set, a weighted average of the set is also
an element in the same set. More formally, if two
consumption bundles, x and x, are in a convex set (in this
4

case, are in R L+ ), then a new bundle, denoted by x and

defined as x = x + (1 )x, where [0, 1], is also a
member of that set (is also in R L+ ).
[Dont forget that x, x and x are all vectors. In
particular each element of x, denoted by x is defined as
x = x + (1 )x .]
Much of the theory that we will study in this class
depends on convexity of consumption sets, but in some
cases this convexity is not needed.

IV. Budget Constraints and the Walrasian Budget Set

In addition to physical constraints, consumers are also
constrained by how much they can afford to purchase. To
define this constraint formally, we need to make two
assumptions:
1. The L commodities (goods and services) are all
traded in a market at given prices, denoted by the
vector p.
2. These prices cannot be influenced by any individual
consumer; that is, all consumers are price takers.

The price vector, like the commodity vector, is a column

vector with L elements in it:
p1
p
p = 2 RL

p
L

Technically, some of these prices could be negative. An

example would be a bad such as pollution; you would
have to pay people to consume it. However, unless
otherwise noted, for simplicity we will assume that
p >> 0, that is p > 0 for all .
The amount of money that the consumer has to purchase
commodities is denoted by w (wealth). Thus the only
consumption bundles that the consumer can afford to
purchase are those that satisfy the budget constraint:
px = p 1x 1 + p 2x 2 + + p L xL w
[In general, in Mas Colell et al. the inner product of two
column vectors is denoted by ; the transpose for the
first column vector is not shown.]
A formal definition of the consumption bundles that the
consumer can afford is:
6

Definition: The Walrasian budget set (also known as

the competitive budget set), which can be denoted as
Bp,w = {x R L+ : px w}, is the set of all feasible
consumption bundles for the consumer who faces prices
p and has a wealth of w.
The consumers problem can be stated as: Given prices
p and wealth w, choose a consumption bundle x from B p,w
.
From now on we will always assume that w > 0.
The following graph shows a Walrasian budget set:

x2

w/p2

{x RL+ : px = w}
Slope = -(p1/p2)

Bp,w

w/p1

x1

The (sub)set Bp,w = {x RL : px = w} is called the

budget hyperplane (note that it has =, not ). (If L = 2,
as in the above diagram, it is a budget line.) It shows the
rate of exchange between goods given their prices.
Question: What happens to the diagram if p 2 decreases?
Another characteristic of the budget hyperplane is that it
is orthogonal (perpendicular) to the price vector p. This is
shown in the following diagram:

x2

w/p2

Bp,w

( x 1 + p1, x 2 + p2)

x = (x - x )
x

w/p1

x1

Let x be any point on the budget hyperplane. Draw the

price vector p from this point. It is orthogonal to any
vector lying within the budget hyperplane. This is the
case because for all points on the hyperplane we have
px = w. Thus p x = w, and for any x on the hyperplane
px = w. Thus px = 0 for any vector x within the
hyperplane.
A final property of the Walrasion budget set Bp,w is that it
is convex, which means that if any bundles x and x are
both elements of Bp,w, then x = x + (1-) x is also in
9

Bp,w (where 0 1). This is easy to show. First, since

both x and x R L+ (both are nonnegative), then x R L+ .
Second, the assumption that both x and x are in B p,w
means that px w and px w, which in turn implies that
(px) + (1-)px w + (1-)w = w. Finally, px =
p[x + (1-) x] = (px) + (1-)px w, so x is in Bp,w.
Final note: There are other types of budget sets that
consumers may face in the real world that are not
Walrasion, due to (for example) taxes and overtime
payments that lead to kinks in the budget set. These may
not be convex; an example is given on p.22 of Mas Colell
et al.

