You are on page 1of 22

Properties of Walrasian Demand

Econ 2100 Fall 2017

Lecture 5, September 12

Problem Set 2 is due in Kellys mailbox by 5pm today

Outline
1 Properties of Walrasian Demand
2 Indirect Utility Function
3 Envelope Theorem
Summary of Constrained Optimization
gi (x) 0
When x is solves max f (x) subject to then
i = 1; ::; m
m
X
rL (x ; ) = rf (x ) i rgi (x ) = 0
i =1
and gi (x ) 0, i 0, and i gi (x ) = 0 for i = 1; :::; m.

Important details:
If the better than set or the constraint sets are not convex: big trouble.
If functions are not dierentiable: small trouble.
If the geometry still works we can nd a more general theorem (see convex
analysis).
When does this fail?
If the constraint qualication condition fails.
If the objective function is not quasi concave.
This means you must check the second order conditions when in doubt.
Walrasian Demand

Denition
Given a utility function u : Rn+ ! R, the Walrasian demand correspondence
x : Rn++ R+ ! Rn+ is dened by
x (p; w ) = arg max u(x) where B(p; w ) = fx 2 Rn+ : p x w g:
x2B p;w

When the utility function is quasi-concave and dierentiable, the First Order
Conditions for utility maximization say:
utility budget X non negativity
r( ) budget r( ) non negativity r( )=0
function constraint constraint constraints constraints
So, at a solution x 2 x (p; w ):
@L (x ) @u (x )
= w xi + i = 0 for all i = 1; ::; n
@xi @xi
Marginal Rate of Substitution

Suppose we have an optimal consumption bundle x where some goods are

consumed in strictly positive amounts.
Then, the corresponding non-negativity constraints hold and the correspoding
multipliers equal 0.
At such a solution x , the rst order condition is:
@u (x )
= w pi for all i such that xi > 0
@xi

This expression implies

@u(x )
@xj pj
@u(x )
= for any j; k such that xj ; xk > 0
pk
@xk
This is the familiar condition about equalizing marginal rates of substituitons
to price ratios.
Marginal Utility Per Dollar Spent
At an optimal consumption bundle x where some goods are consumed in
strictly positive amounts, the rst order condition is:
@u (x )
= w pi for all i such that xi > 0
@xi

Another way to rearrange it gives

@u(x ) @u(x )
@xj @xk
= for any j; k such that xj ; xk > 0
pj pk
the marginal utility per dollar spent must be equal across goods.
@u (x ) @u (x )
If not, there are j and k for which @xj
< pkk
@x
pj
DM can buy p"j less of j, and p"k more of k, so the budget constraint still holds,
and
by Taylors theorem, the utility at the new choice is 0 1
@u (x ) @u (x )
@u (x ) " @u (x ) " B @xk @xj C
u (x )+ + +o (") = u (x )+" @ A+o (")
@xj pj @xk pk pk pj

which implies that x is not an optimum.

Think about the case in which some goods are consumed in zero amount.
Demand and Indirect Utility Function
The Walrasian demand correspondence x : Rn++ R+ ! Rn+ is dened by
x (p; w ) = arg max u(x) where B(p; w ) = fx 2 Rn+ : p x w g:
x2B (p;w )

Denition
Given a continuous utility function u : Rn+ ! R, the indirect utility function
v : Rn++ R+ ! R is dened by
v (p; w ) = u(x ) where x 2 x (p; w ):

The indirect utility function measures changes in the optimizedvalue of the

objective function as the parameters (prices and wages) change and the
consumer adjusts her optimal consumption accordingly.
Results
The Walrasian demand correspondence is upper hemi continuous
To prove this we need properties that characterize continuity for
correspondences.
The indirect utility function is continuous.
To prove this we need properties that characterize continuity for
correspondences.
Berges Theorem of the Maximum
The theorem of the maximum lets us establish the previous two results.

