Вы находитесь на странице: 1из 141

Microeconomic Theory

ECON 501
Lecture Notes on Competitive Equilibrium in
Pure Exchange Economies
1
Mehmet Barlo
Sabanc University
1
These lecture notes are prepared from various sources (while honoring due credit) to be used in Eco-
nomics 501-502 course at the Sabanc University, and they are not intended for sale. Please contact the
author for permission to use these lecture notes elsewhere. Email barlo@sabanciuniv.edu for additional
comments and questions.
Contents
1 Pure Exchange Economies 1
2 Preferences 4
2.1 Completeness, Reexivity and Transitivity . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Continuity of Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Debreus Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Desirability Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Convexity of Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Relation among the Assumptions on Preferences . . . . . . . . . . . . . . . . . . . . 15
3 Pareto Optimality and the Core 23
3.1 Pareto Optimality and Individual Rationality . . . . . . . . . . . . . . . . . . . . . 23
3.2 The Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Pareto optimality and the Planners Problem . . . . . . . . . . . . . . . . . . . . . . 27
4 The Price Mechanism 33
4.1 Budget Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.1 Properties of the Budget Correspondence . . . . . . . . . . . . . . . . . . . . 35
Continuity of the Budget Correspondence . . . . . . . . . . . . . . . . . . . 38
4.2 Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2.1 Derivation of Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
i
CONTENTS ii
Demand with Cobb-Douglas Preferences . . . . . . . . . . . . . . . . . . . . 55
Comparative Statics with Cobb-Douglas and Fixed Income . . . . . . . . . . 56
Further Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5 Competitive Equilibrium 60
5.1 Computing Competitive Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.2 Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.1.3 Example 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.1.4 Example 4: Cobb-Douglas Utilities . . . . . . . . . . . . . . . . . . . . . . . 72
5.1.5 Example 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Existence of Competitive Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2.1 The Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3 Welfare Properties of Competitive Equilibria . . . . . . . . . . . . . . . . . . . . . . 94
5.3.1 First Fundamental Theorem of Welfare Economics . . . . . . . . . . . . . . . 94
5.3.2 Second Fundamental Theorem of Welfare Economics . . . . . . . . . . . . . 96
5.3.3 Competitive Equilibria and The Planners Problem . . . . . . . . . . . . . . 99
5.4 Competitive Equilibrium and The Core . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.4.1 Existence of The Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.5 Debreu Scarf Limit Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6 Appendix: Basic Properties of Topological and Metric Spaces 109
7 Appendix: Continuity of Correspondences 117
8 Appendix: Theorem of Maximum 123
9 Appendix: Fixed Point Theorems 125
9.1 Brouwers Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
CONTENTS iii
9.2 Kakutanis Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
10 Appendix: Finite Normal Form Games and Nash Equilibrium 129
11 Appendix: Convexity and Separating Hyperplane Theorems 133
1 Pure Exchange Economies
In this chapter we are going to concentrate on pure exchange economies. After describing the
environment, we will present the denition of general (competitive) equilibrium and analyze the
properties of its components. Finally, this chapter will establish its existence.
The notes presented here are based on Debreu (1959), and Hildenbrand and Kirman (1976).
Let the commodity space be given by R
M
, and the consumption space X by R
M
+
= y R
M
:
y
k
0, k = 1, . . . , M.
This requirement restricts attention to situations where there are only M < goods each of
which are perfectly divisible. Hence, cases with innitely many goods (thus, cases that involve non-
bounded time aspect) are ignored. Moreover, each good being assumed to be perfectly divisible,
eliminates some of quite interesting cases. Such a case could be tires, because half a tire is not good
for any kind of car, and may only be used in harbors.
We aim to describe agents behavior via their preferences over consumption bundles. A pref-
erence relation is a binary comparison, i.e. for each x and y a preference relation consists of a
comparison. (Note that this does not imply that each x and y must be comparable, indeed a
preference relation in which such a point occurs is allowed.)
The set of agents is given by N = 1, . . . , n, and each of them have preferences (to be dened
below) given by
i
on X, where x
i
y for x, y X means that agent i thinks that consumption
bundle x is at least as good as consumption bundle y (weakly prefers x to y).
Denition 1 A preference relation, , is a binary relation dened on X, i.e. X X,
whose graph is given by Graph() = (x, y) X X [ x y.
1
Economics 501-2 by Mehmet Barlo: 1 Pure Exchange Economies 2
As an example consider X R
+
and . Now we have Graph() = (x, y) R
2
+
[ x y.
The following notation is needed for future analysis:
Denition 2 For x, y X let
1. x ~y if and only if x y and not y x, and we say x is strictly preferred to y;
2. x y if and only if x y and y x, and we say x is indierent to y;
3. the upper-contour set of x is given by |(o(x) = y X [ y x; and o|(o(x) = y
X [ y ~x denes the strict upper-contour set of x.
4. the lower-contour set of x is given by L(o(x) = y X [ x y; and oL(o(x) = y
X [ x ~y denes the strict lower-contour set of x.
5. the indierence set of x is 1(x) = y X [ x y.
In words, the upper contour set of a consumption bundle x X R
M
+
given to player i is the
set of all consumption bundles each of which is weakly preferred to x. The strict upper contour set
is those in which the preference relation is strict. Similar considerations apply to the denitions of
the lower contour and strict lower contour sets.
Now we are ready to dene a pure exchange economy:
Denition 3 (Pure Exchange Economy) A pure exchange economy c with consumption
space X and the set of agents N, is
c = e
i
,
i

iN
,
where e
i
X and
i
X X denote the agent is initial endowment and preference relation,
respectively.
That is, a pure exchange economy is dened with a nitely many and perfectly divisible goods,
nitely many agents each of whom possesses a preference relation and a given initial endowment
vector. That is summarized in the above denition.
Economics 501-2 by Mehmet Barlo: 1 Pure Exchange Economies 3
In words, when we talk about pure exchange economies, we imagine an island in the Bahamas
which does not have any access to the outside world. Each agent possesses some goods (good or
bad depends on the tastes given by
i
) given by their initial endowments. If there were no trade,
each has to eat his own initial endowment by himself. But, with trade, he could exchange some
of his goods with others so that after the exchange everybody gets better o.
In formal economics, what we seek is an arrangement of allocations from which further trade
should not be benecial to the agents. Because if it were, the point under question can hardly be
called a solution. Inherently, we believe in the rationality of agents, and therefore, we expect the
solution of such an exchange situation one with that particular feature, namely the requirement
that the solution must be one from which further trade should not benecial to the agents.
These ideas, formalized in chapter 3, are captured by the notion of the core, which in essence
requires that the solution is one such that no coalitions of agents should be able to be strictly
better o by switching jointly to another feasible allocation. Consequently, whatever bargaining
is going on, whatever power distributions are in eect, we expect the solution to this economy be
an allocation in the core. Because if it is not, then that means there are some agents who are not
using an opportunity that they should have. And that is clearly against our belief that the solution
should not violate any rationality considerations.
Before going into the analysis of the core, we will rst concentrate on assumptions on preferences
that will be needed in our analysis.
2 Preferences
In order to employ various results from mathematics some assumptions (axioms) on preference
relations are needed. Some of those axioms might be stronger than desired, which would result that
the model we are going to analyze in this chapter not be suitable for situations for which some of
these axioms are violated.
2.1 Completeness, Reexivity and Transitivity
Denition 4 is said to be reexive if for all x X, x x.
Denition 5 is said to be transitive if for all x, y, x X,
x y and y z x z.
Denition 6 A binary relation on X X is a preorder if it is both reexive and transitive.
Denition 7 is said to be complete if for all x, y X, either x y or y x, or both.
Note that sometimes the assumption that is a complete preorder is called rationality. I
personally dont agree with that denition, because after all I might not know my preferences
comparing Beatles St. Peppers Lonely Hearts Club Band and Help! (an example of incomplete
preferences), and that does not mean that I am indierent between the two. It just means that I
do not prefer one over the other, and am not indierent between them.
For a concrete example consider X = R
2
+
and =. It should be obvious to the reader that
this binary relation is transitive, reexive but not complete. This is because of the following: For
4
Economics 501-2 by Mehmet Barlo: 2 Preferences 5
any x, y X (recall that X is the same as R
M
+
), x is greater or equal to x, thus, x x, hence these
preferences are reexive. Moreover, for any x, y and z in X, x y and y z implies x z, leading
to the conclusion that for all x, y and z in X with x y and y z, we have x z. However, this
relation, , is not complete. To see this consider M = 2 and two points x = (1, 2) and y = (2, 1).
We cannot say that x y or x y.
Similar problems appear with transitivity as well. That is, it is not dicult to come up with
intuitive preference ordering that violate transitivity. Those examples generally involve preferences
over risky, or time dependent assets.
Analysis of preference relations when completeness (or transitivity) is dropped is a modern hot
research topic.
From now on unless otherwise stated we will keep the following assumption:
Assumption 1 Preference relation given by X X is a complete preorder. That is, is
reexive, complete and transitive.
2.2 Continuity of Preferences
The next assumption will require that preferences should not feature any jumps.
In other words, continuity eliminates cases when an agents preferences may involve a drastic
change. In order to see that, consider the following example. Suppose that there are 3 goods, the
rst being coee, the second tea and the last sugar. Now assume that the preferences of the agent
under analysis is such that he prefers coee to tea as long as none of it has any sugar. Yet, he
cannot take any sugar in his coee no matter what happens, because then he hates it. (Of course
we also need to assume that this agent is like a tasting machine where he can detect very tiny
amounts of sugar in his coee.) Therefore, for any natural number n our agent strictly prefers y
n
given by (0, 1, 1/n) to x
n
= (1, 0, 1/n) (that is he strictly prefers getting one tea with 1/n sugar to
one coee with the same amount of sugar). It should be obvious that x
n
converges to x = (1, 0, 0)
Economics 501-2 by Mehmet Barlo: 2 Preferences 6
and y
n
to y = (0, 0, 1). Finally, even though for any natural number n, we have y
n
~x
n
, in the limit
this relation is reversed so that x ~y. Thus, these preferences are not continuous.
Denition 8 is continuous if for all x X, both |(o(x) and L(o(x) are closed.
The following Proposition will establish equivalent denitions for continuity based on closedness
and openness of sets and sequences, concepts that the reader is assumed to be familiar with from
textbooks such as Kolmogorov and Fomin (1970) and Rudin (1976):
Proposition 1 The following statements are equivalent:
1. is continuous;
2. for all x X, both o|(o(x) and oL(o(x) are open;
3. Graph() is closed;
4. for all x, y X with y ~x, there exists
1
,
2
> 0 with
[x x

[ <
1
and [y y

[ <
2
y

~x

;
5. for any sequence x
n
, y
n

nN
X X with x
n
y
n
for all n N, and (x
n
, y
n
) (x, y), we
have that x y.
Proof. The proof will involve the following steps:
Step 1. is continuous if and only if for all x X, both o|(o(x) and oL(o(x) are open.
Proof. From basic mathematical analysis, it is well known that a set A X is open if and only
if XA is closed. Noting that for any x X, Xo|(o(x) = L(o(x) and XoL(o(x) = |(o(x),
nishes the proof.
Step 2. is continuous if and only if Graph() is closed.
Proof. Recall that Graph() = (x, y) X X [ x y. Suppose that is continuous. Then
for any y X, x X : x y = |(o(y) is closed. Moreover, for any x X, y X : x y =
L(o(x) is closed. Thus, because that x and y were arbitrary, Graph() is closed.
Economics 501-2 by Mehmet Barlo: 2 Preferences 7
Now, suppose that Graph() is closed. Then for any y X, |(o(y) = x : x y is closed.
Moreover, for any x X, L(o(x) = y : x y is also closed. Thus, is continuous.
Step 3. is continuous if and only if for all x, y X with y ~x, there exists
1
,
2
> 0 with
[x x

[ <
1
and [y y

[ <
2
y

~x

.
Proof. Suppose is continuous. Then for any x, y with y ~x, it must be that y o|(o(x),
and x oL(o(y). Because of the second step, we know that both o|(o(x) and oL(o(y) are open.
Hence, again from basic mathematical analysis we know that there exists
1
> 0 and
2
> 0, such
that B

1
(x) = x

X : [xx

[ <
1
is contained in oL(o(y), and B

2
(y) = y

X : [yy

[ <
2

is contained in o|(o(x). Thus, for all [x x

[ <
1
and [y y

[ <
2
we know that y

~x
and y ~x

, because y

2
(y) o|(o(x) and x

1
(x) oL(o(y). Let
1
,
2
> 0 be xed.
Repeating the same argument for any y

with xed
2
, and thus y

~x, we obtain the existence of

2,1
with
2
>
2,1
> 0 such that y

~x

for any x

2,1
(x); and, for any x

with xed
1
, and
thus y ~x

, we obtain the existence of


1,2
with
1
>
1,2
> 0 such that y

~x

for any y

1,2
(y).
Letting
1
=
1,2
, and
2
=
2,1
delivers the required result.
1
Suppose that there exists x, y with y ~x, but there are no
1
,
2
> 0 such that for all x

and y

,
[x x

[ <
1
and [y y

[ <
2
, implies y

~x

. Then, using the above technique, it can be shown


that these would contradict with the openness of o|(o(x) or oL(o(y).
Step 4. is continuous if and only if for any sequence x
n
, y
n

nN
X X with x
n
y
n
for all
n N, and (x
n
, y
n
) (x, y), we have that x y.
Proof. Suppose for a contradiction that is continuous, but there exists a sequence x
n
, y
n

nN

X X with x
n
y
n
for all n N, and (x
n
, y
n
) (x, y), we have that y ~x. Therefore, by step 3
we know that there exists
1
> 0 and
2
> 0 such that x

and y

, [x x

[ <
1
and [y y

[ <
2
,
implies y

~x

. Let N be high enough so that for all n > N, x


n
B

1
(x) and y
n
B

2
(y). Thus, by
step 3 we know that y
n
~x
n
for all n > N, a contradiction to x
n
y
n
for all n N.
1
Generally, such a tedious check is not needed because this technique presented above is quite standard, and is
implied by the openness of sets.
Economics 501-2 by Mehmet Barlo: 2 Preferences 8
Conversely, assume that for any sequence x
n
, y
n

nN
X X with x
n
y
n
for all n N, and
(x
n
, y
n
) (x, y), we have that x y. Take any x X, and we will show that L(o(x) is closed.
From basic mathematics, recall that a set is closed if and only if every convergent sequence in that
set, converges in that set. Take any sequence y
n
in L(o(x) converging to y. I need to show that
y L(o(x). Dene a sequence by x
n
= x and y
n
with y
n
L(o(x), and lim
n
y
n
= y. Because of
our hypothesis, we have that x y, y L(o(x), hence L(o(x) is closed. Closedness of the upper
contour sets can easily be shown using a similar argument.
Instead of continuity we could have worked with less restrictive notions:
Denition 9 is upper-semi continuous if for all x X, |(o(x) is closed (alternatively
oL(o(x) is open). And is lower-semi continuous if for all x X, L(o(x) is closed (alter-
natively o|(o(x) is open).
The reader should note that a preference ordering is continuous if and only if it is both upper-
semi continuous and lower-semi continuous.
2.3 Debreus Representation Theorem
Our next task is to present the phenomenal result known as Debreus representation theorem, which
establishes that under very weak assumptions, there exists a continuous real valued function that
represents an agents preferences.
Denition 10 Given preference ordering on XX we say that a real valued function u : X R
represents if and only if
x y u(x) u(y).
Proposition 2 If a real valued function u : X R represents some preference relation , and
f : R R is a strictly increasing function
2
, then v : X R dened by v = f u also represents
2
We say that a function f : R R is a strictly increasing if x > y f(x) > f(y), x, y R.
Economics 501-2 by Mehmet Barlo: 2 Preferences 9
the same preferences . The assertion fails if f is a non-decreasing function
3
.
Exercise 1 Prove Proposition 2.
Now we are ready for the famous Representation Theorem by Debreu (1959).
Theorem 1 (Debreu (1959)) Suppose that preferences X X are continuous complete
preorders. Then there exists a continuous real valued function u : X R which represents .
Proof. The proof of this Theorem is beyond the scope of this course, and hence, is omitted.
We refer the reader to page 56 of Debreu (1959).
Under the light of the following denition, the next Proposition is easier to be obtained:
Denition 11 A function f : X R is upper-semi continuous if for all R, x X [
f(x) is a closed set in X. Similarly, it is lower-semi continuous if for all R,
x X [ f(x) is a closed set in X.
Lemma 1 A function f : X R, where (X, ) is a metric space, is continuous
4
if and only if it
is both upper-semi continuous and lower-semi continuous.
Exercise 2 Prove this lemma.
Proposition 3 Suppose that u : X R represents a given preference relation . Then the
following must hold:
1. is a complete preorder;
2. if u : X R representing is continuous so is .
3. if u : X R representing is upper-semi continuous so is .
4. if u : X R representing is lower-semi continuous so is .
3
We say that a function f : R R is a non-decreasing function if x y f(x) f(y), x, y R.
4
A function f : X R

is continuous if for all x


n

nN
X with x
n
x, we have f(x
n
) f(x).
Economics 501-2 by Mehmet Barlo: 2 Preferences 10
Exercise 3 Prove the previous Proposition.
Exercise 4 Suppose that the preference ordering ~ on R
2
+
is given by the following: For any
x, y R
2
+
, x ~y if either x
1
> y
1
, or x
1
= y
1
and x
2
> y
2
. This relation is called the lexicographic
order, and is basically the simplied version of the alphabetical order.
1. Is this relation reexive, complete and transitive? Is this relation continuous? Prove your
answers.
2. For any given x X nd 1(x) with this order.
3. Prove that there is no a continuous utility function u(x
1
, x
2
) representing this preference re-
lation.
The utility function that Theorem 1 implies is ordinal. That is what matters is the ranking and
the particular value of the utility function does not have any particular meaning apart from it being
employed to rank alternatives.
The following Proposition elaborates more on that point:
Proposition 4 Suppose a function u : X R represents preferences given by , and let f : R R
be a strictly increasing function (i.e. r > r

for r, r

R if and only if f(r) > f(r

)). Then
v : X R dened by v = f u (i.e. v(x) = f(u(x)) for all x X) also represents .
Proof. We need to prove that v = f u represents . In order to do that take any x, y X
and without loss of generality suppose that x y. (Thus, we need to show that v(x) v(y).) We
already know that u(x) u(y) because u represents . Since f is strictly increasing, we must have
v(x) = f(u(x)) f(u(y)) = v(y), which delivers the required conclusion.
Therefore, the ranking given by and represented by u is preserved under a strictly increasing
transformation. This, in turn, means that the utility gure does not mean anything apart from being
a tool for ranking. In order to see that consider the following: Suppose M = 1 and u is an ordinal
utility function with u(100) = 1, u(50) = 1/2 and nally u(0) = 0. Now consider f dened by
Economics 501-2 by Mehmet Barlo: 2 Preferences 11
f(r) = r
1000
for each r R
+
. Because of the above Proposition we now that v = f u also represents
the same preferences and v(100) = 1, v(50) =
_
1
2
_
1000
, and v(0) = 0. Note that the ranking is still
the same. But as it should be obvious, the particular value of u(x) does not mean much because
if it were the following would hold: under v the dierence between v(100) v(50) < v(50) v(0)
implies going from 50 to 100 is more desirable than going from 0 to 50. But under u because that
u(100) u(50) = u(50) u(0) going from 50 to 100 is seen the same as going from 0 to 50. Thus,
u and v do not represent the same cardinal preferences.
2.4 Desirability Assumptions
Now we are going to consider assumptions on which will relate the physical amount of consumption
to preferences.
Denition 12 A preference relation XX is locally non-satiated if for all x X and all
> 0, there exists y X with |y x| < and y ~x.
The important thing to note regarding local non-satiation is that satiated preferences and thick
indierence curves are ruled out. Because with thick indierence curves, any point inside a thick
indierence curve would have a close by neighborhood so that for this agent each point in that
neighborhood is indierent to the others.
Denition 13 is monotone if x y implies x ~y. is weakly monotone if for all x ,=
y X with x y, we have x y. Moreover, is strongly monotone if for all x, y X with
x y and x ,= y, we have x ~y.
Strong monotonicity, a rather strong assumption, says that no matter what ones level of con-
sumption, a bundle with slightly more of anything is preferred to what one has. Whereas, weak
monotonicity says that in such cases the agent should not be worse o.
The following easier Representation Theorem can be found in standard microeconomics text-
books, and hence, is left as an exercise.
Economics 501-2 by Mehmet Barlo: 2 Preferences 12
Proposition 5 Suppose that preferences are continuous complete preorders and monotone. Then
there exists a continuous non-decreasing utility function representing them.
Exercise 5 Prove this Proposition. (Hint: Check Mas-Colell, Whinston, and Green (1995).)
2.5 Convexity of Preferences
These assumptions about convexity of preferences will entail the idea that individuals prefer bundles
in which commodities are fairly evenly distributed to those which are concentrated on a few goods.
That is why an agent with convex preferences should prefer orta sekerli kahve.
We will have three sets of assumptions for convexity, weak convexity, convexity and strong
convexity of preferences. The denitions and results are due to section 7 of chapter 4 of Debreu
(1959).
Denition 14 (Weak Convexity) If x
2
x
1
then x
2
+ (1 )x
1
x
1
for any (0, 1).
Note that this denition allows for thick indierence curves, hence a weakly convex preference
relation might be violating local non-satiation. In fact an agent with weakly convex (and with
convex) preferences might be indierent between all the feasible consumption bundles. But these
preferences would not be strictly convex with our denitions.
Proposition 6 is weakly convex if and only if for every x X, |(o(x) is convex.
Proof. Suppose that is weakly convex, and I aim to prove that for all x X, |(o(x)
is convex. That is, given any x X, for any y, z in |(o(x), y + (1 )z |(o(x), i.e.
y +(1)z x for all [0, 1]. Note that due to weak convexity of , for any y, z with (without
loss of generality) y z, we already have y + (1 )z z. Due to transitivity, we can conclude
that because z |(o(x), y + (1 )z z x, hence, y + (1 )z |(o(x).
Conversely, suppose that for all x X |(o(x) is convex, and I will prove that is weakly
convex. Pick any x, y with y x. Thus, y |(o(x). Also note that x is also trivially in |(o(x).
Economics 501-2 by Mehmet Barlo: 2 Preferences 13
Hence, because that |(o(x) is convex, for any [0, 1], y +(1 )x |(o(x), thus, y +(1
)x x. Hence, is weakly convex.
Denition 15 (Convexity) If x
2
~x
1
then x
2
+ (1 )x
1
~x
1
for any (0, 1).
In other words, is convex if a possible consumption bundle x
2
is strictly preferred to another
x
1
, then their weighted average with arbitrary positive weights in (0, 1) is strictly preferred to x
1
.
Finally the last notion of convexity is the following:
Denition 16 (Strong-Convexity) If x
2
x
1
with x
1
,= x
2
, then x
2
+ (1 )x
1
~x
1
for any
(0, 1).
In other words, is strongly convex if two possible consumption bundles x
1
, x
2
are indierent,
then their weighted average with arbitrary positive weights in (0, 1) is strictly preferred to x
1
.
The following gives us a related notion in terms of functions.
Denition 17 A function f : X R is said to be:
1. concave if for all [0, 1] and every x, y X, f(x + (1 )y) f(x) + (1 )f(y).
2. strictly concave if for all (0, 1) and every x, y X, f(x + (1 )y) > f(x) + (1
)f(y)
3. quasi-concave if for all [0, 1] and every x, y X, f(x +(1 )y) minf(x), f(y).
4. strictly quasi-concave if for all (0, 1) and every x, y X, f(x + (1 )y) >
minf(x), f(y).
Proposition 7 Suppose that is a continuous complete preorder and is represented by a continuous
utility function u. Then:
1. If u is concave, then is weakly convex;
2. If u is strictly concave, then is strictly convex;
Economics 501-2 by Mehmet Barlo: 2 Preferences 14
3. Every concave function is also quasi-concave, but the reverse is not true;
4. Every strictly concave function is also strictly quasi-concave, but the reverse is not true.
5. is weakly convex if and only if u is quasi-concave;
6. is strictly convex if and only if u is strictly quasi-concave;
Proof. The proof will have the following steps:
Step 1. If u is concave, then is weakly convex.
Proof. Let x, y X with x y. Need to show that x + (1 )y y for all [0, 1]. By
hypothesis (i.e. concavity of u), for any [0, 1] we have u(x +(1 )y) u(x) +(1 )u(y),
and because x y, u(x) u(y). Thus, u(x) +(1 )u(y) u(y) +(1 )u(y) = u(y). Thence,
u(x+(1 )y) u(y) implying that (due to representation) x+(1 )y y for all [0, 1].
Step 2. If u is strictly concave, then is strictly convex.
Proof. Let x, y X be such that x y. I need to show that x+(1)y ~y x. By hypothesis,
u(x) = u(y), and because of strict concavity of u, for any (0, 1) it must be that u(x+(1)y) >
u(x) + (1 )u(y) = u(x) = u(y). Thus, due to representation, x + (1 )y ~y x.
Step 3. Every concave function is also quasi-concave, but the reverse is not true.
Proof. Suppose u is concave. Thus, for any [0, 1], and for any x, y X, u(x+(1)y)
u(x) + (1 )u(y) minu(x), u(y) + (1 ) minu(x), u(y) = minu(x), u(y). Hence, u is
quasi-concave.
For a counter example consider X = R
2
+
, and u(x
1
, x
2
) = x
10
1
x
10
2
. This function is clearly not
concave, yet is quasi-concave.
Step 4. Every strictly concave function is also strictly quasi-concave, but the reverse is not true.
Proof. The proof is left to the reader, because it involves the same argument as in the proof of
step 3. Moreover, the example given there also works for this situation.
Step 5. is weakly convex if and only if u is quasi-concave.
Economics 501-2 by Mehmet Barlo: 2 Preferences 15
Proof. Assume that is weakly convex, and without loss of generality pick any x, y X with
x y, thus, minu(x), u(y) = u(y). I need to prove that for all [0, 1], u(x + (1 )y)
minu(x), u(y) = u(y). Because that is weakly convex, for all [0, 1], x+(1)y y, thus,
by representation, u(x + (1 )y) u(y) = minu(x), u(y), implying that u is quasi concave.
For the reverse direction, assume that u is quasi concave, and consider any x, y with x y.
I need to show that for all [0, 1], x + (1 )y y. Because that u is quasi concave, and
x y implies minu(x), u(y) = u(y), we have u(x + (1 )y) minu(x), u(y) = u(y), and by
representation this means that x + (1 )y y.
Step 6. is strictly convex if and only if u is strictly quasi-concave.
Proof. The proof is left to the reader, because it involves the same argument used in the proof
of step 5.
2.6 Relation among the Assumptions on Preferences
First of all it should be mentioned that a preference relation generally is thought of being complete,
reexive and transitive. While preferences not satisfying one or more of these assumptions are
denitely quite interesting, it needs to be noted that dealing with them becomes technically very
dicult. That is why in this section (and in the course) you will not be able to go deeper into that
subject. For the curious reader should want to read the introductions of Ok (2002) and Dubra,
Maccheroni, and Ok (2004).
Second, the assumption of continuity is a very convenient one when it comes to maximization
issues. Because after all continuity is one of the most important ingredients to make sure that an
optimum is obtained. Indeed, we will employ the following Theorem on more than a few instances:
every continuous function on a compact set achieves its maximum. Dealing with continuity requires
tools that are covered in a principles of mathematical analysis course. There are lots of great books
on that subject, and the following two are widely regarded as being two of the best introductory
textbooks into that subject: Rudin (1976), Kolmogorov and Fomin (1970).
Economics 501-2 by Mehmet Barlo: 2 Preferences 16
In this course we can handle the relation between the desirability assumptions that we have
seen so far. Later when we go into convexity and strict convexity, we also will be dealing with their
relation to the existing desirability assumptions.
The student should understand the requirement of these assumptions very well in order to deal
with them in a proper manner.
The assumption of local non-satiation says that every consumption bundle has to possess a
strictly better rival in any of the neighborhoods one can consider. In order to understand an
assumption I always nd it helpful to consider the case when that assumption is not satised.
In this particular case, if a preference relation is not locally non-satiated, then there must be a
consumption bundle which does not have a strictly better rival in any neighborhoods you can
consider. Therefore, if the preferences are given so that they involve a thick indierence curve, then
any consumption bundle residing in the thick indierence curve (that is that point must be strictly
inside, and not on the boundary) would not have a strictly better rival when the neighborhoods
one considers is suciently close to that point.
Similarly, the assumption of monotonicity requires that when an agent gets strictly more of
all the goods, he should get strictly better o. Note that monotonicity does not say about what
happens between two consumption bundles (that is, it does not make any claims about which
one should be chosen) when the agent does not get more from all the goods. Again in order to
understand this assumption consider a preference relation what is not monotone. Then, it must be
the case that there exists at least two consumption bundles one providing strictly more in all the
coordinates than the other; and the agent does not strictly prefer the one that gives strictly more
consumption.
Finally, the assumption of strict monotonicity implies that each good is valuable (which generally
is referred to as each good being good! ). In particular it says that when the agent is confronted with
a choice between a consumption bundle that provides at least as much as another but more in
one coordinate, the agent has to choose the rst one. Thus, if a preference relation is not strictly
Economics 501-2 by Mehmet Barlo: 2 Preferences 17
monotone, then there exists two dierent consumption bundles rst one greater equal to the second,
and the agent does not strictly prefer the rst one over the second.
As a concrete example to see these consider the following preference relation:
Example 1 Let M = 2 and the preferences are given as follows: When x
1
, x
2
, y
1
, y
2
> 3
x y x
1
x
2
> y
1
y
2
.
On the other hand, if the assumption of x
1
, x
2
> 3, and not y
1
, y
2
> 3, then no matter what the
values are x ~y. Finally, when x
1
, x
2
, y
1
, y
2
3, then x y.
In this example I wish to over the assumptions we have seen so far. First of all, for any x, the
denition of the preferences implies that x x, thus, this relation is reexive.
How about completeness? Do we have that for all x, y in R
2
+
, either x y or y x or both. Let
us start by saying yes. Thus, take any x and y. Thus, without loss of generality either one of the
following 3 cases can happen: (1) x
1
, x
2
, y
1
, y
2
> 3; or (2) x
1
, x
2
> 3, and not y
1
, y
2
> 3 (Actually,
symmetrically we also have the case where y
1
, y
2
> 3, and not x
1
, x
2
> 3, but changing x and y
handles this case with the second one we have already written. That is why I wrote without loss of
generality in the previous sentence.); or (3) x
1
, x
2
, y
1
, y
2
3. If case 1 were to happen, then since
x
1
x
2
and y
1
y
2
are both real numbers, it is either x
1
x
2
y
1
y
2
, or x
1
x
2
y
1
y
2
, or both.
Thus, for all x, y satisfying x
1
, x
2
, y
1
, y
2
> 3 these preference relation is complete. Next consider the
case when x, y are such that x
1
, x
2
> 3, and not y
1
, y
2
> 3. In this case we already know that x ~y,
thus, x y and not y x. Hence, when x, y satisfy x
1
, x
2
> 3, and not y
1
, y
2
> 3, the preference
relation is complete. Finally, for the case when x, y are such that x
1
, x
2
, y
1
, y
2
3, we already know
that x y. Thus, x y and y x. Consequently, this preference relation is complete in that case
as well.
When we check for transitivity, we need to consider any x, y, z in R
2
+
such that x y and y z.
From this we need to conclude that no matter what the levels of x, y, z are that x z. Thus, take
any x, y, z with x y and y z. The rest of the argument should follow as demonstrated above
checking cases.
Economics 501-2 by Mehmet Barlo: 2 Preferences 18
This preference relation is continuous. In order to see this take any x and y, and any two
sequences x
n

