Micro Prelim Solutions

Joseph Steinberg
May 5, 2009
Contents
1 Fall 2008, I.1 (Werner, Pratt theorem) 2
2 Fall 2008, I.2 (Werner, Prot rationalization) 3
3 Spring 2008, I.1 (Werner, Risk) 4
4 Spring 2008, I.1 (Werner, Reavealed preference) 5
5 Fall 2007, I.1 (Werner, Homothetic preferences & Homogeneous utility functions) 6
6 Fall 2007, I.2 (Werner, Risk) 7
7 Fall 2007, II.1 (B. Allen, FWT) 8
8 Fall 2007, II.2 (B. Allen, Sonnenscheins conjecture) 9
9 Fall 2007, III.1 (Aldo, Nash equilibrium, dominance 11
10 Spring 2007, I.1 (Werner, Continuity of preferences) 14
11 Spring 2007, I.1 (Werner, Pratts theorem) 15
12 Spring 2007, II.1 (B. Allen, Generic economies) 16
13 Spring 2007, II.2 (B. Allen, FWT/Continuum of agents) 18
14 Spring 2007, III.1 (Aldo, Extensive form games) 20
15 Fall 2006, I.1 (Werner, Prot function and supply correspondence) 21
16 Fall 2006, I.2 (Werner, Expected utility) 23
17 Fall 2006, II.1 (B. Allen, Nonconvexities) 24
18 Fall 2006, II.1 (B. Allen, Generic economies) 26
19 Fall 2006, III.1 (Aldo, Perfect equilibrium) 28
20 Fall 2005, I.1 (Werner, Risk) 28
21 Fall 2005, III.1 (Aldo, Nash equilibrium) 31
1
1 Fall 2008, I.1 (Werner, Pratt theorem)
Denition. Let v be a twice-dierentiable and strictly increasing von Neumann-Morgenstern utility func-
tion. The Arrow-Pratt measure of (absolute) risk aversion is
A(x) =
v

(x)
v

(x)
.
Denition. Let v be a von Neumann-Morgenstern utility function. Given xed deterministic level of
initial consumptrion w, the risk compensation for additional risky consumption plan z with E[ z] = 0 is
(w, z, dened by
v(w (w, z)) = E[v(w + z)].
Part (a)
Theorem (Pratts theorem). Let v
1
, v
2
be two strictly increasing, C
2
, von Neumann-Morgenstern utility
functions with risk compensation
1
,
2
and absolute risk aversion measures A
1
, A
2
, respectively. Then the
following three conditions are equivalent:
(A) A
1
(x) A
2
(x) for every x R.
(B)
1
(w, z)
2
(w, z)) for every w R and every risky consumption plan z with E[ z] = 0.
(C) v
1
is a concave transformation of v
2
, i.e., v
1
(x) = f(v
2
(x)) for every x R, where f is concave and
strictly increasing.
Part (b)
(i) Proof that (A) implies (C). Suppose (A) holds. Then A
1
(x) A
2
(x), x R. Dene f by f(t) =
v
1
(v
1
2
(t)) for evert t. Then
f

(t) =
v

1
(t)(v
1
2
(t))
v

2
(v
1
2
(t))
and
f

(t) =
v

1
(x) (v

2
(x)v

1
(x))/v

2
(x)
[v

2
(x)]
2
where x = v
1
2
(t). We can rewrite this as
f

(t) = [A
2
(x) A
1
(x)]
v

1
(x)
[v

2
(x)]
2
.
Since (A) holds, f

(t) 0, t R. So f is concave, and thus (A) implies (C).

(ii) Proof that (C) implies (B). Suppose (C) holds. Let w R and let z be a risky consumption plan with
E[ z] = 0. By denition of
1
,
v
1
(w
1
(w, z)) = E[v
1
(w + z)].
Since v
1
= f(v
2
) and f is concave, Jensens inequality implies that
v
1
(w
1
(w, z)) = E[v
1
(w + z)] = E[f(v
2
(w + z))] f(E[v
2
(w + z)]).
By denition of
2
,
f(E[v
2
(w + z)]) = f(v
2
(w
2
(w, z))) = v
1
(w
2
( z)).
Combining the last two inequalities, we have
v
1
(w
1
(w, z)) v
1
(w
2
( z)).
Since v
1
is strictly increasing, it must be that
1
(w, z)
2
(w, z). Thus (C) implies (B).
2
Part (c)
Let v
1
(x) =
x
1
1
for (0, 1) and let v
2
(x) = e
x
for > 0. Then
A
1
(x) =
v

1
(x)
v

1
(x)
=
x
1
x

=
x

x
1+
=

x
.
and
A
2
(x) =
v

2
(x)
v

2
(x)
=

2
e
x
e
x
=

2
e
x
e
x
= .
Then for x >

, A
1
(x) < A
2
(x), and for x <

, A
1
(x) > A
2
(x). Thus neither v
1
nor v
2
is more risky than
the other in the sense of Pratts theorem.
2 Fall 2008, I.2 (Werner, Prot rationalization)
Part (a)
Weak axiom of prot maximization (WAPM): The set of observations p
t
, y
t

T
t=1
satises the WAPM
if
p
t
y
t
p
t
y
s
, s, t.
Part (b)
Suppose that there exists a closed, convex production set Y that prot-rationalizes the observations. Then
y
t
Y for all t and p
t
y
t
= max
yY
p
t
y for all t = 1, . . . , T. Take s, t 1, . . . , T. Then since y
t
, y
s
Y
and p
t
y
t
p
t
y, y Y , we have p
t
y
t
p
t
y
s
. Thus p
t
, y
t

T
t=1
satises the WAPM. How does Y being
closed and convex matter for this direction?
Now suppose that these observations satisfy the WAPM. Then p
t
y
t
p
t
y
s
, s, t. Let Y be the convex
hull of y
1
, . . . , y
T
, i.e.,
Y =
_
y R
n
: y =
T

t=1

t
y
t
, (
1
, . . . ,
T
)
T
_
where

T
=
_
(
1
, . . . ,
T
) :
t
0, t,
T

t=1

t
= 1
_
.
Then Y is convex by denition. Take y
n
Y such that y
n
y. Then y
n
is a convex combination of
y
1
, . . . , y
T
for all n, i.e.,
y
n
=
T

t=1

t
n
y
t
for (
1
n
, . . . ,
T
n
)
T
. Note that
T
is closed. Then (
1
n
, . . . ,
T
n
) (
1
, . . . ,
T
)
T
, which implies
that
y
n
y =
T

t=1

t
y
t
.
Then y Y , so Y is closed. We know p
t
y
t
p
t
y
s
, s, t. Then
p
t
y
t

s=1

s
(p
t
y
s
), t, (alpha
1
, . . . ,
T
)
T
.
We can rewrite this as
p
t
y
t
p
t
_

s=1

s
y
s
_
, t, (alpha
1
, . . . ,
T
)
T
.
3
This implies that
p
t
y
t
p
t
y, t, y Y.
In other words, p
t
y
t
= max
yY
p
t
y. Thus Y prot-rationalizes p
t
, y
t

T
t=1
.
3 Spring 2008, I.1 (Werner, Risk)
Part (a)
Let v : R
+
R be a von Neumann-Morgenstern utility function dened as v(x) = ( x)
2
where is
greater than the largest possible value than either y or z can take. Then v(x) is concave and nondecreasing
for all possible values of y and z. Suppose y is more risky than z. We want to show that var(y) var(z), i.e,
E[y
2
] (E[y])
2
E[z
2
] (E[z])
2
.
Since E[z] = E[y], we just need to show that
E[z
2
] E[y
2
].
Since y is more risky than z and v is continuous, nondecreasing and concave, it must be that E[v(z)]
E[v(y)], i.e.,
E[( z)
2
] E[( y)
2
].
We can rewrite this as
E[
2
+ 2z z
2
] E[
2
+ 2y y
2
].
Since E(y) = E(z), several terms cancel out and we are left with
E[z
2
] E[y
2
].
Multiplying by 1 and reversing the inequality gives us
E[y
2
] E[z
2
].
Thus var(y) var(z).
Part (b)
I assume that the support of both y and z is the closed interval [a, b]. Suppose that y is more risky than z
and z is more risky than y. Since y is more risky than z, then by denition z second-order stochastically
dominates y, i.e.,
_
w
a
F
z
(t) dt
_
w
a
F
y
(t) dt, w [a, b].
Similarly, since z is more risky than y,
_
w
a
F
y
(t) dt
_
w
a
F
z
(t) dt, w [a, b].
Then
_
w
a
F
z
(t) dt =
_
w
a
F
y
(t) dt, w [a, b].
Does this imply that F
z
(t) = F
y
(t) for all t [a, b]?
4
Part (c)
Dene the random variables y and z as
y =
_

_
2 p = 1/2
3 p = 1/4
7 p = 1/4
and
z =
_

_
1 p = 1/4
3 p = 1/4
4 p = 1/4
6 p = 1/4
Note that E[y] = E[z] = 3.5.
Let v
1
(x) = ln(x). Then v
1
is continuous, nondecreasing and concave, and we have
E[v
1
(y)] = 1/4 ln(84) > 1/4 ln(72) = E[v
1
(z)].
Then y is not more risky than z.
Now let v
2
(x) = (10x)
2
, which is also continuous, nondecreasing and concave (at least on the relevant
range). Then
E[v
2
(y)] = 46.5 < 45.5E[v
2
(z)].
Thus z is not more risky than y.
4 Spring 2008, I.1 (Werner, Reavealed preference)
Part (a)
Generalized weak axiom of revealed preference (GWARP): The set of observations p
t
, x
t

