CONSUMER CHOICE

1. THE CONSUMER CHOICE PROBLEM

1.1. Unit of analysis and preferences. The fundamental unit of analysis in economics is the economic agent.
Typically this agent is an individual consumer or a rm. The agent might also be the manager of a public
utility, the stockholders of a corporation, a government policymaker and so on.
The underlying assumption in economic analysis is that all economic agents possess a preference ordering
which allows them to rank alternative states of the world.
The behavioral assumption in economics is that all agents make choices consistent with these underlying
preferences.
1.2. Denition of a competitive agent. A buyer or seller (agent) is said to be competitive if the agent
assumes or believes that the market price of a product is given and that the agents actions do not inuence
the market price or opportunities for exchange.
1.3. Commodities. Commodities are the objects of choice available to an individual in the economic sys-
tem. Assume that these are the various products and services available for purchase in the market. Assume
that the number of products is nite and equal to L ( =1, ..., L). A product vector is a list of the amounts of
the various products:
x =
_

_
x
1
x
2
.
.
.
x
L
_

_
The product bundle x can be viewed as a point in R
L
.
1.4. Consumption sets. The consumption set is a subset of the product space R
L
, denoted by X
L
R
L
,
whose elements are the consumption bundles that the individual can conceivably consume given the phys-
ical constraints imposed by the environment. We typically assume that the consumption set is X = R
L
+
= {x
R
L
: x

0 for = 1, ... , L}.

1.5. Prices. We will assume that all L products are traded in the market at dollar prices that are publicly
quoted. Howthey are determined will be discussed later. The prices are represented by a price vector
p =
_

_
p
1
p
2
.
.
.
p
L
_

_
R
L
For nowassume that all prices are strictly positive, i.e. p

>> 0. We will also assume that all consumers

are price takers in the sense that they cannot inuence the price at which they buy or sell a product.
Date: October 10, 2005.
1
2 CONSUMER CHOICE
1.6. Income or wealth. Assume that each consumer has wealth equal to m
i
or the representative consumer
has wealth m.
1.7. Affordable consumption bundles. We say that a consumption bundle x is affordable for the represen-
tative consumer if
p x = p
1
x
1
+ p
2
x
2
+ + p
Lx
L
m (1)
If x is also an element of R
L
+
, then the set of feasible consumption bundles is x R
L
+
: px m. This is
called a Walrasian budget set and is denoted B
p,m
.
1.8. Preferences. We assume a preference relation over products with the following properties
1: complete in that for all x
1
, x
2
X, we have x
1
x
2
or x
2
x
1
(or both)
2: transitive in that x
1
, x
2
, x
3
X, if x
1
x
2
and x
2
x
3
then x
1
x
3
.
3: locally nonsatiated in that for every x
1
X and every > 0, there is x
2
X such that ||x
2
- x
1
||
and x
2
x
1
.
4: continuous in that for any sequence of pairs
{ (x
n
1
, x
n
2
) }

n=1
with x
n
1
x
n
2
n,
x
1
= lim
n
x
n
1
, and x
2
= lim
n
x
n
2
,
we have x
1
x
2
.
1.9. Existence of a utility function. Based on the preferences dened in 1.8, there exists a continuous utility
function v(x) that represents in the sense that x
1
x
2
iff v(x
1
) v(x
2
).
1.10. Convexity. We often assume that preferences are convex in the sense that if x
1
x
2
, then for 0
1, x
1
+ (1-)x
2
x
1
. This implies that indifference curves are convex to the origin. If the utility function is
quasi-concave, then the indifference curves will be convex and vice versa.
2. THE UTILITY MAXIMIZATION PROBLEM
2.1. Formal Problem. The utility maximization problem for the consumer is then as follows
max
x0
u = v(x)
s.t. px m
(2)
where we assume that p >> 0, m > 0 and X = R
L
+
.
This is called the primal preference problem. If we have smooth convex indifference curves and an inte-
rior solution, then the problem can be solved using standard Lagrangian techniques. Alternatively, Kuhn-
Tucker methods can be used. The Lagrangian function is given by
L = v(x) (
n
i=1
p
i
x
i
m) (3)
The rst order conditions are
v
x
i
p
i
= 0, i = 1, 2, . . . , n

n
i=1
p
i
x
i
+ m = 0
(4)
The value of is the amount by which L would increase given a unit relaxation in the constraint (an
increase in income). It can be interpreted as the marginal utility of expenditure. This units of this are of
CONSUMER CHOICE 3
course arbitrary. The solution to 2 is given by x(p,m) = g(p,m). These functions are called Marshallian
demand equations. Note that they depend on the prices of all good and income. Based on the structure
of preferences and the nature of the optimization problem, they will have certain properties which we will
discuss shortly.
2.2. Cobb-Douglas Example. Consider a utility function given by
u = v(x) = x
1
1
x
2
2
(5)
We usually assume that
i
> 0. To maximize utility subject to a budget constraint we obtain we set up a
Lagrangian function.
L = x
1
1
x
2
2
[ p
1
x
1
+ p
2
x
2
m] (6)
Differentiating equation 6 we obtain
L
x
1
=

1
x
1
1
x
2
2
x
1
p
1
= 0 (7a)
L
x
2
=

2
x
1
1
x
2
2
x
2
p
2
= 0 (7b)
L

= [ p
1
x
1
+ p
2
x
2
] + m = 0 (7c)
Take the ratio of the 7a and 7b to obtain
1 x

1
1
x

2
2
x1
2 x

1
1
x

2
2
x2
=
p
1
p
2

1
x
2

2
x
1
=
p
1
p
2
(8)
We can now solve the equation for the quantity of good 2 as a function of the quantity of good 1 and the
prices of both goods. Doing so we obtain
x
2
=