V. Demand Functions
We can define a consumers Walrasian demand
correspondence, x(p, w), as a relationship that assigns a
set of chosen consumption bundles for every possible
value of p and w. In theory, it is possible that, for a given
(p, w) pair, there is more than one consumption bundle.
However, we will usually assume that there is a single
unique consumption bundle for each (p, w) pair, and so
we can call this relationship the Walrasian demand
function.
10

For the rest of this lecture (and the next lecture) we will
make two assumptions about the Walrasian demand
correspondence: it is homogenous of degree zero and it
satisfies Walras law. Lets define these two concepts:
Definition: The Walrasian demand correspondence
x(p, w) is homogenous of degree zero if, for any > 0,
x(p, w) = x(p, w) for any values of p and w.
The intuition here is that the consumer does not suffer
from money illusion; if prices and wealth increase or
decrease by the same proportion then there is no reason to
change his or her consumption bundle. More formally,
such a change in prices and wealth does not change the
consumers feasible consumption bundles (that is Bp,w =
Bp,w) and thus his or her choice should not change.
Definition: The Walrasian demand correspondence
satisfies Walras law if, for every p >> 0 and w > 0,
px(p, w) = w.
This simply states that the consumer spends all the wealth
that he or she has, which is reasonable to assume for most
people (most people do not reach a satiation point). It
is sometimes called the adding up restriction. This
assumption should be interpreted flexibly. In particular,
for a consumer who spends money over several time
periods during his or her life all, we are assuming is that
11

all money (resources) are spent by the end of his or her

lifetime, not in every time period. Thus saving in an
earlier time period to spend in a later one is perfectly
consistent with Walras law.
See the bottom of p.23 of Mas-Colell et al. for a fairly
easy exercise that checks whether Walras law holds for a
specific type of a demand function.
We will see in later lectures that both homogeneity of
degree zero and Walras law hold under a wide variety of
circumstances for demand functions that are derived from
utility maximization. For now we will focus on what can
be learned from demand functions that satisfy these two
conditions, without asking exactly where the demand
functions come from.
We also assume for the rest of this lecture that x(p, w) has
a single bundle for each possible (p, w) pair. That is, we
are working with demand functions, not demand
correspondences. This allows us to express x(p, w) as a
vector of demand functions for individual commodities:
x 1 (p, w)
x (p, w)

x(p, w) = 2

x (p, w)

L
12

Note: This approach of discussing demand functions

without reference to any underlying preferences or utility
function can be interpreted as an application of the
choice rules approach discussed in Lecture 1.
Comparative Statics
Many aspects of human behavior that economists (and
others) would like to understand can be modeled as
changes in demand functions in response to changes in
prices (p) or wealth (w).
First we will consider wealth effects (often called income
effects). Lets fix prices at p and see how the consumers
demand for each good changes as wealth increases. For
the entire vector of goods, x, this can be depicted as the
consumers Engel function, denoted by x( p , w). The
changes in demand for all L goods as w increases is called
the wealth expansion path (also known as the income
expansion path), and the derivative of the Engel function
for good , x ( p , w)/w is called the wealth effect (also
known as the income effect).
This figure shows the wealth expansion path (denoted by
E p) for two goods:

13

x2

w > w > w

Bp,w

Ep

Bp,w

Bp,w

x( p , w)
x( p , w)
x( p , w)
x1

Perhaps the first way to classify commodities according to

their wealth effects is whether the consumer increases or
decreases consumption of the good as his or her wealth
increases. A commodity is normal at the price-wealth
pair (p, w) if x (p, w)/w 0. The other possibility is
that the commodity is inferior at the price-wealth pair
(p, w), which is defined as if x (p, w)/w < 0. If a
commodity is normal for all possible price-wealth pairs
then we say it is a normal good. If every good is a
normal good than we say demand is normal.

14

Question: Is it also possible for every good to be an

inferior good?
We will sometimes use matrix notation to show the
wealth effects for all L goods:
x 1 (p, w)
w
x (p, w)
2

RL
D w x(p, w) =
w

x
(p,
w)
L

Now lets turn to price effects, which is how demand

changes as prices vary. To keep things simple, suppose
that there are only two commodities. We can show an
Economics 101 demand curve if we fix w and p 1 and
see how the demand for good 2 changes as p2 changes:

15

p2

x2( p1 , p2, w)
x2( p1 , p2, w)
x2( p1 , p2, w)
x2
Question: Is p 1 > or < p 1?
Another useful way to show price effects in a diagram is
an offer curve, which shows how the demand for two
goods changes as the prices change. In the following
diagram, w and p 1 are held constant while p2 changes:

16

x2

p2 < p2 < p2

w/p2
w/p2

x( p1 , p2, w)
w/p2

x( p1 , p2, w)
x( p1 , p2, w)
w/p1

x1

In this diagram, as p 2 decreases the demand for x 2

increases, which is what one would expect. The impact
on x 1 is ambiguous, since price and income effects work
in the opposite direction (this will discussed more in later
lectures). Returning to x 2, it is theoretically possible, but
empirically very rare, for the demand for x 2 to decline as
p 2 decreases. Such a good is called a Giffen good; see
p.27 of Mas-Colell et al. to see a diagram showing this.
Sometimes it is convenient to denote the LL matrix of
price effects as D px(p, w):
17

x 1 (p, w)
x 1 (p, w)

p
p L
1

D px(p, w) =

x
(p,
w)
x
(p,
w)

L
L

p L
p1

Implications of Homogeneity and Walras Law

Both homogeneity and Walras Law impose restrictions
on the price and wealth effects on demand functions.
Lets start with homogeneity. Recall that it implies that
x(p, w) x(p, w) = 0 for all > 0. The first property
can be derived by differentiating this with respect to and
then setting equal to 1:
Proposition 2.E.1: If the Walrasian demand function
x(p, w) is homogenous of degree zero, then for all p and
w:
x (p, w)
x (p, w)
pk +
w = 0 for = 1, 2, L
k =1
w
p k
L

This can be written more compactly in matrix notation:

D px(p, w)p + D w x(p, w)w = 0
18

Partial intuition: increases in prices include an implicit

reduction in real income and this will tend to reduce
consumption, but an increase in wealth usually has the
opposite effect.
The first expression of Proposition 2.E.1 can be written in
elasticity form. First, we need to define elasticities:
k(p, w) =

pk
x (p, w)
x (p, w)
p k

w (p, w) =

x (p, w)
w
x (p, w)
w

Elasticities show the percentage change in the demand

for good in response to a (small) percentage change in
the price of good k (which could be good ) or wealth. In
particular, k(p, w) (x /x )/(p k/p k) and w (p, w)
(x /x )/(w/w). Elasticities are useful because they
have no units (the units cancel out). But note that
elasticities could vary over p and w; there is no reason to
expect them to be constant as prices and wealth change.
Here is Proposition 2.E.1 in elasticity form:
L

k(p, w) + w(p, w) = 0 for all = 1, 2, L

k =1

19

This form reflects the homogeneity assumption: the same

percentage increase in prices and wealth will lead to no
change in consumption.
Now lets consider two implications of Walras law on
the impact of prices and wealth on demand for x. First,
differentiating the expression px(p, w) = w with respect
to prices gives:
Proposition 2.E.2 (Cournot aggregation condition): If
the Walrasian demand function satisfies Walras law, then
for all p and w:
L

=1

x (p, w)
+ x k(p, w) = 0 for k = 1, 2, L
p k

This can be written in matrix notation as:

pD px(p, w) + x(p, w)T = 0T
[The T notation indicates a transpose, which simply
means that a column vector becomes a row vector. Also,
the notation implies that the vector in front of it, in
this case p, is also a row vector.]
Next, differentiating pw(p, w) with respect to w gives:
20

Proposition 2.E.3 (Engel aggregation condition): If the

Walrasian demand function satisfies Walras law, then for
all p and w:
L

=1

x (p, w)
=1
w

This can be written in matrix notation as:

pD w x(p, w) = 1
These two propositions are quite intuitive:
1. Total expenditure cannot change in response to a
change in prices (you need to cut back on something
when the price of good k increases).
2. Total expenditure must change by an amount equal to
any change in wealth.
Both of these propositions can be expressed in elasticity
terms (b (p, w) = p x (p, w)/w: the budget share):
L

b (p, w)k(p, w) + b k (p, w) = 0

=1

(Cournot)

b (p, w)w(p, w) = 1

=1

(Engel)
21