T
t=1
satises the GWARP if
p
t
x
s
p
t
x
t
p
s
x
s
p
s
x
t
, s, t.
Part (b)
Suppose that locally non-satiated utility function u rationalizes p
t
, x
t

T
t=1
. Since u rationalizes these obser-
vations, it must be that (i) for all t = 1, . . . , T, u(x
t
) u(x) for all x such that p
t
x p
t
x
t
.
Pick t 1, . . . , T and take x such that p
t
x < p
t
x
t
. Then we can nd > 0 small enough so that
for all y with |y x| < , p
t
y < p
t
x
t
. By local non-satiation of u, for each such there exists y with
|y x| < and u(y) > u(x). By (1), u(x
t
) u(y), so it must be that u(x
t
) > u(x). Thus we also have (ii)
for all t = 1, . . . , T, u(x
t
) > u(x) for all x such that p
t
x < p
t
x
t
.
Take s, t such that p
t
x
s
p
t
x
t
. Then by (i), u(x
t
) u(x
s
). Suppose that p
s
x
t
< p
s
x
s
. Then (ii)
implies that u(x
s
) > u(x
t
) which is a contradiction. Therefore it must be that p
s
x
t
p
s
x
s
. So we have
p
t
x
s
p
t
x
t
p
s
x
s
p
s
x
t
, s, t.
Thus the GWARP holds.
Part (c)
Let u(x) = 1 for all x R
n
+
, i.e., u is constant. Then for any x, y R
n
+
such that u(y) > u(x), so u is
globally satiated and thus also locally satiatied. Let n = 2 and T = 2. Let the observations of prices and
consumption bundles be as follows:
p
1
= (1, 2), x
1
= (1, 2) p
2
= (2, 1), x
2
= (2, 1).
5
Then we have
p
1
x
1
= 5
p
1
x
2
= 4
p
2
x
1
= 4
p
2
x
2
= 5
So p
1
x
2
< p
1
x
1
but p
2
x
2
> p
2
x
1
. Then the GWARP does not hold.
Part (d)
First, dene the relations R (weak) and P (strict) between observation x
t
and commodity bundle x as
p
t
x p
t
x
t
x
t
Rx
p
t
x < p
t
x
t
x
t
Px
I will write (x
t
Rx) and (x
t
Px) to denote the negations of these relations.
Theorem (Afriat). Observations p
t
, x
t

T
t=1
are rationalized by a locally non-satiatied utility function if and
only if these observations satisfy the Generalized axion of revealed preference (GARP):
x
t1
Rx
t2
, x
t2
Rx
t3
, . . . , x
tn1
Rx
tn
(x
tn
Px
t1
), t1, . . . , tn 1, . . . , T.
5 Fall 2007, I.1 (Werner, Homothetic preferences & Homogeneous
utility functions)
Denition. u() is a utility representation of _ if for all x, x

, x _ x

u(x) u(x

).
This question could be answered in two ways. First, the statement we are asked to prove is false.
Lexicographic preferences satisfy all assumptions of the question but have no utility representation at all.
The other possible answer would be to add the assumption of continuity and prove existence of a utility
representation (and then show that it is homogeneous of degree one).
First answer. Let _ be a lexicographic preference relation on R
2
+
, i.e., x _ y if x
1
y
1
or (x
1
= y
1
and
x
2
y
2
). This preference relation is clearly reexive, transitive and complete. Let x, y R
2
such that x y
and x ,= y. Then either x
1
> y
2
or (x
1
= y
2
and x
2
> y
2
). In either case, x y, so _ is strongly monotone.
Now let x, y R
2
+
such that x y. Then x
1
= y
1
and x
2
= y
2
. Let > 0. Then x
1
= y
1
and x
2
= y
2
,
so x y. Therefore _ is homothetic.
Suppose for contradiction that u : X R such that u is a utility representation of _. For each x R
+
,
we can nd a rational number r(x) such that u(x, 2) > r(x) > u(x, 1). Let x, x

R
+
such that x

> x.
Then
u(x

, 2) > r(x

) > u(x

, 1) > u(x, 2) > r(x) > u(x, 1).

Thus r(x

) > r(x), so we have a one-to-one map r : R

+
Q. This is a contradiction since R
+
is uncountable
and Q is countable. Therefore _ has no utility representation.
Second answer (additional assumption of continuity). Let _ be a reexive, transitive, complete, strictly
monotone, and continuous preference relation on R
2
+
. Suppose that there exists a utility representation
u : R
2
+
R of _ that is homogeneous of degree 1. Let x, y R
2
+
such that x y and let > 0. Since u
is a utility representation of _, u(x) = u(y). Then u(x) = u(y). Since u is homogeneous of degree one,
u(x) = u(y). This implies that x y, so _ is homothetic.
Now suppose that _ is homothetic. Let x R
l
+
. Let e = (1, 1, . . . , 1) R
++
. Then e R
l
+
R
+
.
Choose such that e x. Since _ is monotone, e _ x. Also note that 0e x, so x _ 0e.
6
Dene A
+
and A

as
A
+
= R
+
: e _ x
A

= R
+
: x _ e
(1)
Since _ is continuous, A
+
and A

are closed. A
+
and 0 A

, so both sets are also nonempty. By

completeness of _,
R
+
= A
+
A

.
Since R
+
is connected, it cannot be separated into two disjoint closed sets, so A
+
A

,= . Let (x)
A
+
A

. Then (x)e x.
Now suppose that

A
+
A

such that

,= (x). Then either

> (x) or (x) >

. If the
former case is true, then since _ is monotone,

e (x)e x. If the latter, then

e (x)e x. In
either case, it cannot be true that

A
+
A

. Therefore (x) is unique. Let y R

l
+
, and suppose that
y _ x. Then by denition of (x), we have
(y)e y _ x (x)e.
By transitivity of _, (y)e _ (x)e, and so by monotonicity of _, (y) (x).
Now suppose that (y) (x). Then by monotonicity of _, (y)e _ (x)e. By denition of (x), we
have
y (y)e _ (x)e x.
By transitivity of _, y _ x. Thus we have y _ x (y) (x), so u(x) = (x) is a utility representation
of _. Note that u : R
l
+
R
+
. For any x R
l
+
, we can write u
1
([0, (x)]) and u
1
([(x), )) as
u
1
([0, (x)]) = y R
l
+
: (x) u(y) = y R
l
+
: (x)e _ y
u
1
([(x), )) = y R
l
+
: u(y) (x) = y R
l
+
: y _ (x)e
By continuity of _, both of these sets are closed. So we have for every x R
l
+
, [0, (x)] and u
1
([0, (x)])
are closed, and [(x), ) and u
1
([(x), )) are closed. Thus by the topological denition of continuity,
u(x) = (x) is continuous. Thus we have shown that _ has a continuous utility representation.
Let > 0. By denition of u, u(x) = (x) such that (x)e x. Since _ is homothetic, (x)e x.
Using the denition of (x) from the proof of existence of u, this implies that (x) = (x). Using the
denition of u again, this gives us u(x) = u(x), so u is homothetic.
Thus we have shown that _ is homothetic if and only if it has a utility representation that is homogeneous
of degree 1.
6 Fall 2007, I.2 (Werner, Risk)
Denition. Random variable Z is more risky than random variable Y if E(Y ) = E(Z) and E[v(Y )]
E[v(Z)] for every continuous, nondecreasing and concave function v. In other words, every risk-averse agent
prefers Y to Z.
Part (a)
Note that
E(Y + Z) =
1
E(Y + Z[Y = y
1
) +
2
E(Y + Z[Y = y
2
) =
1
y
1
+
2
y
2
= E(Y ).
Let v be concave. We can write E[v(Y + z)] as
E[v(Y + Z)] =
1
E[v(Y + Z)[Y = y
1
] +
2
E[v(Y + Z)[Y = y
2
].
By Jensens inequality,
E[v(Y + Z)[Y = y
1
] v[E(Y + Z[Y = y
1
)]
7
and
E[v(Y + Z)[Y = y
2
] v[E(Y + Z[Y = y
2
)].
Thus
E[v(Y + Z)]
1
v[E(Y + Z[Y = y
1
)] +
2
v[E(Y + Z[Y = y
2
)].
Since E[Z[Y = y
1
] = E[Z[Y y
2
] = 0, this implies that
E[v(Y + Z)]
1
v(y
1
) +
2
v(y
2
) = E[v(Y )].
Therefore Y + Z is more risky than Y .
Part (b)
Given the specication of the problem, E(Y +2Z) = E(Y +Z) = E(Y ). Note that Y +Z =
1
2
Y +
1
2
Y +2Z.
Again, let v be concave. Then
E[v(Y + Z)] = E
_
v
_
1
2
Y +
1
2
Y + 2Z
__
E
_
1
2
v(Y ) +
1
2
v(Y + 2Z)
_
=
1
2
E [v(Y )] +
1
2
E [v(Y + 2Z)]