2
x
1
p
1

1
p
2
(9)
Now substitute 9 in 7c to obtain
L

= [ p
1
x
1
+ p
2
x
2
] + m = 0
p
1
x
1
+ p
2

2
x
1
p
1

1
p
2
= m

p
1

1
x
1

1
+

2
x
1
p
1

1
= m

p
1
x
1

1
(
1
+
2
) = m
x
1
=

1
(
1
+
2
)
m
p
1
(10)
Similarly for x
2
so that we have
4 CONSUMER CHOICE
x
2
=

2
(
1
+
2
)
m
p
2
(11)
Note that demand for the kth good only depends on the kth price and is homogeneous of degree zero in
prices and income. Also note that it is linear in income. This implies that the expenditure elasticity is equal
to 1. This can be seen as follows.
x
1
=

1
(
1
+
2
)
m
p
1

x
1
m
m
x
1
=
_

1
(
1
+
2
)
1
p
1
_
_
m
1
(1 + 2 )
m
p1
_
= 1
(12)
We can nd the value of the optimal u by substitution
u = x
1
1
x
2
2
=
_
m
p
1
_

1

1
+
2
__
1
_
m
p
2
_

2

1
+
2
__
2
= m
1+2
p
1
1
p
2
2

1
1

2
2
(
2
+
2
)
12
(13)
This can also be written
u = x
1
1
x
2
2
=
_
m
p
1
_

1

1
+
2
__
1
_
m
p
2
_

2

1
+
2
__
2
=
_

1

1
+
2
_
1
_

2

1
+
2
_
2
_
m
p
1
_
1
_
m
p
2
_
2
(14)
For future reference note that the derivative of the optimal u with respect to m is given by
u = m
1+2
p
1
1
p
2
2

1
1

2
2
(
2
+
2
)
12
u
m
= (
1
+
2
)m
1+21
p
1
1
p
2
2

1
1

2
1
(
2
+
2
)
12
= m
1+21
p
1
1
p
2
2

1
1

2
1
(
2
+
2
)
112
(15)
We obtain by substituting in either the rst or second equation as follows

1
x
11
1
x
2
2
p
1
= 0
=

1
x
11
1
x
2
2
p
1

2
x
1
1
x
21
2
p
2
= 0
=

2
x
1
1
x
21
2
p
2
(16)
CONSUMER CHOICE 5
If we now substitute for x
1
and x
2
, we obtain
=

1
x
11
1
x
2
2
p
1
x
1
=
m
p
1
_

1

1
+
2
_
x
2
=
m
p
2
_

2

1
+
2
_
=

1
_
m
p
1
_

1

1
+
2
__
11
_
m
p
2
_

2

1
+
2
__
2
p
1
=

1
m
1+21
p
11
1
p
2
2

11
1

2
2
(
1
+
2
)
112
p
1
= m
1+21
p
1
1
p
2
2

1
1

2
2
(
1
+
2
)
112
(17)
Thus is equal to the derivative of the optimal u with respect to m.
To check for a maximum or minimum we set up the bordered Hessian. The bordered Hessian in this
case is
H
B
=
_

2
L(x

)
x
1
x
1

2
L(x

)
x
1
x
2
g(x

)
x
1

2
L(x

)
x
2
x
1

2
L(x

)
x
2
x
2
g(x

)
x
2
g(x

)
x
1
g(x

)
x
2
0
_

_
(18)
We compute the various elements of the bordered Hessian as follows
6 CONSUMER CHOICE
L = x
1
1
x
2
2
[p
1
x
1
+p
2
x
2
m]
L
x
1
=
1
x
11
1
x
2
2
p
1
L
x
2
=
2
x
1
1
x
21
2
p
2

2
L
x
2
1
= (
1
)(
1
1)x
12
1
x
2
2

2
L
x
1
x
2
=
1

2
x
11
1
x
21
2

2
L
x
2
2
= (
2
)(
2
1)x
1
1
x
22
2
g
x
1
= p
1
g
x
2
= p
2
The derivatives of the constraints are constants. The bordered Hessian is given by
H
B
=
_

_
(
1
)(
1
1)x
12
1
x
2
2

1

2
x
11
1
x
21
2
p
1

2
x
11
1
x
21
2
(
2
)(
2
1)x
1
1
x
22
2
p
2
p
1
p
2
0
_

_
(19)
To nd the determinant of the bordered Hessian expand by the third row as follows
|H
B
| = (1)
4
p
1

2
x
11
1
x
21
2
p
1
(
2
)(
2
1)x
1
1
x
22
2
p
2

+ (1)
5
p
2

(
1
)(
1
1)x
12
1
x
2
2
p
1

2
x
11
1
x
21
2
p
2

+ 0
= p
1

2
x
11
1
x
21
2
p
1
(
2
)(
2
1)x
1
1
x
22
2
p
2

p
2

(
1
)(
1
1)x
12
1
x
2
2
p
1

2
x
11
1
x
21
2
p
2

= p
1
p
2

2
x
11
1
x
21
2
p
2
1
(
2
)(
2
1)x
1
1
x
22
2
p
2
2
(
1
)(
1
1)x
12
1
x
2
2
+p
1
p
2

2
x
11
1
x
21
2
= 2p
1
p
2

1
2x
11
1
x
21
2
p
2
1
(
2
)(
2
1)x
1
1
x
22
2
p
2
2
(
1
)(
1
1)x
12
1
x
2
2
(20)
For a maximum we want equation 20 to be positive. Rewriting it we obtain
CONSUMER CHOICE 7
2p
1
p
2

2
x
11
1
x
21
2
p
2
1
(
2
)(
2
1)x
1
1
x
22
2
p
2
2
(
1
)(
1
1)x
12
1
x
2
2
> 0 (21)
We can also write it in the following convenient way
2p
1
p
2