nN
, and y
n

nN
with x
n
x and y
n
y and x
n
y
n
for all n N. In order
to prove continuity we need to show that x y. Without loss of generality, again there are three
cases as in the previous paragraphs: (1) x
1
, x
2
, y
1
, y
2
> 3; or (2) x
1
, x
2
> 3, and not y
1
, y
2
> 3;
or (3) x
1
, x
2
, y
1
, y
2
3. In the rst case, for n high enough since x
n
and y
n
converges to x and y
respectively, x
n
and y
n
are both such that x
n
1
, x
n
2
, y
n
1
, y
n
2
> 3. Thus, because that x
n
y
n
for n high
enough we know that x
n
1
x
n
2
y
n
1
y
n
2
. Thus in the limit x
1
x
2
y
1
y
2
, enabling the conclusion
that x y. In the second case x
1
, x
2
> 3, and not y
1
, y
2
> 3, again for n high enough x
n
1
, x
n
2
> 3,
and not y
n
1
, y
n
2
> 3. By the denition of the preferences, then, it is easy to see that x ~y. Finally
for the third case when x
1
, x
2
, y
1
, y
2
3, we already know that x y (because x y means x y
and y x). Thus, in all these cases we have x y, enabling the conclusion that these preferences
are continuous.
These preferences are not locally non-satiated: In order to see that consider x = (1, 1). Now
any point y in its close by neighborhoods (with a radius less than or equal to 2 actually) will have
y
1
, y
2
3. Thus, by the denition of these preferences, there is no y in a close by neighborhood of
x = (1, 1) such that y ~x. Hence, these preferences are not locally non-satiated.
Because of the following two propositions these preference are not monotone and strictly mono-
tone.
Proposition 8 Suppose that a preference relation dened on R
M
+
is monotone. Then it is locally
non-satiated.
Proof. Suppose is monotone. Take any x, and consider any close by neighborhoods of
it. It is easy to see that any close by neighborhoods of x will contain y with y
m
> x
m
for all
m = 1, . . . , M. Because of monotonicity we know that y ~x. This, in turn, establishes that is
locally non-satiated, because any close by neighborhood of x contains a y with y ~x.
Proposition 9 Suppose that a preference relation dened on R
M
+
is strictly monotone. Then it
is both locally non-satiated and monotone. But the reverse conclusions do not hold.
Economics 501-2 by Mehmet Barlo: 2 Preferences 19
Proof. Suppose is strictly monotone. Thus, for any x and y with x y, it is true that
x y and x ,= y. Because of strict monotonicity we know that x ~y. Thus, these preferences are
monotone, because for all x and y with x y, we have x ~y. Now the conclusion that any strictly
monotone preference relation has to be locally non-satiated follows from the above and previous
Proposition.
The example to show that monotonicity does not imply strict monotonicity is the Leontie
preferences that we have done in the lectures.
For the rest of the section we assume that preferences are complete preorders. But not necessarily
continuous. We will consider more technicalities and some dicult situations.
Let us start with continuity and monotonicity.
Obviously, weak monotonicity (monotonicity) does not imply monotonicity (strong monotonic-
ity). In order to see the rst relation consider a preference ordering under which all x, y X
are x y. This relation is weakly monotonic, but is neither locally non-satiated nor monotonic
(strongly monotonic). In order to see the second relation consider Leontie preferences, which are
monotone, but not strongly monotone. As noted above, it is clear that strong monotonicity implies
monotonicity: In order to see this consider x, y X with x y. Since the preference ordering is
strongly monotone, and x y implies x y and x ,= y, we have x ~y, thus is monotone. By
the same logic strong monotonicity implies weak monotonicity, which is left to verify by the reader.
Whether or not monotonicity implies weak monotonicity is a nontrivial question which is not that
interesting. In fact, when a preference relation does not satisfy continuity one can nd examples of
preferences where they are not weakly monotone, but monotone. The reader is asked to be aware of
this complication, which could make a nice bonus question. However, if preferences are continuous,
monotonicity implies weak monotonicity. To see this, consider x ,= y, x y. Since preferences are
monotone for each n N, x
n
dened by x
n
= (1 +
1
n
)x x y, is strictly preferred to y. Since
the preference ordering is continuous, x y thus, is weakly monotone.
For convexity assumptions same kind of complications due to continuity might arise: Although
Economics 501-2 by Mehmet Barlo: 2 Preferences 20
at rst reading it looks like convexity implies weak convexity, that is not true in general. However,
that relation is restored if is continuous.
The following example will display that convexity does not imply weak convexity.
Example 2 Consider a preference relation R
2
+
R
2
+
dened by: x ~y if and only if either
x
1
> y
1
, or x
1
= y
1
and x
2
,= y
2
= k, k > 0 (k is xed, say k = 4); x y if and only if either
x
1
= y
1
and x
2
, y
2
,= k, or x
1
= y
1
and x
2
= y
2
= k.
We should note that this relation is not continuous. Moreover, take x, y such that x
1
= y
1
,
and x
2
> k > y
2
, thus, x y. There is a (0, 1) such that x
2
+ (1 )y
2
= k. Hence,
x, y ~x + (1 )y, thus, this is not weakly convex (and not strictly convex). On the other
hand, it is an easy exercise to show that this is convex. Indeed, for any x, y, x ~y can be satised
only when x
1
> y
1
, or x
1
= y
1
but y
2
= k. For any (0, 1), in the rst case x + (1 )y
will be such that x
1
+ (1 )y
1
> y
1
, thus, by the denition of these preferences we must have
x +(1 )y ~y. In the second case, for any (0, 1), we must have x
1
+(1 )y
1
= x
1
= y
1
,
and x
2
+ (1 )y
2
,= k. Thus, x + (1 )y ~y.
Proposition 10 Suppose that is continuous and convex. Then it is weakly convex.
Proof. Suppose not. Let x
1
, x
2
X be such that x
2
~x
1
. It must be shown that A = x
[x
1
, x
2
] [ x
1
~x is empty. Note that A cannot contain a single point, since its complement in
[x
1
, x
2
] is the set x [x
1
, x
2
] : x x
1
closed by the assumption of continuity. Thus, if A were
not empty it would own at least two dierent points x

, x

. However, x
1
~x

implies by convexity
x

~x

, and x
2
~x

implies again by convexity x

~x

. A contradiction would thus obtain.


As stated before weak convexity allows for preferences that violate local non-satiation (even
when continuity is satised). In that sense convexity is stronger, that is convex and continuous
preferences must obey local non-satiation.
Proposition 11 Suppose that is convex and continuous. Then it is locally non-satiated.
Economics 501-2 by Mehmet Barlo: 2 Preferences 21
Proof. Without loss of generality assume that there is not a global satiation point, i.e. for
every x X there is y X such that y ~x. Because, otherwise, trivially local non-satiation cannot
be obtained. Take any x X. We need to show that there exits a x

in the neighborhood of x
such that x

~x. Now pick any y X with y ~x. Thus as shown above all the points in the line
segment (x, y) are strictly preferred to x. Thus, pick x

such that it is (1) close to x; and (2) is in


(x, y).
Similar complications do arise in the interaction between convexity and strong convexity, that
is strong convexity does not necessarily imply convexity. That holds only whenever the preferences
are continuous.
Proposition 12 Suppose that is continuous and strongly convex. Then it is convex.
Proof. We refer the reader to page 61 of Debreu (1959).
The following is an explicitly dened example of which are strongly convex but not convex:
Example 3 Consider a preference relation R
2
+
R
2
+
dened by: x ~y if and only if either
x
1
> y
1
; or x
1
= y
1
and x
2
, y
2
2, and [x
2
1[ < [y
2
1[; or x
1
= y
1
and x
2
, y
2
> 2 and
x
2
> y
2
. That is x y if and only if either x
1
= y
1
and x
2
= y
2
, or x
1
= y
1
and x
2
, y
2
are such that
[x
2
1[ = [y
2
1[.
We should note that this relation is not continuous.
Therefore, in the rst case the hypothesis of strict convexity is not satised at all (that is there
are no x, y with x ,= y and ...). In the second case, the mixture with (0, 1) when x
1
= y
1
would
have [(x
2
+ (1 )y
2
) 1[ < [x
2
1[ = [y
2
1[. Thus this relation is strictly convex.
Moreover, take x, y such that x
1
= y
1
, and x
2
= 3, y
2
= 1, thus, x ~y (therefore x y). Let
= 1/2. Note that z = 1/2x + 1/2y involves z
1
= x
1
= y
1
and z
2
= 2. By the denition of these
preferences it should be clear that y ~z because z
1
= y
1
and [z
2
1[ = 1 > 0 = [y
2
1[. This shows
that these preferences are not convex.
It is worthwhile to see that strong convexity, unlike convexity, does not allow for linear segments
in indierence curves. Moreover, it is intuitively hard to justify strong convexity, but not convexity.
Economics 501-2 by Mehmet Barlo: 2 Preferences 22
Another important relation one has to keep in mind is obtaining strict monotonicity with the
use of monotonicity, strict convexity of preferences which need to be continuous complete preorders.
Proposition 13 Suppose that is continuous complete preorders satisfying monotonicity and
strict convexity. Then is strongly monotone.
Proof. Take any x, y X with x y and x ,= y. I need to show that under the hypothesis of
the Proposition, x ~y in order to obtain strong monotonicity.
Recall that for any x, y X, x y means that x
k
> y
k
for all k = 1, . . . , M.
Clearly, y ~x is not possible. Because that would violate monotonicity, since due to continuity
we would have the existence of > 0 but small, and (1 )y ~x. Then, x y, and y ~x, a
contradiction to monotonicity.
Thus, if x ~y is not true, the only situation to be worried about is that x y. But then, for
any (0, 1), x + (1 )y ~x due to strict convexity of the preferences. Consequently, because
that x y and x ,= y, for any 1 > 0, x (1 )(x + (1 )y). Thus, we reach the
required contradiction to monotonicity, because (by continuity) there exists > 0, strictly positive
but small, such that x (1 )(x + (1 )y) and (1 )(x + (1 )y) ~x.
3 Pareto Optimality and the Core
First we will introduce notions of welfare for pure exchange economies.
In this chapter preferences are assumed to be complete preorders.
For notational purposes for any given pure exchange economy c =
i
, e
i
)
iN
, let the set of
feasible allocations be given by F(c), i.e.
F(c) =
_
x X
n
:

iN
(x
i
e
i
) 0
_
.
3.1 Pareto Optimality and Individual Rationality
The rst notion is individual rationality. In general it says that an allocation is individually rational
if it does not make none of the agents strictly worse o than the initial situation. Otherwise, as
economists we should expect agents to oppose to such kind of allocations.
Denition 18 Given a pure exchange economy c =
i
, e
i
)
iN
, an allocation x X
n
is individ-
ually rational if for all players i = 1, . . . , n we have
x
i

i
e
i
. (3.1)
We denote the set of individually rational allocations of a pure exchange economy c by IR(c).
The reader is encouraged to visualize in the Edgeworth box the representing a pure exchange
economy, the set of feasible and individually rational points.
The second notion of welfare that we are going to analyze is due to an Italian economist, Pareto.
We will present 2 versions of this notion. The regular one, in words, will say that none of the agents
23
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 24
can be made strictly better o without harming some other agent(s). The weak one, on the other
hand, will tell us that the test will involve trying to determine whether or not all the agents can
be made strictly better o. Note that as economists, we would not think that non-Pareto optimal
allocations as nice because by its denition, this allocation is not likely to be observed, because
it is in the interests of all the agents to change it.
Denition 19 Given a pure exchange economy c =
i
, e
i
)
iN
, a feasible allocation x X
n
is
1. Pareto optimal if there is no allocation y F(c) such that for all i N
y
i

i
x
i
,
and for some j N
y
j
~
j
x
j
,
2. weakly Pareto optimal if there is no other allocation y F(c) such that for all i N
y
i
~x
i
.
We denote the set of Pareto optimal and weakly Pareto optimal allocations of a pure exchange
economy c by PO(c) and WPO(c), respectively.
It is clear that if an allocation x X
n
is Pareto optimal, then it is weakly Pareto optimal. But
the reverse is not true. This is to be veried by the reader, as required in the following exercise:
Exercise 6 Show that for all pure exchange economies c, where preference are complete preorders,
PO(c) WPO(c). Prove by giving a concisely specied example that the converse does not hold
even if preferences are assumed to be monotone, continuous and weakly convex.
The following Proposition will be quite useful in the future, and it gives us a sucient condition
when for a given c, we have PO(c) = WPO(c).
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 25
Proposition 14 When preferences are strongly monotone and continuous complete preorders then
PO(c) = WPO(c).
Proof. I will prove that under these hypothesis, for any exchange economy c, WPO(c)
PO(c), because then this would imply PO(c) = WPO(c) (we already know from the above that
PO(c) WPO(c)). That is, I will prove that any weakly Pareto optimal allocation is also Pareto
optimal. In particular, I will show that if an allocation is not Pareto optimal, then it cannot be
weakly Pareto optimal (that is, the proof will be done by counter positive).
Consider an allocation x X
N
and assume that it is not Pareto optimal. Thus, there exists
y F(c) such that y
i

i
x
i
for all i N and y
j
~
j
x
j
for at least 1 j N.
Let S N be given by j N : y
j
~
j
x
j
, i.e. the set of agents who are strictly better o under
y. Note that this set is non-empty, because x is not Pareto optimal.
Consider the following allocation y X
N
: y
j
= (1)y
j
for all j S; and y
i
= y
i
+

#N\S

jS
y
j
for all i N S. Due to continuity, there exists (0, 1] but small, so that y
j
~
j
x
j
for all j S.
Moreover, for any i N S, due to strict monotonicity and transitivity, y
i
~y
i

i
x
i
. Thus, y is
such that for every player i N, y
i
~
i
x
i
.
The only step left to nish to show that x is not weakly Pareto optimal is to prove that y F(c),
i.e. y is feasible. This follows because:

iN
y
i
=

jS
y
j
+

iN\S
y
i
= (1 )

jS
y
j
+

iN\S
y
i
+
#N S
#N S

jS
y
j
=

iN
y
i

iN
e
i
,
where the last inequality follows from y F(c).
Exercise 7 (Bonus) Give an explicitly constructed example of a pure exchange economy with at
least 2 goods with strictly positive total endowments and preferences given by complete preorders,
for which:
1. there are no Pareto optimal allocations;
2. there is only one Pareto optimal allocation.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 26
3.2 The Core
In a pure exchange economy Pareto optimality answers what kind of allocations are plausible when
the group of all of the agents considers possible deviations. On the other hand, individual rationality
considers only single agent deviations, that is such allocations cannot be improved upon by a
deviation done by a single agent.
The consideration of group deviations (not necessarily the whole group) leads us to the notion of
the core. In words, a feasible allocation will be in the core of an economy, if there are no coalitions
with a feasible group deviation opportunity.
Denition 20 The core of an pure exchange economy c =
i
, e
i
)
iN
, denoted by C(c), is the set
of feasible allocations x X
n
such that there is no coalition S 2
N
and y
S
X
|S|
such that
1. y
S
is resource feasible for S, i.e.

iS
(y
S
i
e
i
) 0; and
2. y
S
i
~
i
x
i
for all i S.
For notational convenience let o 2
N
. It is simply the set of all coalitions that can be
formed out of the society given by N. Again for notational convenience for any S o let
F(c [ S) =
_
y
S
X
|S|
:

jS
(y
S
j
e
j
) 0
_
.
Clearly, F(c [ N) = F(c).
Whenever an allocation x X
n
is not in the core, there must be a coalition S o, that we refer
to as the blocking coalition and an allocation y
S
X
|S|
which is feasible for S, and renders strictly
higher utility for all the members in the blocking coalition. In other words, for any x / C(c) with
S o blocking it via the use of allocation y
S
X
|S|
, we say that x can be improved upon by
coalition S o via y
S
X
|S|
.
Therefore, for any x C(c), the coalitions S = N, and S = i for i N, are not blocking.
Consequently, any allocation in the core is both weakly Pareto optimal and individually rational.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 27
However, the reverse direction does not hold, that is there might be individually rational and weakly
Pareto optimal allocations that are not in the core. These are summarized in the following Theorem:
Theorem 2 For all exchange economies c, C(c) IR(c) WPO(c). Moreover, the containment
might be strictly whenever n > 2. In the case of n = 2, if agents preferences are given by continuous
complete preorders satisfying strong monotonicity, then C(c) = IR(c) PO(c).
Exercise 8 Prove this Theorem (also by identifying a concrete, well-dened example).
Whether or not the core of an exchange economy is empty is an important question that we
need to answer. Answering this question directly, on the other hand, would make us go deeper into
cooperative game theory. Therefore, we will deal with that question and show the existence of the
core later in the course.
The following exercise is for the motivated student:
Exercise 9 (Bonus) Give explicitly specied examples of 2 agent 2 good pure exchange economies
with strictly positive total endowments and preferences given by complete preorders, satisfying the
following properties:
1. the core of the economy is empty;
2. even though there exists a Pareto optimal allocation and a core allocation, there are no allo-
cations that are both Pareto optimal and in the core.
3.3 Pareto optimality and the Planners Problem
In this section, we will be describing an alternative way of nding the set of Pareto optimal allo-
cations. Indeed, the set of Pareto optimal allocations, under some assumptions, will be given as
maximizers of a certain program, which is call the Social Planners Problem.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 28
I have to stress that this formulation is used extensively in modern macroeconomics, and provides
not only important insight, but also a very useful tool for us to identify the set of Pareto optimal
allocations.
The result that we wish to obtain is as follows:
1
Consider a pure exchange economy c. Then, an allocation x

X
n
is Pareto optimal if and
only if there exists

(N) (i.e.

i
[0, 1] for all i N and

iN

i
= 1) and x

solves the
social planners problem for c at

, where it is is given by
max
xF(E)

iN

i
u
i
(x
i
). (SP(c [

))
Of course, this result does not hold in general. The assumptions that we need for this result to hold
(the proofs are given later in this section) is as follows: Every agents preferences are represented
by a continuous, strictly increasing, and concave utility function u
i
: X R.
The proof of this result involves the use of an important Theorem (Minkowskis Separating
Hyperplane Theorem, Theorem 3 given below, and it is also used in the proof of the Second Fun-
damental Theorem of Welfare Economics, which we will see later in the course).
Theorem 3 Let C be a convex set of R

, < , and y /

C, where

C denotes the interior of C.