1
2
E [v(Y + Z)] +
1
2
E [v(Y + 2Z)]
where the last line comes from the result in part (a). Rearranging, we see that
1
2
E[v(Y + Z)]
1
2
E[v(Y + 2Z)]
so
E[v(Y + Z)] E[v(Y + 2Z)].
Therefore Y + 2Z is more risky that Y + Z.
7 Fall 2007, II.1 (B. Allen, FWT)
Part (a)
We need the following denitions before proceeding with the theorem:
Denition. An allocation is a vector x R
ln
.
Denition. Trader is budget set is B
i
(p, e
i
) = x
i
X
i
: p x
i
p e
i
.
Denition. Trader is demand is x
i
(p, e
i
) = x
i
B
i
(p, e
i
) : x
i
_
i
x

i
x

i
B
i
(p, e
i
).
Denition. An allocation x is feasible if x
i
X
i
for all i 1, . . . , n and

n
i=1
x

n
i=1
e
i
.
Denition. An allocation x is (strongly) Pareto optimal if it is feasible and there is no other feasible
allocation x

such that x

i
_ x
i
for all i 1, . . . , n and x

j
x
j
for some j 1, . . . , n.
Denition. A competitive equilibrium is an allocation x

and a price vector p

R
l
+
such that:
(i) x

i
X
i
for all i 1, . . . , n.
(ii) x

i
x
i
(p

, e
i
) for all i 1, . . . , n.
8
(iii) x

is feasible.
Denition. Preference relation _
i
is locally nonsatiated if for all x
i
and all > 0, x

i
such that |x

x
x
i
| <
and x

i

i
x
i
.
Theorem (First welfare theorem). Consider a pure exchange economy with l goods and n traders indexed
by i 1, . . . , n, each of which having endowment e
i
R
l
and preferences _
i
which are complete and
continuous preorders on their respective consumption sets X
i
. Let (x

, p

) be a competitive equilibrium in
this economy. Then if _
i
are locally nonsatiated for all i, x

is Pareto optimal.
Part (b)
Proof. Let (x

, p

) be a competitive equilibrium in this economy and suppose for contradiction that x

is not
Pareto optimal. Then there exists another feasible allocation y R
ln
such that y
i
_
i
x

i
for all i 1, . . . , n
and y
j

j
x

j
for at least one j. Since x

j
x
j
(p

, e
j
) it must be that y
j
, B(p

, e
j
), i.e., p

y
j
> p

e
j
.
Now suppose for contradiction that for some i, p

y
i
< p

e
i
. Then y
i
B
i
(p

, e
i
). Since x

i
x
i
(p

, e
i
),
this means that x

i
_
i
y
i
, so y
i

i
x

i
. By local nonsatiation of _
i
, for all > 0 y

i
such that |y

i
y
i
| <
and y

i

i
y
i
. Pick small enough so that p

i
< p

e
i
. Then y

i
B
i
(p

, e
i
). But continuity of _
i
implies
that y

i

i
x

i
, which contradicts the fact that x

i
x
i
(p

, e
i
). Therefore p

y
i
p

e
i
for all i.
Since p

y
i
p

e
i
for all i and p

y
j
> p

e
j
, we can see that

n
i=1
p

y
i
>

i=1
p

e
i
. Since
p

,= 0, this implies that

n
i=1
y
i
>

i=1
e
i
, so y is not feasible. This is a contradiction, so x

is Pareto
optimal.
Part (c)
(i) Convexity of preferences is not one of the assumptions of the theorem so this complication doesnt
aect the proof. Thus the theorem still holds.
(ii) As long as the LNS assumption still holds, the answer for (i) applies here as well. Note that weak
convexity is compatible with thick indierence curves (an example of preferences that are not locally
nonsatiated).
(iii) If preferences are nonsatiated but not locally nonsatiated we cannot prove that p

y
i
p

e
i
for all i.
More specically, given the supposition (for contradiction) that p

y
i
< p

e
i
, we may not be able to
nd an small enough that there exists y

i
such that |y

i
y
i
| < , y

i

i
y
i
, and p

i
< p

e
i
. Thus
the theorem may not hold with this complication.
(iv) ??? Not entirely sure ???
Part (d)
The FWT tells us that under fairly weak assumptions the set of competitive equilibrium allocations is a
subset of the set of Pareto optimal allocations. In other words, competitive equilibrium is a mechanism for
arriving at a Pareto optimal allocation. This implies that CE allocations are non-wasteful. However, it
is important to note that the FWT does not say anything about the existence of competitive equilibria. It
is not logically inconsistent to say that the FWT holds when there are zero competitive equilibria. In fact,
this is really a trival case of the theorem; if CE = , then it must be that CE PO since is a subset of
every set.
8 Fall 2007, II.2 (B. Allen, Sonnenscheins conjecture)
Part (a)
Theorem (Sonnenscheins conjecture). Dene

as

=
_
p :
p
j
p
k
, j, k
_
9
Let Z :

R
n
be a continuous function that is homogeneous of degree zero and satises Walras law.
Then there is a pure exchange economy of n consumers whose aggregate demand is represented by Z on the
domain P

.
Part (b)
(Z is a continuous function)
Lemma. The budget set B
i
(, e
i
) :

+
is a continuous (u.h.c. and l.h.c.) correspondence.
Proof. Fix e
i
R

+
. By observation, B(p, e
i
) is compact for all p

, i.e., B(, e
i
) is compact-valued on

.
Let p
n

such that p
n
p

and let x
n
X such that x
n
B(p
n
, e
i
), n. Since p
n

++
and p
n
p , > 0 such that p
n,i
, n. Then every B(p
n
, e
i
) contains only commodity bundles in
which the quantity of each commodity is nite. Thus x
n
is bounded, so there exists a subsequence x
n
k

such that x
n
k
x. We know that x
n
k
B(p
n
k
, e
i
), n
k
. Then p
n
k
x
n
k
p
n
k
e
i
. By continuity of the
dot product, p x p e
i
. Then x B( p, e
i
). Therefore B(, e
i
) is u.h.c.
Again, let p
n

such that p
n
p

. Let x B( p). We want to show that x

n
such that
x
n
x and x
n
B(p
n
, e
i
), n.
Case 1: Suppose p x < p e
i
. Then p x p e
i
= p ( x e
i
) < 0. Construct x
n
so that x
n
= x, n.
Then x
n
x and p
n
x
n
= p
n
x. Note that p ( x e
i
) is a continous function of p. Since p ( x e
i
) < 0
and p
n
p, N > 0 such that for all n N, p
n
( x e
i
) < 0. Then for n N, p
n
x < p
n
e
i
. Since
x
n
= x, n, we have p
n
x
n
< p
n
e
i
. Thus x
n
B(p
n
, e
i
), n N. Since x
n
x B( p, e
i
), B(, e
i
) is
l.h.c. in this case.
Case 2: Suppose p x = p e
i
. Construct x
n
so that x
n
=
_
pnei
pn x
_
x, n. Then
p
n
x
n
= p
n

_
p
n
e
i
p
n
x
_
x =
_
p
n
e
i
p
n
x
_
p
n
x = p
n
e
i
.
Thus x
n
B(p
n
, e
i
), n. Since p
n
p and p x = p e
i
,
x
n

_
p e
i
p x
_
x = x.
Thus B(, e
i
) is l.h.c. in this case as well.
Since B(, e
i
) is u.h.c. and l.h.c., B(, e
i
) is continuous.
Lemma. Let _
i
be a complete, continuous and monotone preorder and let e
i
R

+
. Then x
i
(, e
i
) is a
nonempty, compact-valued and u.h.c. correspondence on

.
Proof. Given the assumptions on preferences, for all i I, _
i
has a continuous utility representation u
i
:
X R. Then we can express demand x
i
(p, e
i
) as
x
i
(p, e
i
) = argmax
xB(p,ei)
u
i
(x).
Since e
i
R

+
, B(, e
i
) is nonempty and compact-valued on

. By the previous lemma, B(, e

i
) is continuous
on

. Thus the theorem of the maximum implies that x

i
(, e
i
) is nonempty, compact-valued and u.h.c. on

.
Lemma. If _
i
is strictly convex, then x
i
(p, e
i
) is single-valued.
Proof. Let x x
i
(p, e
i
) and suppose for contradiction that y x
i
(p, e
i
) such that x ,= y. Since x, y
x
i
(p, e
i
), x
i
y. Let (0, 1). Since p x p e
i
and p y p e
i
,
p e
i
(p x) + (1 )p y = p [x + (1 )y].
So x +(1 )y B(p, e
i
). By strict convexity of _
i
, x + (1 )y
i
x. This contradicts x x
i
(p, e
i
), so
it must be that x
i
(p, e
i
) is single-valued.
10
With these helpful results out of the way, we can move on to what we want to prove.
Proposition. Continuity of Z requires that _ be complete, continuous, monotone and strictly convex for
all i I.
Proof. Given the assumptions on preferences, the lemmas above imply that each consumers demand x
i
(p, e
i
)
is u.h.c. and single-valued. These two properties together are equivalent to continuity, so each consumers
demand is a continuous function. Since Z is just the sum of the consumers demand functions (minus a
constant for the endowments), Z is also a continous function.
(Z is homogeneous of degree zero)
Proposition. Z is homogeneous of degree zero
Proof. Consumer is budget set is
B
i
(p, e
i
) = x R

+
: p x p e
i
.
Since any x that satises the constraint p x p e
i
also satises (p) x (p) e
i
, B
i
(p, e
i
) = B
i
(p, e
i
).
Thus every consumers budget set is homogeneous of degree zero in prices. This implies that each consumers
demand
x
i
(p, e
i
) = x B(p, e
i
) : x _
i
y, y B(p, e
i
)
is also homogeneous of degree zero in prices. Since Z is the sum of the consumers demands minus a constant,
Z is also homogeneous of degree zero.
(Z satises Walras law)
Lemma. If a preorder _ is monotone then it is locally nonsatiated.
Proof. Let x X and let > 0. let e denote the unit vector in R

. Let y = x +

2

e. Then y x. Since _
is monotone, y x. By construction, |y x| =

2
< . Thus _ is locally nonsatiated.
Lemma. If _
i
is locally nonsatiated then p x = p e
i
, p

, x x
i
(p, e
i
).
Proof. Suppose not. Then p

and x x
i
(p, e
i
) such that p x ,= p e
i
. By denition of x
i
(p, e
i
),
p x p e
i
, so it must be that p x < p e
i
. Then > 0 such that for all y B