2
x
11
1
x
21
2
+
2
p
2
1
x
1
1
x
22
2

2
2
p
2
1
x
1
1
x
22
2
+
1
p
2
2
x
12
1
x
2
2

2
1
p
2
2
x
12
1
x
2
2
> 0
(22)
To eliminate the prices we can substitute from the rst-order conditions.
p
1
=

1
x
11
1
x
2
2

p
2
=

2
x
1
1
x
21
2

This then gives

2
_

1
x
11
1
x
2
2

__

2
x
1
1
x
21
2

2
x
11
1
x
21
2
+
2
_

1
x
11
1
x
2
2

_2
x
1
1
x
22
2

2
2
_

1
x
11
1
x
2
2

_2
x
1
1
x
22
2
+
1
_

2
x
1
1
x
21
2

_2
x
12
1
x
2
2

2
1
_

2
x
1
1
x
21
2

_2
x
12
1
x
2
2
> 0
(23)
Multiply both sides by
2
and combine terms to obtain
2
2
1

2
2
x
312
1
x
322
2
+
2
1

2
x
312
1
x
322
2

2
2

2
1
x
312
1
x
322
2
+
1

2
2
x
312
1
x
322
2

2
1

2
2
x
312
1
x
322
2
> 0
(24)
Now factor out x
312
1
x
322
2
to obtain
x
312
1
x
322
2
_
2
2
1
2
2
+
2
1

2

2
2

2
1
+
1

2
2

2
1

2
2
_
>0
x
312
1
x
322
2
_

2
1

2
+
1

2
2
_
>0
(25)
With positive values for x
1
and x
2
the whole expression will be positive if the last term in parentheses is
positive. Then rewrite this expression as
_

2
1

2
+
1
2
2
_
> 0 (26)
Now divide both sides by
2
1

2
2
(which is positive) to obtain
_
1

2
+
1

1
_
> 0 (27)
8 CONSUMER CHOICE
3. THE EXPENDITURE (COST) MINIMIZATION PROBLEM
3.1. Basic duality formulation. The fundamental (primal) utility maximization problem is given by
max
x0
u = v(x)
s.t. px m
(28)
Dual to the utility maximization problem is the cost minimization problem
min
x 0
m = px
s.t. v(x) = u
(29)
3.2. Marshallian and Hicksian demand functions. The solutionto equation 29 gives the Hicksian demand
functions x = h(u, p). The Hicksian demand equations are sometimes called compensated demand equa-
tions because they hold u constant. The solutions to the primal and dual problems coincide in the sense
that
x = g (p, m) = h(u, p) (30)
3.3. Indirect objective functions. We can substitute the optimal levels of the decision variables as functions
of the parameters back into the objective functions to obtain the indirect objective functions. For the primal
problem this gives
u = v (x
1
, x
2
, , x
n
) = v [g
1
(m, p), g
2
(m, p), . . . , g
n
(m, p) ] = (m, p) (31)
This is called the indirect utility function and species utility as a function of prices and income. We can
also write it as follows
(m, p) = max
x
[v(x) : px = m] (32)
The indirect utility function for the Cobb-Douglas utility function is given by
u = x
1
1
x
2
2
=
_
m
p
1
_

1

1
+
2
__
1
_
m
p
2
_

2

1
+
2
__
2
=
_

1

1
+
2
_
1
_

2

1
+
2
_
2
_
m
p
1
_
1
_
m
p
2
_
2
(33)
For the Cobb-Douglas utility function with multiple inputs, the indirect utility function is given by
CONSUMER CHOICE 9
= v (x (p, m) ) =
n

i=1
x
i
i
=
n

i=1
_

i

n
j=1

j
m
p
i
_
i
=
n

i=1
_
m

n
j=1

j
_
i _

i
p
i
_
i
=
_
m

n
j=1

j
_

n
k=1
k
n

i=1
_

i
p
i
_
i
(34)
For the dual problem the indirect objective function is
m =
n
j=1
p
j
h
j
u, p ) = c(u, p) (35)
This is called the cost (expenditure) function and species cost or expenditure as a function of prices and
utility. We can also write it as follows
c(u, p) = min
x
[p x : v(x) = u] (36)
3.4. Inversion of (m, p) and c(u, p). Because c(u, p) = m, we can rearrange or invert it to obtain u as
a function of m and p. This will give (m, p). Similarly inversion of (m ,p) will give c(u, p). These
relationships between the utility maximization cost minimization problems are summarized in gure 1
FIGURE 1. Utility Maximization and Cost Minimization
3.5. Properties of the cost function. The cost function in the consumer problemhas a number of analogous
to its properties in the production problem.
10 CONSUMER CHOICE
3.5.1. C.1. The cost function is nondecreasing in p, increasing in u, and increasing in at least one p.
Let p
1
p
2
. Let x
1
be the cost minimizing input bundle with p
1
and x
2
be the cost minimizing input
bundle with p
2
. Then p
2
x
2
p
2
x
1
because x
1
is not cost minimizing with prices p
2
. Now p
1
x
1
p
2
x
1
because p
1
p
2
by assumption so that
C(p
1
, y) = p
1
x
1
p
2
x
1
p
2
x
2
= C(p
2
, y)
Nonsatiation guarantees that the function will be increasing in u. Let V(u
0
) be the set of all bundles that
are equivalent to or preferred to bundles that provide utility level u
0
. Now let u
1
u
2
. Because V(u
1
) is a
subset of V(u
2
) if u
1
u
2
then
C(u
1
, p) = min
x
{px : x V (u
1
)} min
x
{px : x V (u
2
)} = c(u
2
, p)
The point is that if we have a smaller set of possible xs to choose from then cost must increase.
3.5.2. C.2. Positively linearly homogenous in p
C(u, p) = C(u, p), p > 0.
Let the cost minimization problem with prices p be given by
C(u, p) = min
x
{px : xV (u)}, u DomV, p > 0, (37)
where
DomV = {u R
1
+
: V (u) = }
The x vector that solves this problem will be a function of u and p, and is usually denoted h(u,p). This is
the is Hicksian demand function. The cost function is then given by
C(u, p) = p h(u, p) (38)
Now consider the problem with prices tp (p >0)