Then there exists R

0 such that for all x C


y x.
Notice that, the requirement for this Proposition is not the convexity of preferences, but the con-
cavity of the utility function representing these preferences. This is rather not a usual requirement,
and is due to the techniques that must be used in the proof of this result.
1
Note that for any nite set E = a
1
, a
2
, . . . , a
|E|
, (E) denotes the simplex formed on E, i.e. the set of all
probability distributions on E. Formally,
(E) =
_
_
_
p R
|E|
: p
k
[0, 1], and
|E|

k=1
p
k
= 1
_
_
_
.
p
k
, k = 1, . . . , [E[, denotes the weight (or probability) that p assigns to the kth component of E.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 29
Given weights = (
i
)
iN
(N), where (N) denotes the set of probability distributions on
the nite set of players N,
i
[0, 1] is often referred to as the bargaining weight of player i in the
Social Planners Problem.
Let us start with the formalities: For the rest of this section, for convenience assume that agents
preferences are continuous complete preorders, satisfying strict monotonicity.
For what follows, the following set will be quite useful:
|
E
= u R
n
: u
i
[u
i
(0), u
i
(x
i
)], i N for some x PO(c). (3.2)
Note that |
E
is well dened because u
i
(0) exists. (ln utility function for example does not satisfy
this assumption.)
Under these assumptions (strict monotonicity in particular), PO(c) is non-empty, thus, so is
|
E
for all c. To see that why PO(c) is non-empty, consider the following: For any given exchange
economy c, we can dene the following allocation x by x
i
= 0 for all i ,= 1 and x
1
=

jN
e
j
. Such
an allocation is clearly Pareto optimal (due to strict monotonicity), since all the goods belong to
agent 1.
In words, for a given economy c, |
E
contains the set of all utility vectors that can be obtained
from all distributions of total endowments. Thus, it can be called utility possibility set. That is,
due to continuity of the utility functions u(F(c)) = |
E
, where
u(F(c)) = u R
n
: there exists x F(c) with u
i
= u
i
(x
i
) for all i N.
Let us have a simple example to understand what that set is: Say there are 2 agents and 2 goods,
and both agents have a utility function given by u
i
(x
i,1
, x
i,2
) = 2 (x
i,1
+ x
i,2
). Say the amount of
endowments is e
1
= (1, 0) and e
2
= (0, 1). Here, any allocation that does not waste any good, i.e.
any x R
22
+
with x
1
+x
2
= (1, 1), is Pareto optimal (note that there is no dierence between weak
Pareto optimality and the regular one, because these preference are strictly monotone). Hence, |
E
is simply given by (u
1
, u
2
) : u
1
+ u
2
4. The reader is asked to either visualize or draw this
particular |
E
.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 30
Now the following Lemma will establish that |
E
is a convex set whenever agents preferences
are weakly convex.
Lemma 2 Suppose that agents preferences are represented by a continuous, strictly increasing
2
and concave utility functions. Then for every exchange economy c, |
E
is a convex subset of R
n
.
Proof. Pick u and u

both in |
E
. I will show that for any [0, 1], u + (1 )u

|
E
.
Let x and x

both in PO(c) be the associated Pareto optimal allocations (x the Pareto optimal
allocation require for u, and x

the one for u

), and x [0, 1]. Since for all i N


u
i
+ (1 )u

i
u
i
(x
i
) + (1 )u
i
(x

i
) u
i
(x
i
+ (1 )x

i
),
for the proof to nish, it suces to show that there exits x PO(c) such that u
i
( x
i
) u
i
(x
i
+(1
)x

i
) for all i N. If the allocation x+(1)x

is Pareto optimal, we are done. If it is not (recall


that due to strict monotonicity, weak Pareto optimality and Pareto optimality coincide), dene x
as follows: x solves max
xF(E)
u
1
( x
1
) subject to u
i
( x
i
) = u
i
(x
i
+ (1 )x

i
) for all i ,= 1. Note that,
the constraint set is non-empty because x + (1 )x

is an element. Because that the constraint


set is compact and the the utilities are continuous, there exists a solution x. Note that player 1 is
obtaining at least as much utility as under x+(1)x

(but maybe more), while the others are not


worse o. Hence, by strict monotonicity x is Pareto optimal. Thus, since u
i
+ (1 )u

i
u
i
( x
i
)
for all i N and x PO(c), u
i
+ (1 )u

i
|
E
.
We are ready for the main result of this section:
Proposition 15 Given a pure exchange economy c where every agents preferences are strictly
monotone, convex, continuous and complete preorders represented by a continuous and concave
utility function u
i
: X R; an allocation x

X
n
is Pareto optimal if and only if there exists

(N) and x

solves the social planners problem for c at

, where it is is given by
max
xF(E)

iN

i
u
i
(x
i
). (SP(c [

))
2
Remember that a function f : R
M
R is strictly increasing if for all x, y R
M
with x y and x ,= y we have
f(x) > f(y).
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 31
Proof. First o, for the converse direction in order to render a proof by counter-positive, x
c suppose that x

/ PO(c). Thus, there is y F(c) with u


i
(y
i
) > u
i
(x

i
) for all i N since
under these assumptions by Proposition 14 we have PO(c) = WPO(c). Thus, for all (N),
u
i
(y
i
) > u
i
(x

i
), rendering the conclusion that there is no

(N) such that x

solves
SP(c [

). Thus we have proven that if x

solves SP(c [

) at some

(N), then x

PO(c).
To prove that if x

PO(c), then there exists a

(N) such that x

solves SP(c [

), we
will employ Lemma 2 and the hyperplane separation result of Theorem 3. Let x

PO(c). Note
that x

|
E
, because if it were, by the dening property of |
E
, there is another Pareto optimal
allocation that provides strictly higher utilities to all the agents, a contradiction to the (weak) Pareto
optimality of x

. Moreover, by Lemma 2, |
E
is convex and non-empty. Thus, by the hyperplane
separating Theorem, Theorem 3, there exists

R
n
with

,= 0 such that

u(x

u for
all u |
E
. Moreover,

0 (i.e.

i
0 for all i N), and in order to provide a continuous
reading the particular proof of this step is given in the footnote.
3
Thus, letting

(N) be
dened by

j
=

j
nishes the proof.
The following notion of welfare has will bring individual rationality and Pareto optimality under
the same umbrella. In fact, the reader should note that these notions are pretty similar as long as
deviations from a given allocation is considered. Given an allocation for a pure exchange economy,
c, considering individual deviations (and the objection of an agent to his portion of that allocation
3
Due to Proposition 4, which says that increasing transformation of utility functions does not matter for repre-
sentation of preferences, we may assume that u
i
(0) > 0 for all i N. If you will, just add a suciently high real
number to the lowest level of utilities given by the utility of consuming 0, and due to strict monotonicity, all other
consumption bundles would give utilities that are higher. If

0 was not the case, that means there exists at least


one agent j N with

j
< 0. Fix such an agent j, and consider u dened by u
j
= u
j
(0) > 0, and u
i
= u
i
(x

i
) for
all i ,= j. By the dening property of |
E
, because that u(x

) u and x

being Pareto optimal, u |


E
. We already
know that by the separating hyperplane Theorem, Theorem 3, that

u(x

u because u |
E
. Doing simple
arithmetics then shows:

u(x

u =

u(x

j
u
j
(0),
hence, because that u
j
(0) > 0, we have

j
0, a clear contradiction to

j
< 0.
Economics 501-2 by Mehmet Barlo: 3 Pareto Optimality and the Core 32
based on the comparison of his welfare under that allocation with that under his initial endowment)
leads us to the notion of individual rationality. On the other hand, if the society as a whole is
considered (and objections to that allocation is solely are based on what the whole society could have
done under the restriction of feasibility) leads us to the notion of Pareto optimality. Moreover, we
could consider these simultaneously, and identify the set of Pareto optimal and individually rational
allocations by the following optimization problem (of course under the hypothesis of Proposition 15)
as follows: An allocation x

X is Pareto optimal and individually rational, if there is

(N)
such that x

solves the following problem:


max

iN

i
u
i
(x
i
) (3.3)
s.t. x F(c) IR(c).
The proof of this observation is omitted since it consists of repeating the same arguments.
4 The Price Mechanism
Walras (1874-1877) formulated a solution to the trade problem that prevails in a pure exchange
economy not by considering a result emerging from a bargaining structure (which is exactly what
was done by Edgeworth (1881)), but instead, looking at a certain mechanism, which we call the
price mechanism.
Every agent in a pure exchange economy observes the same level of prices, and it will be assumed
that none of the agents are able or aware of their ability to inuence these prices. This is called the
price taking behavior assumption.
Therefore, each agent will go to the market with all their endowments (leaving or hiding some
of them at home in order to consume after they return from the market is not allowed), sell them
using the price given to them, and then identify the optimal consumption bundles they can aord.
It is often thought that the price taking behavior assumption is justied when there are many
consumers in the economy. While having some merit, this is not sucient to justify this assump-
tion. Because even when there are many consumers, if the economy has a family called the House
of Saud (and the King of Saudi Arabia, the largest oil producer in the world, comes from that
family), or Bill Gates, or J. P. Morgan inside, and these agents are aware of their power and
can contemplate manipulating the market, the assumption of many agents does not suce. In-
deed, mathematical economists have shown such counterexamples even with innitely many agents.
Therefore, what we need is to have many alike agents, none of which has a critical control on a
critical resource. And to me that assumption is far from being realistic, especially when one sees
that %99 of the wealth of U.S.A. is in the hands of %1 of the population in that country.
33
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 34
Leaving these observations aside, and pretending to believe in the price taking assumption, lead
to the following considerations: The budget set, and the demand.
4.1 Budget Set
In this section we are going to treat the budgetary constraints the agents face, as a set (which also
can be viewed as a set-valued function, i.e. a correspondence; which will be handled in the further
parts of this section). For the demand and competitive equilibrium, the result we are going to derive
in this section that says that the budget correspondence behaves well, is an essential one.
Moreover, note that as far as the analysis of general equilibrium is concerned, the level of
prices does not matter in a pure exchange economy. What matters is the relative prices. That is
established in the following discussion.
Denition 21 The budget set is of player i in a pure exchange economy c with prices p R
M
+
is
B
i
(p, e
i
) = x
i
X : p (x
i
e
i
) 0.
It is an easy exercise to see that for p, p

R
M
+
, with p = p

, where R
++
, B
i
(p, e
i
) =
B
i
(p

, e
i
). In other words, the budget set is homogeneous of degree 0 in prices. That is why in our
analysis we may restrict attention to p , where is the M 1 dimensional simplex (that is
p
k
[0, 1] for all k = 1, . . . , M, and

M
k=1
p
k
= 1), because we can always let =
1

M
k=1
p
k
. Thus, for
all k = 1, . . . , M, (1) p
k
[0, 1] and (2)
M
k=1

k
p
k
= 1; consequently, p .
The following Proposition is essential for future results:
Proposition 16 For every pure exchange economy c, B
i
(p, e
i
) is a non-empty, compact (closed
and bounded) subset of X for every pure exchange economy c with strictly positive prices p 0.
Moreover, it is also convex, i.e. for every x
i
, y
i
B
i
(p, e
i
), x
i
+ (1 )y
i
B
i
(p, e
i
) for every
[0, 1].
Proof. Because that for all p 0, 0 is in B
i
(p, e
i
), B
i
(p, e
i
) is non-empty for every p 0.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 35
Let p 0, and let
K
p
= max
k=1,...,M
p e
i
p
k
.
Then a circle with its center at the origin, 0, and radius given by K
p
contains B
i
(p, e
i
), thus, B
i
(p, e
i
)
is bounded.
It is closed because for any convergent sequence x
n

nN
B
i
(p, e
i
) with lim
n
x
n
= x, it must
be that p x
n
p e
i
implies in the limit, p x p e
i
since the dot product and greater or equal
to operations are both continuous. Thus, x B
i
(p, e
i
), and hence, B
i
(p, e
i
) is closed.
Because B
i
(p, e
i
) is both bounded and closed, it is compact.
Finally, for convexity, assume that x
i
, y
i
B
i
(p, e
i
). Then, p (x
i
+(1 )y
i
) = (p x
i
) +(1
)(p y
i
) p e
i
for every [0, 1], because x
i
, y
i
B
i
(p, e
i
). Thus, (x
i
+ (1 )y
i
) B
i
(p, e
i
),
as was to be shown.
This nishes the proof of Proposition 16
4.1.1 Properties of the Budget Correspondence
In this section we are going to treat the budgetary constraints the agents face, as a set-valued
function (i.e. a correspondence). It is appropriate to remind the reader that for the existence of
equilibrium, the results we are going to derive in this section (saying that the budget correspondence
behaves well) are essential.
First let me introduce the notion of correspondences (set-valued functions) to the reader. The
reader should already be aware that a function is a mapping which maps every point in the domain
to a single point in the range. A correspondence, on the other hand, is a mapping which maps every
point in the domain to a subset of the range. That is, trivially every function is a singleton-valued
correspondence.
As an example consider the following correspondence F that maps X into subsets of Y , and the
notation we are going to abide by is: F : X Y . So, every point x X is mapped to F(x) Y .
Let X = [0, 1] and Y = [0, 1]. As a rst example consider F : X Y dened by F(x) = [0,
1
2
x].
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 36
Thus, for any x [0, 1], F(x) = [0, x/2]. Another example is G : X Y dened by
G(x) =
_

_
x if x ,= 1/2,
0, 1 otherwise.
Note that apart from the exception of x = 1/2, G(x) is given by x. But when x = 1/2, there are
two points in G, 0 and 1.
The following denition will be used for the rest of the course:
Denition 22 Let F be a correspondence from X into Y , i.e. F : X Y . We say that, F is:
1. non-empty valued, if for all x X, F(x) ,= ;
2. closed-valued, if for all x X, F(x) Y is a closed set in Y ;
3. compact-valued, if for all x X, F(x) Y is a compact set in Y ;
4. convex-valued, if for all x X, F(x) Y is a convex set in Y .
The denition of continuity of correspondences requires more care, and will be dealt later in this
section. Moreover, these notes contain an appendix for the motivated and interested reader.
One possible interpretation of the budget is to see it as a correspondence, where the arguments
are p, e. I.e., B(p, e) : R
M
+
X X, where for all (p, e) R
M
+
X, B(p, e) X.
Note that because for any > 0, B
i
(p) = B
i
(p), we can always let =
1

M
k=1
p
k
. Thus, instead
of using a price vector p R
M
+
0, we could use p =
1

M
k=1
p
k
p, and p . Hence, the domain of
the budget correspondence (the set of prices to be considered) can be restricted (without any loss
of generality) to .
Denition 23 For a given economy c = (
i
, e
i
)
iN
, the budget correspondence of player i for a
given endowment e
i
is a set-valued function, B
i
: X X dened by
B
i
(p [ e) = x X : p (x e) 0. (4.1)
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 37
Whenever we do not need to index across the players and keep e X
n
xed we are going to
abuse notation and let the budget correspondence be given by B(p).
The following Theorem is essential, and lists the properties of the budget correspondence: (Recall
that

int(), and thus, when the domain of the budget correspondence is restricted to only
strictly positive prices, there is no loss of generality to consider

as the set of prices that are


allowed.)
Theorem 4 B :

X is non-empty valued, homogeneous of degree 0 in prices


1
, compact-valued,
convex valued, and continuous.
Proof. We are going to present the proof using separate lemmas in order for the result be more
tractable.
Lemma 3 B :

X is non-empty valued.
Proof. The origin, 0 X, is in B(p) for all p , thus the result follows.
Lemma 4 B :

X is homogeneous of degree 0 in prices.


Proof. Follows from the discussion following denition 21.
Lemma 5 B :

X is convex valued.
Proof. Let p

and x, x

B(p). Then for any [0, 1], and x = x + (1 )x

p ( x e) = p((x + (1 )x

) e)
= p ((x e) + (1 )x

e) = p (x e) + (1 )p (x

e) 0,
since p (x e) 0 and p (x

e) 0. Thus, x B(p).
Lemma 6 B :

X is compact valued.
1
I.e. for any > 0, B
i
(p) = B
i
(p).
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 38
Proof. Pick p

, and let (p) = min


k=1,...,M
p
k
, and dene the (p)-constraint simplex

(p)

by

(p)
= p : p
k
(p), for all k = 1, . . . , M. (4.2)
The reader is asked to verify that for any p

,
(p)
is non-empty
2
, convex, and compact.
Now we can dene K
(p)
R
++
by
K
(p)
= 2 max
k=1,...,M
p e
p
k
, (4.3)
since p
k
(p) for all k = 1, . . . , M. Thus, for p

, B(p) /
(p)
, where /
(p)
= x X : x
k

K
(p)
, k, is a compact subset of X,
3
since it is both closed and bounded. Thus, the remaining
task to complete to proof is to show that B(p) /
(p)
is closed. Take any sequence x
n

n
B(p),
with x
n
x. Thus, for all n N, p (x
n
e) 0. Since all the operations involved (the dot
product, and the operation) are continuous, p (x e) 0, thus, x B(p), showing that B(p)
is closed. As p

was arbitrary, result follows.


Finally, continuity of the budget correspondence follows from Lemmas 7 and 8 which will be
presented after the introduction and discussion of continuity of correspondences.
Continuity of the Budget Correspondence
Because that our attention is restricted to nite dimensional Euclidian spaces, namely R
M
, the
continuity of functions is easy: We know from basic mathematical analysis that a function f :
R
M
R
K
is continuous if and only if: (1) for every sequence x
n
R
M
with x
n
x (i.e. for any
> 0, there exists N

N such that for all n > N

we have |x
n
x| < ), we have f(x
n
) f(x);
OR (2) for any open set E R
K
, f
1
(E) = x R
M
: f(x) E R
M
is open in R
M
.
Let us analyze if we can modify these denitions (even with being restricted to nite dimensional
Euclidian spaces) and obtain a nice working denition for the continuity of correspondences.
2
Because for any p

, p

dened by p

k
=
1
M
is in
(p)
.
3
The curious reader should prove this step by using Theorem 26 from appendix 6.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 39
In fact, let us start with the rst denition, the one that involves sequences in the domain of
the correspondence. That denition would then read: F : R
M
R
K
is continuous if and only if for
every sequence x
n
R
M
with x
n
x, we have F(x
n
) F(x). The problem with that denition
is that the convergence of F(x
n
) to F(x) is unfortunately not trivial at all. This is because, we are
talking about two sets. In fact, F(x
n
) F(x), for F(x
n
), F(x) R
K
means that for any > 0,
there exists N

N such that for all n > N

we have d(F(x
n
), F(x)) < . So we need to have a
distance notion on subsets of R
K
, because without it, F(x
n
) F(x) does not make any meaning.
In mathematics, when attention is restricted to nite dimensional Euclidian spaces, to my knowl-
edge the only distance notion used is the Hausdor distance (metric), which will be dened below:
(The formalities presented below are taken from Berge (1963).) Let A and B be two non-empty
sets in R
K
, and write
(A, B) = sup
xA
inf
yB
d(x, y),
(B, A) = sup
yB
inf
xA
d(x, y).
The numerical function dened by
(A, B) = max(A, B), (B, A)
is called a Hausdor Metric.
As an example consider the following two sets: Let A = [0, 1] and B = [3, 5]. Then in order
to nd (A, B), consider any x A, and solve inf
yB
d(x, y) for that given x. It should be clear
that no matter what x A is, the closest member of B is 3, thus, inf
yB
d(x, y) = 3 x. Then,
sup
xA
3 x is clearly given by 3 when x = 0 A. Thus, (A, B) = 3. Now, in order to compute
(B, A), x any y B, and and solve inf
xA
d(x, y) for that given x. Clearly, the closest member
of A for any y B is 1, thus, inf
xA
d(x, y) = y 1. Hence, sup
yB
inf
xA
d(x, y) = sup
yB
(y 1),
thus is equal to 4, and the supremum is given by y = 5. Thus, (B, A) = 4. Thence, (A, B) = 4.
This metric, while being very useful in many situations, is not well suited for our analysis because
of the following reason: This metric can only be used on compact sets (not even closed ones; to see
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 40
this let A = 0 and B = R, it can be easily shown that (A, B) does not exists even though both
of these sets are closed): To see this, let A = (0, 1) and B = [0, 1]. Clearly, because that A B we
have (A, B) = 0. Moreover, for (B, A) x any y B. If y A, clearly we have inf
xA
d(x, y) = 0.
If y / A, then y is equal to either 1 or 0. In any case, inf
xA
d(x, y) = 0. Thus, (B, A) = 0. This is
a contradiction to being a metric on all (bounded) subsets of R
K
, because A ,= B but (A, B) = 0.
The following presented and proven in Lemma 9 of appendix 7 makes this observation precise:
as dened above is a metric for the family of non-empty and closed subsets of a compact metric
space X.
Therefore, because that the budget correspondence is a mapping from into X, it is not
necessarily compact valued on . To see this, consider 2 goods with prices p = (0, 1), and it is
easy to see that the resulting budget set (therefore, the budget correspondence for these prices) is
not compact. Thus, this distance notion as it is is not well suited for our purposes. But, it is a
very useful way of dening continuity of correspondences whenever the correspondence is question
is compact-valued, because then the Hausdor distance is well dened.
Next, consider a modication to the second denition of continuous functions: F : R
M
R
K
is continuous if and only if for any open set E R
K
, F
1
(E) R
M
is open in R
M
. That again
sounds promising, yet we have to ask ourselves about the meaning of F
1
(E). When attention
is restricted to functions (singleton-valued correspondences) its meaning is clear: F
1
(E) = x
R
M
: F(x) E. But when F(x) is not a single point but a set, would the following be sucient?
F
1
(E) = x R
M
: F(x) E. To my knowledge, the answer is armative, even though I have
not personally seen a proof. That is why, it would be great if any one of you could provide a clean
proof of this (the details about what I mean by being sucient, will be given later).
In the literature, there are two notions of continuity: upper-hemi continuity and lower-hemi
continuity, and they involve a denition of continuity via open sets. More technical details can be
found in the appendix 7.
First, the upper-hemi continuity:
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 41
Figure 4.1: The Graph of F
Denition 24 Let F : R
M
R
K
, M, K < , be a correspondence mapping R
M
into R
K
. We
say that F is upper-hemi continuous if for all x
0
R
M
, for each open set G R
K
containing
F(x
0
) R
K
, there exists a neighborhood U
x
0
R
M
such that
x U(x
0
) F(x) G.
Consider the following example (remember that I think that in order to understand a notion, it is
best to see a counter example rst): Let M = K = 1 and consider a correspondence F : [0, 1] [0, 2]
dened by:
F(x) =
_

_
[1/2, 1] if x < 1/2,
[3/4, 1] otherwise.
The graph of F is given in gure 4.1. To see why this correspondence is not upper-hemi continuous,
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 42
consider x
0
= 1/2, and G = (0.70, 1.1) and note that F(1/2) = [3/4, 1] G. Now, for any
neighborhood of x
0
= 1/2, denoted by U(x
0
), it must be the case that there are x

U(x
0
) with
x

< 1/2. Thus, for all such x

we have F(x

) = [1/2, 1], hence, it is not the case that F(x

) G.
Moreover, the reader should note that even though F is not upper-hemi continuous, it is compact-
valued.
An important result, which is not that dicult to establish (and the reader is asked to do it) is:
Proposition 17 Every non-empty valued, singleton-valued, upper-semi continuous correspondence
is a continuous function.
Next, I dene lower-hemi continuity:
Denition 25 Let F : R
M
R
K
, M, K < , be a correspondence mapping R
M
into R
K
. We
say that F is lower-hemi continuous if for all x
0
R
M
, for each open set G R
K
with
G F(x
0
) ,= , there exists a neighborhood U
x
0
R
M
such that
x U(x
0
) F(x) G ,= .
Again, let me consider the following counter-example: Let M = K = 1 and consider a corre-
spondence F : [0, 1] [0, 2] dened by:
F(x) =
_

_
3/4 if x < 1/2,
[1/2, 1] otherwise.
The graph of F is given in gure 4.2. To see why this correspondence is not lower-hemi continuous,
consider x
0
= 1/2, and G = (0.80, 0.90) and note that F(1/2) = [3/4, 1] G, thus, GF(1/2) ,= .
Now, for any neighborhood of x
0
= 1/2, denoted by U(x
0
), it must be the case that there are
x

U(x
0
) with x

< 1/2. Thus, for all such x

we have F(x

) = 3/4, hence, it is not the case that


for all x U(x
0
) (particularly, take x

), we have F(x) G ,= , because indeed F(x

) G = .
Moreover, the reader should again note that even though F is not lower-hemi continuous, it is
compact-valued.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 43
Figure 4.2: The Graph of F
Whether or not the rst (second) example is lower-hemi continuous (upper-hemi continuous) is
left for the reader as an exercise.
Denition 26 Let F : R
M
R
K
, M, K < , be a correspondence mapping R
M
into R
K
. We say
that F is continuous if it is both upper-hemi continuous and lower-hemi continuous.
The following is a bonus question for the motivated and interested reader. By doing this question
in a nice and concise fashion, you would be helping me, because I would be including that proof
into these lecture notes. This question goes back to the suciency discussion I had when I was
talking about dening continuity of correspondences directly with open sets:
Exercise 10 (Bonus) Is the following true or false? F : R
M
R
K
, M, K < , is continuous if
and only for any open set E R
K
, F
1
(E) = x R
M
: F(x) E R
M
is open in R
M
.
These denitions for the continuity of correspondences are nice, yet, it is always much more useful
to obtain equivalent denitions in terms of sequences. This is because (at least for me), working
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 44
with sequences are often more appealing. Moreover, because that our analysis is restricted to nite
dimensional Euclidian spaces (therefore, the rst axiom of countability holds, because open balls
constitute a countable base for the usual topology on nite dimensional Euclidian spaces), obtaining
alternative denitions for these continuity notions is not that dicult.
I should remind the reader that all the required technicalities can be found in appendix 7, which
assumes that the reader has working knowledge of basic topological and metric spaces. Here, I will
be trying to get our job done without going into too much technicalities.
The following result renders an equivalent denition of upper-hemi continuity:
Proposition 18 Let F : X Y be compact-valued, and X R
M
and Y R
K
with M and K
both nite integers. F is upper-hemi continuous at x if and only if for every sequence x
n

n
X
and y
n
Y with x
n
x and y
n
F(x
n
), there is a convergent subsequence of y
n
with limit in
F(x).
In order to understand this Proposition, let us consider the following example: F : [0, 1]
R dened by F(x) = 1/x for all x > 0; and F(0) = 0. Note that this is a singleton-valued
correspondence, i.e. a function. I know that this function is not upper-hemi continuous at x
0
= 0
(you might check it for yourselves), and aim to show it using this Proposition.
Consider x
0
= 0. Then for any sequence x
n

n
[0, 1] with x
n
x, any y
n

n
such that
y
n
F(x
n
) must be such that y
n
= 1/x
n
(because our correspondence is singleton-valued). Thus,
it cannot be the case that any subsequence of y
n
converges to 0 = F(x).
This denition, while being informative, is still not the best for our purposes. Therefore, we
continue in the search of a more user friendly alternative.
At this point it is appropriate to point out that we now can connect these notions of continuity
with the Hausdor continuity, whenever F can be assumed to be compact valued. In fact, we have
an equivalence result, that is a compact valued correspondence is continuous if and only if it is
Hausdor continuous. We refer the motivated reader to Theorem 32 of appendix 7 for full details
of this result.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 45
Figure 4.3: The Graph of F is not closed
If Y , the range of F, can be assumed to be compact, we have a much nicer alternative (equivalent)
denition of upper-hemi continuity. Indeed, many economists (even some formal ones) do not know
much about the distinction of these two denitions, and pretend that the following is the most
primitive one. But before that, I would have to dene the notion of closed graphness.
Denition 27 Let F : X Y , and X R
M
and Y R
K
with M and K both nite integers. We
say that F has a closed graph if Graph(F) = (x, y) XY : y F(x) is closed. Alternatively,
F has a closed graph if for all x X and all x
n

nN
X with x
n
x, and every y
n

nN
Y
with y
n
y, and y
n
F(x
n
) for all n N, implies y F(x).
This denition is very simple, because all it asks is whether or not the graph of F is closed. To
see this real fast, consider the correspondence that we dealt with before after the introduction of
the upper-hemi continuity, the one depicted in gure 4.1. Let the sequences be as given in gure
4.3, i.e. x
n

n
such that x
n
1/2, and y
n

n
with y
n
F(x
n
) for all n and y
n
0.60. Clearly,
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 46
even though y
n
F(x
n
) for all n and x
n
x and y
n
y, we do not have y F(x), thus, the
graph of F is not closed.
The following Proposition identies situations in which we can use closed-graphness as an alter-
native denition of upper-hemi continuity:
Proposition 19 Let F : X Y be compact-valued, and X R
M
and Y R
K
with M and K
both nite integers, and Y compact. Then F is upper-hemi continuous if and only if it has a closed
graph.
Note that compactness of Y is essential for this result. The reader could see this by considering
an example that we already have worked out: F : [0, 1] R dened by F(x) = 1/x for all x > 0;
and F(0) = 0. Note that this is a singleton-valued correspondence, i.e. a function. We already have
worked out that this correspondence is not upper-hemi continuous. But the reader should be able
to easily see that it is both compact-valued and has a closed graph. The only hypothesis missing is
the compactness of the range of this correspondence.
With these technical (working) knowledge, we are ready to prove that the budget correspondence
is upper-hemi continuous on

, i.e. on the domain given by strictly positive prices:


Lemma 7 B :

X is upper-hemi continuous.
Proof. First let us demonstrate that the budget correspondence has a closed graph. Pick any
p

and take any p


n

n
with p
n
p. Since p
n
converges to p, there exists M such that for
all n > M, p
n

. Therefore, without loss of generality we may assume that p


n

, and for
any x
n
X with x
n
B(p
n
) there is x X with x
n
x. The reason is that: if necessary we
may take subsequences (and re-index) x
n
so that x
n
x for some x X, because any arbitrary
sequence of x
n

n
, lives in a compact set / such that /
(p
n
)
/, because p
n

n
lives in

, implying
we have the existence of such a / and , where is associated with / such that is the minimal
number such that /

is contained in /. Thus x
n

n
possesses a convergent subsequence with limit
point x. Moreover, note that / contains /
(p
n
)
for all n, since (p
n
) for all n > M. Due to
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 47
the continuity of the dot product operation and the operation involved in B, x B(p). Since
p

was arbitrary this establishes that the budget correspondence has a closed graph. However
the proof is not over.
To conclude that the budget correspondence is upper-hemi continuous we need to show the
following statement: If B :