(x) (the open ball of radius

around x), p y < p e
i
. Then for all y B

(x), y B(p, e
i
). By local nonsatiation of _
i
, x

(x) such
that x

i
x. This contradicts x x
i
(p, e
i
), so it must be that p x = p e
i
x x
i
(p, e
i
).
Proposition. The result follows from the two lemmas above.
Part (c)
to come later
9 Fall 2007, III.1 (Aldo, Nash equilibrium, dominance
Part (a)
I will state some denitions and prove some lemmas to do this part.
Denition. The vector = (
1
(a
i
)
a
i
A
i )
iI
) is a perturbation vector if
i
(a
i
) > 0, i I, a
i
A
i
and

a
i
A
i

i
(a
i
) < 1, i I.
Denition. Given a NFG G and a perturbation vector , the perturbed game G

has the same players

utilities and action sets, but the players mixed strategies are constrained to the sets
S
i

= s
i
S
i
: s
i
(a
i
)
i
(a
i
), a
i
A
i
.
The constrained set of mixed strategy proles is S

=
iI
S
i

.
11
Denition. The set of Nash equilibria in the perturbed game G

is
NE(G

) = s S

: u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i

.
Denition. A mixed strategy prole s is a perfect equilibrium if there exists a sequence of perturbation
vectors
n
such that
i
(a
i
) 0, i I, a
i
A
i
, and there exists a sequence s
n
NE(G
n
) such that
s
n
s. In other words, a perfect equilibrium is the limit of equilibria in some sequence of perturbed games.
Lemma. In a nite game, for every perturbation vector , NE(G

) ,= .
Proof. Clearly, S

is nonempty, compact and convex. The constrained best response correspondence is

BR
i

( s) = s
i
S
i

: u
i
(s
i
, s
i
) u
i
(t
i
, s
i
), t
i
S
i

.
We just need to verify that BR
i

(s) is nonempty, closed, convex and u.h.c. for all s S

. ll this in,
same as for normal BR correspondence. Then BR

= BR
i

also has those properties, so Kakutani implies

existence of a xed point. This is basically exactly the same as proving existence of a Nash equilibrium in
an unconstrained game.
Lemma. In any nite game, the set of perfect equilibria is nonempty.
Proof. For any sequence of perturbation vectors
n
0, we can use the previous theorem to get s
n

NE(G
n
). Then s
n
is a sequence in S. And since S is compact, there exists a convergent subsequence
s
n
k
s S. Then s is a perfect equilibrium by denition.
Lemma. A perfect equilibrium is a Nash equilibrium.
Proof. Let s be a perfect equilibrium. By denition, there exists a sequence of perturbation vectors
n
0
and a sequence s
n
NE(G
n
) such that s
n
s. We want to show that s NE(G), i.e.,
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i
.
We have
u
i
( s
i
n
, s
i
n
) u
i
(t
i
, s
i
n
), i I, t
i
S
i
n
.
For any t
i
S
i
we can nd a sequence t
i
n
S
i
n
such that t
i
n
t
i
. Then we have
u
i
( s
i
n
, s
i
n
) u
i
(t
i
n
, s
i
n
)
so the conclusion follows from continuity of u
i
.
So we now know that the set of perfect equilibria is nonempty, and that every perfect equilibrium is a
Hash equilibrium.
Proposition. In a perfect equilibrium, no players assigns a positive probability to a weakly dominated pure
strategy.
Proof. Suppose for contradiction that there exists a perfect equilibrium s such that s
i
(a
i
k
) > 0 for some
weakly dominated strategy a
i
k
A
i
. Since s is a perfect equilibrium,
n
such that
1.
i
n
(a
i
) > 0, i I, a
i
A
i
, n,
2.

a
i
A
i
i
n
(a
i
) < 1, i I, n,
3.
i
n
(a
i
) 0, i I, a
i
A
i
,
and n, s
n
NE(G
n
) such that s
n
s. Then s
i
n
(a
i
k
) s
i
(a
i
k
) > 0. Since
i
n
(a
i
k
) 0, s
i
n
(a
i
k
) >
i
n
(a
i
k
)
for n suciently large. Let
n
= s
i
n
(a
i
k
)
i
n
(a
i
k
). Then
n
> 0 for n suciently large.
Let t
i
S
i
weakly dominate a
i
k
. Dene the mixed strategy r
i
n
by
r
i
n
(a
i
k
) =
i
n
(a
i
k
) +
n
t
i
(a
i
k
)
r
i
n
(a
i
j
) = s
i
n
(a
i
j
) +
n
t
i
(a
i
j
), a
i
j
,= a
i
k
12
Note that

a
i
A
i
r
i
n
(a
i
) = 1. Then
u
i
(r
i
n
, s
i
n
) u
i
( s
n
) =

aA
Pr
s
i
n
(a
i
)r
i
n
(a
i
)u
i
(a)

aA
Pr
s
i
n
(a
i
) s
i
n
(a
i
)u
i
(a)
=

aA
Pr
s
i
n
(a
i
)
_
r
i
n
(a
i
) s
i
n
(a
i
)

u
i
(a)
=

aA
Pr
s
i
n
(a
i
)
_
_
_
_

a
i
j
A
i
,a
i
j
=a
i
k

n
t
i
(a
i
j
, a
i
)
_
_
+ (
i
n
(a
i
k
) +
n
t
i
(a
i
k
) s
i
n
(a
i
k
))u
i
(a
i
k
, a
i
)
_
_
=

aA
Pr
s
i
n
(a
i
)
_
_

n
_
_

a
i
j
A
i
,a
i
j
=a
i
k
t
i
(a
i
j
, a
i
)
_
_
+
n
(t
i
(a
i
k
) 1)u
i
(a
i
k
, a
i
)
_
_
=
n

aA
Pr
s
i
n
(a
i
)
_
_
_
_

a
i
j
A
i
,a
i
j
=a
i
k
t
i
(a
i
j
, a
i
)
_
_
+ (t
i
(a
i
k
) 1)u
i
(a
i
k
, a
i
)
_
_
=
n

aA
Pr
s
i
n
(a
i
)
__

a
i
A
i
t
i
(a
i
, a
i
)
_
u
i
(a
i
k
, a
i
)
_
=
n

aA
Pr
s
i
n
(a
i
)[u
i
(t
i
, a
i
) u
i
(a
i
k
, a
i
)]
Since t
i
weakly dominates a
i
k
, u
i
(t
i
, a
i
) u
i
(a
i
k
, a
i
) for all a
i
A
i
, and b
i
A
i
such that
u
i
(t
i
, b
i
) > u
i
(a
i
k
, b
i
). Since s
n
S
n
, s
n
is completely mixed. Then Pr
s
i
n
(a
i
) > 0 for all a
i
A
i
.
Since
n
> 0, this implies that u
i
(r
i
n
, s
i
n
) > u
i
( s
n
). But this contradicts the fact that s
n
NE(G
n
).
Therefore it must be that no player plays a weakly dominated strategy with positive probability.
Thus we have shown that the set of perfect equilibria is nonempty, a perfect equilibrium is a Nash
equilibrium, and in every perfect equilibrium no player plays a weakly dominated strategy with positive
probability. QED.
Part (b)
Lemma. The following three statements are equivalent:
(A) s is a perfect equilibrium (as in denition in part (a)).
(B) There exists a sequence of completely mixed strategy proles s
n
s such that s
i
is a best response to
s
i
n
for all i I and all n suciently large.
(C) There exists a sequence
n
0 such that
n
> 0 and a corresponding sequence of
n
-perfect strategies
s
n
such that s
n
s.
Proof. We will show that A C B A. First, suppose that s is a perfect equilibrium. Then there exists
a sequence of perturbation vectors
n
0 and a corresponding sequence of perturbed-game Nash equilibria
s
n
NE(G
n
) such that s
n
s. Since I and A are nite, we can dene
n
= max
iI,a
i
A
i
i
n
(a
i
). By
construction, s
i
n
BR
i
n
( s
n
), so any pure strategy a
i
that is not a best response to s
n
will be played with
probability
i
n
(a
i
). Thus each s
n
is
n
-perfect, so A C.
Now suppose that

t satises C. Then
n
0 and a corresponding sequence

t
n
of
n
-perfect strategies
such that

t
n

t. Since I and A are nite, we can nd d > 0 such that for all a
i
support(

t
i
),

t
i
n
(a
i
) d.
So for n large enough, d >
n
, which implies that any a
i
support(

t
i
) is a best response to

t
i
n
. This in turn
implies that

t
i
is a best response to

t
i
n
. Thus C B.
Finally, suppose r satises B. Then a sequence r
n
of completely mixed strategies such that r
n
r
and r
i
is a best response to r
i
n
for all i I and all n suciently large. We want to show that r is a perfect
13
equilibrium, i.e., a sequence of perturbation vectors
n
0 and a sequence r
n
NE(G
n
) such that
r
n
r. Dene
n
by

i
n
(a
i
) =
_
1/n a
i
support( r
i
)
r
i
n
(b
i
) b
i
, support( r
i
)
Then
i
n
(a
i
) > 0, i I, a
i
A
i
, n, and for n suciently large,