C(y, tp) = min

x
{tpx : x V (u)}, u DomV, p > 0
= t min
x
{px : x V (y)}, y DomV, p > 0
(39)
The x vector that solves this problemwill be the same as the vector which solves the problemin equation
37, i.e., h(u,p). The cost function for the revised problem is then given by

C(p, tp) = tp h(u, p) = tC(u, p) (40)

CONSUMER CHOICE 11
3.5.3. C.3. C is concave and continuous in w
To demonstrate concavity let (p, x) and (p, x) be two cost-minimizing price-consumption combinations
and let p= tp + (1-t)p for any 0 t 1. Concavity implies that C(u, p) tC(u, p) + (1-t) C(u, p). We can
prove this as follows.
We have that C(u, p) = p x= tp x+ (1-t)p x where x is the optimal choice of x at prices p Because
x is not necessarily the cheapest way to obtain utility level u at prices p or p,we have p x C(u, pw)
and p x C(u, p) so that by substitution C(u, p) tC(u, p) + (1-t) C(u, p). The point is that if p x and
p x are each larger than the corresponding term is the linear combination then C(u, p) is larger than the
linear combination.
Rockafellar [11, p. 82] shows that a concave function dened on an open set (p > 0) is continuous.
3.6. Shephards Lemma.
3.6.1. Denition. If indifference curves are convex, the cost minimizing point is unique. Then we have
C(u, p)
p
i
= h
i
(u, p) (41)
which is a Hicksian Demand Curve
3.6.2. Constructive proof using the envelope theorem. The cost minimization problem is given by
C(y, w) = min
x
px : v(x) u = 0 (42)
The associated Lagrangian is given by
L = px (v(x) u) (43)
The rst order conditions are as follows
L
x
i
= p
i

v
x
i
= 0, i = 1, . . . , n (44a)
L

= (vx u) = 0 (44b)
Solving for the optimal xs yields
x
i
(u, p) = h
i
(u, p) (45)
with C(u,p) given by
C(u, p) = px(u, p) = ph(u, p) (46)
If we now differentiate 46 with respect to p
i
we obtain
C
p
i
=
n
j=1
p
j
x
j
(u, p)
p
i
+ x
i
(u, p) (47)
From the rst order conditions in equation 44a (assuming that the constraint is satised as an equality)
we have
p
j
=
v
x
j
(48)
12 CONSUMER CHOICE
Substitute the expression for p
j
from equation 48 into equation 47 to obtain
C
p
i
=
n
j=1

v(x)
x
j
x
j
(u, p)
p
i
+ x
i
(u, p) (49)
If >0 then equation 44b implies[v(x)-u] = 0. Nowdifferentiate equation 44b with respect to p
i
to obtain

n
j=1
v(x(u, p))
x
j
x
j
(u, p)
p
i
= 0 (50)
which implies that the rst term in equation 49 is equal to zero and that
C(u, p)
p
i
= x
i
(u, p) (51)
3.7. The cost function and Marshallian demand functions. If we substitute the indirect utility function in
the Hicksian demand functions obtained via Shephards lemma in equation 41, we get x in terms of m and
p. Specically
x
i
= x
i
(u, p) = h
i
(u, p) = h
i
[(m, p), p] = g
i
(m, p) = x
i
(m, p) (52)
3.8. Cobb-Douglas Example. The utility function is given by
v(x
1
, x
2
, . . . , x
n
) =
n

i=1
x
i
i
= x
1
1
x
2
2
x
3
3
(53)
First set up the Lagrangian problem
L =
n
i=1
p
i
x
i
(
n

i=1
x
i
i
u ) (54)
The rst order conditions are as follows
L
x
i
= p
i

_

i
x
1
1
x
2
2
x
i 1
i1
x
i 1
i
x
i + 1
i +1

= 0, i = 1, . . . , n (55a)
= p
i


i
v
x
i
= 0, i = 1, . . . , n (55b)
L

=
n

i=1
x

i
+ u = 0 (55c)
(55d)
Taking the ratio of the ith and jth equations we obtain
p
i
p
j
=

i
x
j

j
x
i
(56)
We can now solve the equation for the jth quantity as a function of the ith quantity and the ith and jth
prices. Doing so we obtain
x
j
=

j
x
i
p
i

i
p
j
=

j
x
1
p
1

1
p
j
(57)
CONSUMER CHOICE 13
where we treat the rst good asymmetrically and solve for each demand for a good as a function of the
rst. Now substituting in the utility function we obtain
v =
n

j=1
x
j
j
=
n

j=1
_

j
x
1
p
1

1
p
j
_
j
(58)
Because x
1
, p
1
and
1
do not change with j, they can be factored out of the product to obtain
u =
_
x
1
p
1

1
_

n
j=1
j
n

j=1
_

j
p
j
_
j
(59)
We then solve this expression for x
1
as a function of u and the other xs. To do so we divide both sides
by the product term to obtain
x

n
j=1
j
1
_
p
1

1
_

n
j=1
j
=
u

n
j=1
_
j
pj
_
j
(60)
We now multiply both sides by
_
1
p1
_

n
j=1
j
to obtain
x

n
j=1
j
1
=
_
1
p1
_

n
j=1
j
u

n
j=1
_
j
pj
_
j
(61)
If we now raise both sides to the power
1

n
j=1
j
we nd the value of x
1
x
1
=
_

1
p
1
_
_
_
u

n
j=1
_
j
pj
_
j
_
_
1

n
j=1

j
(62)
Similarly for the other x
k
so that we have
x
k
=
_

k
p
k
_
_
_
u

n
j=1
_
j
pj
_
j
_
_
1

n
j=1

j
(63)
Now if we substitute for the ith x in the cost expression we obtain
14 CONSUMER CHOICE
C =
n
i=1
p
i
_