Theorem (Theorem of the Maximum)

If f : X ! R is a continuous function and ' : Q ! X is a continuous
correspondence with nonempty and compact values, then
the mapping x : Q ! X dened by
x (q) = arg max f (x)
x 2'(q)
is an upper hemicontinuous correspondence and
the mapping v : Q ! R dened by
v (q) = max f (x)
x 2'(q)
is a continuous function.

Berges Theorem is useful when exogenous parameters enter the optimization

problem only through the constraints, and do not directly enter the objective
function.
This happens in the consumers problem: prices and income do not enter the
utility function, they only aect the budget set.
Properties of Walrasian Demand
Proposition
If u(x) is continuous, then x (p; w ) is upper hemicontinuous and v (p; w ) is
continuous.

Proof.
Apply Berges Theorem:
If u : Rn+ ! R a continuous function and B (p; w ) : Rn++ R+ ! Rn+ is a
continuous correspondence with nonempty and compact values. Then:
(i): x : Rn++ R+ ! Rn+ dened by x (p; w ) = arg maxx 2B (p;w ) u(x) is an upper
hemicontinuous correspondence and
(ii): v : Rn++ R+ ! R dened by v (p; w ) = maxx 2B (p;w ) u(x) is a continuous
function.
We need continuity of the correspondence from price-wage pairs to budget
sets.
We must show that B : Rn++ R+ ! Rn+ dened by
B (p; w ) = fx 2 Rn+ : p x w g
is continuous and we are done.
Continuity for Correspondences

Reminder from math camp.

Denition
A correspondence ' : X ! Y is
upper hemicontinuous at x 2 X if for any neighborhood V Y containing
'(x), there exists a neighborhood U X of x such that '(x 0 ) V for all
x 0 2 U.
lower hemicontinuous at x 2 X if for any neighborhood V Y such that
'(x) \ V 6= ;, there exists a neighborhood U X of x such that
'(x 0 ) \ V 6= ; for all x 0 2 U.
A correspondence is upper (lower) hemicontinuous if it is upper (lower)
hemicontinuous for all x 2 X .
A correspondence is continuous if it is both upper and lower hemicontinuous.
Conditions for Continuity (from Math Camp)

The following su cient conditions are sometimes easier to use.

Proposition (A)
Suppose X Rm and Y Rn . A compact-valued correspondence ' : X ! Y is
upper hemicontinuous if, and only if, for any domain sequence xj ! x and
corresponding range sequence yj such that yj 2 '(xj ), there exists a convergent
subsequence fyjk g such that lim yjk 2 '(x).

Note that compactness of the image is required.

Proposition (B)
Suppose A Rm , B Rn , and ' : A ! B. Then ' is lower hemicontinuous if, and
only if, for all fxm g 2 A such that xm ! x 2 A and y 2 '(x), there exist
ym 2 '(xm ) such that ym ! y .
Continuity for Correspondences: Examples

Exercise
Suppose ' : R ! R is dened by:
f1g if x < 1
'(x) = :
[0; 2] if x 1
Prove that ' is upper hemicontinuous, but not lower hemicontinuous.

Exercise
Suppose ' : R ! R is dened by:
f1g if x 1
'(x) = :
[0; 2] if x > 1
Prove that ' is lower hemicontinuous, but not upper hemicontinuous.
Continuity of the Budget Set Correspondence

Question 1, Problem Set 3; due next Tuesday.

Show that the correspondence from price-wage pairs to budget sets,
B : Rn++ R+ ! Rn+ dened by
B(p; w ) = fx 2 Rn+ : p x wg
is continuous.

First show that B(p; w ) is upper hemi continuous.