a
i
A
i
i
n
(a
i
) < 1, i I, and most
importantly,
i
n
(a
i
) 0, i I, a
i
A
i
. We then have r
i
n
(a
i
) r
i
(a
i
) > 0, a
i
support( r
i
) and
r
i
n
(b
i
) =
i
n
(b
i
), b
i
, support( r
i
). By assumption, r
i
is a best response to r
i
n
for all i I and all n
suciently large. Then for all i I, r
i
n
BR
i
n
( r
i
n
), the
n
-constrained best response set. Then we have
r
i
n
NE(G
n
) for n suciently large. Therefore B A.
Proposition. If s is perfect, then s is undominated.
Proof. Suppose not. Then for some i I, t
i
S
i
such that s
i
is weakly dominated by t
i
. Then
u
i
( s
i
, a
i
) u
i
(s
i
, a
i
), a
i
A
i
and b
i
A
i
such that
u
i
( s
i
, b
i
) < u
i
(s
i
, b
i
).
Since s is a perfect equilibrium, condition (B) in the lemma above implies that there exists a sequence s
n
of
fully mixed strategies such that s
n
s and for n suciently large, s
i
BR
i
(s
n
) for all i I. In particular,
this holds for the player i who plays a weakly dominated strategy. This means that for n suciently large,
u
i
( s
i
, s
i
n
) u
i
(t
i
, s
i
n
), t
i
S
i
.
Thus for n suciently large,
u
i
( s
i
, s
i
n
) u
i
(s
i
, s
i
n
)
where s
i
is the strategy that weakly dominates s
i
. But since s
n
are fully mixed, Pr
s
i
n
(b
i
) > 0 for b
i
dened above, so we should have
u
i
( s
i
, s
i
n
) =

a
i
A
i
Pr
s
i
n
(a
i
)u
i
( s
i
, a
i
) <

a
i
A
i
Pr
s
i
n
(a
i
)u
i
(s
i
, a
i
) = u
i
(s
i
, s
i
n
).
Therefore we have a contradiction.
Again, the set of perfect equilibria is a nonempty subset of the set of all Nash equilibria. We have shown
that in every nite game there exists a perfect equilibrium in which no player uses a dominated strategy.
QED.
Part (c)
Consider the game in the gure below. Player 2 has only one action and always receives 0 payo (thus all
of his actions are best responses). However, every action a
1
n
A
1
is dominated by action a
1
n+1
, so player 1
has no best respones. Therefore there are no Nash equilibria in this game.
10 Spring 2007, I.1 (Werner, Continuity of preferences)
Denition. A preference relation _ is transitive if for all x, y, z X, x _ y and y _ z implies that x _ z.
Denition. A preference relation _ is complete if for all x, y X, x _ y, y _ x, or both.
Proof. For simplicity, let P(x) denote the preferred-to-x set y X : y _ x and let L(x) denote the lower
contour set y X : y _ x.
Suppose C1 holds. Let x X and let x
n
be the constant sequence with x
n
= x n. Consider any
sequence y
n
such that y
n
y and y
n
P(x) n. Then y
n
_ x n. By denition of x
n
, this implies
that y
n
_ x
n
n. By C1,
y = lim
n
y
n
_ lim
n
x
n
= x.
14
Player 1
Player 2
a
2
1
a
1
1
_
1, 0
_
a
1
2
_
1
1
2
, 0
_
a
1
3
_
1
3
4
, 0
_
a
1
4
_
1
7
8
, 0
_
.
.
.
.
.
.
Figure 1: An innite game
Thus y P(x), so P(x) is closed. Similarly, consider any sequence z
n
such that z
n
z and z
n
L(x) n.
Then z
n
_ x n. Again, by denition of x
n
this implies that z
n
_ x
n
n. By C1,
z = lim
n
z
n
_ lim
n
x
n
= x.
Thus z L(x), so L(x) is closed. Therefore C1 implies C2.
Now suppose C2 holds. Let x, y X and let x
n
and y
n
be sequences in X such that x
n
x,
y
n
y, and y
n
_ x
n
n. Suppose for contradiction that C1 doesnt hold, i.e., x y. By C2, P(x) is closed,
so
P(x)
c
= y X : y x
is open. Note that y P(x)
c
, so there exists an
1
> 0 such that B
1
(y), the open ball of radius
1
around
y, is a subset of P(x)
c
. Since y
n
y, N
1
> 0 such that n N
1
, y
n
B
1
(y) P(x)
c
. Then n N
1
,
y
n
x. Similarly, C2 implies L(y) is closed, so
L(y)
c
= x X : x y
is open. Again, x L(y)
c
, so there exists an
2
> 0 such that B
2
(x) L(x)
c
. Since x
n
x, N
2
> 0 such
that n N
2
, x
n
B
2
(x) L(x)
c
. Then n N
2
, x
n
y.
Let N = max(N
1
, N
2
). Then we have shown that x y
N
and x
N
y. Then x L(y
N
)
c
and
y P(x
N
)
c
. Both of these sets are open. By an open ball argument similar to the one used above, there
exists an M N such that for all n M, x
n
L(y
N
)
c
and y
n
P(x
N
)
c
. In other words, x
n
y
N
and
x
N
y
n
for all n M. Recall that by construction of x
n
and y
n
, y
n
_ x
n
n. Then y
N
_ x
N
. But
then transitivity implies that x
n
y
n
which is a contradiction. Therefore it must be that x _ y, so C2
implies C1.
11 Spring 2007, I.1 (Werner, Pratts theorem)
Denition. The risk compensation for a risky claim z such that E[z] = 0 and deterministic initial
consumption w is (w, z) that solves E[v(w + z)] = v(w (w, z)).
Denition. If v is twice-dierentiable and strictly increasing, the Arrow-Pratt measure of risk aversion is
A(w)
v

(w)
v

(w)
.
Theorem (Pratts theorem). Let v
1
, v
2
be two C
2
, strictly increasing vN-M utility functions with
1
,
2
,
and A
1
, A
2
respectively. Then the following conditions are equivalent:
(i) A
1
(w) A
2
(w) for every w R.
(ii)
1
(w, z)
2
(w, z) for every w R and every risky plan z with E[z] = 0.
15
(iii) v
1
is a concave transformation of v
2
, i.e., v
1
(w) = f(v
2
(w)) for every w R, where f is concave and
strictly increasing.
Part (i)
Dene the function v
1
: R R by v
1
(w) = v(w + w) where w 0 for all w R. Then v
1
is also a
strictly increasing, twice-dierentiable vN-M utility function. Risk compensation
1
(w, z) is dened by
E[v
1
(w + z)] = v
1
(w
1
(w, z))
which is the same as
E[v(w + w + z)] = v(w + w
1
(w, z))
so
1
(w, z) = (w + w, z). We also have
A
1
(w) =
v

1
(w)
v

1
(w)
=
v

(w + w
v

(w + w
= A(w + w).
By the Pratts theorem, A
1
(w) A(w) for all w R if and only if
1
(w, z) (w, z) for every w R
and every risky claim z with E[z] = 0. By the above results concerning A
1
(w) and
1
(w, z), this implies
that A(w + w) A(w) for every w R if and only if (w + w, z) (w, z) for every w R and every
risky claim z with E[z] = 0. Thus A is weakly decreasing in w if and only if is weakly decreasing in w for
every z with E[z] = 0.
Part (ii)
If we let w 0 instead of w 0 in the denition of v
1
in part (i), the same proof used above implies that
A is weakly increasing in w if and only if is weakly increasing in w for every z with E[z] = 0. Combined
with the original result in part (i), this implies that A is constant for all w R if and only if (w, z) is
independent of w for every z with E[z] = 0. Thus it suces to show that v(w) = e
w
is, up to an increasing
linear transformation, the only strictly-increasing, twice-dierentiable vN-M utility function for which A(w)
is constant.
Let v be a strictly increasing, twice-dierentiable vN-M utility function with A(w) = > 0 for all w R.
Then A(w) =
v

(w)
v

(w)
= . We can solve this dierential equation:

(w)
v

(w)
=
d
dw
[ln v

(w)] =
ln v

(w) = w + C
1
v

(w) = e
w+C1
v(w) =
1

e
w+C1
+ C
2
v(w) =
1

e
C1
e
w
+ C
2
So v is an increasing linear transformation of e
x
. Thus any strictly increasing, twice-dierentiable vN-M
utility function with constant risk aversion must be an increasing linear transformation of e
x
. This
completes the proof.
12 Spring 2007, II.1 (B. Allen, Generic economies)
Part (a)
Sucient conditions for Z to be a smooth function are: For all i I,
u
i
is C
2
16
u
i
is strictly dierentiably monotone (Du
i
(x) 0, x)
u
i
is strictly dierentiably concave (D
2
u
i
(x) is negative denite x)
u
i
satises the boundary condition
cly R

++
: u
i
(y) u
i
(x) R

++
= , x.
In order to guarantee that each individuals prefefences are represented by a continuous utility function, we
require that each individuals preference relation be complete, continuous, and monotone.
What about endowments?
Part (b)
The generic approach uses techniques from analysis and dierential topology to study the frequency of badly
behaved economies (those with unstable numbers of equilibria, a continuum of equilibria, etc.). It begins
with the assumption that Z is smooth (and thus implicity assumes the conditions on part (a)). We consider
aggregate excess demand as a function of both prices and endowments. To avoid dimensionality issues, we
consider aggregate excess demand in the rst 1 commodities. Let

Z(p, e) denote this function. By Walras
law, if

Z(p, e) = 0, then aggregate excess demand in the last commodity is zero as well.
Let E P denote the set of pairs of prices and endowments. Note that E P has dimension n+ 1.
It can be shown that 0 is a regular value of