i
p
i
_
_
_
u

n
j=1
_
j
pj
_
j
_
_
1

n
j=1

j
= (
n
i=1

i
)
_
_
u

n
j=1
_
j
pj
_
j
_
_
1

n
j=1

j
= (
n
i=1

i
) u
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
(64)
3.9. The indirect utility function and Hicksian demands. If we substitute C(u,p) in the Marshallian de-
mands, we get the Hicksian demand functions
x
i
= x
i
(m, p) = g
i
(m, p) = g
i
[C(u, p), p] = h
i
(u, p) = x
i
(u, p) (65)
3.10. Roys identity. We can also rewrite Shephards lemma in a different way. First write the identity
(C(u, p), p] = u (66)
Then totally differentiate both sides of equation 66with respect to p
i
holding u constant as follows
[C(u, p), p]
m
C(u, p)
p
i
+
[C(u, p), p]
p
i
= 0 (67)
Rearranging we obtain
C(u, p)
p
i
=

[C(u,p),p]
pi
[C(u,p),p]
m
= g
i
(m, p) (68)
where the last equality follows because we are evaluating the indirect utility function at income level m.
Figure 2 makes these relationships clear.
4. PROPERTIES OF INDIRECT UTILITY FUNCTION
The indirect utility function has the following properties.
4.1. .1. (m,p) is nonincreasing in p, that is if p p, (m,p) (m,p).
4.2. .2. (m,p) is nondecreasing in m, that is if m m, (m,p) (m,p).
4.3. .3. (m,p) is homogeneous of degree 0 in (p, m) so that (tm,tp) = (m,p) for t > 0.
4.4. .4. (m,p) is quasiconvex in p; that is {p: (m,p) < } is a convex set for all .
4.5. .5. (m,p) is continuous for all p > 0, m > 0.
CONSUMER CHOICE 15
FIGURE 2. Demand, Cost and Indirect Utility Functions
5. DISCUSSION AND PROOFS OF PROPERTIES OF INDIRECT UTILITY FUNCTION
5.1. .1. (m,p) is nonincreasing in p, that is if p p, (m,p) (m,p).
If prices go up, indirect utility cannot increase. Let B = {x: px <m} and B = {x: px <m} for p >p. Then
B is contained in B. Therefore the maximum of v(x) over B is at least as great as the maximum of v(x) over
B
5.2. .2. (m,p) is nondecreasing in m, that is if m m, (m,p) (m,p).
If income goes up, indirect utility cannot decrease. Let B = {x: px <m} and B = {x: px <m} for m > m.
Then B is contained in B. Therefore the maximumof v(x) over B is at least as great as the maximumof v(x)
over B.
5.3. .3. (m,p) is homogeneous of degree 0 in (p, m) so that (tm,tp) = (m,p) for t > 0.
This is called the absence of money illusion. If prices and income are multiplied by the same positive
number, the budget set does not change. Specically,
n

i=1
(tp
i
) x
i
= tm

i=1
p
i
x
i
= m
With the same budget set, the utility maximization problem has the same solution.
5.4. .4. (m,p) is quasiconvex in p; that is {p: (m,p) < } is a convex set for all .
Suppose p and p are such that (m,p) , (m,p) . Let p = tp + (1-t)t. We want to show that
(m,p) . Dene the budget sets:
16 CONSUMER CHOICE
B = {x : px < m}
B

= {x : p

x < m}
B

= {x : p

x < m}
We can showthat any x in B must be in either B or B; that is that B B B. Assume not; then x is such
that tpx + (1-t)px |leq m, but Px > m and px >m. These two inequalities can be written as
tpx > tm
(1 t)p

x > (1 t)m
(69)
Summing the two expressions in equation 69 we obtain
tpx + (1 t)p

x > m
But this contradicts the original assumption that x is in neither B or B.
Now note that
(m, p

) = max
x
v(x), such that x B

max
x
v(x), such that x (B B

); because B B

because (m, p) and (m, p

)
5.5. .5. (m,p) is continuous for all p > 0, m > 0.
By the theorem of the maximum (given below) (m,p) is continuous for p > 0, m > 0. In the utility
maximization problem, f(x,) in the theorem of the maximumis the utility function. It does not depend on
. The constraint set is those values of x that are in the budget set as parameterized by p and m. So in
this case is (p,m). The indirect utility function (m,p) is M() while the ordinary demand functions x(m,p)
are m(). The utility function is continuous by assumption. The constraint set is closed. If p > 0 and m > 0,
the constraint set will be bounded. If some price were zero, the consumer might want to consume innite
amounts of this good. We rule that out.
Theorem 1 (Theorem of the Maximum). Let f(x,) be a continuous function with a compact range and suppose
that the constraint set ()is a non-empty, compact-valued, continuous correspondence of . Then
(i) The function M() = max
x
{f(x, ) : x () } is continuous
(ii) The correspondence m() = {x () : f(x, ) = M()} is nonempty, compact valued and upper
semi-continuous.
Proof: See Berge [1, p. 116].
6. MONEY METRIC UTILITY FUNCTIONS
6.1. Denition of m(p,x). Assume that the consumption set X is closed, convex, and bounded from below.
The common assumption that the consumption set is X = R
L
+
= {x R
L
: x