X has a closed graph and is compact and non-empty valued, then


B is upper-hemi continuous.
Using Proposition 18 we know that in order to prove upper-hemi continuity of the budget
correspondence it is enough to show that for all p

and for all p


n

with p
n
p, and for
all x
n

n
X with x
n
B(p
n
), there exists a convergent subsequence of x
n

n
with limit point
in B(p). To proceed, suppose not. That is, there exists p

and p
n

with p
n
p, and
x
n

n
X with x
n
B(p
n
), such that x
n

n
does not possess any subsequence with limit point in
B(p). Therefore, we already know that x
n
must have a convergent subsequence x
n
m

m
with limit
point x, since x
n

n
/, a compact set as was shown above. But by hypothesis x / B(p), i.e.
p (x e) > 0. However, this is an impossibility since for p
n
m
(x
n
m
e) 0 for all n
m
N.
Now, let us turn our attention to lower-hemi continuity. In this case, obtaining an alternative
denition with sequences is actually easier. In fact, because that the nite Euclidian spaces satisfy
the rst axiom of countability, we right away have:
Theorem 5 Let F : X Y , and X R
M
and Y R
K
with M and K both nite integers. Then
F is lower-hemi continuous if for all x X and all x
n

nN
X that converges to x, and every
y F(x), there exist y
n

nN
Y with y
n
F(x
n
) for all n N and y
n
y.
The readers are encouraged to apply this denition (with sequences) to the examples that were
discussed above.
Now we are ready for the next and nal result of this chapter:
Lemma 8 B :

X is lower-hemi continuous.
Proof. Due to results provided in appendix 7, rst axiom of countability is satised in this
setting. Thus we may use the following denition:
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 48
Figure 4.4: The Budget Correspondence is lower-hemi continuous
Denition 28 B :

X is lower-hemi continuous if for all p

and all p
n

nN
that
converges to p, and every x B(p), there exist x
n

nN
X with x
n
B(p
n
) for all n N and
x
n
x.
There are two cases to consider: (1) p (x e) < 0, and (2) p (x e) 0.
For the rst case, construct the sequence of x
n

n
by
x
n
=
_

_
x if n > M,
e otherwise .
Clearly, x
n
x and x
n
B(p
n
) by construction.
For the second case (in this case, the following gure, gure 4.3 might help ) let x
n

n
be dened
by
x
n
=
_

_
x if p
n
(x e) 0,
w
n
x otherwise ,
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 49
where w
n
[0, 1] solves maxw : p
n
(wx
n
e) 0. Since [0, 1] is compact and the maximized
function is continuous, x
n
is well dened. Note that by construction, x
n
x and x
n
B(p
n
).
The following exercises (which are some of the ones given in the appendix 7) would make the
reader more familiar with these continuity notions:
Exercise 11 Let F : X Y , and X R
M
and Y R
K
with M and K both nite integers, and
Y compact. Analyze the relations between the following conditions. That is, if there is any prove the
implication relations between them. Otherwise, render explicitly dened examples. (1) upper-hemi
continuity, (2) lower-hemi continuity, (3) closed graph, and (4) compact valued.
Exercise 12 Give explicitly dened correspondences from [0, 1] into [0, 1], which satisfy all but one
of the following assumptions: (You are asked to go over 5 cases!) (1) upper-hemi continuous, (2)
lower-hemi continuous, (3) convex valued, (4) nonempty valued, (5) compact valued. (Can you
have a upper hemi continuous correspondence be not compact valued? Prove your answer, or give
an example.)
Exercise 13 Analyze the following correspondences and determine whether or not they are (1)
upper-hemi continuous (2) lower-hemi continuous and (3) closed graph; and whether or not Propo-
sition 19 holds in those examples. Moreover for the singleton-valued correspondences, indicated by
(

), determine if they are (1) upper-semi continuous or (2) lower-semi continuous or obeying none
of those continuity denitions:
1. F : [0, 100] R
+
dened by F(x) = n for x [n, n + 1], and n 0, . . . , 100.
2. (

) F : [0, 100] R
+
dened by F(x) = n for x [n, n + 1), and n 0, . . . , 100.
3. (

) F : [0, 100] R
+
dened by F(x) = n for x (n, n + 1], and n 0, . . . , 100.
4. F : [0, 100] R
+
dened by F(x) = n for x (n, n + 1), and n 0, . . . , 100.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 50
5. (

) F : R
+
R
+
dened by
F(x) =
_

_
1/x if x > 0
0 if x = 0
6. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
[x +
1
4
, x +
3
4
] if x = 5
7. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
(x +
1
4
, x +
3
4
) if x = 5
8. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
[x 1, x + 2] if x = 5
9. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
(x 1, x + 2) if x = 5
4.2 Demand
In this section we are going to dene and analyze important properties of demand. First we are
going to dene demand without assuming representation, but with preferences.
Denition 29 The (individual) demand of agent i, at prices p 0, is
T
i
(p) = x
i
B
i
(p, e
i
) : x
i

i
y
i
, for all y
i
B
i
(p, e
i
). (4.4)
For those of you who know game theory, note that the demand is nothing but an agents best
response to a given price vector.
As we have done with the budget, we can view the demand as a correspondence (set-valued
function), and obtain the following:
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 51
Denition 30 The (individual) demand correspondence, T :

X is dened by
T(p) = x B(p) : x y, for all y B(p). (4.5)
Now the rst result on the existence of demand:
Theorem 6 Suppose that are upper semi continuous complete preorders. Then T :

X is
non-empty valued.
Proof. Let p

, and take any y

B(p), and dene


T

(p) = x X : x y

B(p),
and note that T

(p) ,= since y

(p). Since is upper semi continuous


4
, and T

(p) is the
intersection of a closed set, and a compact set B(p), thus T

(p) is compact.
Before going further, I have to show you an important Theorem from mathematical analysis:
The formal statement of that Theorem is given in the appendix 6: We say that a system of subsets
A

( not necessarily countable) of a set X is said to satisfy the nite intersection property
if every nite intersection
n
k=1
A

k
is non-empty. The Theorem, Theorem 24, is: If a system of
compact subsets A

of a set X satises the nite intersection property, then

is non-
empty.
Let k N be arbitrary and take any y

k
j=1
from X such that y

1
y

2
. . . y

k
(re-index
if necessary). Note that T

1
T

2
. . . T

k
and all are non-empty, thus
k
j=1
T

j
= T

1
,= .
Thus, T

is a system of compact subsets of a set X satisfying the nite intersection property


given in denition 38. Thus Theorem 24 applies and hence letting the index set for be given by
= B(p) delivers:

x B(p) : x y

= x B(p) : x y, for all y B(p)


= T(p).
4
Thus, |(o(y

) closed.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 52
Thus, due to p

being arbitrary, T :

X is non-empty valued.
Note that the result above works for any e X even if e = 0, because then trivially for any
p

, T(p) = e.
Note that when preferences are continuous complete preorders, then we have a utility represen-
tation, and therefore,
T
i
(p) = arg max
x
i
B
i
(p,e
i
)
u
i
(x
i
).
In words, it says that the agent i is maximizing his utility, subject to the constraint ensuring that
the bundle he wishes to consume resides in his budget set. And the demand is the set consumption
bundles that solve this problem. Moreover, this problem is often referred to as the consumers
problem.
The reader should be aware that he is expected to be able to identify the demand very eciently
in various types of examples.
The below, is an easier result stating and proving the existence of demand under the assumption
that the preferences are continuous complete preorders.
Proposition 20 Suppose that
i
are continuous complete preorders. Then T
i
(p) ,= for every
p 0.
Proof. Under these hypothesis, the budget set is compact, and the prefences are represented
by a continuous real valued function. Therefore,
T
i
(p) = arg max
x
i
B
i
(p,e
i
)
u
i
(x
i
).
Hence, we are maximizing a continuous function on a compact set, hence, by the famous Weierstrass
Theorem, there exists a solution.
Next we will present a stronger result which requires some structure on the preferences. We
will need to assume that are continuous complete preorders, and we will employ the Theorem of
Maximum, Theorem 33, which is due to Berge (1963).
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 53
Theorem 7 Suppose that are continuous complete preorders. Then T :

X is non-empty
valued, compact valued and is upper hemi continuous.
Proof. Let us remind the reader of the Theorem of Maximum in order to make this proof more
readable (and not ip that many pages).
Berges Theorem of Maximum, 1959. Let X R
m
and Y R
k
, let f : Y R be a
continuous function, and let F : X Y be a non-empty-valued, compact-valued and continuous
correspondence. Then the function h : X R dened by h(x) = maxf(y) [ y F(x) is
continuous, and the correspondence dened by (x) = y F(x) [ f(y) = h(x) is non-empty-
valued, compact-valued, and upper-hemi-continuous correspondence.
The proof will be just relabeling our setup in order to use the Theorem of Maximum.
Note that since is a continuous complete preorder, by Theorem 1 there exists a continuous
real valued function u : X R representing . Hence we see that u = f,

= X, X = Y and
nally B = F in terms of the notation given in the Theorem of Maximum. Therefore, the function
in this setting is x B(p) [ u(x) = maxu(x

) [ x

B(p), consequently, is nothing but T(p).


Since X = R
M
+
is a metric space with the Euclidian metric, and B :

X is a non-empty-valued,
compact-valued and continuous correspondence by Theorem 4, Theorem of Maximum applies, hence
the result follows.
The following Proposition will summarize additional properties of demand the demand corre-
spondence:
Proposition 21 For given e 0 and e ,= 0, let be a continuous complete preorder which satises
local-non-satiation:
1. Walras Law holds: for all p

and all x(p) T(p), we have


p (x(p) e) = 0.
2. The indirect utility function v :

R dened by v(p) = u(x(p)) for p

and
x(p) T(p) is continuous.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 54
3. T :

X is homogeneous of degree 0 in prices.


5
4. If is weakly convex, then T :

X is convex valued.
5. If is strongly convex, then T :

X is singleton-valued (i.e. is a function), and is


continuous.
Proof. The proof is done in 5 steps:
1. Let p

and suppose that for some x B(p). Thus for < p (e x), and > 0, there
exists x

B(p) with x

~x(p) since is locally non-satiated. Consequently, x / T(p).


2. The indirect utility correspondence is v :

R dened by v(p) = r R : r =
u(x(p)), x(p) T(p), is singleton valued since for all x(p), x

(p) T(p), u(x(p)) = u(x

(p)) for all


p

, because otherwise it would contradict to the denition of T :

X. Moreover, the above


given denition implies that by the Theorem of Maximum, v :

R is a continuous function.
3. Since for all p

B(p) = x X : p (x e) 0 = x X : tp (x e) 0 = B(tp),
for t > 0, it follows B :

X is homogenious of degree 0. Thus, so is T :

X.
4. Suppose that is weakly convex, thus, by Proposition 6, for all x(p), x

(p) T(p), because


that |(o(x(p)) = |(o(x

(p)) (note that this is due to u(x(p)) = u(x

(p))) is convex, x = x(p) +


(1)x

(p), where [0, 1], is such that x x(p) x

(p). Moreover, since by Theorem 23 B :


X is convex valued, x B(p), therefore x T(p).
5. Suppose that is strongly convex, and let p

. If there were x(p), x

(p) T(p) with


x(p) ,= x

(p), we already know that x(p) x

(p), thus, by the strong convexity of , for any


(0, 1), x(p) + (1 )x

(p) ~x(p) x

(p). This contradicts to x(p), x

(p) T(p) since by


Theorem 23 B :

X is convex valued.
Moreover, since we already know that by Theorem 7 T is upper hemi continuous. Because that
it is also singleton valued, due to Theorem 31 from appendix 7 (or Proposition 17 for the previous
5
A correspondence F : X Y is homogeneous of degree > 0 if F(tx) = t

F(x).
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 55
section) result follows.
This nishes the proof.
4.2.1 Derivation of Demand
In this section, I will be considering some preferences, and obtain the demand correspondence.
Let us start with an easy case:
Demand with Cobb-Douglas Preferences
Cobb-Douglas preferences are widely used in applied microeconomics, for reasons that you will see.
These preferences are given by: For any consumption bundle x R
M
+
, u : X R is dened by:
u(x) = x

1
1
x

2
2
. . . x

M
M
,
where = (
k
)
M
k=1
(M). That is,
k
[0, 1], and

M
k=1

k
= 1.
Note that the restriction of is not a serious one, because we can take a monotone
transformation and obtain in the simplex. Formally, for any with
k
0 and ,= 0, let

be
dened by

k
=

k

M
k

=1

k

. It is then easy to see that

(M).
Let e
i
0 and e
i
,= 0 (note that when e
i
= 0 the demand trivially equals 0 for any price vector),
and assume that p

. Then in order to identify the demand correspondence (actually by already


know that the demand will be a function, because of Proposition 21, since these preferences are
strictly convex and strictly monotone), one must solve the following program:
max x

1
1
. . . x

M
M
subject to
p (x e
i
) 0.
Because that our utility function is a strictly concave utility function, and is also dierentiable,
we may resort to the LaGrangian method to solve this problem:
L = x

1
1
. . . x

M
M
(p (x e
i
)) .
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 56
Then, because that
L
x
k
=
k
u
x
k
p
k
, by equating it to 0 (now xs are given stars because from
now one they are the optimal values), and also doing the same for l ,= k,

k
u
x

l
u
x

l
=
p
k
p
l

l
x

l
x

k
=
p
k
p
l
,
thence,
x

k
=

k

1
p
1
p
k
x

1
for all k = 1, . . . , M. The applied microeconomists often call the rst good as we have done here,
the numeraire good.
It is a big mistake to take the derivative of L with respect to , because after all, is not a
choice of the consumer. This mistake is very common, but still, a mistake!
What needs to be done is to apply Walras law: We know that these preferences are locally non-
satiated (indeed, they are actually strictly monotone for all x 0), thus we must have p x

= p e
i
.
Hence,
p e
i
= p x

=
M

k=1
p
k
x

k
=

1
p
k
p
1
p
k
x

1
=
p
1

1
x

k
=
p
1

1
x

1
,
thus, it can be easily shown that
x

k
=

k
p
k
(p e
i
) .
In other words, the demand correspondence is a function and given by (suppressing e
i
for obvious
reasons)
T
i
(p) =
_

1
p
1
(p e
i
) , . . . ,

M
p
M
(p e
i
)
_
. (4.6)
Comparative Statics with Cobb-Douglas and Fixed Income
Although we will not be going into the details of comparative statics, I still nd it helpful to talk
about them in order for you to understand what the aim of that analysis is. And, the demand
function of the Cobb-Douglas given in equation 4.6, is a very good example of that.
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 57
When agents have a xed income, money (whatever it is), at this level of the course, the student
should be able to show right away that, the Cobb-Douglas demand is given by:
T
i
(p) =
_

1
p
1
W, . . . ,

M
p
M
W
_
, (4.7)
where W 0 denotes the monetary endowment that the agent has. (In pure exchange economies,
W = p e
i
.)
The initial question that one could ask about the behavior of demand is about what happens
to the demand to good k when the price of good k increases. Indeed, when the demand for good
k decreases, then we say that the law of demand (downward sloping feature) holds. With Cobb-
Douglas utilities, this requires checking
D
i,k
(p)
p
k
which is given by

k
p
k
2
W < 0, whenever W,
k
> 0.
Thus, the law of demand holds with Cobb-Douglas utilities. The simplicity of agents behavior with
changes in prices is one of the important reasons why Cobb-Douglas preferences are widely used.
Indeed, the reader is asked to interpret the term from this perspective.
However, this simplicity might not be the case with other preferences. Indeed, when the demand
of good k increases when the price of good k increases (keeping W constant), we call such a good
a Gien good. Below you are asked to nd a clear example of such a situation:
Exercise 14 Consider an agent having W > 0, and 2 goods. Specify a preference relation, which is
a continuous locally non-satiated and complete preorder, and prove that with these preferences one
of the goods is a Gien good.
Another immediate question that can be considered concerns agents behavior with changes in
W. We say that a good k is a normal good, when its consumption increases as W increases. It is
an inferior good, when the demand to good k decreases when W increases.
In the case of Cobb-Douglas, this exercise is very simple:
D
i,k
W
=

k
p
k
> 0 whenever
k
> 0. Note
that the agents behavior to changes in W does not depend on the level of W. That is why the
Cobb-Douglas case this popular, because of the lack of income eects.
Again, this might not be the case with other utility functions, and below you are asked to
identify an inferior good:
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 58
Exercise 15 Consider an agent having W > 0, and 2 goods. Specify a preference relation, which is
a continuous locally non-satiated and complete preorder, and prove that with these preferences one
of the goods is an inferior good.
Another interesting question is about how the demand of one good chances with changes in the
price of another good. Indeed, in elementary microeconomics, if this eect is positive (that is one
buys more of good k when the price of good l increases), these two goods are called substitutes.
On the other hand, if this eect is negative (i.e. one buys less of good k when the price of good l
increases), then these two goods are called complements.
With Cobb-Douglas preferences, this question again has a very simple answer:
D
i,k
p
l
= 0, for
all l ,= k. Thus, with Cobb-Douglas the change in the price of another good, does not aect the
demand of the original good.
Note that when we consider pure exchange economies, this is not true, because, W = p e
i
.
Thus,
D
i,k
p
l
=

k
p
k
e
i,l
. However, the aect of the change of the price of good l, still gives a very
simple eect to the demand to good k, and this eect is constant in p
l
.
Further Examples
In this section, I am assigning some interesting exercises to the reader.
Exercise 16 (Quadratic Utilities) Consider 2 goods, e
i
> 0, p

and the following utility


function:
u(x) = x
1
+ ln x
2
.
Identify the demand correspondence. Analyze how the demand of each of these goods change with
changes in the prices of itself and the other good. Does the demand into the second good depend on
the endowments (or the total money, p e
i
)?
Exercise 17 (Lexicographic Preferences) Consider 2 goods, e
i
> 0, p

and the preferences


are given by Lexicographic orderings where the rst good takes the precedence. Identify the demand
Economics 501-2 by Mehmet Barlo: 4 The Price Mechanism 59
correspondence. Analyze how the demand of each of these goods change with changes in the prices
of itself and the other good. Analyze the income eects.
Exercise 18 Consider 2 goods, e
i
> 0, p

and the preferences are given by following utility


function:
u(x) = 2x
1
| +x
2
|.
Identify the demand correspondence. Analyze how the demand of each of these goods change with
changes in the prices of itself and the other good. Analyze the income eects.
Exercise 19 () Consider 2 goods, e
i
> 0, p

and the preferences are given by following utility


function:
u(x) = (x
1
|)
1/2
(x
1
|)
1/2
.
Identify the demand correspondence. Analyze how the demand of each of these goods change with
changes in the prices of itself and the other good. Analyze the income eects. (Hint: Can you use
the LaGrangian analysis in this setting?)
5 Competitive Equilibrium
This equilibrium concept is most probably the most important one in economics which signicant
and inuential policy decisions are based on it.
It is rst formulated by Walras (1874-1877), who proved the existence of it counting number
of equations. The existence was established by Debreu (1952) (including production), which is a
part of his book Debreu (1959), and re-presented in Debreu (1983). His existence proof is based
on the phenomenal achievement of Nash (1950), who proves the existence of Nash equilibria in
nite normal form games using Kakutanis Fixed Point Theorem due to Kakutani (1941), which
itself is a cleaned up version of the xed point theorem hidden in von Neumann (1937) which
is re-presented in von Neumann and Morgernstern (1944) also containing the famous Expected
Utility Theorem. Arrow and Debreu (1954) generalizes the competitive equilibrium to the case of
uncertainty including production, which is the very basic model in modern macroeconomics.
The basic idea in competitive equilibrium, restricting ourselves to economies of pure exchange, is
the following: Each agent comes to the market place with his endowment. There he observes a price
vector declared by the Walrasian auctioneer, who tries to minimize the excess demand (the goods
demanded minus the total goods in the economy), and the agent takes the price vector as given
even if he might own a considerable proportion of some good, and the set of agents is nite. Upon
observing the price, each agent determines the net trade they desire. If the resulting allocation is
such that the excess demand is positive in any coordinate (thus feasibility is violated) the Walrasian
auctioneer, declares another price vector. This continues until the excess demand is non-positive.
Such a price vector, and allocation prole is called a competitive equilibrium.
60
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 61
Edgeworth in his book Edgeworth (1881) presents another approach for the solution of a pure
exchange economy based on bargaining. Imagine that all the agents come to the market and start
to bargain on possible allocations. By the use of axioms on the welfare of agents, such as Pareto
optimality, individual rationality (participation), an agreeable allocation is described. Relating that
construction with the competitive equilibrium, involving a price mechanism, is the very basis of the
Theory of Welfare Economics.
Now let us render the denition of a competitive equilibrium:
Denition 31 (Competitive Equilibrium) A competitive equilibrium for an economy c =
(
i
, e
i
)
iN
is a price vector p

R
M
+
0 and an allocation x

= (x

i
)
iN
R
Mn
+
such that
1. [Optimality] for all i N, x

i
T
i
(p

); and
2. [Feasibility] markets clear, i.e. x

F(c), that is

iN
_
x

i,k
e
i,k
_
0 for all k = 1, . . . , M. (5.1)
The set of all competitive equilibria for a given c is denoted by CE(c).
We need to dene the excess demand correspondence formally:
Denition 32 The excess demand correspondence Z : X is dened by
Z(p) =
_
z(p) R
M
: z(p) =

iN
(x
i
(p) e
i
), for x
i
(p) T
i
(p), i
_
. (5.2)
If T : X
n
, where T =
iN
T
i
, is singleton valued, then so is Z : X, and in fact the
denition reduces to Z(p) =

iN
(T
i
(p) e
i
).
For any set E R
M
, when we write E x for x R
M
, we mean that for all y E, y x.
The following observation is a consequence of our denitions:
Proposition 22 p

is a competitive equilibrium price if and only if Z(p

) 0.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 62
Proof. This proof only involves manipulation of our denitions. It is obvious from the denition
that if (p

) is a competitive equilibrium price then Z(p

) 0. For the converse, pick any z(p

)
Z(p

) and assume that Z(p

) 0, hence, z(p) 0. Thus, it is clear that the feasibility condition


of the competitive equilibrium is satises. Since the denition of z(p

) involves x
i
(p

) T(p

),
optimality is also guaranteed.
5.1 Computing Competitive Equilibria
In this section, I will be going over some examples (from earlier courses) of pure exchange economies,
and in each, will demonstrate how to identify the set of competitive equilibria.
5.1.1 Example 1
Consider a pure exchange economy with free disposal (i.e. agents can get rid of their goods) and
with two commodities, k = 1, 2; and two traders i = a, b. Each trader i has the consumption set
R
2
+
and the endowments are given by e = (e
a
, e
b
) = ((0, 4), (4, 0)). Suppose that the preferences of
agent a is represented by the following utility function u
a
: R
2
+
R, dened by for any x R
2
+
u
a
(x
1
, x
2
) = minx
1
+ 2x
2
, 2x
1
+ x
2
.
The preferences of the other agent is given by the following utility function u
b
: R
2
+
R, dened
by for any x R
2
+
u
b
(x
1
, x
2
) = x
1
+ x
2
.
We will illustrate our answers to the following with a clearly labeled Edgeworth Box diagram to
scale (possibly using the same box for more than one answer).
First let us start with the set of Weak Pareto and Pareto optimal allocations.
Please refer to gure 5.1. In that gure the black line represents mister as indierence curve; the
dashed line that of mister b. The set of Weak and Pareto optimal allocations coincide (because both
these agents have preferences that are continuous complete preorder satisfying strict monotonicity)
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 63
Figure 5.1: The Edgeworth Box and Weak Pareto and Pareto optimal allocations.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 64
Figure 5.2: The Competitive Equilibrium.
and are diagonal in that gure and are given by the allocations residing on the diagonal of the
Edgeworth box. That is,
PO(c) = WPO(c) = x F(c) [ x
a,1
= x
a,2
, and x
b,1
= x
b,2
.
The set of competitive equilibrium is: (p

, x

) = ((1, 1), ((2, 2), (2, 2))). To see this, consider


gure 5.2:
Formally, the argument required is: Without loss of generality let p = (1, p
2
) (recall that what
matters is relative prices).
If p
2
= 1, then mister a maximizes his utility at x = (2, 2) because it is the point where his
expansion path intersects the budget line. Mister b, on the other hand, maximizes his utility at any
allocation such that they add up to 4. In particular, (2, 2) is a member of his demand. Therefore,
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 65
we have a competitive equilibrium: (p

, x

) = ((1, 1), ((2, 2), (2, 2))).


If p
2
< 1 then mister b would maximize his utility at 4/p
2
of good 2 and 0 of good 1. Since
4/p
2
> 4 (because that p
2
< 1 what he demands is not feasible, thus, there are no competitive
equilibria when p
2
< 1.
If p
2
> 1, then mister b demands (4, 0) and mister a demands where the diagonal intersects
the budget line and thus this intersection is clearly not at (4, 0). Thus, there are no competitive
equilibria when p
2
> 1.
Consequently, the unique competitive equilibrium is (p

, x

) = ((1, 1), ((2, 2), (2, 2))).


Finally, let us do the Core as well: The set of Core allocations are equal to individually rational
and Weakly Pareto optimal ones when there are only 2 agents. Thus, are equal to:
C(c) = x WPO(c) : x
a,1
[4/3, 2] .
5.1.2 Example 2
Consider a pure exchange economy with free disposal (i.e. agents can get rid of their goods) and
with two commodities, k = 1, 2; and two traders i = a, b. Each trader i has the consumption set
R
2
+
and the endowments are given by e = (e
a
, e
b
) = ((1, 7), (9, 1)). Suppose that the preferences of
agent a is represented by the following utility function u
a
: R
2
+
R, dened by for any x R
2
+
u
a
(x
1
, x
2
) = min2x
1
+ x
2
, x
1
+ 2x
2
.
The preferences of the other agent is given by the following utility function u
b
: R
2
+
R, dened
by for any x R
2
+
u
b
(x
1
, x
2
) = (minx
1
, x
2
)
45
.
First, the set of Weakly Pareto and Pareto optimal allocations.
Refer to gure 5.3: The set of Pareto optimal allocations are given by the dashed line, and are
equal to
PO(c) = x F(c) : x
b,1
= x
b,2
.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 66
Figure 5.3: The Weak Pareto and Pareto optimal allocations.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 67
Figure 5.4: The Competitive Equilibria.
Moreover, the set of Weakly Pareto optimal allocations is given in that gure by the shaded area
(I was too lazy to dene this set with notation!).
Note that in this example the set of weakly Pareto optimal allocations is not equal to the set
of Pareto optimal ones. To see this consider the following allocation x = ((0, 0), (10, 8)). Clearly
this allocation is weakly Pareto optimal because there are no feasible ways to increase both of the
agents utilities at the same time. Yet, this allocation is not Pareto optimal, because the following
feasible allocation provides same returns to mister b but gives strictly higher utilities to mister 1:
x

= ((2, 0), (8, 8)).