Z. Dene M =

Z
1
(0), the preimage of zero under

Z, as the
equilibrium manifold. Since 0 is a regular value of

Z, the regular value theorem implies that M is a smooth
manifold of dimension m. Dene the map T : M E as the projection of M onto the space of endowments,
i.e., T(p, e) = e. We will dene a regular economy as a regular value of T. Similarly, a critical economy
is a critical value of T. It is straightforward to show that T is smooth, and thus Sards theorem implies
that the set of critical economies has Lebesgue measure zero. In other words, the set of regular economies is
generic (this result is where the name of the generic approach comes from).
After showing that the set of regular economies is generic, we can use the inverse function theorem and
the implicit function theorem to show that a regular economy has a nite number of equilibria, and on a
neighborhood of each equilibrium, the equilibrium price vector depends smoothly on the endowments. Thus
regular economies are well-behaved, in that they do not have continuous sets of equilibria, nor is the number
of equilibria unstable. In other words, the set of badly-behaved economies has measure zero.
The conclusions of the generic approach are important for several reasons. First, these results act as a
sort of rebuttal to Sonnenscheins conjecture, which states that any continuous function dened on a compact
subset of that is homogeneous of degree zero and satises Walras law represents aggregate excess demand
for some pure exchange economy. So while Sonnenscheins conjecture tells us that there exist economies
that are extraordinarily badly behaved, the generic approach tells us that the set of such economies has
measure zero, i.e., virtually all economies are well-behaved. Second, the generic approach tells us that most
economies are stable in the sense that a small perturbation of the endowment vector will have little eect
on the equilibrium price(s). Anything else?
Part (c)
Im not exactly sure what she means by the equilibrium price correspondence. Is it the projection of T
1
onto the space of prices? If so, we could say that it is a local dieomorphism around regular values, allowing
us to use the inverse function theorem. . .
Part (d)
As stated in part (b), a generic economy (here an economy is dened by the initial endowment vector)
has a nite number of equilibrium price vectors, and each one depends smoothly on the endowment vector.
Again, the economic implications of this are that virtually all economies are well-behaved, so Sonnenscheins
conjecture has very little bite.
17
Part (e)
If we just want Z to be continuous, we only require that individual preferences be complete, continuous,
monotone, and strictly convex. To ensure that Z satises Walras law, we just need individual preferences
to be locally nonsatiated. Monotonicity takes care of this.
Part (f)
If Z, and thus

Z is just C
0
, the regular value theorem no longer applies, so we cannot be sure that M is a
smooth manifold. This implies that we cannot be sure the the projection of M onto the space of prices is
smooth either.Is there something else to this? This cant be all she wants.
Part (g)
In addition to the answer for (f), Sards theorem no longer applies, so we cannot be sure that the set of
well-behaved economies is generic. In other words, we cannot say much about the properties of an arbitrary
generic economy. Again, not sure what she wants here.
13 Spring 2007, II.2 (B. Allen, FWT/Continuum of agents)
Part (a)
Theorem (First welfare theorem). Let c = (_
i
, e
i
)
iI
be a pure exchange economy with goods and n
traders indexed by I = 1, . . . , n, each of which having endowment e
i
R

+
and preferences _
i
with are
complete, continuous, and locally nonsatiatied preorders on R

+
. Then if (x

, p

) is a competitive equilibrium
in this economy, x

is Pareto optimal.
Part (b)
Proof of FWT. Let (x

, p

) be a competitive equilibrium in this economy and suppose for contradiction that

x

is not Pareto optimal. Then there exists another feasible allocation y R

n
such that y
i
_
i
x

i
for all
i I and y
j

j
x

j
for at least one j I. First, note that since x

is a competitive equilibrium, it must be

that x

i
x
i
(p

, e
i
) (trader is demand correspondence at price p

), dened as
x
i
(p

, e
i
) = x

i
B(p

, e
i
) : x

i
_
i
x
i
, x
i
B(p

, e
i
)
where B(p

, e
i
) is trader is budget set
B(p

, e
i
) = x
i
R

+
: p

x
i
p

e
i
.
Since x

j
x
j
(p

, e
j
) (trader js demand correspondence at price p

), it must be that y
j
, B(p

, e
j
)
(trader js budget set), i.e., p

y
j
> p

e
j
.
Now suppose for contradiction that for some i I, i ,= j, that p

y
i
< p

e
i
. Then y
i
B(p

, e
i
). Since
x

i
x
i
(p

, e
i
) and y
i
_
i
x

i
, this implies that y
i

i
x

i
. By local nonsatiation of _
i
, for all > 0, y

i
such
that |y

i
y
i
| < and y

i

i
y
i
. Since the dot product is a continuous function, we can pick small enough
so that p

i
< p

e
i
. Then y

i
B(p

, e
i
). Continuity of _
i
implies that y

i

i
x

i
, which contradicts the
fact that x

i
x
i
(p

, e
i
). Therefore it must be that p

y
i
p

e
i
, i.
Since p

y
i
p

e
i
, i and p

y
j
> p

e
j
, we have

iI
p

y
i
>

iI
p

e
i
.
Since p

,= 0, this implies that

iI
y
i
>

iI
e
i
.
Thus y is not feasible. This is a contradiction, so it must be that x

is Pareto optimal.
18
Part (c)
A complete preference relation is one that can compare all possible commodity bundles. This property
implies that the agent is able to determine which of two commodity bundles he prefers, regardless of what
those two bundles are.
A continuous preference relation is one that maintains its ranking between two commodity bundles when
those bundles are slightly perturbed. This property implies that the benet from each commodity does not
change drasticly when the amount of the commodity is changed slightly.
Lastly, local nonsatiation means that in each neighborhood of a commodity bundle, no matter how small
the radius, there exists another commodity bundle that is strictly better. This property implies that the
agent can always make himself better o given slightly more resources; when his choice set expands (no
matter how slightly) he will always nd a newly aordable alternative that he strictly prefers to his choice
given the smaller choice set.
Part (d)
Let the set of traders be I = [0, 1]. In this context, an allocation is a vector-valued (Lebesgue) integrable
function mapping from I to R

+
. Then the initial endowments are now represented by a function e : I R

+
,
i.e., e(i) is trader is endowment. For the sake of simple notation, I will write
_
x to denote
_
x(i) di. Each
individual traders preference relation is the same as in the discrete case. However, since we have a continuum
of agents, I will let preferences be dened as a map from I to R, which I will use to denote the space of
complete, continuous preorders on R

+
. Then trader is preference relation is _ (i). To ensure measurability
where we need it, we will require that for all allocations x, y, the set i : x(i) _ (i) y(i) I is measurable.
Budget sets and demand correspondences are also the same as before: trader is budget set is
B(p, i) = x R

+
: p x p e(i)
and his demand is
x(p, i) = x B(p, i) : x _ (i) y, y B(p, i).
Denition. A feasible allocation is an allocation x : I R

+
such that
_
x =
_
e.
Denition. An allocation x : I R

+
is Pareto optimal if another feasible allocation y : I R

+
such
that:
y(i) _ (i) x(i), i, and
S I such that y(j) _ (j) x(j), j S and S has positive measure.
Denition. A competitive equilibrium is an allocation/price vector pair (x

, p

) such that x

is feasible
and x

(i) x(p

, i), i I.
Proposition. Suppose that _ (i) is monotone for all i I (in addition to being complete and continuous
as given by the problem). If (x

, p

) is a competitive equilibrium, then x

is Pareto optimal.
Sketch of proof. Suppose not. Then there exists a feasible allocation y such that y(i) _ (i) x

(i), i and
y(j) _ (j) x

(j), j S, where S I has positive measure. By the exact same logic as in the proof of
the FTW in the context of a discrete set of traders, p

y(i) p

e(i), i and p

y(j) > p

e(j), j S.
Then since all allocations are Lebesgue integrable and S has positive measure, we should be able to write
p

_
y > p

_
e. This would imply that y is not feasible - a contradiction. Thus x

should be Pareto
optimal.
Note that if the set S had measure zero and p y(i) = p e(i), i I S, then p

_
y = p

_
e even
though p

y(j) > p

e(j), j S. Thus we need S to have positive measure in order to show that y is not
feasible.
Part (e)
Dont know much about OLG models, at least in the GE context.
19
14 Spring 2007, III.1 (Aldo, Extensive form games)
Part (a
An linear extensive form game is:
A set I = 1, . . . , n of players.
Nature.
A set X of nodes.
A set Z X of nal nodes.
An initial node o X.
Player utilities u
i
: Z R, i I.
A partial order over X where:
x y means x comes before y.
is reexive, assymetric, and transitive.
For all x X, o x.
For all x X, the set of predecessor nodes p(x) y : y x is linearly ordered by .
For all x X, the set of immediate predecessor nodes is
P(x) y p(x) : y

s.th. y ,= y

, y ,= x, z y

x.
For all x X, the set of immediate successor nodes is
S(x) y X : y x, y

s.th. y y

x.
Z = x X : S(x) = .
Player partition P = (P
1
, . . . , P
n
).
Information partitions U
i
of P
i
, i I.
Choice sets C
u
, u U
i
, i I.
such that #[p(z) u] 1 for all z Z, for all i I, and all i = u U
i
.
Part (b)
Let z Z be a nal node. Since the game is linear, for all i I and all u U
i
, #(p(z) u) 1. Note that
p(z) is unique and thus there exists a unique set of choices in
iI
(
uU
i C
u
) for the players that leads to
z. Let ( s
i
, s
i
) represent the pure strategy prole that plays these choices.
Let b
i
B
i
be a behavioral strategy for player i. Given a pure strategy s
1
of player is opponents, the
probability induced on nal node z Z is
Pr
(b
i
,s
i
)
(z) =

uU
i
b
i
u
( s
i
u
)

uU
i
s
i
u
( s
i
u
).
where s
i
u
(c) = 1 if s
i
u
= c and s
i
u
(c) = 0 if s
i
u
,= c. Dene the mixed strategy
i
as

i
(s
i
) =
i
_
(s
i
u
)
uU
i
_
=

uU
i
b
i
u
(s
i
u
), s
i
S
i
20
where s
i
S
i
is a pure strategy of player i and s
i
u
C
u
is the choice associated with that pure strategy at
information set u. Note that
i
(s
i
) 0 by construction, and

s
i
S
i

i
(s
i
) =

s
i
S
i

uU
i
b
i
u
(s
i
u
) = 1
why??? So
i
(S
i
), i.e.,
1
is indeed a mixed strategy for player i.
The probability induced by (
i
, s
i
) on z is
Pr
(b
i
,s
i
)
(z) =
i
( s
i
)

uU
i
s
i
u
( s
i
u
)
=

uU
i
b
i
u
( s
i
u
)

uU
i
s
i
u
( s
i
u
)
= Pr
(b
i
,s
i
)
(z)
So for any s
i
S
i
,
i
induces the same probability on nal nodes as b
i
. Thus
i
is equivalent to b
i
.
Part (c)
Consider the game in gure 2. There are 2 players plus nature, and player 1 has an information set u in
which he does not recall what happened previously. Let