Then show that B(p; w ) is lower hemi continuous.
In both cases, use the propositions in the previous slide (and
Bolzano-Weierstrass).
There was a very very similar problem in math camp.
Properties of Walrasian Demand
Denitions
Given a utility function u : Rn+ ! R, the Walrasian demand correspondence
x : Rn++ R+ ! Rn+ is dened by
x (p; w ) = arg max u(x) where Bp;w = fx 2 Rn+ : p x w g:
x 2B p;w
and the indirect utility function v : Rn++ R+ ! R is dened by
v (p; w ) = u(x (p; w )) where x (p; w ) 2 arg max u(x):
x 2B p;w

Properties of Walrasian demand:

if u is continuous, then x (p; w ) is nonempty and compact.
x (p; w ) is homogeneous of degree zero: for any > 0, x ( p; w ) = x (p; w )
if u represents a locally nonsatiated %, then p x = w for any x 2 x (p; w ).
if u is quasiconcave, then x (p; w ) is convex.
if u is strictly quasiconcave, then x (p; w ) is unique.
If u(x ) is continuous, then x (p; w ) is upper hemicontinuous and v (p; w ) is
continuous.
Next, comparative statics: How does Walrasian Demand change as income
and/or (some) prices change?
Implicit Function Theorem
This is the main tool to perform comparative static analysis (also with constraints)
Same as math camp.

Theorem (Implicit Function Theorem)

Suppose A is an open set in Rn+m and f : A ! Rn is continuously dierentiable.
Let Dx f be the n n derivative matrix of f with respect to its rst n arguments.
If f (x; q) = 0n and Dx f (x; q) is nonsingular, then there exists a neighborhood B of
q in Rm and a unique continuously dierentiable g : B ! Rn such that
g (q) = x and f (g (q); q) = 0n for all q 2 B
Moreover,
1
Dq g (q) = [Dx f (x; q)] Dq f (x; q):

@fi
Notation: (Dx f )ij = @xj

IFT gives a way to write, locally, the xs as dependent on qs via a smooth

(dierentiable) function.
We use this in the consumers utility maximization problem by setting f ( ) to
be the FOC for utility maximization;
then g ( ) is the Walrasian demand, and q is a price vector and income.
Bordered Hessian

For comparative statics he following object is useful.

Denition
Let m = fi : xi (p; w ) > 0g and reindex Rn so these m dimensions come rst. The
bordered Hessian H of u with respect to its rst m dimensions is
2 3
0 u1 u2 um
6 u1 u11 u21 um1 7
6 7 where ui = @u
0 (Dx u)> 6 u2 u12 u22 u 7 @xi
H= 2 = 6 m2 7 ,
@2 u
Dx u Dxx u 6 .. .. .. . . 7 and u ij =
4 . . . .. .. 5 @xi @xj

um u1m u2m umm

This only considers demand functions which are not zero and then takes rst
and second derivatives of the utility function with respect to those goods.
Dierentiability of Walrasian Demand

Proposition
Suppose u is twice continuously dierentiable, locally nonsatiated, strictly
quasiconcave, and that there exists " > 0 such that xi (p; w ) > 0 if and only if
xi (p; w ) > 0, for all (p; w ) such that k(p; w ) (p; w )k < ".a If H is nonsingular at
(p; w ), then x (p; w ) is continuously dierentiable at (p; w ).
a This condition is automatically satised if xi > 0 for all i , by continuity of x .

Proof.
Question 2, Problem Set 3.

These are su cient conditions for dierentiability of x (p; w ), yet we do not

have axioms on % that deliver a continuously dierentiable utility.
We also assume demand is strictly positive (locally), but this is the very object
we want to characterize.
Proof.
We want to use the Implicit Funtion Theorem. Things we know right away:
u is strictly quasiconcave ) x is a function.
u is continuously dierentiable ) the rst order conditions are well dened.
u is locally nonsatiated ) the budget constraint binds.
x is strictly positive in the rst m commodities ) ignore corresponding
nonnegativity constraints.
It now su ces to show that x is dierentiable in the rst m prices, and w ,
and ignore the last n m commodities.a
The remainder of the proof uses IFT to show that x is dierentiable as
desired;
Fill in the details.
a There is a neighborhood of (p; w ) where consumption is zero for the last n m commodities,
and these constant dimensions will have no eect on the dierentiability of x .
Comparative Statics
Without constraints this is the same as math camp

Problem
Let : Rn Rm ! R, written (x; q); DM chooses x to maximize ; while q are
parameters she does not control.