Next, we identify the set of Competitive Equilibria. Consider gure 5.4:
Let p = (1, p
2
), p
2
0. The only equilibrium is the one that is depicted in gure 5.4. In any
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 68
Figure 5.5: The Core.
other price, the demands do not coincide and are feasible. In this competitive equilibrium, the price
ratio is so that the slope of the budget line is equal to the slope of the lower part of mister as
indierence curve. Thus, his demand includes what mister b is demanding at those prices, namely
the intersection of his expansion path with the budget line.
Core allocations are given by the shaded area in gure 5.5:
5.1.3 Example 3
Consider a pure exchange economy with free disposal (i.e. agents can get rid of their goods) and
with two commodities, j = 1, 2; and two traders i = a, b. Each trader i has the consumption set
R
2
+
and the endowments are given by e = (e
a
, e
b
) = ((0, 4), (4, 0)). Each trader i has preferences on
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 69
Figure 5.6: The Weak Pareto and Pareto optimal allocations.
R
2
+
represented by a utility function u
i
: R
2
+
R dened by
u
a
(x
1,a
, x
2,a
) = 10x
1,a
(x
2,a
2)
2
and u
b
(x
1,b
, x
2,b
) = 10x
2,b
(x
1,b
2)
2
.
First, Pareto and Weak Pareto optimality:
Consider gure 5.6:
Mister 1 is achieving a utility of -4 at his endowment. Moreover, he is indierent between his
endowment and only to (0, 0). On the gure his indierence curves are black. The same kind of
reasoning applies for mister 2. Indeed, he too is indierent between his endowment and only (0, 0)
viewed from his corner (when viewed from mister 1s corner that point is (4, 4)).
First of all we should note that the set of Pareto optimal and weakly Pareto optimal allocations
cannot be strictly inside the Edgeworth box. The intuitive reason is the movements of the indier-
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 70
ence curves which tells us that player a would rather have good 1, and player b good 2. Formally,
take any allocation strictly inside the Edgeworth box. Then the allocation x = ((4, 0), (0, 4)) strictly
Pareto dominates such an allocation. The formal proof is left to the interested reader.
Moreover, the set of Pareto optimal and weakly Pareto optimal allocations cannot be on the
upper and left sides of the Edgeworth box. (Those are marked with R1 and R2 on the gure.) That
is all allocations in x F(c) : x
1,a
= 0 and x F(c) : x
2,b
= 0 are strictly Pareto dominated
by x = ((4, 0), (0, 4)) (that is both of the agents get strictly higher utilities).
The rest of the Edgeworth box that we need to check is those allocations in x F(c) : x
2,a
= 0
and x F(c) : x
1,b
= 0.
We will rst consider those in x F(c) : x
2,a
= 0, and x
1,a
< 2. Those are marked with
R3 and R6 in gure 5.6. Note that for any such allocation, the allocation of ((2, 0), (2, 4)) makes
both of the agents strictly better o (thus, ((2, 0), (2, 4)) strictly Pareto dominates any allocation
in x F(c) : x
2,a
= 0, and x
1,a
< 2). A similar argument holds for allocations in x F(c) :
x
1,b
= 0, and x
2,b
< 2.
We claim that any allocation in x F(c) : x
2,a
= 0, and x
1,a
2 and x F(c) : x
1,b
=
0, and x
2,b
2 (those sets are marked with R4 and R5 on gure 5.6) are Pareto optimal and weakly
Pareto optimal. In order to see that take any allocation from x F(c) : x
2,a
= 0, and x
1,a
2.
When player as utility is increased, then we must consider an allocation in the same set to the
right of the one we started with. Thus, player bs utility would decrease. Moreover, if player bs
utility is to be increased, then we must move to the left of that point, which would make player as
utility to decrease. Thus, the conclusion follows by using this argument for other points as well.
For the set of competitive equilibria, consider 5.7:
There is a competitive equilibrium when p

= (1, 1) and x

= ((4, 0), (0, 4)). In order to see why,


note that at those prices agent a demands the lower corner of his budget set (because he likes good
1), and agent b demands the upper part of his (because he likes good 2).
In the following, we will show that there are no other competitive equilibria in this economy.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 71
Figure 5.7: The Competitive Equilibrium.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 72
Suppose p = (1, p
2
), p
2
0. If p
2
> 1, then mister a would demand the corner corresponding
to buying only of good 1. Then his demand will be 4 p
2
> 4, hence, not feasible. Thus, there are
no competitive equilibria when p
2
> 1. Similarly, if p
2
< 1, then it should be easy for the reader
to see that, in this case mister would be demanding all good 2 and thus his demand would not be
feasible. Therefore, there are no competitive equilibria when p
2
> 1. Thus, the unique competitive
equilibrium is the one identied above.
In this economy the set of Core allocations (equal to the set of individually rational and Weakly
Pareto optimal ones) is given by the R4 and R5 in gure 5.6, i.e. is exactly equal to the set of
Pareto optimal allocations.
Our nal remark is: Consider the Pareto optimal allocation ((2, 0), (2, 4)). For any price line
that goes through this allocation, depending on the slope, either player a or b would demand a
non-feasible allocation. Thus, that point cannot be supported in competitive equilibrium. Hence,
we have a Pareto optimal allocation that cannot be supported by a competitive equilibrium when
endowments can be redistributed (remember this example after seeing the Second Fundamental
Theorem of Welfare Economics).
5.1.4 Example 4: Cobb-Douglas Utilities
The following is left as an exercise:
Consider a pure exchange economy with free disposal (i.e. agents can get rid of their goods)
and with two commodities, j = 1, 2; and two traders i = a, b. Each trader i has the consumption
set R
2
+
and the endowments are given by e = (e
a
, e
b
) = ((0, 4), (4, 0)). Each trader i has preferences
on R
2
+
represented by a utility function u
i
: R
2
+
R dened by
u
a
(x
1,a
, x
2,a
) = x
1/3
1,a
x
2/3
2,a
and u
b
(x
1,b
, x
2,b
) = 4 ln x
2,b
+ ln x
1,b
.
Illustrate your answers to the following questions with a clearly labeled Edgeworth Box diagram
to scale (possibly using the same box for more than one answer).
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 73
1. Find the set of Pareto optimal allocations. If there is none, show why. Then, nd the set of
Weakly Pareto Optimal allocations. If there is none, show why.
2. Find the set of Walrasian equilibrium allocations. If there is none, show why.
3. Find the set of Core allocations.
5.1.5 Example 5
Consider a pure exchange economy with free disposal (i.e. agents can get rid of their goods) and
with two commodities, j = 1, 2; and two traders i = a, b. Each trader i has the consumption set
R
2
+
and the endowments are given by e = (e
a
, e
b
) = ((0, 4), (4, 0)). The preferences of trader a are
given by the following utility function
u(x
1,a
, x
2,b
) = x
1,a
+ x
2,b
.
The preferences of agent b are the usual lexicographic preferences with good 1 having precedence.
First the Pareto and Weakly Pareto optimal allocations: The reader is kindly asked to refer to
gure 5.8:
Mister a has linear indierence curves, and as was analyzed in the lectures, mister b does not
have any indierence curves because with lexicographic preferences, there are no two separate points
between which an agent can be indierent.
Let us deal rst with the set of weakly Pareto optimal allocations. Given a feasible allocation
making mister a strictly better o requires him to receive a higher sum of goods, and therefore,
(because that we have to do this in a feasible way) mister b would be getting lower of good 1 or
good 2. In any case, since mister b likes both of these goods (but good 1 more than good 2) that
implies mister b will be strictly worse o. Therefore, in this case any feasible allocation (all the
points in the Edgeworth box) is weakly Pareto optimal.
When it comes to the analysis of the Pareto optimal allocations, rst let us deal with an allocation
that is strictly in the Edgeworth box. Such allocations (in which each agent gets a strictly positive
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 74
Figure 5.8: The Edgeworth Box.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 75
amount of each good) are not Pareto optimal. The reason is that for any such allocations (x or x

in
gure 6), we can nd another allocation (y for x, and y

for x

) which player a would be indierent


to but player b would strictly prefer. Thus, any point on the interior of the Edgeworth box is Pareto
dominated (not strictly though). The same reasoning holds for all the points on the lower and right
sides of the box marked R3 and R4. Therefore, the set of Pareto optimal allocations are given by
the R1 and R2.
The competitive equilibria, or alternatively Walrasian equilibria:
A Walrasian equilibrium is p

= (1, 1) and x

= e. First to see why this prole constitutes a


competitive equilibrium, notice that at these prices, player a would demand any point x
1,a
, x
2,a
such
that x
1,a
+ x
2,a
= 4. Thus, e
a
is among the points he demands. Moreover, mister b prefers good 1
to good 2. Thus, at p

= (1, 1), he would prefer to allocate all his wealth to the rst good, and not
consume any of the second good. Therefore, e
b
is his demand when p

= (1, 1).
For the rest, as before let p = (1, p
2
). No matter what the levels of p
2
is, mister b will always
buy only good 1 with all the money he has. He will never buy any of good 2. So no matter what
p
2
0 is e
b
is his demand. On the other hand, player as demand would include e
a
whenever the
price line is more steep than his indierence curve. Thus, for all p
2
1, his demand includes only
consuming good 2, i.e. e
a
. However, when p
2
> 1 his demand would be to get good 1 with all the
money he has. And it should be obvious now that then his demand would be outside the Edgeworth
box, and thus, not feasible.
Consequently, we have many Walrasian equilibrium prices, any price p = (1, p
2
) with p
2
1 is a
competitive equilibrium price. Notice that, however, there exists only one competitive equilibrium
allocation, which is given by e. So in this economy there will be no trade at all.
The only core allocation in this case is e. Because all the other Weakly Pareto optimal allocations
are not individually rational, that is either mister a or b would prefer to sticking to his initial
endowment.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 76
5.2 Existence of Competitive Equilibrium
In this section rst we are going to briey discuss the question of existence for the competitive
equilibrium in a pure exchange economy, and later will have this result stated and proved formally.
This is quite an important part of our analysis, because if a solution concept were not to exist
in an interesting class of formulations, then that solution concept is of no use to us as economists,
because it means that there are interesting cases which cannot be handled by that solution concept.
The rst existence proof was done by Leon Walras, Walras (1874-1877), in which he assumed
that the utility functions are continuous, strictly increasing, at least twice dierentiable, and strictly
quasi-concave (to be precise, the utility function in question can be turned into a strictly concave
one via a strictly monotone transformation). Note that under these assumptions, the rst order
analysis in order to identify the demand is justied, thus, we can consider the LaGrangian rst
order analysis. Note that we are trying to nd M n +M variables, rst part for the allocation of
consumption bundles, and the second for the prices. Using the rst order analysis, using the Walras
Law we would express the demand of each agent for each good as a function of the prices. Thus,
we would have M n equalities. Moreover, because of feasibility (which will hold with equalities
because of the assumption of strict monotonicity) we have M (the number of the goods) equations.
Thus, in total, we have M n +M equations. Moreover, this system of M n +M equations are
linearly independent upto a multiplication of prices (recall the homogeneity of the degree of 0 of the
demand), but as was noted before we can map the prices into the M1 dimensional simplex. Thus,
we have M n +M unknowns and M n +M system of equations that are linearly independent
(because the prices are in the simplex). Hence, a unique solution exists.
In fact this is the method that you are using when solving for competitive equilibria with the
rst order analysis in some of your homeworks, and more elementary microeconomics courses.
On the other hand, it should already be clear to you that, there are situations that do not
involve such a nice formulation. Indeed, the requirement of a dierentiable utility function is quite
a restrictive one.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 77
How do we handle such situations?
The techniques required to handle such situations are conceptually somewhat more dicult. But
with 2 agents and 2 goods the use of a very important Theorem of mathematics, the Intermediate
Value Theorem, suces. It says, if f is a continuous function mapping R into itself, and if there
exists x, y R with x ,= y and f(x) > 0 > f(y), then there exists z R residing between x and y
such that f(z) = 0.
In order to see this, consider the following easy setting. Let there be two agents each of whom
has preferences given by continuous complete preorders satisfying strict monotonicity and strict
convexity, and each of them have strictly positive amount of endowments in both of the goods (i.e.
e
i,k
> 0 for all i, k = 1, 2). Below I will prove the existence of a competitive equilibrium in this
setting using the Intermediate Value Theorem.
Under these assumptions we know that for each of them their demand is unique and upper
hemi-continuous, and thus, it is a continuous function on the set of strictly positive prices. Thus,
the excess demands are also a continuous function on the set of strictly positive prices. Note that
z
k
(p) = ((x
1,k
(p) e
1,k
) + (x
2,k
(p) e
2,k
)) ,
denotes the excess demand for good k = 1, 2, and x
i,k
(p) denotes the demand of agent i = 1, 2 for
good k = 1, 2. Then due to the above we know that x
i,k
(p), and z
k
(p) are both continuous functions
on p 0. By the Walras Law (recall that these preferences are strictly monotone thus are locally
non-satiated and therefore Walras Law holds), it must be that for every p 0
p
1
(x
1,1
(p) e
1,1
) +p
2
(x
1,2
(p) e
1,2
) = 0,
p
1
(x
2,1
(p) e
2,1
) +p
2
(x
2,2
(p) e
2,2
) = 0.
Summing up the right hand sides and the left hand sides of the two equations above delivers
p
1
(x
1,1
(p) e
1,1
) +p
1
(x
2,1
(p) e
2,1
) +p
2
(x
1,2
(p) e
1,2
) + p
2
(x
2,2
(p) e
2,2
) = 0,
p
1
((x
1,1
(p) e
1,1
) + (x
2,1
(p) e
2,1
)) + p
2
((x
1,2
(p) e
1,2
) + (x
2,2
(p) e
2,2
)) = 0,
p
1
z
1
(p) +p
2
z
2
(p) = 0,
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 78
thus we reach the following conclusion: For all p 0
p
1
z
1
(p) = p
2
z
2
(p). (5.3)
Now, because that the demand and the budget sets are homogeneous of degree zero (only relative
prices matter), without loss of generality we may let p
1
= and p
2
= 1. Thus, condition 5.3
becomes
z
1
(, 1) = z
2
(, 1).
Recall that by Proposition 22, and the above condition, all I need to prove (for the existence of a
competitive equilibrium) is that there exists p

0 (or alternatively,

> 0) such that z


1
(p

) = 0,
because then due to condition 5.3, I automatically have z
2
(p

) = 0 as well, and by Proposition 22, p

would be a competitive equilibrium price vector. Due to that regard, I will employ the Intermediate
Value Theorem as follows:
Note that, when > 0 is suciently small, arbitrarily close to 0, then because of e
i,k
> 0 for
all i, k = 1, 2 (each agent has some of every good in their initial endowments), and each agents
preferences being strictly monotone, the excess demand for good 1 must be strictly positive. For-
mally, for any natural number K N, there exists
K
> 0 such that z
1
(
K
, 1) > K. If this does not
hold, then it would contradict to the excess demand being continuous and preferences being strictly
monotone. Then, let > 0 be such that z
1
(, 1) > 0. By the same reasoning, for suciently high
(good 1 becomes very expensive), it must be that the excess demand for good 1 is strictly negative,
because then due to strict monotonicity the agents would demand the second good. Thus, there
exists > 0, such that z
1
( , 1) < 0. Then, because that z
1
is a continuous function on p 0, by
the Intermediate Value Theorem there exists

(, ) such z
1
(

, 1) = 0. Moreover, by condition
5.3, z
2
(

, 1) = 0, thus, z(

, 1) = 0, hence, p

= (

, 1), is a competitive equilibrium price.


It is appropriate to point that this proof has rather been quite a special case, two agents and
two goods.
The proof for the existence of a competitive equilibrium with more than 2 agents and more than
2 goods, on the other hand, uses a very similar reasoning, but requires a multidimensional version
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 79
of the Intermediate Value Theorem, which are called Fixed Point Theorems. Indeed the following
Theorem is the key in the existence proof:
Theorem 8 (Brouwers Fixed Point Theorem, 1912) If C is a non-empty, convex and com-
pact subset of R
m
, m N, and f : C C is a continuous function, then f has a xed point, i.e.
there exists a x such that f(x) = x.
Note that when the dimension is 1, i.e. m = 1, then all that this Theorem says is that there
exists x such that f(x) x = 0. Thus, letting g(x) = f(x) x, is indeed giving use the Intermediate
Value Theorem. That is why Brouwers Fixed Point Theorem is a multidimensional generalization
of the Intermediate Value Theorem.
In the next section, we will see in detail that Brouwers Fixed Point Theorem, even though very
essential, is not enough to get the existence result that we wish to have, which is:
Theorem 9 (Existence Theorem of Arrow and Debreu (1954)) Let c =
i
, e
i
)
iN
be a -
nite agent and nite and perfectly divisible good pure exchange (with free disposal) economy with
the following assumptions:
1. e
i
0 for all i N;
2.
i
is a continuous complete preorder for every i N;
3.
i
satises strict monotonicity for every i N;
4.
i
is weakly convex for every i N.
Then there exists (p

, x

) CE(c) with p

0.
This Theorem lists all the conditions needed to guarantee the existence of a competitive equi-
librium. You should observe that we have to have strict monotonicity for this Theorem to work.
That should have been obvious to the capable reader from the proof I have given you in the 2 agent
2 good case, especially when I was proving that there exists a price for which the excess demand
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 80
for the rst good must be negative and or positive. That part in the proof of this Theorem does
not change. We have to rely on strict monotonicity to use the multidimensional versions of the
Intermediate Value Theorem (i.e. the Fixed Point Theorems). But the good thing we get out of
strict monotonicity is also that p

0, that is the prices have to be strictly positive. Another


important assumption is weak convexity. It is needed in order to have the demand behave well, i.e.
satisfy some kind of generalized version of the continuity. In general we already know that demand
is not a function because it might be not singleton valued. Yet, if the preferences are weakly convex,
then the demand must be convex, and in addition to that the assumptions on continuity help us to
obtain some form of continuity for the demand set (is not a single point). Another remark I would
like to do regarding the assumptions is that we do not need strict convexity. Weak convexity is
enough.
Note that this Theorem does not say that if one of the assumptions listed above is violated, then
there are no competitive equilibria. In fact, there may be. In previous section you already have seen
such examples, that is examples of pure exchange economies which have at least one competitive
equilibria, but one or more of the assumptions above were not holding. For example, a symmetric
formulation with Leontie preferences has competitive equilibria, but strict monotonicity is not
satised.
The reason why Brouwers Fixed Point Theorem does not suce to obtain this existence result is
because in that Fixed Point Theorem, the excess demand correspondence needs to be a continuous
function (singleton-valued) on the whole simplex which includes the corners. But that then would
not be tting the assumption of strict monotonicity.
These are presented below; and this, rather restrictive, result is to be considered as a warm-up
for the more general existence Theorem.
Theorem 10 Let Z : X be a continuous function satisfying Walras Law. Then there exists
p

with Z(p

) 0.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 81
Proof. Dene : by
(p) =
_
p
1
+ maxZ
1
(p), 0
1 +

M
k=1
maxZ
k
(p), 0
, . . . ,
p
M
+ maxZ
M
(p), 0
1 +

M
k=1
maxZ
k
(p), 0
_
. (5.4)
Note that since Z is continuous on and by equation 5.4, is also continuous on , which is
a non-empty convex compact. Thus by Brouwers Fixed Point Theorem, Theorem 34, there exists
p

such that p

= (p

).
Thus,
p

k
=
k
(p

) =
p

k
+ maxZ
k
(p

), 0
1 +

M
l=1
maxZ
l
(p

), 0
.
By Walras Law p Z(p) = 0, thus,
0 =
M

k=1
p

k
Z
k
(p

) =
M

k=1
p

k
+ maxZ
k
(p

), 0
1 +

M
l=1
maxZ
l
(p

), 0
Z
k
(p

)
=
1
1 +

M
l=1
maxZ
l
(p

), 0
M

k=1
(p

k
+ maxZ
k
(p

), 0) Z
k
(p

)
=
M

k=1
p

k
Z
k
(p

) +
M

k=1
maxZ
k
(p

), 0Z
k
(p

) =
M

k=1
maxZ
k
(p

), 0Z
k
(p

),
makes us conclude that Z(p

) 0.
Note that we need the preferences (as complete preorders) to be bounded whenever the budget
set is not compact (i.e. whenever p ), as the continuous demand function is well dened for all
p , not only for p

. Moreover, preferences need to be strongly convex locally non satiated


whenever the budget set is compact. It is obvious that these conditions are rather very restrictive,
and exactly determining them or whether or not there are cases satisfying these assumptions, in
my opinion, are not worth the eort.
Because that Brouwers Fixed Point Theorem does not suce to obtain the general existence
Theorem, Arrow and Debreus Theorem 9, we need to consider a generalization of Brouwers Fixed
Point Theorem to correspondences: The Kakutani Fixed Point Theorem.
This theorem from Kakutani (1941) is essential in the theory of economics. It is the standard
tool to prove existence of an equilibrium.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 82
Figure 5.9: The Kakutani Fixed Point Theorem: C = [0, 1].
Theorem 11 (Kakutanis Fixed Point Theorem, 1941) If C is a non-empty, convex, and
compact subset of R
m
, and F : C C is a nonempty and convex valued, upper-hemi continu-
ous correspondence, then F has a xed point, i.e. there exists x C such that x F(x).
To understand and to appreciate this Theorem, consider the following gure 5.9
If any of the hypothesis of this Theorem is not satised, then the guarantee of the existence of
a xed point cannot be given.
Below, I will not deal with the properties of C, being non-empty convex and compact; rather,
will analyze the hypothesis of this Theorem on the correspondence. For the below, suppose that
C = [0, 1].
Suppose that F may be non-empty valued. This is a rather trivial case, and the following
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 83
Figure 5.10: F not convex-valued.
example does what we want: F dened by F(x) = 1/2 for all x ,= 1/2, and F(x) = when x = 1/2.
Trivially, this correspondence does not possess a xed point.
Next, convex valuedness: Dene F : C C by F(x) = 1 if x 1/2 and F(x) = 0 if x 1/2.
This correspondence is upper hemi-continuous (because, it has a closed graph, and its range is
compact and itself is compact valued), non-empty valued. But has no xed points as can be
observed in gure 5.10
Finally, upper hemi-continuity: In this setting upper hemi-continuity (due to the hypothesis of
the Kakutani Fixed Point Theorem) is equivalent to closed graphness whenever F is compact valued.
The example given in gure 5.11 identies a correspondence which is not upper hemi-continuous
(because does not have a closed graph) but is non-empty valued and convex valued: For a concisely
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 84
Figure 5.11: F is not upper hemi-continuous.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 85
dened example, consider F dened by F(x) = 1 if x < 1/2 and F(x) = 0 if x 1/2.
5.2.1 The Proof
The following Existence Theorem, taken from Mas-Colell, Whinston, and Green (1995), will cover
more interesting cases in terms of assumptions on preferences and endowments. The set of assump-
tions needed for it to hold will be presented in the Proposition that follows this Theorem.
Theorem 12 Let e 0 and Z :

X satisfy the following:


1. is a continuous function (in the interior of the price simplex); and
2. the Walras Law holds (in the interior of the price simplex), i.e. for all p

, p Z(p) = 0;
and
3. is bounded below (in the interior of the price simplex), i.e. for all p

there exists b R
with Z(p) b where b = (b, . . . , b); and
4. obeys the following boundary condition (BC): for any p
n

with p
n
p , the
sequence |Z(p
n
)|
n
is unbounded.
Then there exists p

such that Z(p

) = 0.
Proof. Let us start by dening the xed point correspondence to which Kakutanis Fixed Point
Theorem, Theorem 35 from appendix 9 will be applied.
Dene : by
(p) =
_

_
q : q Z(p) = max
q
q Z(p) if p

q : q p = 0 if p .
(5.5)
Obviously, since is non-empty, convex and compact all we need before applying Kakutanis Fixed
Point Theorem is to show that : is non-empty valued, convex valued and is upper hemi
continuous.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 86
Claim 1 : is non-empty valued.
Proof. There are two cases, either (1) p

or (2) p .
For the rst case, take any p

, and note that the denition tells us (p) = arg max


q
qZ(p).
Since is non-empty and compact, and the objective function is continuous, by Theorem 28
(p) ,= .
The second case: Pick any p , let J
1
, J
2
1, . . . , M, with J
1
J
2
= be dened by
J
1
= k : p
k
= 0 and J
2
= k : p
k
> 0. Now note that q dened by q
k
=
1
|J
1
|
for all k J
1
and
q
k
= 0 for all k J
2
is in . Moreover, p q = 0 since for all k either p
k
or q
k
(and not both) is
equal to 0. Thus, q (p).
Claim 2 : is convex valued.
Proof. Again we have two cases, either (1) p

or (2) p .
When considering the rst case, pick any p

and any q

, q

(p) and any [0, 1].