1
= (p
1
(L, ), p
2
(L, r), p
3
(R, ), p
4
(R, r)) = (0, 1/2, 1/2, 0)
and let s
2
= A. Then the probability distribution induced on nal nodes by (
1
, s
2
) is
Pr
(
1
,s
2
)
(z
1
, z
2
, z
3
, z
4
, z
5
, z
6
) = (1/4, 1/4, 0, 1/4, 1/4, 0).
Suppose there exists a behavioral strategy b
1
that is equivalent to
1
. Then b
1
and
1
must induce the
same probability on all nal nodes when the opponent plays s
2
. Since Pr
(
1
,s
2
)
(z
1
) = 1/4, it must be
that b
1
(L) = 1/2, which means that b
1
(R) = 1/2 as well. Then since Pr
(
1
,s
2
)
(z
2
) = 1/4, it must be that
b
1
() = 1, which means that b
1
(r) = 0. But then Pr
(b
1
,s
2
)
(z
5
) = 0 which contradicts the fact that b
1
is
equivalent to
1
.
15 Fall 2006, I.1 (Werner, Prot function and supply correspon-
dence)
Denition. The prot function

is dened as

(p) = sup
yY
p y.
Denition. The supply correspondence s

is
s

(p) = y

Y : p y

p y, y Y .
Part (a)
I will make use of the theorem of the maximum so I will state it here:
Theorem. Let X R
n
and Y R
m
. Let : Y X be a nonempty, compact-valued, continuous
correspondence and let f : X Y R be a continuous function. Then
h(x) = max
y(x)
f(x, y)
is a continuous function and
g(x) = argmax
y(x)
f(x, y)
is a nonempty, compact-valued, u.h.c. correspondence.
21
Nature
1
1/2
z
1
L
1
R
z
2

z
3
r
2
1/2
1
A
z
4

z
5
r
z
6
B
u
Figure 2: A linear game that is not of perfect recall
Note that in this case, Y is closed and bounded (i.e., compact). Then since the dot product is a
continuous function it must achieve a maximum on Y , so we can replace the sup with max in the
denition of the prot function, i.e,

(p) = max
yY
p y.
We can think of Y as a nonempty, compact-valued, continuous (since its constant) correspondence of p.
Then by the theorem of the maximum,

is a continuous function of p.
To see that

is convex, let p, p

R
n
and let [0, 1]. Let p

= p + (1 )p

and let y

(p

).
Then

(p

) = p

= [p + (1 )p

] y

= [p y

] + (1 )[p

]
[max
yY
p y] + (1 )[max
yY
p

y]
=

(p) + (1 )

(p

)
Thus

is convex.
Part (b)
By the theorem of the maximum (see part (a)), s

is a u.h.c. correspondence. Since Y (the range of s

)
is compact, this implies that s

has a closed graph. We can also prove the result directly. Let p
n
R
n
such that p
n
p R
n
and let y
n
s

(p
n
) such that y
n
y. We want to show that y s

( p). Since
y
n
s

(p
n
), n,
p
n
y
n
p
n
y, y Y.
22
Since p
n
p and y
n
y, continuity of the dot product implies that
lim
n
[p
n
y
n
] = [ lim
n
p
n
] [ lim
n
y
n
] = p y p y = [ lim
n
p
n
] y = lim
n
[p
n
y], y Y.
Then y s

( p), so s

has a closed graph.

Part (c)
This is a direct result of the envelope theorem but I will prove it here. Suppose

is dierentiable at p.
This implies that s

(p) is a singleton. Let f(p, y) = p y. Then

(p) = max
yY
f(p, y) = f(p, s

(p)).
By the chain rule,
D

(p) = D
p
(f(p, y

(p))) = D
p
(f(p, y))[
y=s

(p)
+ D
y
f(p, s

(p))D
p
s

(p).
But D
y
(f(y, p) = 0 is a FOC of the maximization problem, so we have
D

(p) = D
p
(f(p, y))[
y=s

(p)
.
By construction, D
p
f(p, y) = y, so
D

(p) = s

(p).
Part (d)
Assume

and s

are dierentiable at p. By part (c), D

(p) = s

(p). Dierentiating again, we have

D
2

(p) = Ds

(p).
Since

is a convex function, it must be that D

2

(p) is positive semi-denite, i.e., all leading principal

minors of D
2

(p) must be non-negative. Then all leading principal minors of Ds

(p) must be non-negative.

This implies that
s

i
(p)
p
i
0.
16 Fall 2006, I.2 (Werner, Expected utility)
Denition. Utility function U has a state-separable representation if there exists functions v
s
: R
R, s = 1, . . . , S such that
U(c) U(c

) =
S

s=1

s
v(c
s
)
S

s=1

s
v
s
(c

s
).
Denition. Utility function U has a concave expected utility representation if there exists a concave
function v : R R such that
U(c) U(c

) E[v(c)] E[v(c

)].
Part (a)
If S 3, U has a state-separable representation if and only if it satises the sure-thing axiom, i.e.,
U(c
s
, x) U(d
s
, x) U(c
s
, y) U(d
s
, y)
for all c, d R
S
, all x, y R, and all s S, where c
s
denote the consumption plan c with c
s
replaced by
x. See http://econ.ucsb.edu/
~
mkapicka/RG/Werner-Axiomatization.pdf.
23
Part (b)
Suppose U has a state-separable representation and is risk averse with respect to . Consider...
Finish this!
17 Fall 2006, II.1 (B. Allen, Nonconvexities)
Part (a)
Denition. Trader is budget set is
B(p, e
i
) = x X
i
: p x p e
i
.
Denition. Trader is demand is
x
i
(p, e
i
) = x B(p, e
i
) : x _
i
y, y B(p, e
i
).
Denition. A competitive equilibrium in this economy is an allocation and price vector pair (x

, p

)
such that p

0,

iI
x

iI
e
i
, and x

i
x
i
(p

, e
i
), i.
Part (b)
Denition. Trader is preference relation _
i
is locally nonsatiated if for all x X
i
and all > 0, y X
i
such that y ,= x, |y x| < , and y
i
x.
Part (c)
(i) Some (or all) traders preference relations can be nonconvex. This can be interpreted as. . .
(ii) Some (or all) traders consumption sets can be nonconvex. The economic interpretation for this would
be that some goods are indivisible, e.g., cars, houses, etc.
(iii) Anything else?
Part (d)
(i) TRUE: Nonconvexity of preferences can cause a discontinuity that may prevent existence of competitive
equilibrium.
Nonconvex preferences can cause individual demand (and thus aggregate excess demand) to be a
correspondence with non-convex values rather than a continuous (single-valued) function. In this case,
equilibrium may fail to exist.
Note that nonconvex preferences cause the assumptions of all 4 of the existence theorems (very easy,
extended very easy, easy, and extended easy) we covered in class to fail. The very easy and easy
existence theorems require that aggregate excess demand be a continuous function (which requires strict
convexity of all consumers preference relations). The extended VEET and extended EET require that
aggregate excess demand be a convex-valued correspondence (which requires all consumers preferences
to be weakly convex).
The following example illustrates these concepts.
Example. Consider a pure exchange economy with two commodities, x and y, with prices p and 1p
respectively. Suppose there are two consumers, each with endowment (1, 1) and preferences represented
by the following utility functions:
u
1
(x, y) = min(x, y)
and
u
2
(x, y) = x
2
+ y
2
.
24
Note that this implies that consumer 2s preference relation is nonconvex.
Consumer 1s budget set is
B
1
(p, e
1
) = x R
2
+
: px + (1 p)y 1.
Her demand is
x

1
(p, e
1
) = argmax
(x,y)B1(p,e1)
min(x, y) = (1, 1).
Consumer 2s budget set is
B
2
(p, e
2
) = x R
2
+
: px + (1 p)y 1
and her demand is
x

2
(p, e
2
) = argmax
(x,y)B2(p,e2)
x
2
+ y
2
=
_

_
(1/p, 0) p < 1/2
(2, 0), (0, 2) p = 1/2
(0, 1/(1 p)) p > 1/2
Note that consumer 2s demand is a correspondence and x

2
(1/2, e
2
) is nonconvex.
Looking at x

1
(p, e
1
) and x

2
(p, e
2
), we can see this economy has no competitive equilibrium. If p < 1/2,
then x

1
(p, e
1
) = (1, 1) and x

2
(p, e
1
) = (1/p, 0). Thus aggregate demand for commodity x is 1+1/p > 2,
and aggregate demand for commodity y is 1 < 2. Thus Z(p) ,= 0, i.e., p < 1/2 is not a CE price vector.
Similarly, if p > 1/2, then x

1
(p, e
1
) = (1, 1) and x

2
(p, e
2
) = (0, 1/(1 p)). Then aggregate demand for
x is 1 < 2 and aggregate demand for y is 1 + 1/(1 p) > 2, so p > 1/2 is not a CE price vector.
Finally, if p = 1/2, then x