Comparative Statics
We want to know how DM adjusts her optimal choice x (q) when the parameters
q change. What is the derivative of x (q)?

If is strictly concave and dierentiable, and smooth, the Implicit Function

Envelope Theorem Without Constraints
Dene: x (q) = arg max (x; q): The maximizer solves the rst order condition:
f (x; q) = Dx (x; q) = 0>
n : The Jacobian of f is Dx f (x; q) = Dxx (x; q) , and is
nonsingular (why?).

Reminder from math camp

By IFT: x (q) is continuously dierentiable close to a solution x (q); q, and:
1 1
Dq x (q) = [Dx f (x (q); q)] Dq f (x (q); q) = [Dxx (x (q); q)] Dqx (x (q); q)
| {z }| {z }
n n n m

Using the Chain Rule, the change of (x (q); q) is:

=0
z }| {
Dq (x (q); q) = Dq (x; q)j q=q + Dx (x; q)j q=q Dq x (q) = Dq (x; q)j q=q
x=x (q) x=x (q) x=x (q)

the second order eect of how the maximizer x responds to q is irrelevant; only
the rst order eect of how q changes the objective function evaluated at the
xed maximizer x (q) matters. This observation is sometimes called the Envelope
Theorem.
@2
@x @q@x
If x and q are scalars, Dq x (q) becomes @q = @2
@x 2
Comparative Statics With Constraints
There are k equality constraints, Fi (x; q) = 0 with each Fi smooth.

The maximization problem now becomes:

x (q) = arg max (x; q)
F (x;q)=0k

Assume constraint qualications are met and form the Lagrangian:

>
L( ; x; q) = (x; q) F (x; q):
The derivative of the Lagrangian is:
0 1
1 k
z }| {
B F (x; q) C
B C
B C
f ( ; x; q) D( ;x) L( ; x; q) = B C
| {z } B > C
@Dx (x; q) D F (x;
|{z} | {z }A
x q)
1 (k +n) | {z }
1 n 1 k k n

Fix q 2 Rm and let and x be the corresponding maximum;

We can now use IFT on the FOC f ( ; x ; q) = 0>
k +n (it works because the
Jacobian is non singular).
Comparative Statics With Constraints
By IFT, x and are implicit functions of q in a neighborhood of q, and
1
Dq ( (q); x (q)) = D( ;x ) f ( ; x ; q) Dq f ( ; x ; q)
We know f is the derivative of the Lagrangian:
F (x; q)
f ( ; x ; q) = >
Dx (x; q) Dx F (x; q)
So, we can gure out that
0 1
k k k n
z}|{ z }| {
B C
B 0 Dx F (x ; q) C
B C
D( ;x) f ( ; x ; q) = B C
| {z } B D F (x ; q)> D 2 (x ; q) D [ > D F (x ; q)]C
@ x x x A
(k +n) (k +n) | {z } | xx {z }
n k n n
and 0 1
k m
z }| {
B Dq F (x (q); q) C
B C
B C
Dq f ( ; x ; q) = B C
| {z } BD 2 (x ; q) > C
Dq ( Dx F (x ; q)> )A
@ qx
(k +n) m | {z } | {z }
n m n m
Envelope Theorem With Constraints
Now we can gure out the change in the objective function.
Using the Chain Rule, the change of (q) = (x (q); q) is:
Dq (x (q); q) = Dq (x; q)jq=q, x=x (q) + Dx (x; q)jq=q, x=x (q) Dq x (q)

Because the constraint hold, F (x; q) = 0k and thus

Dx F (x; q)Dq x (q) =Dq F (x; q)
>
From FOC we get Dx (x; q) = Dx F (x; q)
Thus,
> >
Dx (x; q)jq=q, x=x (q) Dq x (q) = Dx F (x; q)Dq x (q) = Dq F (x; q)

Envelope Theorem With Constraints

Then, the change in the objective function due to a change in q is:
>
Dq (x (q); q) = Dq (x; q)jq=q;x=x (q) ( (q)) Dq F (x; q)jq=q;x=x (q)