Due to the denition of given by equation 5.5, note that q

Z(p) = q

Z(p). Now note that for


q = q

+ (1 )q

,
q Z(p) = (q

+ (1 )q

) Z(p) = (q

Z(p)) + (1 ) (q

Z(p))
= q

Z(p) = q

Z(p)
= max
q
q Z(p),
thus, q (p).
The same consideration in case 2 is done with similar techniques. Pick any p and any
q

, q

(p) and any [0, 1]. Due to the denition of given by equation 5.5, note that
q

p = q

p = 0. Now note that for q = q

+ (1 )q

,
q p = (q

+ (1 )q

) p = (q

p) + (1 ) (q

p) = q

p = q

p = 0,
thus, q (p).
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 87
Claim 3 : is upper hemi continuous.
Proof. Note that in order to use closed graphness, an easier tool, in this proof by applying
Proposition 30, rst we need to establish that : is compact valued, since is compact.
Thus, it suces to prove that : is closed valued, i.e. for all p , (p) is closed.
Again we have 2 cases, either (1) p

or (2) p . In the rst case the observation that


is compact valued is an immediate consequence of Berges Theorem of Maximum, Theorem 33.
For the second case, take any q(n)
n
(p) with q
n
q

, where p . Thus for all n N,


q(n) p = 0. Since all the operations involved are continuous, q

p = 0, thus, q

(p).
Having established that : is compact valued, by Proposition 30 it suces to prove
that has a closed graph. That is we need to establish that for all p
n

n
, with p
n
p, and
q
n

n
with q
n
(p
n
) for all n and q
n
q, implies that q (p).
Again we need to consider two cases: either (1) p

or (2) p .
For the rst case, i.e. when p

we know that for any p


n

n
, with p
n
p, there exists
N N with p
n

for all n > N. Thus, for all such n, by the dening property of we have
q
n
Z(p
n
) q

Z(p
n
) for all q

. Because that p

, there exists N

such that for all n > N

,
p
n

. Thus, without loss of generality assume that p


n

n
lives in

. Therefore, there exists


> 0 such that p
n

, where

= p : p
k
, k = 1, . . . , M. Note that

is
non-empty convex and compact for > 0 small. Thus, p
n

n
is a sequence residing in a compact
set. Moreover, because that Z is a continuous function from

into X, and

is compact, we have
Z(

) = z(p) : p

is a compact subset of X. Hence, Z(p


n
) lives in Z(

), a non-empty
compact set. Hence, because that the dot product and greater or equal to sign being continuous,
we have q
n
Z(p
n
) q

Z(p
n
) for all q

implies (we now may take double limits because both


p
n
and Z(p
n
)
n
live in compact spaces), q Z(p) q

Z(p) for all q

, thus, q (p).
Now the second case, i.e. when p will involve a little more complicated observation. There
we need to use the boundary condition.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 88
Now take any good k with p
k
> 0. We shall argue that for n large, we have q
n,k
= 0, thus,
q
k
= 0, consequently implying such a q is in (p). Since p
k
> 0 there exists > 0 such that p
n,k
>
for n suciently large. If in addition we have p
n
, then q
n,k
= 0 by the denition of . On the
other hand if p
n

, since Z is bounded from below and obeys condition BC, for n N large we
have
Z
k
(p
n
) < max
j=1,...,M
Z
j
(p
n
). (5.6)
This is because of the following: as p
n
, by condition of BC, we have that the right hand side
of equation 5.6
max
j=1,...,M
Z
j
(p
n
) .
And using the Walras Law
1
and feasibility the left hand side of the equality can be bounded as
follows:
Z
k
(p
n
)
1

p
n,k
Z
k
(p
n
) =
1

j=k
p
n,j
Z
j
(p
n
) <
1

j=k
p
n,j
_

iN
e
i,j
_

_
max
l=1,...,M

iN
e
i,l
_

j=k
p
n,j
<
1

_
max
l=1,...,M

iN
e
i,l
_
Thus having established equation 5.6 we may conclude that by the denition of , and p
n

with
p
n,k
> , we have q
n,k
= 0.
Claim 4 : has a xed point.
1
Note that since by Walras Law we have for all p

, p Z(p) = 0. Now when p by the boundary condition


p Z(p) is not well dened, as the righthand side of the last equality is innity. But in this case we already are
working under the hypothesis that p

, thus
0 = p Z(p) =
M

j=1
p
j
Z
j
(p)
= p
k
Z
k
(p) +
M

j=k
p
j
Z
j
(p).
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 89
Proof. Since : is a non-empty valued, convex valued, and upper hemi continuous
correspondence with non-empty convex and compact range and domain, Kakutanis Fixed Point
Theorem, Theorem 35 applies, consequently, there exists p

with p

(p

).
Claim 5 p

such that p

(p

) is in

.
Proof. Suppose not, then p

, thus, we have
p

(p

) = q : p

q = 0,
implying for p

, p

= 0, an impossibility.
Claim 6 Z(p

) = 0.
Proof. Since p

, be Walras Law we have p

Z(p

) = 0. Therefore if we were to have


Z
k
(p

) < 0 for some k = 1, . . . , M, this would imply that Z


k
(p

) > 0 for some other k

,= k. Thus
such a price p

, for q (p

), has q
k
= 0 and q
k
> 0, hence, (p

) , contradicting to p

such
that p

(p

) is in

.
This last step nishes the proof of the Theorem.
Now we are going to analyze the conditions on the primitive assumptions to obtain the hypothesis
of Theorem 12. They are described in the following Proposition:
Proposition 23 Suppose that for all i the following hold:
1. e
i
0;
2.
i
is a continuous, complete preorder, which satises monotonicity and strong convexity.
Then Z(p) : X satises the following:
1. Z is a continuous function (in the interior of the price simplex); and
2. For Z, the Walras Law holds (in the interior of the price simplex), i.e. for all p

,
p Z(p) = 0; and
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 90
3. Z is bounded below (in the interior of the price simplex), i.e. for all p

there exists b R
with Z(p) b where b = (b, . . . , b); and
4. Z obeys the following boundary condition (BC): for any p
n

with p
n
p , the
sequence |Z(p
n
)|
n
is unbounded; and
5. Z homogenious of degree 0.
Exercise 20 Prove Proposition 23. Note that in that proof you need to show that if a preference
order is a continuous, complete preorder, which satises monotonicity and strong convexity, then it
is strongly monotone (the reader should already have realized the close connections between condition
(BC) and strong monotonicity).
Exercise 21 Show by example (give the utility function and render the explicit computation) that
the boundary condition, p
n
p |Z(p
n
)| (where is the price simplex, p
n
and p
are price vectors and Z is the excess demand function) is not equivalent to the statement if the
price of a good goes to 0, then the excess demand for that good goes to .
Now we consider a more general Existence Theorem which does not require the excess demand
correspondence to be singleton valued:
Theorem 13 Suppose that e
i
0 for all i, and Z

X is an excess demand correspondence


satisfying:
1. upper hemi continuity, non-empty valuedness, and convex-valuedness; and
2. Walras Law, i.e. for all z(p) Z(p) we have p z(p) = 0; and
3. is bounded below, i.e. for all z(p) Z(p) there exists b R with z(p) b where b =
(b, . . . , b) R
M
; and
4. obeys the following boundary condition BC: for any p
n

, with p
n
p , and for
any z(p
n
) where z(p
n
) Z(p
n
) for all n N, we have |z(p
n
)
n
is an unbounded sequence.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 91
Then, there exists p

and z(p

) Z(p

) such that z(p

) = 0.
Proof. Let > 0, and let

= p : p
k
, k = 1, . . . , M. Then we know that
C

=
p
Z(p) is compact, since Z is upper hemi continuous and compact valued on

(the
interested and curious reader is asked to verify this step, after all it is a great exercise). Since
C

may not be convex we need to take the convex hull


2
of it, so let

C

= con C

. Now

C

is
compact, convex and non-empty, as the operation of convexication preserves compactness. Now
dene f

:

C

by
f

(z) arg max


q
e
q z. (5.7)
Due to Berges Theorem of Maximum, f

is non-empty valued, compact-valued and upper hemi


continuous. Moreover, because that the objective function is linear, it is also convex valued. Now
dene

by

(p, z) = Z(p) f

(z). (5.8)
Since, by construction

is non-empty valued, convex valued and upper hemi


continuous, Kakutanis Fixed Point Theorem, Theorem 35 applies, hence, there exists ( p, z) with
( p, z)

( p, z).
Now let
n
0, and let p
n
, z
n

n
be the sequence of corresponding xed points. Note that
since the sequence p
n

n
lives in a non-empty compact set, it has a convergent subsequence, so
without loss of generality we may let p

be such that p
n
p

.
Now we need to argue that p

/ . Suppose it were, thus since p

,= 0 (after all p

), there
exists good k with p

k
> 0. Without loss of generality assume that k = 1. Then for suciently large
n we have
z
n,1
< max
j=1,...,M
z
n,j
.
To see that note that the right hand side is unbounded due to e 0 and condition BC; but, the
left hand side of the last inequality can be bounded by 0 as follows: As p

1
> 0 there exists > 0
2
We refer the reader to appendix 11 for more on convex hulls.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 92
such that p

1
> > 0, and p
n,1
> for n large, hence
z
n,1

p
n,1

z
n,1
=
1

j=1
p
n,j
z
n,j

max
l=1,...,M

iN
e
i,l

j=1
p
n,j

max
l=1,...,M

iN
e
i,l

.
Thus for n large, by the denition of f

we have p
n,1
=
n
. Since
n
0 this implies p
n,1
0,
contradicting to p

1
> 0, and p
n,1
p

1
.
Now note that because p

, there exists

such that p

. Thus, let z

be such that
(p

, z

(p

, z

).
The remaining step to establish is to show that z

= 0. Suppose not, without loss of generality


assume that z

1
> 0 (because we cannot have z

0 due to the Walras Law). Then, z

k
< 0
for some k ,= 1, thus, because for any q f

(z

)
1
we have q
1
> 0, and q
k
= 0, it must be that
f

(z

) . But, recall that p

(z

) due to p

being a xed point of

. Thus, we obtain a
clear contradiction, p

and p

.
Again we will discuss the primitive assumptions needed for Theorem 13 in the following Propo-
sition, which establishes the existence result of Arrow Debreus Theorem 9:
Proposition 24 Suppose that for all i the following hold:
1. e
i
0;
2.
i
is a continuous, complete preorder, which satises strong monotonicity and weak convexity.
Then Z(p) : X satises the following:
1. is upper hemi continuous, non-empty valued, and convex valued; and
2. Walras Law, i.e. for all z(p) Z(p) we have p z(p) = 0; and
3. is bounded below, i.e. for all z(p) Z(p) there exists b R with z(p) b where b =
(b, . . . , b) R
M
; and
4. obeys the following boundary condition BC: for any p
n

, with p
n
p , and for
any z(p
n
) where z(p
n
) Z(p
n
) for all n N, we have |z(p
n
)
n
is an unbounded sequence.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 93
Exercise 22 Prove Proposition 24.
Exercise 23 Consider a pure exchange economy with consumption sets R
M
+
. Assume that for
all traders i = 1, . . . , N, the initial endowment vectors are strictly positive, i.e. e
i
R
M
++
, and
preferences are represented by log linear utility functions u
i
: R
M
+
R dened by
u
i
(x
i
) =
i,1
log x
i,1
+ . . . +
i,M
log x
i,M
with
i,k
> 0 for all i = 1, . . . , N and every k = 1, . . . , M with

M
k=1

i,k
= 1 for each i. Does this
economy satisfy the assumptions of Theorem 10 and/or Theorem 12 and/or Theorem 13? Prove
your answer.
Exercise 24 Prove that if f is a weakly increasing function from [0, 1] to [0, 1], i.e. x y
f(x) f(y), then f has a xed point. Note that, we dont need continuity for this result.
Exercise 25 Consider the correspondence : [0, 1] [0, 1] dened by (x) = 1 if x < 1 and
(1) = [0, 1). Does it have a xed point? Why or why not? Draw its graph.
Exercise 26 A normal for game is given by

N, (S
i
)
iN
, (u
i
)
iN
_
, where N is the set of players,
S
i
is agent is set of strategies, and u
i
:
iN
S
i
R is player is payo function. A pure strategy
prole s

is a Nash equilibrium if for all i = 1, . . . , N; we have u


i
(s

i
, s

i
) u
i
(s
i
, s

i
) for all s
i
S
i
.
For more on Nash equilibrium we refer the reader to appendix 10.
Consider the class of pure exchange economies with m traders and M commodities, such that
each traders consumption set is R
M
+
, and each trader has a strictly positive endowment vector.
1. Dene a game that includes an articial player (the Walrasian auctioneer) and that has the
property that the set of Nash equilibrium allocations is equal to the set of Walrasian equilibrium
allocations. Prove the equality. Be sure to state explicitly any assumptions you need to make
about preferences or endowments for this equivalence.
2. Prove the following theorem, a modied version of Nashs existence theorem done by Debreu:
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 94
Theorem 14 Assume that N < ; and S
i
is a non-empty convex and compact set for all
i N; and the best response BR
i
is -valued, convex-valued and upper hemi continuous for
all i N. Then an equilibrium exists.
3. Using the equivalence relation shown in a, and existence result given in b, state Debreus
existence theorem in terms of the aggregate excess demand function Z. In order to do that
you need to dene and identify the conditions required on agents preferences for the validity
of this existence theorem.
5.3 Welfare Properties of Competitive Equilibria
In the following sections we are going to relate the concepts of Pareto optimality (together with
individual rationality) and the Core with competitive (Walras) equilibrium.
5.3.1 First Fundamental Theorem of Welfare Economics
The question whether or not regulation is needed in economics is essential. Adam Smith tells us that
whenever each agent is trying to do the best for himself (only), it is as if these interests are arranged
by an invisible hand in order to maximize the whole welfare of the economy. This argument is
very important because if one believes in it, the consequences would be to oppose regulation and
intervention into the economy by a third party (government).
The First Fundamental Theorem of Welfare Economics establishes that in the case of Walrasian
general equilibrium analysis (note that we have this result for production economies as well, but
there for reasons of convenience we will present it for exchange economies) Adam Smiths argument
holds. As you can imagine, the policy implications of this Theorem is enormous.
Later research displays that this Theorem fails to hold when the following tree of the main
assumptions of the formulation of general equilibrium is weakened.
The rst is about the price-taking behavior. In fact the Prisoners Dilemma is a very well
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 95
known example which establishes that whenever agents have the ability to use all their strategic
power, equilibria might consists of non-Pareto optimal allocations.
The second assumption that one has to be careful about concerns the informational setup of the
model. Akerlof establishes that when symmetric information about the preferences of agents are
dropped, equilibria might again involve non-Pareto optimal points.
The third assumption is that the welfare of every agent depends only on his own consumption.
That is if agents get some returns from the consumption of other agents, then again this Theorem
may fail to hold. Two typical examples in which the First Fundamental Theorem of Welfare
Economics does not hold in this setting could be: (1) the utility of an agent depends on his
consumption, and inversely on the variance of the consumption vector of every agents (this agent
wants sort of more equality in consumption, but still gets payos from his own consumption); and
(2) one of the goods under analysis is a public good, a good such that its consumption by one agent
does not exclude the consumption of others (e.g. security).
Now the Theorem:
Theorem 15 (First Fundamental Theorem of Welfare Economics) For any exchange econ-
omy c, where agents preferences are given by continuous complete preorders satisfying local non-
satiation, CE(c) PO(c).
Proof. Suppose that (x

, p

) CE(c). Thus by the denition of competitive equilibrium (each


agent maximizing his utility)
x
i
~
i
x

i
p

x
i
> p

e
i
.
By continuity and local non-satiation this condition implies
x
i

i
x

i
p

x
i
p

e
i
.
For any allocation x that Pareto dominates x

, we must have p

x
i
p

e
i
for all i N, and this
same inequality holding strictly for some j N. Thus,

i
p

x
i
>

i
p

e
i
. If x were feasible, we
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 96
would have a contradiction, because then

i
x
i


i
e
i
would imply (since p

0 and p

,= 0 by
the denition of a competitive equilibrium)

i
p

x
i

i
p

e
i
.
5.3.2 Second Fundamental Theorem of Welfare Economics
This Theorem that my dear late Professor Murat R. Sertel used to refer to as the fundamental
design theorem links Pareto optimal allocations with the concept of competitive equilibria. In
particular, it displays when a Pareto optimal allocation of an economy can be obtained as a com-
petitive equilibrium of another economy which consists of the same agents, same preferences but a
redistribution of initial endowments. It is a design issue in the sense that of a planner who wants
to obtain a Pareto optimal allocation in competitive equilibrium (as its mechanism) it displays how
to redistribute the initial endowments.
The following version and its proof is from Mas-Colell, Whinston, and Green (1995).
Theorem 16 (Second Fundamental Theorem of Welfare Economics) Assume that the pref-
erences of agents are weakly convex, continuous complete preorders satisfying strong monotonicity.
Then for every c =
i
, e
i
)
iN
, and x

PO(c) with x

0, x

CE(c

) where c

=
i
, x

i
)
iN
.
Note that in words this Theorem says that, under these hypothesis any Pareto optimal allocation
of a given economy can be obtained in competitive equilibrium (through the price mechanism)
when redistributions of the initial endowments are allowed. Thus, it might very well be that the
new competitive equilibrium is not individually rational for some of the players who are comparing
their payos against their payos in the initial endowments of the original economy. Hence, many
view this Theorem as the Theorem of the Revolution, whatever revolution they are talking about
is obviously not that important. And that is also why, a normative economist inevitably must
consider the laws governing inheritance (miras) because they are the laws governing how the initial
endowments are distributed. If those rules were like those in the Ottoman Empire where until the
second Tanzimat there were no private ownership allowed to Turkish people, then the result is not
same as the one we have nowadays, where inheritance is done by only paying some portion of the
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 97
assets inherited to the government as taxes. Obviously, these are deep issues, and without taking
any sides I urge you to think about them.
The assumptions required for this proof needs a little understanding. In the proof of this
Theorem, we would have to use the Existence Theorem. This is because, this Theorem needs the
existence of a competitive equilibrium in order to be able to say that a Pareto optimal allocation of
the original economy can be supported as a competitive equilibrium. That is why the assumptions
needed for the proof of this Theorem is the same as the assumptions listed in the Existence Theorem,
Theorem 9. And no additional assumptions are needed. (Note that here the endowments of the
new economy the one obtained by a redistribution would be the Pareto optimal allocation of
the original economy, thus, we need the assumption that this bundle is strictly positive in every
coordinate for the Pareto optimal allocation of the original economy.)
Proof of the Second Fundamental Theorem of Welfare Economics. First note that
in this proof we need the existence of a competitive equilibrium, thus, we need the additional
assumptions of the preferences being continuous complete preorder with weak convexity and strong
monotonicity, and x

0.
Let x

PO(c), thus o|(o


i
(x

i
) is a non-empty, convex and open subset of R
n
(check out Propo-
sition 6 to recall the denition of a strict upper contour set). Dene o|(o(x

) =

i
o|(o
i
(x

i
),
again a non-empty, convex and open subset of R
M
. Note that x

i
x

i
is not in o|(o(x

) since
x

is Pareto optimal, because if x

o|(o(x

) then there would be feasible redistribution that


would make everybody strictly better o (remember that under the hypothesis of this Theorem
PO(c) = WPO(c)). Therefore by the Separating Hyperplane Theorem, Theorem 39 given in the
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 98
footnote below
3
, there exists p

R
M
0 such that
p

(y x

) 0 for all y o|(o(x

). (5.9)
In the following claims we will establish that (p

, x

) is a competitive equilibrium for the economy


given by c

=
i
, x

i
)
iN
.
Claim 7 p

0.
Proof. Fix k = 1, 2, . . . , m, and let q
(k)
R
M
; be dened by q
(k)
j
= 0 for all j ,= k, and
q
(k)
k
= 1. Since
i
obeys strong monotonicity for all i, x

+q
(k)
o|(o(x

). Thus by equation 5.9,


p

( x

+ q
(k)
x

) = p

q
(k)
= p

k
0.
Claim 8 If x
i
~
i
x

i
for some i N, then p

x
i
p

i
.
Proof. Fix i N, and let x X
n
be such that x
i
~
i
x

i
, and x
j
= x

j
for all j ,= i. Since each
agent has strictly monotone preferences, we can identify another allocation (by a redistribution)
x

X
n
, such that x

= x, and x

j
~
j
x

j
for all j N (this step is left as an exercise to the reader).
Thus, x

= x o|(o(x

), hence,
p

= p

x =

j
p

x
j
= p

(x
i
x

i
) + p

.
Thus, rendering the conclusion that p

(x
i
x

i
) 0.
Claim 9 If x
i
~
i
x

i
for some i N, then p

x
i
> p

i
.
3
The following Theorem, known as Minkowskis Separating Hyperplane Theorem. The version we have is presented
from Hildenbrand and Kirman (1976). For other separation theorems and related results the reader is encouraged
to check Chapter 11 of Rockafellar (1970): Let C be a convex set of R

, and y /

C, where

C denotes the interior of
C. Then there exists R

0 such that for all x C


x y.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 99
Proof. Suppose not. Hence, suppose that there is an agent i and x
i
~
i
x

i
but p

x
i
= p

i
(by the previous claim it cannot be p

x
i
< p

i
). Since each agents preferences are continuous
there exits (0, 1) such that x
i
~
i
x

i
. However, since all the entries are non-negative and by
the previous claim we know that p

(x
i
) = (p

x
i
) p

i
= p

x
i
, the desired contradiction is
imminent, because since < 1 it must be that p

x
i
< p

i
.
Noting that the above claims establish that (p

, x

) CE(c

) where c

=
i
, x

i
)
iN
nishes
the proof.
5.3.3 Competitive Equilibria and The Planners Problem
Recall Proposition 15 was telling us that when every agents preferences are represented by a
continuous, strictly increasing, and concave utility function u
i
: X R (note that 0 X), then an
allocation x

X
n
is Pareto optimal if and only if there exists

(N) and x

solves the social


planners problem for c at

, where it is is given by
max
xF(E)

iN

i
u
i
(x
i
). (SP(c [

))
In this section, we will have the connection of the competitive equilibrium allocations and
the social planners problem, by using the First and Second Fundamental Theorems of Welfare
Economics.
Due to that regard, let us restrict attention to situations in which every agents preferences are
represented by a continuous and concave utility function u
i
: X R. Adding the assumption
that e
i
0, then makes sure that the hypothesis of the Existence Theorem, Theorem 9, the First
Fundamental Theorem of Welfare Economics, Theorem 15, the Second Fundamental Theorem of
Welfare Economics, Theorem 16, and Proposition 15 are all satised. The reader is asked to verify
that step. Thus, using these observations I present the following result:
Theorem 17 Let c =
i
, e
i
)
iN
be a pure exchange economy with nitely many agents and nitely
many goods. Moreover, assume that for every agent i N his references,
i
, are represented by
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 100
a continuous, strictly increasing, and concave utility function. Then, if x

0 is in CE(c), then
there exists

(N) such that x

solves SP(c [

). Moreover, if x

0 solves SP(c [

) at
some

(N), then x

CE(c

) where c

=
i
, x

i
)
iN
.
Proof. Suppose that x

0 is in CE(c). Then by the First Fundamental Theorem of Welfare


Economics, x

PO(c), and therefore, by Proposition 15, there exists

(N) such that x

solves SP(c [

). Conversely, suppose that x

0 solves SP(c [

) at some

(N), then by
Proposition 15 x

PO(c) and x

0, therefore, by the Second Fundamental Theorem of Welfare


Economics, x

CE(c

) where c

=
i
, x

i
)
iN
.
Note that this is a very important result for two reasons.
The rst is about the conceptual information that this result provides. In fact, it says that
competitive equilibria can be viewed as a decentralized version of the social planners problem
where the decentralization is done through the use of prices which have a one to one relation with
the bargaining weights that is used in the social planners problem. So therefore, every competitive
equilibrium that involves some prices can be viewed as a result of a social planners problem in which
the bargaining weights used are coming from the prices. (Therefore, one can identify the bargaining
weights needed to get that competitive equilibrium.) Conversely, every strictly positive solution
vector of a social planners problem with a given vector of bargaining weights, can be associated
with a competitive equilibrium with redistributions in which the prices are to be determined from
the bargaining weights employed.
Because of these there is one to one relation between the social planners problem and the
competitive equilibrium. But notice that the requirements needed for this result is quite restrictive.
Quasi-concave utilities (therefore, convex preferences) do not suce. We have to have either concave
utility functions or cardinal payos (as opposed to ordinal ones that we keep working with).
The second reason for why this Theorem is very important regards its extensive use in modern
macro economics. Generally, solving for competitive equilibria is quite dicult in the innitely
repeated versions of the pure exchange economies. Before you ask, yes all these results continue to
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 101
hold in that setting as well. But to see that you have to take either Masters level Macro Economics,
or look into the rst 4 chapters of Stokey, Lucas, and Prescott (1989). But in these settings, solving
the social planners problem is much more easy, and once we have these 4 Theorems working (the
3 Theorems and 1 Proposition listed in the proof), solving the social planners problem suces to
identify the competitive equilibria. Because then it means that, the solution to the social planners
problem can be decentralized via a price mechanism.
Indeed, one of the most important existence proofs for the competitive equilibria of innitely
repeated pure exchange economies, uses this method (having the existence as a standing hypothesis
in the proof of the Second Fundamental Theorem of Welfare Economics for that setup), and is often
referred to as the Ichishii approach.
5.4 Competitive Equilibrium and The Core
It is not dicult to prove the following result:
Proposition 25 Every competitive equilibrium allocation is individually rational.
Proof. Because that for any p 0, e
i
B
i
(p, e
i
), for all x
i
T
i
(p) it must be that x
i

i
e
i
.
Thus, the result follows from the denition of a competitive equilibrium.
Hence, because of this result and the First Fundamental Theorem of Welfare Economics, we know
that every competitive equilibrium allocation must be individually rational and Pareto optimal.
But is it in the core?
This is a critical question, because recall that the core is the set of allocations that we would
expect out of any bargaining structure between our rational agents in this pure exchange economies.
That is, every point in the core are allocations from which no further trade is not benecial for any
subcoalition of agents.
Thus, if the competitive equilibrium allocations were not to be a subset of the core allocations,
then that means there exists a competitive equilibrium allocation and a subcoalition of agents
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 102
who would get strictly higher payos when they jointly deviate to another allocation which can
be achieved via a feasible redistribution of their endowments. That in turn suggests that that
particular competitive equilibrium allocation is not an allocation from which further trade is not
benecial for any subcoalition of agents. That would have been quite a negative observation.
Luckily the following Theorem due to Shapley will establish that every competitive equilibrium
allocation must be in the core (the reverse direction clearly does not hold). That implies that
whenever the competitive equilibrium is implemented no subgroup of the society would want to
leave the society and exploit a feasible redistribution among themselves. Thus, every competitive
equilibrium allocation is an allocation from which further trade is not benecial for any subcoalition
of agents. For me, this is one of the strongest justications (maybe the strongest) of the use of
competitive equilibrium.
Theorem 18 For every exchange economy c in which agents preferences are given by continuous
complete preorders satisfying local non-satiation, CE(c) C(c).
Proof. Suppose x