0 for = 1, ... , L} is more than

sufcient for this purpose. Assume that the preference ordering satises the properties given in section 1.8.
Then for all x X, let BT(x) = {yBT| y x}. For the price vector p, the money metric m(p,x) is dened by
m(p, x) = min
y 0
py
s.t. y BT(x)
(70)
CONSUMER CHOICE 17
If p is strictly greater than zero and if x is a unique element of the least cost commodity bundles at prices
p, then m(p,x) can be viewed as a utility function for a xed set of prices. It can also be dened as follows.
m(p, x) = C(u(x), p) (71)
The money metric denes the minimum cost of buying bundles as least as good as x. Consider gure 3
FIGURE 3. Utility Maximization and Cost Minimization
x
i
x
j
z
x
m p, x

p
2
All points on the indifference curve passing through x will be assigned the same level of m(p,x), and all
points on higher indifference curves will be assigned a higher level.
6.2. Example of a money metric utility function. Consider the Cobb-Douglas utility function
v(x
1
, x
2
, . . . , x
n
) = A
n

i=1
x
i
i
= A x
1
1
x
2
2
x
3
3
(72)
The cost function associated with this utility function is given in equation 64, which we repeat here.
18 CONSUMER CHOICE
C = (
n
i=1

i
)
_
_
v
A

n
j=1
_
j
pj
_
j
_
_
1

n
j=1

j
= (
n
i=1

i
) v
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
(73)
To obtain the money metric we replace v in equation 73 with v(x) from equation 72.
C = (
n
i=1

i
) v
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
m(p, x) = (
n
i=1

i
)
_
A
n

i=1
x
i
i
_ 1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
= (
n
i=1

i
)
_
A
n

i=1
_
x
i
p
i

i
_
i
_ 1

n
j=1

j
(74)
7. MONEY METRIC INDIRECT UTILITY FUNCTIONS
7.1. Denition of (p,p
0
,m). For the price vectors p and p
0
and income m, the money metric indirect utility
function is dened by
(p, p
0
, m) = c((p
0
, m), p) (75)
The indirect money metric utilty function denes the minimum cost of buying bundles at prices p that
yield uility at least as large as than obtained when prices are p
0
and income is m. The money metric indirect
utility function is sometimes called the indirect compensation function.
7.2. Example of an indirect money metric utility function. Consider the Cobb-Douglas utility function
from equation 72
v(x
1
, x
2
, . . . , x
n
) =
n

i=1
x
i
i
= x
1
1
x
2
2
x
3
3

The cost function associated with this utility function is given in equation 64, which we repeat here.
C = (
n
i=1

i
) v
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
(76)
To obtain the money metric we replace v in equation 76 with (p,m) from equation 34 which we also
repeat here.
(p, m) =
_
m

n
j=1

j
_

n
k=1
k
n

i=1
_

i
p
i
_
i
(77)
Making the substitution we obtain
CONSUMER CHOICE 19
C = (
n
i=1

i
) v
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
m(p, x) = (
n
i=1

i
)
_
_
_
m

n
j=1

j
_

n
k=1
k
n

i=1
_

i
p
0
i
_
i
_
_
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
=
_

n
j=1

j
_
_
m

n
j=1

j
_
n

j=1
_ _

j
p
0
j
_
j
_
1

n
j=1

j
_
_
n

j=1
_
p
j

j
_
j
_
_
1

n
j=1

j
= m
n

j=1
_ _

j

j
_
j
_
1

n
j=1

j
_
_
n

j=1
_
p
j
p
0
j
_
j
_
_
1

n
j=1

j
= m
_
_
n

j=1
_
p
j
p
0
j
_
j
_
_
1

n
j=1

j
(78)
8. PROPERTIES OF DEMAND FUNCTIONS
Demand functions have the following properties
8.1. Adding up or Walras law.
n

i=1
p
i
h
(
u, p) =
n

i=1
p
i
g
i
(m, p) = m (79)
8.2. Homogeneity.
h
i
(u, p ) = h
i
(u, p ) = g
i
( m, p ) = g
i
(m, p) (80)
The Hicksian demands are derivatives of a function that is homogeneous of degree one, so they are
homogeneous of degree zero. Eulers theorem then implies that
n

j=1
h
j
(u, p)
p
j
p
j
= 0 (81)
If all prices and income are multiplied by a constant t > 0, the budget set does not change and so the
optimal levels of x(m,p) do not change. We can also write this in differential form using the Euler equation.
n

j=1
g
j
(m, p)
p
j
p
j
+
g
j
(m, p)
m
m = 0

j=1
g
j
(m, p)
p
j
p
j
=
g
j
(m, p)
m
m
(82)
20 CONSUMER CHOICE
8.3. Symmetry. The cross price derivatives of the Hicksian demands are symmetric, that is, for all i = j
h
j
(u, p)
p
i
=
h
i
(u, p)
p
j
(83)
This is clear fromthe denition of the Hicksian demands as derivatives of the cost function. Specically,
C(u, p)
p
i
= h
i
(u, p) (84)
so that

2
C(u, p)
p
j
p
i
=
h
i
(u, p)
p
j
and

2
C(u, p)
p
i
p
j
=
h
j
(u, p)
p
i
(85)
by Youngs theoremon the equality of cross-partials.
8.4. Negativity. The nxn matrix formed by the elements
hi(u,p)
pj
is negative semi-denite, that is, for any
vector z, the quadratic form
n

i=1
n

j=1
z
i
z
j
h
i
(u, p )
p
j
0 (86)
If the vector z is proportional to p, then the inequality becomes an equality and the quadratic form is
zero. This means the matrix is negative semi-denite. This follows from the concavity of the cost function.
If we denote
hi(u,p)
pj
by s
ij
, then we can write the entire matrix of cross partial derivatives as S = s
ij
. This
then implies that
z

Sz 0 (87)
By the properties of a negative semi-denite matrix, this means that s
ii
0, or that the Hicksian demand
functions have a slope which is non-positive. This follows from concavity of cost, and does not require
convex indifference curves.
8.5. The Slutsky equation. If we differentiate equation 65 with respect to p
j
and then substitute from
Shephards lemma for
C(u,p)
pj
,, we obtain
x
i
= x
i
(m, p) = g
i
[C(u, p), p] = h
i
(u, p) = x
i
(u, p)
x
i
(u, p) = h
i
(u, p) = g
i
[C(u, p), p]

x
i
(u, p)
p
j
= s
ij
=
h
i
(u, p)
p
j
=
g
i
(m, p)
m
C(u, p)
p
j
+
g
i
(m, p)
p
j
=
g
i
(m, p)
m
x
j
+
g
i
(m, p)
p
j
=
x
i
(m, p)
m
x
j
+
x
i
(m, p)
p
j
(88)
The last term in equation 88 is the uncompensated derivative of x
i
with respect to p
j
. To compensate for
this, an amount, x
i
, times
gi
m
must be added on. We can also write equation 88 as follows
x
i
(m, p)
p
j
=
x
i
(u, p)
p
j

x
i
(m, p)
m
x
j
(m, p) (89)
CONSUMER CHOICE 21
The rst term is called the substitution effect, the second termthe income effect. Notice that

h
i
(u, p)
p
i

>

x
i
(m, p)
p
i

(90)
when the i
th
good is normal.
9. INTEGRABILITY
In section 8 we determined that a systemof demand equations satises
(i) Walras Law
(ii) Homogeneity of degree zero in prices and income
(iii) Symmetry
(iv) Negativity
The integrability question has to do with whether a system of equations satisfy these four properties
whether there exists a utility function from which this system can be derived. The answer to this question
is yes as indicated in the following theorem.
Theorem 2. A continuously differentiable function x which maps R
n
+
into the real line is the demand function gen-
erated by some increasing, quasiconcave utility function if (and only if, when utility continuous, strictly increasing,
and strictly quasiconcave) it satises Walras law, symmetry and negativity.
This question is typically posed in terms of the cost function. One shows that a system of Hicksian
demand functions satisfying Walras law, symmetry and negativity has an associated cost function from
which it can be derived. Once one has the cost function, it can be used to obtain the utility function.
Shephards lemma (equation 41) states that
C(u, p)
p
i
= h
i
(u, p) (91)
Ordinary (Marshallian demand equations can be written in terms of Hicksian demand equations (see
equation 65) as follows
x
i
(m, p) = x
i
(C(u, p), p) = h
i
(u, p) = x
i
(u, p) (92)
Combining equations 91 and 92 we can then write
C(u, p)
p
i
= h
i
(u, p) = x
i
(u, p) = x
i
(C(u, p), p), i = 1, 2, . . . , k (93)
Now suppose we have a system of demand functions x
i
(m, p) for i = 1, 2, . . . , k. Now pick some point
x
0
(m,p
0
) and assign it an arbitrary level of utility u
0
. Also assume that
C(u
0
, p
0
) = p
0
x(m
0
, p
0
) (94)
If a cost function which generated this systemof demand functions exists, then it must satisfy the system
of partial differential equations given by
C(u
0
, p)
p
i
= h
i
(u
0
, p) = x
i
(C(u
0
, p), p), i = 1, 2, . . . , k (95)
If this systemof partial differential equations has a solution, then x(m,p) is the demand systemgenerated
the cost function C(u, p). In order to understand the conditions for such a system to have a solution,
consider the derivative of equation 95 with respect to p
j
.
22 CONSUMER CHOICE

2
C(u
0
, p)
p
j
p
i
=
x
i
(C(u
0
, p), p)
C
C(u
0
, p)
p
j
+
x
i
(C(u
0
, p), p)
p
j
(96)
Substituting for
C(u
0
,p)
pj
using Shephards lemma and writing m in place of C we obtain

2
C(u
0
, p)
p
j
p
i
=
x
i
(m, p)
m
x
j
(m, p) +
x
i
(m, p)
p
j
(97)
Notice that the left hand side of equation 97 is symmetric in i and j by Youngs theorem. But if the left
hand side of 97 is symmetric the right hand side of 97 must also be symmetric. This implies that the right
hand side of 97 being symmetric is a necessary condition for the system of partial differential equations in
95 to have a solution. Frobenius theoremstates that symmetry of the right hand side of 97 is also sufcient
for 95 to have a solution. Then notice that the right hand side of 97 is the Slutsky matrix associated with the
demand systemx(m,p). So symmetry of the Slutsky matrix is necessary and sufcient for a function C(u, p)
to exist, from which x(p,m) can be derived. The question remaining is whether the function C(u,p) which
solves 95 is a proper cost function.
We therfore need to verify that the properties of the cost function stated in subsection 3.5 hold for the
function which solves the systemin equation 95. These properties are
1. Nondecreasing in p, increasing in u, and increasing in at least one p.
2. Positively linearly homogenous in p
3. Concave and continuous in w
C(u, p) as a solutionto 95 will be nondecreasing in p because Shephards lemma shows that the derivative
of C(u, p) with respect to any price is Hicksian demand which in nonnegative. The other properties in items
1 and 3 can be similarly shown. C(u,p) will be concave if it has a Hessian matrix which is negative semi-
denite. But the Hessian of C(u,p) is just the Slutsky matrix. So if a system of demand equations has a
negative semi-denite Slutsky matrix, then the solution to the system partial differential equations in 95
will be concave.
A continuously differentiable function x which maps R
n
+
into the real line is the demand function The
bottom line is that a system of demand functions that satises Walras law, symmetry and negativity is
consistent with some increasing, quasiconcave utility function.
CONSUMER CHOICE 23
10. SOME NOTES ON FUNCTIONS, CORRESPONDENCES, AND FUNCTIONAL STRUCTURE
10.1. Functions. By a function we mean a rule that assigns to each element in a set X, a unique element
{f(x)} in another set Y. Consider the set X = R
1
and Y = R
1
and the rule f(x) = 3x. For any real x, the function
assigns a unique real number in the set Y. We often use the following notation for a function
f : X Y
The set X is called the domain of the function f. The set of values taken by f, that is, the set y Y: (x)
[y = f(x)] is called the range of f. The range of f will generally be smaller than Y. Consider the case where X
is the rational numbers and Y is the real numbers. The function f(x) = 3x will not cover all members of Y. A
function whose range is all of Y is said to be onto Y. If A is a subset of X, the image under f of A is dened
to be the set of elements in Y such that y = f(x) for some x in A. This is denoted as f[A] and formally given
by
f [A] = {y Y : (x) [x A and y = f (x)]}
The function f is onto Y iff Y = f[X]. If B is a subset of Y, the inverse image f
1
[B] is the set of x in X for
which f(x) is in B. Formally
f
1
[B] = {x X : f (x) B}
The function f is onto Y iff the inverse image of each nonempty set of Y is nonempty. Consider again
the example where X is the rational numbers. There are elements of Y such that the is no element of X
that could generate them under the function f(x) = 3x. A function f: X Y is called one-to-one if f(x
1
) =
f(x
2
) only when x
1
= x
2
. Consider the function mapping the real line into the real line where f(x) = x
2
. The
function is not one-to-one since f(-3) = f(3). The function is also not onto since there is no real x for which
f(x) = -25.
10.2. Correspondences. Let X and Y be two sets. If with each element of X we associate a subset (X) of Y,
we say that the correspondence x (X) is a mapping of Xinto Y. The set (x) is called the image of x under
the mapping . If the set (x) always consists of a single element, we say that is a function. Consider as
an example the case where X is R
1
+
and Y is R
1
Now consider the mapping (x) = y R
1
: y -