/ C(c) but x

CE(c) for a contradiction (also let p

be the associated
equilibrium price). Since x

CE(c), x

F(c), and since x

/ C(c), there exists S o and


y
S
F(c [ S), such that y
S
j
~
j
x

j
for all j S. Therefore, by local non-satiation and x

CE(c),
for all j S
y
S
j
~
j
x

j
p

y
S
j
> p

e
j
.
Consequently,

jS
p

(y
S
j
e
j
) > 0. Since y
S
F(c [ S), we have a contradiction, because
then

jS
y
S
j


jS
e
j
would imply (since p

0 and p

,= 0 by the denition of a competitive


equilibrium)

jS
p

(y
S
j
e
j
).
Note that when agents preferences are required to be strictly monotone, the set of weak Pareto
and Pareto optimal allocations coincide. Then Shapleys Theorem is a generalization of the First
Fundamental Theorem of Welfare Economics. But in general, as you can note, Shapleys Theorem
and the First Fundamental Theorem of Welfare Economics does not have strong monotonicity, but
local non satiation. Thus, weak Pareto and Pareto optimal allocations need not be the same, but
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 103
still they have to agree on the set of competitive equilibrium allocations. That is, even though
the set of weak Pareto and Pareto optimal allocations might be dierent, WPO(c) CE(c) =
PO(c) CE(c). In words, when preferences are given by continuous complete preorders satisfying
local non-satiation, there cannot be a competitive equilibrium that is in the set of Core allocations
but not Pareto optimal ones (generally, for arbitrary allocations, as you know this is a possibility
from your homeworks).
5.4.1 Existence of The Core
Now that we have the necessary technology, let us come back to the existence question concerning the
core. The following Theorem, even though not the most general existence result, would demonstrate
conditions under which the core of an exchange economy is non-empty.
Theorem 19 For all exchange economies c such that for all i, e
i
0 and
i
is a continuous
complete preorder satisfying strong monotonicity and weak convexity, we have C(c) ,= .
Proof. Due to Theorems 9, and 18 result follows.
5.5 Debreu Scarf Limit Result
One of the basic problems in the theory of the core concerns the question about the kinds of
economies where the dierence between the core and competitive equilibrium allocations is small.
In other words, under which circumstances does cooperative barter and price competition through
decentralized markets leads essentially to the same result?
Debreu and Scarf (1963) (also reprinted in Debreu (1983)) provides an answer to this question
in the case when the size of the economy increases in a certain fashion. In particular, they consider
replica economies, economies in which each agent gets cloned symmetrically as the size of the
economy gets bigger.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 104
They demonstrate that as the replication process goes to innity, the only allocation that remains
in the core is exactly equal to the competitive equilibrium allocation.
Now we are going to see this result in full detail:
Denition 33 Given an economy c =
i
, e
i
)
iN
, let the rst replica of it be c
1
c. Then for
any r N, the rth replica then is c
r
=
_

i,h
, e
i,h
)
iN
_
r
h=1
, where
i,h
=
i
, and e
i,h
= e
i
for all
h = 1, . . . , r.
The rst question we have to analyze is whether or not the competitive equilibrium of c remains
the same in c
r
, r N.
Proposition 26 If x CE(c), then for any r N, x
(r)
CE(c
r
) where x
(r)
i,h
= x
i
for all h =
1, . . . r, whenever for all i N,
i
are continuous complete preorders satisfying monotonicity and
strict convexity.
Proof. Let ( p, x) CE(c), and note that T
p
i
(c) = x
i
. For any r N, since for all h = 1, . . . , r,

i,h
=
i
, and e
i,h
= e
i
, we have x
i
= T
p
i,h
(c
r
). Moreover,

i
x
i,h
= r

i
x
i
r

i
e
i
=

i
e
i,h
, thus, feasibility is also satised.
Next we will present a condition which will be used to check that as the size of the economy
gets bigger, the structure of the change is well-behaved.
Denition 34 An allocation x X
n
satises the equal treatment property, if e
i
= e
j
and

i
=
j
for i, j N implies x
i
= x
j
.
Note that here the denition rendered was for a particular allocation, but we could continue to
dene the equal treatment property for an allocation rule: An allocation rule f : c f(c), where
f(c) X
n
is the set of allocations for the economy c, satises the equal treatment property if for
every exchange economy c, any allocation in f(c) satises the equal treatment property.
The following exercise is aimed to shed more light on the equal treatment property.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 105
Exercise 27 Does the following allocation rules satisfy the equal treatment property? Prove your
claim either analytically, or by identifying clearly and concisely stated examples.
1. Feasibility;
2. Individual rationality;
3. Pareto optimality;
4. Core;
5. No-envy
4
.
The following Proposition will tell us that the competitive equilibrium allocation rule satises the
equal treatment property whenever agents preferences are assumed to be strictly convex, monotone
and continuous complete preorders.
Proposition 27 Assume that agents preferences are given by strictly convex, monotone and con-
tinuous complete preorders. Then the competitive equilibrium allocation rule satises the equal
treatment property.
Proof. Suppose not, i.e. there exists an economy c (satisfying the hypothesis of the Proposition)
such that there exists i, j N with e
i
= e
j
and
i
=
j
, but x

i
,= x

j
where (x

, p

) CE(c).
By monotonicity (implying the Walras Law) we have p

i
= p

e
i
= p

e
j
= p

j
. Thus for
any (0, 1), x
i
= x

i
+(1)x

j
is such that p

x
i
= p

e
i
. This renders the desired contradiction,
since
i
is strictly convex x
i
~
i
x

i
, thus contradicting to the assumption that x

i
D
p

i
(c).
Although the core allocation rule does not satisfy the equal treatment property, in the replica
economies all the clones must get the same allocation whenever agents preferences (in the original
economy) is given by strictly convex, monotone and continuous complete preorders. The following
Proposition formally establishes this observation.
4
Recall that for a given exchange economy c an allocation x X
n
is no-envy if x
i

i
x
j
for all i, j N. Moreover,
the no-envy allocation rule gives for any given economy c the set of no-envy allocations.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 106
Proposition 28 Suppose that c features agents with strictly convex, monotone and continuous
complete preorders. Then for any r N and any allocation x in the core of c
r
, we must have
x
i,h
= x
i,g
, for any h, g with 1 h, g r.
Proof. Let r N and c satisfying the assumptions of the Proposition be given. Moreover,
suppose that x is an allocation in the core of c
r
but there exists type i

for which not all the clones


get the same allocation, i.e. there exists h, g, 1 h, g r, such that x
i

,h
,= x
i

,g
. Now we are
going to consider a coalition consisting of the least happy agent of each type, and show that such
a coalition would block x.
For any type i N, let x
(i)
represent the least desired of the consumption bundles x
i,q
, q =
1, . . . , r according to the common preferences
i
. I.e. x
(i)
y
i
X : x
i,q

i
y
i
, for all q =
1, . . . , r. Then by strict convexity:
1
r
_
r

q=1
x
i,q
_

i
x
(i)
, for all i N,
with strict preference holding for i

. On the other hand,


n

i=1
1
r
_
r

q=1
x
i,q
_
=
1
r
r

q=1
_
n

i=1
x
i,q
_
=
1
r
r
n

i=1
e
i
=
n

i=1
e
i
,
thus this redistribution is feasible.
Now since preferences are continuous there exists > 0, such that
(1 )
1
r
_
r

q=1
x
i

,q
_
~
i
x
(i

)
.
This slack can be redistributed in order to reach the following: The set consisting of one consumer
of each type, each of whom receives a least preferred consumption, would block x, because using
the slack provided by agent i

all the members of this coalition can be made strictly better o.


Now it is time to present and prove the Debreu-Scarf limit result:
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 107
Theorem 20 (Debreu Scarf Limit Theorem) Suppose that c features agents with strictly
convex, monotone and continuous complete preorders; and e
i
0 for all i N. Then if x
(r)
C(c
r
)
for all r N, then x CE(c).
Proof. Let x C(c), and x
(r)
C(c
r
) for all r N. Let
i
be dened as follows:

i
= z
i
R
M
: z
i
+ e
i
~
i
x
i
. (5.10)
Clearly,
i
is non-empty, convex (since each agent has strictly convex preferences) and open (since
each agent has continuous preferences). Moreover, note that the origin 0 /
i
because otherwise
e
i
~
i
x
i
would imply that x / C(c). Let =
iN

i
. Clearly it is open, non-empty, and 0 / .
But might not be convex. Thus, we need to take the convex hull of , thus consider con().
Claim 10 0 / con().
Proof. Suppose 0 con(). Then, there exists (N) and (z
i
)
iN
where z
i

i
for all
i N such that

iN

i
z
i
= 0. Let I = i N :
i
> 0. Now select an integer k which will
eventually tend to +, and let a
k
i
be the smallest integer greater than or equal to
i
k. For each
i I we dene z
k
i
as
z
k
i
=
k
i
a
k
i
z
i
,
and observe that z
k
i
+ e
i
[e
i
, z
i
+ e
i
] and lim
k
z
k
i
+ e
i
= z
i
+ e
i
. The continuity assumption on
preferences implies that z
k
i
+ e
i
~
i
x
i
for suciently high k.
Now particularly for that integer k that we identied above, consider the coalition composed of
a
k
i
members of type i to each one of whom we assign e
i
+ z
k
i
, for each i I. As we already have
pointed out for each i I we have e
i
+ z
k
i
. Moreover, since

iI
a
k
i
(e
i
+z
k
i
) =

iI
(a
k
i
e
i
+ a
k
i
k
i
a
k
i
z
i
) =

iI
(a
k
i
e
i
+ k
i
z
i
) =

iI
a
k
i
e
i
+k

iI

i
z
i
=

iI
a
k
i
e
i
,
(since

iN

i
z
i
=

iI

i
z
i
= 0) the coalitional feasibility assumption is also satised. Thus, such
a coalition would block x
()
where = max
iI
a
k
i
, contradicting to the assumption that x
(r)
C(c
r
)
for all r N.
Economics 501-2 by Mehmet Barlo: 5 Competitive Equilibrium 108
Claim 11 p

R
m
0 such that
p

z 0, for all z con().


Proof. Since con() is a non-empty, convex and open subset of R
m
, and 0 / con(), Minkowskis
Separation Theorem, Theorem 39 from Appendix 11, applies.
Claim 12 p

CE(c).
Proof. If x

i
~
i
x
i
, then clearly (x

i
e
i
) + e
i
~
i
x
i
, thus (x

i
e
i
)
i
con(). Hence,
p

(x

i
e
i
) 0. Since every neighborhood of x
i
contains a bundle which is strictly preferred by
agent i we can do the following: Pick y
n
i
B
(1/2)
n(x
i
) such that y
n
i
~
i
x
i
. Clearly y
n
i
x
i
as n
tends to +, and since p

(y
n
i
e
i
) 0 for all n, p

(x
i
e
i
) 0. But due to feasibility we
have

iN
(x
i
e
i
) = 0, thus Walras Law is satised, i.e. p

(x
i
e
i
) = 0. Moreover, since
x
i
+q
(l)
~
i
x
i
where q
(l)
is as dened in the proof of Claim 7, we have (x
i
+q
(l)
e
i
) +e
i
~
i
x
i
, thus,
(x
i
+ q
(l)
e
i
)
i
con(). Hence, p

(x
i
+ q
(l)
e
i
) 0. And due to previous discussion we
can conclude that p

0.
The argument is virtually complete, since we have demonstrated the existence of a non-negative
price vector p

R
m
+
0 such that for every i: (1) x

i
~
i
x
i
implies p

i
p

e
i
and (2)
p

x
i
= p

e
i
. The remaining customary step in equilibrium analysis is to show that x

i
~
i
x
i
indeed
implies p

i
> p

e
i
. Since e
i
0 there is a non-negative x
0
i
strictly below the budget hyperplane.
If for some x

i
both x

i
~
i
x
i
and p

i
= p

e
i
, the points on the line segment [x
0
i
, x

i
] close enough to
x

i
would be strictly preferred to x
i
and still be strictly below the budget hyperplane, contradicting
to the previous claim. Hence, indeed x

i
~
i
x
i
implies p

i
> p

e
i
.
This completes the proof.
6 Appendix: Basic Properties of
Topological and Metric Spaces
Denitions and some parts of the following discussion is from Kolmogorov and Fomin (1970) and
Rudin (1976), to which we refer the reader for a more elaborate discussion.
Unlike the case in metric spaces, instead of introducing a metric in a given set X, we can go
about things dierently, by specifying a system of open sets in X with suitable properties. After all,
the metric is used to create open sets (neighborhoods) on which the denition of various concepts
rest.
This approach leads to the notion of a topological space, of which metric spaces are rather a
special (although very important) kind.
Denition 35 Given a set X, by a topology in X is meant a system of subsets G X, called
open sets (relative to ), with the following two properties:
1. The set X is itself and the empty set belong to ;
2. Arbitrary (nite or innite) unions

and nite intersections


n
k=1
G

k
of open sets belong
to .
Then a topological space T is T = (X, ), consisting of a set X and a topology in X.
By a closed sets of a topological space T, we mean the complements T G of the open sets G
of T. It follows from the Denition 35 and the duality principle that: (1) the space T itself and
109
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 110
the empty set are closed; and (2) Arbitrary intersections

and nite unions


n
k=1
F
k
of closed
sets of T are closed.
We may introduce the following useful denitions for such abstract spaces:
1. By a neighborhood of a point x in a topological space T we refer to any open set G T
containing x;
2. A point x T is called a contact point of a set M T if every neighborhood of x contains at
least one point of M.
3. A point x T is called a limit point of a set M T if every neighborhood of x contains
innitely many points of M.
4. The set of all contact points of a set M T is called the closure of M, and is denoted by
[M].
5.

M = M ([M
c
] M) denotes the interior of M.
Exercise 28 Prove that for any M T,
1. [M] is closed, and M, [M] and

M have the same contact points; and


2. [[M]] = [M];
3. [M N] = [M] [N];
4. [] = ;
5.

M is open.
Here are some important denitions and results from topological analysis we need to be aware
of:
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 111
Denition 36 Let (X,
x
) and (Y,
y
) be given and f : X Y be a mapping from one topological
space into another. Then we say that f is continuous on X if for all x X, given any neighborhood
V
y
of the point y = f(x), there is a neighborhood U
x
of the point x such that f(U
x
) V
y
.
Theorem 21 Let (X,
x
) and (Y,
y
) be given and f : X Y be a mapping from one topological
space into another. Then we say that f is continuous if and only if f
1
(Y

) = x X [ f(x)
Y

X is an open set in (X,


x
) for all Y

Y open in (Y,
y
).
Proof. Suppose f is continuous on X and let G be any open subset of Y . Choose any point
x f
1
(G), and let y = f(x). Then G is a neighborhood of the point y. Hence, since f is
continuous, there is a neighborhood of U
x
of x such that f(U
x
) G, i.e. U
x
f
1
(G). In other
words, every point x f
1
(G) has a neighborhood contained in f
1
(G). So f
1
(G) is open.
Conversely, suppose f
1
(G) is open whenever G Y is so. Given any point x X, let V
y
be
any neighborhood of the point y = f(x). Then clearly x f
1
(V
y
), and moreover, f
1
(V
y
) is open,
by hypothesis. Therefore U
x
= f
1
(V
y
) is a neighborhood of x such that f(U
x
) V
y
.
Corollary 1 Let (X,
x
) and (Y,
y
) be given and f : X Y be a mapping from one topological
space into another. Then we say that f is continuous if and only if f
1
(Y

) = x X [ f(x)
Y

X is a closed set in (X,


x
) for all Y

Y closed in (Y,
y
).
Denition 37 A set C T = (X, ) is compact if for all G

with G

open for all and


C (

) there exists
1
, . . . ,
k
, k < such that C
_

k
j=1
G

j
_
.
Theorem 22 Every compact set is closed. And every closed subset F of a compact topological space
T is itself compact.
Exercise 29 Prove Theorem 22.
Denition 38 A system of subsets A

of a set T is said to satisfy the nite intersection


property if every nite intersection
n
k=1
A

k
is non-empty.
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 112
Theorem 23 A topological space T is compact if and only if every closed system of subsets of T
with the nite intersection property has non-empty intersection.
Exercise 30 Prove Theorem 23.
Theorem 24 If a system of compact subsets A

of a set T satises the nite intersection prop-


erty, then

is non-empty.
Exercise 31 Prove Theorem 24.
Theorem 25 Let (X,
x
) and (Y,
y
) be given, C X be compact (with respect to
x
and f : X Y
be a continuous mapping. Then f(C) = y Y [ x C f(x) = y is compact with respect to
y
.
Exercise 32 Prove Theorem 25.
Dening a topology in a space T means specifying a system of open sets in T. However, in
many concrete problems it is more convenient to specify, instead of all the open sets, some system
of subset which uniquely determine all the open sets. For example the open sets in a metric space
are precisely those which can be represented as nite or innite unions of open balls. That leads us
to the notion of a base.
Denition 39 A family ( of open subsets of a topological space T is called a base for T if every
open set in T can be represented as a union of sets in (.
Note that the set of all open balls in a metric space R is a base for it. In particular, the set
of all open intervals is a base on the real line. And the set of all open intervals with rational end
points is also a base on the real line, since any open interval can be represented as a union of such
intervals.
Thus we have the following Theorem:
Theorem 26 A set E in R
k
is closed and bounded
1
if and only if E is compact in T = (R
k
,
R
k).
1
E R
k
is closed if E
c
T = (R
k
,
R
k) where
R
k is to be obtained by the set of all open balls; and E R
k
is
bounded if there exists a K N and x E, such that |x

x| < K for all x

E.
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 113
Exercise 33 Prove Theorem 26.
Theorem 27 If f is a continuous mapping of a compact topological space T = (X,
x
) into R
k
,
then f(X) is closed and bounded, thus f is bounded.
Exercise 34 Prove Theorem 27.
Theorem 28 Let (X,
x
) be given, C X be compact (with respect to
x
and f : X R be a
continuous mapping. Then
x C [ f(x) f(x

), x

C ,= .
Exercise 35 Prove Theorem 28.
The concept of a convergent sequence generalizes in a natural way to case of a topological space.
And sequences are essential tools in our studies especially when we deal with metric spaces.
A sequence of points x
n

nN
in a topological space T is said to converge to a point x T
(called the limit of a sequence) if for every neighborhood G(x) of x there is an integer N
G
such that
G(x) contains all points x
n
, n > N
G
.
However, as we will observe below, sequences in topological spaces do not have the important
role (in terms of the denition of closedness, compactness, continuity, ...) that it has for metric
spaces, a notion that the reader should be familiar from analysis courses.
Instead of using the more abstract setting oered by the topological spaces, we may have some
structure on the space we are interested in. Formally, if we have a notion of a distance of points,
we are able to come up with the more tangible notion of metric spaces.
Denition 40 A metric space is a pair (X, ) consisting of a set X and a distance : XX R
with the following properties: For all x, y X
1. (x, y) = 0 if and only if x = y;
2. (Symmetry) (x, y) = (y, x);
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 114
3. (Triangular Inequality) (x, z) (x, y) + (y, z).
Then a metric space R is R = (X, ), consisting of a set X and a metric in X.
With that notion of a metric we now can dene open and closed balls: For any x R let the
open sphere around x be given by
B(x, ) = x

X [ (x, x

) < .
The closed sphere is

B(x, ) = x

X [ (x, x

) .
At this point due to the scope of these lecture notes (after all this is not an analysis course) it
is sucient to remind the reader of the denition of continuity.
Denition 41 Let f : X Y be a mapping, and (X,
x
) and (Y,
y
) be two metric spaces. Then
f is said to be continuous if for all x X, given any > 0, there exists > 0 such that

y
(f(x), f(x

)) < whenever
x
(x, x

) < .
Theorem 29 Let f : X Y be a mapping, and (X,
x
) and (Y,
y
) be two metric spaces. Then f
is continuous if and only if for all x
n

nN
X, with x
n
x we have f(x
n
) f(x).
Exercise 36 Prove Theorem 29.
Note that the notion of a convergent sequence does not play the same basic role for topological
spaces as for metric ones. In fact in the case of a metric space R we have the following essential
result:
Theorem 30 A point x is a contact point of a subset M of a metric space R = (X, ) if and only
if M contains a sequence x
n

nN
M converging to x.
Exercise 37 Prove Theorem 30.
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 115
On the other hand, in the case of a topological space T, this in general is not true. The following
example 4 is about that observation. In other words, a point x can be a contact point of a set M T
without M containing a sequence of converging to x. Like was said above, convergent sequences
are given their rights back if there is a countable neighborhood base at every point x T.
Denition 42 A topological space T is said to obey the rst axiom of countability if every
point x possesses a countable neighborhood base. Moreover, T is said to obey the second axiom
of countability if T possesses a countable base.
An immediate observation is that if T satises the second axiom of countability, then it trivially
satises the rst. Moreover, by simple manipulation of corresponding denitions we have:
Remark 1 If a metric space R has a countable dense
2
subset, then R has a countable base.
The following theorem will ensure that under rst countability convergent sequences regain their
power. That means under the assumption of rst countability, essential results about continuity
and compactness are the same in spirit.
Theorem 31 If a topological space T satises the st axiom of countability, then every contact
point x of a set M T is the limit point of a convergent sequence of points in M.
Proof. Let O be a countable neighborhood base at x, consisting of sets O
n
. It can be assumed
that O
n+1
O
n
(take intersections if necessary). Let x
n
be any point of M in O
n
. Such a point
exists, since x is a contact point of M. Then, obviously, x
n
M converges to x.
The following example is to shed more light on these issues. We refer the reader to Kolmogorov
and Fomin (1970) for more details, and discussions.
Example 4 Let be the system of sets consisting of the empty set and every subset of the closed
unit interval [0, 1] obtained by deleting a nite or countable number of points from X. The curious
reader is encouraged to go through the following steps:
2
Remember that a set M R is dense in a metric space R if its closure, the sets of all contact points in M, equals
R.
Economics 501-2 by Mehmet Barlo: 6 Appendix: Basic Properties of Topological and Metric Spaces 116
1. T = (X, ) is a topological space.
2. T does not satisfy neither the second the second nor the rst axiom of countability.
3. The only convergent sequences in T are the stationary sequences, i.e. the sequences all of
whose terms are the same starting from some index n.
4. The set M = (0, 1] has the point 0 as a contact point, but contains no sequences of points
converging to 0.
7 Appendix: Continuity of
Correspondences
Let X
1
and X
2
be two metric spaces with the metric given by d
i
: X
i
X
i
R. A function f :
X
1
X
2
is continuous if for all x X
1
, and all x
n

nN
X
1
with x
n
x (i.e. lim
n
d
1
(x
n
, x) = 0),
f(x
n
) f(x) (i.e. lim
n
d
2
(f(x
n
), f(x)) = 0).
We can view correspondences (set valued functions) as functions provided that the sets are
interpreted to be points in the range of a correspondence. Thus no modication to continuity
would be required, as long as an appropriate metric is used on sets. The formalities presented
below are taken from Berge (1963). The metric to be used is the Hausdor metric.
Let A and B be two non-empty closed sets in a compact metric space X, and write
(A, B) = supd(x, B) : x A,
(B, A) = supd(y, A) : y B,
where d is such that d(x, B) = infd(x, y) : y B. The numerical function dened by
(A, B) = max(A, B), (B, A)
is called a Hausdor Metric.
Lemma 9 as dened above is a metric for the family of non-empty and closed subsets of a compact
metric space X.
117
Economics 501-2 by Mehmet Barlo: 7 Appendix: Continuity of Correspondences 118
Proof. Note that for any non-empty closed family of sets: (1) (A, B) 0. (2) (A, B) =
0 (A, B) = 0 A B and so by symmetry A = B. And A = B (A, B) = 0 and
(B, A) = 0 thus (A, B) = 0. (3) (A, B) = (B, A). (4) To prove the triangular inequality we
observe that if x A and > 0, there exist points y B and z C such that d(x, B) +(B, C)
d(x, y)+d(x, C) d(x, y)+d(y, z)2 d(x, z)2 d(x, C)2. Since this inequality holds for
all > 0 we have d(x, C) d(x, B)+(B, C), whence (A, C) = sup
xA
d(x, C) (A, B)+(B, C).
In this equality we can interchange A and C without changing the right-handside, and so (A, C)
(A, B) + (B, C).
Exercise 38 Compute the following:
1. ([0, 1], 0);
2. lim
n
([1/n, 1 + 1/n], [2/n, 1 + 2/n]);
3. (
nN
[1/n, 1 1/n], [0, 1]). Based on your answer if (
nN
[1/n, 1 + 1/n], [0, 1]) = 0 does
that mean
nN
[1/n, 1 + 1/n] = [0, 1]? Discuss why is not performing well in this example,
and how such concerns are to be eliminated.
Denition 43 Let X and Y be two metric spaces, H be the family of non-empty compact sets in Y
equipped with the Hausdor metric , F a non-empty valued, compact valued correspondence of X
into Y . Then F is a continuous correspondence F : X Y (i.e. a continuous function F : X H)
if for all x X, and all x
n

nN
X with x
n
x (i.e. lim
n
d
X
(x
n
, x) = 0), F(x
n
)

F(x) (i.e.
lim
n
(F(x
n
), F(x)) = 0).
The following denitions of continuity are easier to check in most instances, and does not require
prociency using the Hausdor metric. Moreover, these topological denitions are more general,
because they do not restrict themselves to compact metric spaces.
Denition 44 Let X and Y be two topological spaces, and F : X Y be a correspondence
mapping X into Y . We say that F is upper-hemi continuous if for all x
0
X, for each open
Economics 501-2 by Mehmet Barlo: 7 Appendix: Continuity of Correspondences 119
set G containing F(x
0
), there exists a neighborhood U
x
0
such that
x U(x
0
) F(x) G.
Denition 45 Let X and Y be two topological spaces, and F : X Y be a correspondence mapping
X into Y . We say that F is lower-hemi continuous if for all x
0
X, for each open set G with
G F(x
0
) ,= , there exists a neighborhood U
x
0
such that
x U(x
0
) F(x) G ,= .
The above topological denitions provide the most general formulation of continuity of corre-
spondences in topological spaces. On the other hand, it is very convenient to work with sequences
to check the continuity of correspondences. But since without any restrictions on the topological
structure sequences are not enough to provide a full answer to continuity, we need to identify the
condition under which convergent sequences are given their rights back. To that regard we need to
remind the reader that under the assumption of rst axiom of countability given in denition 42 we
may proceed further to obtain more tractable denitions of upper and lower hemi continuity. For
more details we refer the reader to appendix 6.
The following denition of lower-hemi continuity only requires that the topological spaces (as
the domain and range of the correspondence in question) obey the rst axiom of countability.
Denition 46 Let X and Y be topological spaces both satisfying the rst axiom of countability.
Then F is lower-hemi continuous if for all x X and all x
n

nN
X that converges to x, and
every y F(x), there exist y
n

nN
Y with y
n
F(x
n
) for all n N and y
n
y.
If the correspondence is compact-valued
1
, then we can use an alternative denition of upper-
hemi continuity with convergent subsequences, provided that X and Y are both topological spaces
satisfying the rst axiom of countability.
1
Let X and Y be two topological spaces, and F : X Y be a correspondence mapping X into Y . We say that
F is compact-valued if F(x) Y is a compact subset of Y .
Economics 501-2 by Mehmet Barlo: 7 Appendix: Continuity of Correspondences 120
Proposition 29 Let X and Y be two topological spaces satisfying the rst axiom of countability,
and F : X Y be a correspondence mapping X into Y . If F is compact-valued then F is upper-
hemi continuous at x if and only if for every sequence x
n
X and y
n
F(x
n
) there is a convergent
subsequence of y
n
with limit in F(x).
Exercise 39 Prove Proposition 29.
Now we shall see that when the range of the correspondence is compact, using denitions 44
and 46 we will have an alternative denition of continuity for correspondence.
Theorem 32 Let X and Y be two metric spaces
2
, H be the family of non-empty compact sets in
Y equipped with the Hausdor metric, F a non-empty valued correspondence of X into Y . Then
F is a single-valued continuous function of X into H if and only if it is an upper-hemi continuous
and lower-hemi continuous correspondence of X into Y .
Proof. We refer the reader to p.126 of Berge (1963).
So by the virtue of this theorem, whenever the range of a correspondence is compact and the
topological spaces as domain and range satisfy the rst axiom of countability, we can restrict
attention to denitions 29 and 46 in order to talk about the continuity of a correspondence. In fact
almost all of the applications we will have the range of the correspondences we are interested in
compact and the domain and range both will satisfy the rst axiom of countability. That led us
to an easier denition of lower-hemi continuity as displayed above in the denition 46. Now with
these additional assumptions we have even an easier denition for upper-hemi continuity than the
one given by 29.
Denition 47 Let X and Y be two topological spaces satisfying the rst axiom of countability,
and F : X Y be a correspondence mapping X into Y . We say that F has a closed graph if
Graph(F) = (x, y) X Y : y F(x) is closed. Alternatively, F has a closed graph if for all
2
Note that by remark 1 both X and Y satisfy the second, hence the rst, axiom of countability
Economics 501-2 by Mehmet Barlo: 7 Appendix: Continuity of Correspondences 121
x X and all x
n

nN
X with x
n
x, and every y
n

nN
Y with y
n
y, and y
n
F(x
n
)
for all n N, implies y F(x).
Proposition 30 Let X and Y be two topological spaces satisfying the rst axiom of countability,
and F : X Y be a non-empty valued, compact-valued correspondence mapping X into a compact
space Y . Then F is upper-hemi continuous if and only if it has a closed graph.
Exercise 40 Prove Proposition 30.
The next result will be useful whenever the correspondence in question is single valued:
Proposition 31 Let X and Y be two topological spaces satisfying the rst axiom of countability,
and F : X Y be a non-empty valued, singleton valued correspondence. Then F is a continuous
function if and only if it has a closed graph.
Exercise 41 Prove Proposition 31.
Exercise 42 Assume that F : X Y is a correspondence and Y is a compact topological space
both satisfying the rst axiom of countability. Analyze the relations between the following conditions.
That is, if there is any prove the implication relations between them. Otherwise, render explicitly
dened examples. (1) upper-hemi continuity, (2) lower-hemi continuity, (3) closed graph, and (4)
compact valued.
Exercise 43 Give explicitly dened correspondences from [0, 1] into [0, 1], which satisfy all but one
of the following assumptions: (You are asked to go over 5 cases!) (1) upper-hemi continuous, (2)
lower-hemi continuous, (3) convex valued, (4) nonempty valued, (5) compact valued. (Can you
have a upper hemi continuous correspondence be not compact valued? Prove your answer, or give
an example.)
Exercise 44 Analyze the following correspondences and determine whether or not they are (1)
upper-hemi continuous (2) lower-hemi continuous and (3) closed graph; and whether or not Propo-
sition 30 holds in those examples. Moreover for the singleton-valued correspondences, indicated by
Economics 501-2 by Mehmet Barlo: 7 Appendix: Continuity of Correspondences 122
(