x. This is a
correspondence from X to Y.
10.3. Functional Structure. Functional structure has to do with the amount of information we have about
a given mapping. Considered in a different fashion, it is about the number of constraints that we know are
imposed on a mapping. For example it we knowthat y = 3x for all elements x X, then we knoweverything
there is to know about the mapping. Alternatively, the mapping is very constrained in that no other rule
satises this mapping. Consider the case where y = ax, but a is not known. In this case we know that the
mapping is linear, but we cannot say much more than that about it. Consider a function which maps R
2
into R
1
. Specically consider the function y = a x
2
1
+ b x
2
2
. We knowthat the function is quadratic and that
it has no linear or constant terms. It is also clear that it is homogeneous of degree 2 in the vector x because
f(3x) = 9f(x). Thus the function is not completely general.
Functional structure relates to the way in which a mapping is constructed and the way in which the
elements of the domain enter the mapping. Consider for example a general mapping from R
3
to R
1
that
has the following form: y = f(x
1
, x
2
, x
3
) = g(x
1
, x
2
) + h(x
3
). While this is not much information about the
mapping, it rules out functions such as y = 3x
1
x
2
+ 5x
1
x
3
+ 8x
2
x
3
.
Functional structure is typically represented in two ways. The earliest and perhaps most common is
through the use of differential constraints on the function. These constraints then imply something about
24 CONSUMER CHOICE
the parent function. The difculty with this approach is that it requires differentiability. The other approach
is to consider structure on the function directly as for example f(x) = F(h(x)) where h is required to be
homogeneous in all the xs.
10.4. Functional Structure and Functional Equations. Functional structure is often related to the solution
of functional equations. Functional equations are equations in which the unknown (or unknowns) are
functions. Such functions can be multi-place in the sense of having more than one argument and can also
deal with several variables. The number of places in the equations need not equal the number of variables.
For example, the famous Cauchy equation f(x+y) = f(x) + f(y) is a function with one place but two variables.
Alinear function such as f(z) = cz would satisfy the equation. As another example consider the cost function
for a multiple output rm. Consider the restriction that the cost of producing the vector of outputs (x+y)
is the sum of the cost of producing either vector individually, i.e., C(w,x+y) = C(w, x) + C(w, y). The
determination of the form that all cost functions satisfying this restriction must take involves solving the
functional equation.
CONSUMER CHOICE 25
11. AGGREGATION ACROSS GOODS
11.1. Separability and Aggregation. Separability is related to the ability to aggregate variables in economic
analysis. For example, is it reasonable to aggregate two or more types of cold cereal together in analyzing
the demand for food. Or can the hours of male and female workers be added together for an analysis of
productivity. Separability is specically concerned with how the rate of substitution between two goods
or factors is affected by levels of other goods or factors. For example the rate of substitution between beef
and pork may or may not be affected by the amount of tofu consumed. If this rate is not affected by tofu
consumption, then some types of aggregation of beef and pork may be possible. While beef and pork may
not be separable from tofu in consumption, they may be separable from shirts. Separability is particularly
important for aggregate analysis where inputs tend to come in generic bundles such as labor, capital and
materials, and goods tend to come in bundles such as housing, food, transportation, entertainment and so
on.
11.2. Differential denition of separability. Consider a function f depending on n variables with f being
is twice differentiable and
f
x
i
> 0 (98)
Then variables x
i
and x
j
are separable from x
k
if and only if

_
f(x)
x
i
f(x)
x
j
_
x
k
= 0 x
n

n
= { (x
1
, x
2
, . . . , x
n
) = x R
n
| x 0
n
and x = 0
n
}
(99)
This says that the marginal rate of substitution between x
i
and x
j
does not depend on the level of x
k
.
This denition is due to Leontief [8] and independently Sono [10].
11.3. Intuition of differential denition of separability. As x
k
changes, the indifference curve when pro-
jected into x
i
, x
j
space will have the same slope at
( x
i
, x
j
)
In gure 4, changing x
3
to x
3
does not change the set of x
1
and x
2
such that u(x
1
, x
2
, x
3
) u( x
1
, x
2
, x
3
).
In gure 5, changing x
3
to x
3
changes the set of x
1
and x
2
such that u(x
1
, x
2
, x
3
) u( x
1
, x
2
, x
3
).
26 CONSUMER CHOICE
FIGURE 4. Separability