), determine if they are (1) upper-semi continuous or (2) lower-semi continuous or obeying none
of those continuity denitions:
1. F : [0, 100] R
+
dened by F(x) = n for x [n, n + 1], and n 0, . . . , 100.
2. (

) F : [0, 100] R
+
dened by F(x) = n for x [n, n + 1), and n 0, . . . , 100.
3. (

) F : [0, 100] R
+
dened by F(x) = n for x (n, n + 1], and n 0, . . . , 100.
4. F : [0, 100] R
+
dened by F(x) = n for x (n, n + 1), and n 0, . . . , 100.
5. (

) F : R
+
R
+
dened by
F(x) =
_

_
1/x if x > 0
0 if x = 0
6. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
[x +
1
4
, x +
3
4
] if x = 5
7. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
(x +
1
4
, x +
3
4
) if x = 5
8. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
[x 1, x + 2] if x = 5
9. F : R
+
R
+
dened by
F(x) =
_

_
[x, x + 1] if x ,= 5
(x 1, x + 2) if x = 5
8 Appendix: Theorem of Maximum
Berges theorem of maximum as stated here is taken from Border (1985). The original version from
Berge (1963) is more general than the following version as it does not restrict attention to nite
dimensional Euclidean spaces. Moreover, the one to be found in Stokey, Lucas, and Prescott (1989),
is more restrictive because they do not allow the domain and range of the correspondence to be of
dierent dimensional Euclidean spaces. The proof of the theorem is taken from Border (1985) as
well.
Theorem 33 (Berges Theorem of Maximum, 1959) Let X R
m
and Y R
k
, let f : Y
R be a continuous function, and let F : X Y be a non-empty-valued, compact-valued and con-
tinuous correspondence. Then the function h : X R dened by h(x) = maxf(y) [ y F(x) is
continuous, and the correspondence dened by (x) = y F(x) [ f(y) = h(x) is non-empty-
valued, compact-valued, and upper-hemi-continuous correspondence.
The following result will be used in the proof of the Theorem of Maximum.
Lemma 10 Let X R
m
, Y R
k
and F, F

: X Y , and dene (F F

) : X Y by
(F F

)(x) = F(x) F

(x). Suppose (F F

)(x) ,= . If F

is closed at x and F is upper-hemi


continuous at x and F(x) is compact, then (F F

) is upper-hemi continuous at x.
Proof. Let U be an open neighborhood of F(x) F

(x), and C = F(x) U


c
. Note that C is
compact and F

(x) C = . Since F

is closed at x if y / F

(x) then we cannot have y


n
y, where
y
n
F

(x
n
) and x
n
x Thus there is a neighborhood U
y
of y and W
y
of x with F

(W
y
) U
c
y
.
123
Economics 501-2 by Mehmet Barlo: 8 Appendix: Theorem of Maximum 124
Since C is compact we can write C V
2
= U
y
1
. . . U
y
n
, so setting W
1
= W
y
1
. . . W
y
n
we
have F

(W
1
) V
c
2
. Now F(x) U V
2
, which is open and so x has a neighborhood W
2
with
F(W
2
) U V
2
, as F is upper-hemi continuous at x. Put W = W
1
W
2
. Then for z W,
(F(z) F

(x)) V
c
2
(U V
2
) U. Thus, (F F

) is upper-hemi continuous at x.
Proof of the Theorem of Maximimum. First note that since F is compact-valued, is
non-empty valued and compact valued (due to Weierstrass theorem). It suces to prove that
is closed at x, for then = F and Lemma 10 implies that is upper-hemi continuous at x.
Let x
n
x, y
n
(x
n
), y
n
y. We wish to show that y (x) and h(x
n
) h(x). Since F is
upper-hemi continuous and compact-valued y F(x). Suppose y / (x). Then there is z F(x)
with f(z) > f(y). Since F is lower-hemi continuous at x, there is a sequence z
n
z, z
n
F(x
n
).
Since z
n
z, y
n
y and f(z) > f(y), the continuity of f implies that eventually for large n,
f(z
n
) > f(y
n
), contradicting with y
n
(x
n
) for all n. Now h(x
n
) = f(y
n
) f(y) = h(x), so h is
continuous at x.
9 Appendix: Fixed Point Theorems
9.1 Brouwers Fixed Point Theorem
Theorem 34 (Brouwers Fixed Point Theorem, 1912) If C is a non-empty, convex and com-
pact subset of R
m
, m N (or homeomorphic to one), and f : C C is a continuous function,
then f has a xed point, i.e. there exists a x such that f(x) = x.
To see intuitively the argument for the proof: Suppose that C is a non-empty, convex and
compact subset of R
m
, m N. Then C is homeomorphic
1
to D

= x R

: |x| 1. Let
S
(1)
= x R

: |x| = 1 be the ( 1)-sphere. If f : D

has no xed point, let


h : D

S
(1)
be the function that maps each x to the intersection of S
(1)
with the ray from
f(x) through x. Then h is continuous, and h(x) = x for all x S
(1)
. This is intuitively impossible.
Formally the Retraction Theorem does the intuitive but complicated formal exhibition:
The Retraction Theorem: There does not exist a continuous h : D

S
(1)
with h(y) = y
for all y S
(1)
.
The following proof of Brouwers Fixed Point Theorem is taken from Border (1985).
Proof of Brouwers Fixed Point Theorem. First let us establish that it is enough to restrict
attention to continuous functions f :

. Suppose that Theorem 34 holds and C R


m
, m N
is homeomorphic to

. Let h :

C be a homeomorphism. Then h
1
f h :

is
continuous, so there exists z

with h
1
f h(z

) = z

. Set z = h(z

). Then h
1
(f(z)) = h
1
(z), so
1
A set X R
m
is homeomorphic to Y R

if there is a bijective continuous function (homeomorphism)


g : X Y such that g
1
: Y X is also continuous.
125
Economics 501-2 by Mehmet Barlo: 9 Appendix: Fixed Point Theorems 126
f(z) = z as h is a bijection.
Hence we need to show that f :

continuous implies that there is a z such that f(z) = z.


Before the proof we need the following denitions and results: A set x
0
, . . . , x
n
R
m
is anely
independent if

n
i=0

i
x
i
= 0 and

n
i=0

i
= 0 imply that
0
= . . . =
n
= 0. An open n-simplex
is the set of all strictly positive convex combinations of n + 1 element anely independent set. A
(closed) n-simplex is the convex hull of an anely independent set of n + 1 vectors. The open
simplex x
0
. . . x
n
(written without commas) is the set of all strictly positive convex combinations of
the x
i
vectors, i.e. x
0
. . . x
n
=

(x
0
, . . . , x
n
). Each x
i
is a vertex of x
0
. . . x
n
and each k-simplex
x
i
0
. . . x
i
k
is a face of x
0
. . . x
n
. For y =

n
i=0

i
x
i
con(x
0
. . . x
n
) let (y) = i :
i
> 0. If
(y) = i
0
, . . . i
k
, then y x
i
0
. . . x
i
k
. This face is called carried of y. It follows that the union of
all the faces of x
0
. . . x
n
is its closure. And for any x, z
m
, x z implies x = z.
Let (x
0
, . . . , x
n
) be simplicially subdivided. Let V denote the collection of all the vertices of
all the subsimplices. (Note that each x
i
V .) A function : V 0, . . . , n satisfying (v) (v)
is called a proper labelling of the subdivision.
Sperners Lemma. 1928 Let (x
0
, . . . , x
n
) be simplicially subdivided and properly labelled by
the function . Then there are an odd number of completely labelled subsimplices in the subdivision.
Let > 0 be given and subdivide

into subsimplices of diameter . Let V be the set of vertices


of the subdivision and dene a labelling function : V 0, . . . , as follows: For v x
i
0
. . . x
i
k
choose (v) i
0
. . . i
k
i : f
i
(v) v
i
. This intersection is non-empty, for if f
i
(v) > v
i
for all
i i
0
. . . i
k
, we would have 1 =

i=0
f
i
(v) >

k
j=0
v
i
j
=

i=0
v
i
= 1, a contradiction where the
second equality follows from v x
i
0
. . . x
i
k
.
Since so dened satises the hypothesis of Sperners Lemma, there exists a completely labelled
subsimplex. That is, there exists a simplex p
0

. . . p

such that f
i
(p
i

) p
i
,i
for each i. Letting 0
we can extract a convergent subsequence (as is compact) of simplices such that p
i

z as 0
for all i = 0, . . . , . Since f is continuous we must have f
i
(z) z
i
for all i = 0, . . . , . Thus as f(z)
and z are in

, f(z) = z.
Economics 501-2 by Mehmet Barlo: 9 Appendix: Fixed Point Theorems 127
9.2 Kakutanis Fixed Point Theorem
This theorem from Kakutani (1941) is essential in the theory of economics. It is the standard tool
to prove existence of an equilibrium.
Theorem 35 (Kakutanis Fixed Point Theorem, 1941) If C is a non-empty, convex, and
compact subset of R
m
, and F : C C is a nonempty and convex valued, upper-hemi continu-
ous correspondence, then F has a xed point, i.e. there exists x C such that x F(x).
It is proved by bootstrapping the Brouwers Fixed Point Theorem, as to be displayed below. An
important generalization is due to Glicksberg (1952)
In order to prove Theorem 35, following Border (1985), we will indeed prove a more generalized
version of it due to Cellina (1969). However, before the statement of this theorem the following
results are due:
Lemma 11 (Cellina, 1969) Let F : X Y be an upper-hemi continuous, nonempty convex and
compact valued correspondence, where X R
m
is compact and Y R
k
is convex. For > 0 dene
F

via F

(x) = con
_

zN

(x)
F(z)
_
, where ^

(x) denotes the -neighborhood around x. Then for


every > 0, there is > 0 such that Graph(F

) ^

(Graph(F)). Note that this does not say


F

(F(x)).
Proof. Suppose not, thus, there is a sequence x
n
, y
n

nN
with (x
n
, y
n
) Graph(F
1/n
) such
that dist((x
n
, y
n
), Graph(F)) > 0, where dist(x, E) = inf[x y[ : y E. Now, (x
n
, y
n
)
Graph(F
1/n
) implies y
n
F
1/n
(x
n
), thus, y
n
con
_

zN
1/n
(x
n
)
F(z)
_
. By Carath`eodorys Theorem
(Theorem 38), there exists (y
0,n
, . . . , y
k,n
)
zN
1/n
(x
n
)
F(z) such that y
n
=

k
i=0

n
i
y
i,n
with

k1
, and y
i,n
F(z
i,n
) for [z
i,n
x
n
[ < 1/n. Since X is compact and F is upper-hemi continuous,
Proposition 29 implies that we can extract convergent (sub)sequences such that x
n
x, y
i,n
y
i
,

n
i

i
, z
i,n
x, for all i, and y =

k
i=0

i
y
i
and (x, y
i
) Graph(F) for all i. Since F is
convex-valued, (x, y) Graph(F), which contradicts dist((x
n
, y
n
), Graph(F)) > 0 for all n.
Economics 501-2 by Mehmet Barlo: 9 Appendix: Fixed Point Theorems 128
Lemma 12 (von Neumanns Approximation Lemma; von Neumann (1937)) Let F : X
Y be an upper-hemi continuous, nonempty convex and compact valued correspondence, where X
R
m
is compact and Y R
k
is convex. Then for any > 0, there is a continuous function f such
that Graph(f) ^

(Graph(F)).
Proof by Hildenbrand and Kirman (1976). Due to Lemma 11, there is > 0 such that
the correspondence F

satises Graph(F

) ^

(Graph(F)). Since X is compact, there exists


x
1
, . . . , x
n
such that ^

(x
i
) is an open cover of X. Choose y
i
F(x
i
). Let f
1
, . . . , f
n
be partition
of unity subordinate to this cover and set g(x) =

n
i=1
f
i
(x)y
i
. Then g is continuous and since f
i
vanishes outside ^

(x
i
), f
i
(x) > 0 implies [x
i
x[ < so g(x) F

(x).
The following theorem (Cellinas Fixed Point Theorem) is presented not only good for proving
Kakutanis Fixed Point Theorem, but also interested economic applications can benet from the
use of Cellinas Fixed Point Theorem for purposes of proving the existence.
Theorem 36 (Cellinas Fixed Point Theorem, 1969) Let K R
m
be nonempty, compact
and convex; and let F : K K. Suppose that there is a closed-valued correspondence : K E
with nonempty compact convex values, where E R
k
is nonempty compact and convex, and a
continuous f : K E K such that for each x K, F(x) = f(x, y) : y (x). Then F has a
xed point, i.e. there is some x in K with x F(x).
Remark 2 Kakutanis xed point theorem is implied by Cellinas Fixed Point Theorem
Proof of Cellinas Fixed Point Theorem. By von Neumanns Approximation Lemma
(Lemma 12), there is a sequence of functions g
n
: K E such that Graph(g
n
) ^
1/n
(Graph()).
Dene h
n
: K K by h
n
(x) = f(x, g
n
(x)). By Brouwers Fixed Point Theorem (Theorem 34)
each h
n
has a xed point x
n
, i.e. x
n
= f(x
n
, g
n
(x
n
)). As K and E are compact we can extract
a convergent subsequence; so without loss of generality, assume x
n
x and g
n
(x
n
) y. Then
( x, y) Graph() as is closed and so x = f( x, y) F( x).
10 Appendix: Finite Normal Form
Games and Nash Equilibrium
This section is to provide the reader for the use of xed point theory in the existence of Nash
equilibrium for normal form games.
Denition 48 An n-person normal form game is =

N, (S
i
)
iN
, (u
i
)
iN
_
, where N = 1, . . . , i
is the set of agents, S
i
is a non-empty set of strategies of player i, and u
i
:
iN
S
i
R is agent
is utility or payo function.
The following will be useful for notation: s = (s
1
, . . . , s
n
) S =
iN
; and s = (s
i
, s
i
) where
s
i
S
i
=
j=i
S
j
. We will restrict attention to nite normal form games, i.e. those with [ S
i
[<
for all i N.
For any nite set X, let (X) = : X [0, 1] [

xX
(x) = 1, and is called the simplex
on X. The set of mixed strategies for agent i is (S
i
), of which
i
is a generic member. Dene,
=
iN
(S
i
), with as its generic element. Similarly, let
i
=
j=i
(S
j
) be the mixed strategy
proles of all players but i. The generic member of
i
is
i
.
Lemma 13 (S
i
), and is non-empty, convex and compact. Moreover, any pure strategy s
i
can
be represented as a mixed strategy
i
that occupies one of the corners of (S
i
).
Exercise 45 Prove Lemma 14.
129
Economics 501-2 by Mehmet Barlo: 10 Appendix: Finite Normal Form Games and Nash Equilibrium130
The expected payo for agent i at a mixed strategy prole is
u
i
() =

sS
_

jN

j
(s
j
)
_
u
i
(s) =

s
i
S
i

i
(s
i
)
_
_

s
i
S
i
_

j=i

j
(s
j
)
_
u
i
(s
i
, s
i
)
_
_
=

s
i
S
i

i
(s
i
)u
i
(s
i
,
i
).
Remark 3 u
i
:
iN
S
i
R is continuous, and linear.
The best response for player i to
i

i
is
B1
i
(
i
) =
i
(S
i
) : u
i
(
i
,
i
) u
i
(

i
,
i
), for all

i
(S
i
) .
Let B1() =
iN
B1
i
(
i
) , and is called the best response correspondence, since it can
be seen as a correspondence B1 : .
Proposition 32 Given any nite normal form game =

N, (S
i
)
iN
, (u
i
)
iN
_
, for all i N,
B1
i
:
i
(S
i
) is convex-valued. Moreover, the best response correspondence B1 :
admits the same property.
Exercise 46 Prove Proposition 32.
Now let us introduce the Nash equilibrium:
The rst known formulation of this equilibrium concept is due to Cournot (1838). Nash (1951)
proves the existence of an equilibrium mixed strategy prole in any nite normal form game.
Denition 49 A mixed strategy prole,

constitutes a Nash equilibrium of a nite normal


form game =

N, (S
i
)
iN
, (u
i
)
iN
_
if for all i N,
u(

i
,

i
) u(
i
,

i
)
for all
i
(S
i
). In other words,

is a xed point of B1 : , i.e.

B1(

).
The following famous theorem is due to Nash (1950).
Theorem 37 Every nite normal form game has a Nash equilibrium.
Economics 501-2 by Mehmet Barlo: 10 Appendix: Finite Normal Form Games and Nash Equilibrium131
In order to prove the existence of Nash equilibrium we will employ the Kakutani Fixed Point
Theorem.
First we should note that by by Proposition 32 for any nite normal form game given by
N, (S
i
)
iN
, (u
i
)
iN
), for all i N, B1
i
:
i
(S
i
) is convex-valued. Moreover, the best
response correspondence B1 : admits the same property.
Now the next task is to prove the other remaining features of the best response correspon-
dence: that it is upper-hemi continuous, and nonempty-valued. This task is obtained using Berges
Theorem of Maximum, Theorem 33.
Proposition 33 For any nite normal form game =

N, (S
i
)
iN
, (u
i
)
iN
_
, for all i N, B1
i
:

i
(S
i
) is nonempty-valued, compact-valued and upper-hemi continuous.
Proof. Letting X and Y be given by
j=i
(S
j
) R
m
and (S
i
) R
k
respectively, and
F :
j=i
(S
j
) (S
i
) be dened by F(
i
) = (S
i
). Note that due to the denition of F and
basic topological properties of (S
i
), F is a compact-valued, and nonempty-valued correspondence.
Moreover, because that F is constant correspondence and is dened as above, it is continuous (exer-
cise). Finally, note that since u
i
is a continuous (remember it is linear as well) function from (S
i
)
into R, B1
i
(
i
) = arg max

i
F(
i
)
u(
i
,
i
) = arg max

i
(S
i
)
u(
i
,
i
) satises the hypothesis
of Theorem 33. Thus, B1
i
is nonempty-valued, compact-valued and upper-hemi continuous. More-
over, since n = [N[ < , B1 : dened by B1() =
iN
B1
i
(
i
) has the same properties.
The nal step is to verify that the best response correspondence B1 : indeed satises
the hypothesis of Kakutanis Fixed Point Theorem, Theorem 35. That in fact is established by
observing that is a nonempty, convex and compact subset of some nite dimensional Euclidean
space, and Propositions 32 and 33. Thus, there exists a Nash equilibrium.
Let us present an alternative proof to Theorem 37, this time using Brouwers Fixed Point
Theorem, Theorem 34.
Economics 501-2 by Mehmet Barlo: 10 Appendix: Finite Normal Form Games and Nash Equilibrium132
Proof of Theorem 37 employing Brouwers Fixed Point Theorem. For and
s
i
S
i
let (, s
i
) = max0, u
i
(s
i
,
i
) u
i
(), and dene a better-response function for i to
be
i
: (S
i
) a continuous function, by
()(s
i
) =

i
(s
i
) +
i
(, s
i
)
1 +

i
S
i

i
(, s

i
)
.
Now we will show that
i
() =
i
if and only if
i
B1
i
(
i
). If
i
B1
i
(
i
) then

i
(, s
i
) = 0 for all s
i
S
i
, so
i
() =
i
. Conversely, if
i
() =
i
, then there is 0 such
that (, s
i
) =
i
(s
i
) for all s
i
S
i
, and if > 0, then it must be the case that all the pure
strategies receiving positive probability in
i
mush have an expected payo strictly greater than
i
,
an impossibility. Thus, = 0, thence, (, s
i
) = 0 for all s
i
S
i
implies u
i
() u
i
(s
i
,
i
for all
s
i
S
i
, consequently,
i
B1
i
(
i
).
Let () = (
1
(), . . . ,
n
()). Then, by the above results : sigma is a continuous
function whose xed points are precisely Nash equilibria. The existence of Nash equilibria now
follows from Brouwers Fixed Point Theorem, Theorem 34.
11 Appendix: Convexity and
Separating Hyperplane Theorems
Denition 50 A set X is said to be convex if for all [0, 1], x = x + (1 )x

is in X,
whenever x, x

X.
For any nite set X, let (X) = : X [0, 1] [

xX
(x) = 1, and is called the simplex
on X.
Lemma 14 For any nite set X, (X) is non-empty, convex and compact.
Exercise 47 Prove Lemma 14.
For any X R
m
, m < , let con(X) denote the convex hull of X, to be the smallest convex
set containing X, that is, the intersection of all the convex sets that contain X.
For an alternative denition of convex hulls we need to use Carath`eodorys Theorem stated
below taken from Rockafellar (1970).
A generalized m-dimensional simplex denotes a set which is the convex hull of m+1 anely
independent points.
Carath`eodorys Theorem states that to identify the convex hull of a m-dimensional set it is not
really necessary to form combinations involving more than m + 1 elements at a time. Once can
limit attention to convex combinations

J
j=1

j
x
j
where J m + 1 for some (
j
)
J
j=1

J1
and
(x
j
)
J
j=1
.
133
Economics 501-2 by Mehmet Barlo: 11 Appendix: Convexity and Separating Hyperplane Theorems 134
Theorem 38 (Carath`eodorys Theorem) Let X R
m
, m < be non-empty. Then x
con(X) if and only if x can be expressed as a convex combination of m + 1 of the points (not
necessarily distinct) in X. In fact C is the union of all the generalized d-dimensional simplices
whose corners belong to X, where d = dim(con(X)).
Proof. We refer the reader to p.155 of Rockafellar (1970).
Thus, alternatively we have the following denition as well:
con(X) =
_
y R
m
: y =
J

j=1

j
x
j
for some (
j
)
J
j=1

J1
and (x
j
)
J
j=1
X
J
_
.
It is easy to note that for any convex X R
m
, con(X) = X.
The following Theorem, known as Minkowskis Separating Hyperplane Theorem. The version
we have is presented from Hildenbrand and Kirman (1976). For other separation theorems and
related results the reader is encouraged to check Chapter 11 of Rockafellar (1970).
Theorem 39 Let X be a convex set of R

, and y /

C. Then there exists R

0 such that
for all x C
x y.
Bibliography
Arrow, K. J., and G. Debreu (1954): Existence of an Equilibrium for a Comptetitive Econ-
omy, Econometrica, 22, 265290.
Berge, C. (1963): Topological Spaces. New York: MacMillan, New York.
Border, K. (1985): Fixed Point Theorems with Applications to Economics and Game Theory.
Cambridge University Press, Cambridge.
Cellina, A. (1969): Aprroximation of Set-Valued Functions and Fixed Point Theorems, Annali
di Mathematica Pure ed Applicata, 4, 1424.
Cournot, A. A. (1838): Researches into the Mathematical Principles of the Theory of Wealth.
New York: MacMillan, 1897.
Debreu, G. (1952): A Social Equilibrium Existence Theorem, Proceedings of the National
Academy of Sciences of the U.S.A., 38, 886893.
(1959): Theory of Value. Wiley, New York.
(1983): Mathematical Economics: Twenty Papers of Gerard Debreu. Cambridge University
Press, Cambridge.
Debreu, G., and H. Scarf (1963): A Limit Theorem on the Core of an Economy, International
Economic Review, 4, 235246.
135
BIBLIOGRAPHY 136
Dubra, J., F. Maccheroni, and E. Ok (2004): Expected Utility Theory without the Com-
pleteness Axiom, Journal of Economic Theory, 115, 118133.
Edgeworth, F. Y. (1881): Mathematical Psychics. Paul Kegan, London.
Glicksberg, I. (1952): A Further Generalization of the Kakutani Fixed Point Theorem with
Applications to Nash Equilibrium Points, Proceedings of the American Mathematical Society, 3,
170174.
Hildenbrand, W., and A. Kirman (1976): Introduction to Equilibrium Analysis. Princeton
University Press, Princeton.
Kakutani, S. (1941): A Generalization of Brouwers Fixed Point Theorem, Duke University
Mathematical Journal, 8, 416427.
Kolmogorov, A., and S. Fomin (1970): Introductory Real Analysis. Dover.
Mas-Colell, A., M. Whinston, and J. Green (1995): Microeconomic Theory. Oxford Uni-
versity Press.
Nash, J. (1950): Non-Cooperative Games, Ph.D. thesis, Princeton University.
Nash, J. F. (1951): Non-Cooperative Games, Annals of Mathematics, 54, 286295.
Ok, E. (2002): Utility Representation of an Incomplete Preference Relation, Journal of Economic
Theory, 104, 429449.
Rockafellar, J. T. (1970): Convex Analysis. Princeton University Press, Princeton.
Rudin, W. (1976): Principles of Mathematical Analysis. Mc Graw Hill, 3rd edn.
Stokey, N., R. Lucas, and E. Prescott (1989): Recursive Methods in Economic Dynamics.
Harvard Press.
BIBLIOGRAPHY 137
von Neumann, J. (1937): A Model of General Economic Equilibrium, Review of Economic
Studies, 13, 19451946.
von Neumann, J., and O. Morgernstern (1944): Theory of Games and Economic Behavior.
New York: John Wiley and Sons.
Walras, L. (1874-1877): Elements dEconomie Politique Pure. Corbaz, Lausanne.

Вам также может понравиться