1
(p, e
1
) = (1, 1) and x

2
(p, e
2
) = (2, 0), (0, 2). Then for both elements of
x

2
(p, e
2
), aggregate excess demand will be nonzero since aggregate demand for one commodity will be
greater than 3 and aggregate demand for the other commodity will be less than 1. Thus p = 1/2 is
not a CE price vector. So for all p [0, 1], p is not a CE price vector, i.e., the economy has no CE.
(ii) TRUE: Nonconvexity of consumption sets can cause discontinuities that prevents existence of a compet-
itive equilibrium. In particular, nonconvex consumption sets lead to a discontinuous individual demand
function or nonconvex demand correspondence, depending on whether or not demand is single-valued.
These properties carry over to aggregate excess demand, which causes all four of the usual theorems
of existence of equilibrium to fail. For the VEET or the EET, discontinuous aggregate excess demand
prevents the use of Brouwers xed point theorem. For the extended versions of those theorems, non-
convexity of the aggregate excess demand correspondence prevents the application of Kakutanis xed
point theorem. The next example illustrates this problem.
Example. Consider a pure exchange economy with two commodities and two consumers, each of
which with consumption set X
i
= Z
2
+
(both commodities must be consumed in non-negative, integer
quantities). Let both consumers have endowment og (1, 1). Let both consumers have strictly convex
preferences represented by separable utility functions of the form
u
i
(x, y) = v
i
(x) + w
i
(y)
where v
i
and w
i
are strictly convex. In particular, suppose that v
i
(0) = w
i
(0), i = 1, 2,
0 < w
i
(x/2) < v
i
(x) < w
i
(x), x > 0.
If we normalize prices as usual, both consumers have the budget set
B
i
(p, e
i
) = (x, y) Z
2
+
: px + (1 p)y px.
25
For p < 1/2, both consumers will demand 2 units of x since neither can aord 2 units of y and they both
prefer 2 units of x to 1 unit of y. For p 1/2, both consumers will demand 2 units of y since they prefer
y to x for equal amounts of both commodities. Thus both consumers have the same discontinuous
demand function
x

i
(p, e
i
) =
_
(2, 0) p < 1/2
(0, 2) p 1/2
So for p < 1/2, aggregate excess demand for x is 2 and aggregate excess demand for y is 2. For
p 1/2, aggregate excess demand for x is 2 and aggregate excess demand for y is 2. Thus there is
no competitive equilibrium.
Part (e)
(i) The problems caused by nonconvex preferences for existence of competitive equilibria do in fact dis-
appear when the economy is large. To see this, rst consider the example in (d)(i), and suppose
that the economy now has two identical consumers of type 2 plus one consumer of type 1 (which
means that aggregate endowment of each commodity is 3). At p = 1/2, x

1
(p, e
1
) = (1, 1) and
x

2
(p, e
2
) = (2, 0), (0, 2). Then if we give one consumer of type 2 the allocation (2, 0) and we give the
other (0, 2), then aggregate demand for each commodity is 1 + 2 = 3. Thus aggregate excess demand
for both commodities is zero, so p = 1/2 is a CE price vector in this modied economy.
More generally, suppose we have an exchange economy with I types of consumers. Suppose type i
has nonconvex preferences. Then excess demand for this type z
i
(p) is a correspondence. In a replica
economy with R consumers of each type, the average excess demand correspondence for type i is
z
R
i
(p) =
1
F
R

r=1
z
i
(p) =
1
R
z
i1
+ . . . + z
iR
: z
ir
z
i
(p), r = 1, . . . , R.
Consider the economy of gure 3 below and let p = p (the price vector shown in the gure). We can
see that as R , z
R
1
( p +e
1
lls the entire segment between the points x
1
and x

1
. In particular, for
any [0, 1] and any R, we can nd a
R
[0, R] such that

a
R
R

1
R
.
By putting a
R
consumers x
1
and Ra
R
consumers at x

1
, we can get average consumption for type 1
of
x
R
1
=
a
R
R
x
1
+
_
1
a
R
R
_
x

1
.
By making R large enough we can get arbitrarily close to x
1
+ (1 )x

1
. We can see that for
any p 0, as R , z
R
i
(p) converges to the convex hull of z
i
(p). Therefore at the limit, the
excess aggregate demand correspondence will be convex-valued and we will be able to apply one of the
extended existence theorems to get existence of an equilibrium. In this sense, as R , the economy
must have an allocation and price vector the constitute an approximate equilibrium.
If we go even further and allow for a continuum of traders rather than a countable set, existence of
competitive equilibrium can be proven without any assumption about convexity of preferences. The
intuition for this is more or less the same as the above argument, taken to its logical conclusion when
the set of traders gets even larger than a countably innite set. See Aumann (19660) for the proof.
(ii) Nonconvexity of consumption sets can cause discontinuities that prevent existence of equilibrium re-
gardless of the size of the economy. Looking at the example in (d)(ii), we can see that regardless of
the number of agents of each type, there will never be an equilibrium.
18 Fall 2006, II.1 (B. Allen, Generic economies)
Part (a)
See the answer for Spring 2007, II.1, part (b).
26
O
1
O
2
e
x
1
x

1
_
2
_
1
Figure 3: An economy with nonconvex preferences
Part (b)
(i)
Denition. A continuous function u
i
: R

++
R is concave if for every x, y R

++
and every
[0, 1],
u
i
(x + (1 )y) u
i
(x) + (1 )u
i
(y).
(ii) For _
i
to be represented by a continuous utility function, it is necessary that _
i
be complete, transitive,
continuous, and monotone. If _
i
has these properties, we can nd a continuous utility representation
as follows.
For every x R

++
, let A
+
= R
+
: e _
i
x and let A

= R
+
: e _
i
x (where
e = (1, 1, . . . , 1) R

). Completeness of _
i
implies that A
+
A

= R
+
. Continuity of _
i
implies
that A
+
and A

are closed. Monotonicity of _

i
implies that both sets are nonempty: x e A

and x

e A
+
. Connectedness of R
+
implies that A
+
A

,= . Denote A
+
A

as u
i
(x).
Transitivity of _
i
ensures that u
i
(x) is unique, i.e., u
i
is a function. Finally, continuity of _
i
implies
that u
i
is continuous by using a topological continuity argument (continuity of _ ensures that the
preimage sets u
1
([0, u
i
(x)]) and u
1
([u
i
(x), ]) are closed for all x R

++
).
Note that it is possible for a discontinuous utility function to represent a continuous preference relation.
However, continuity of _
i
guarantees existence of a continuous utility representation.
For u
i
to be concave, we require that. . .
(iii)
Denition. Preference relation _
i
is weakly convex if for all x, y R

++
and all [0, 1],
y _
i
x y + (1 )x _
i
x.
Denition. Preference relation _
i
is convex if for all x, y R

++
and all (0, 1),
y
i
x y + (1 )x
i
x.
Denition. Preference relation _
i
is strictly convex if for all x, y R

++
and all (0, 1),
(y _
i
x AND y ,= x) y + (1 )x
i
x.
(iv) Finish this!
(v) Finish this!
(vi) Finish this!
27
19 Fall 2006, III.1 (Aldo, Perfect equilibrium)
Denition. Let G = (I, (A
i
, u
i
)
iI
) be a normal form game. A mixed strategy prole s is a perfect
equilibrium if there exists a sequence of perturbation vectors
n
such that e
i
n
(a
i
) 0, i I, a
i
A
i
,
and there exists a sequence s
n
NE(G
n
) such that s
n
s. In other words, a perfect equilibrium is the
limit of equilibria in a sequence of perturbed games G
n
that converges to the game G.
Part (a)
Let s be a perfect equilibrium. Then there exists a sequence of perturbation vectors
n
such that
i
n
(a
i
)
0, i I, a
i
A
i
and a sequence s
n
NE(G
n
) such that s
n
s. We want to show that s NE(G),
i.e.,
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i
.
Since s
n
NE(G
n
), n, we have
u
i
( s
i
n
, s
i
n
) u
i
(t
i
, s
i
n
), i I, t
i
S
i
n
where
S
i
n
= s
i
S
i
: s
i
(a
i
)
i
n
(a
i
), a
i
A
i
, i I.
Take i I and t
i
S
i
. We can nd a sequence t
i
n
S
i
n
such that t
i
n
t
i
. Then we have
u
i
( s
i
n
, s
i
n
) u
i
(t
i
n
, s
i
n
).
Since u
i
is continuous, s
n
s, and t
i
n
t
i
, we have
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
).
Since i I and t
i
S
i
were picked arbitrarily, we have shown that
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i
.
Thus s NE(G).
Part (b)
Let s
n
be a sequence of perfect equilibria such that s
n
s S. We want to show that s is also a perfect
equilibrium. Since s
n
is a perfect equilibrium for all n, there are sequences of perturbation vectors
kn
such
that
kn
0 and sequences s
kn
NE(G

kn
) of perturbed game equilibria such that s
kn
s
n
, n. Take
> 0. Then for each n we can nd s
Kn
such that | s
Kn
s
n
| < /2. Since s
n
s, N > 0 such that for
n N, | s
n
s| < /2. Then the triangle inequality implies that for n N,
| s
Kn
s| | s
Kn
s
n
| +| s
n
s| < .
Then s
Kn
s. Since s
Kn
is a sequence of pertubed game equilibria converging to s, we have shown that s
is a perfect equilibrium. Therefore the set of perfect equilibria is closed.
20 Fall 2005, I.1 (Werner, Risk)
Denition. Random variable Z is more risky than random variable X if E(X) = E(Z) and X second-
order stochastically dominates Z. Note that we have two equivalent denitions of second-order stochastic
dominance:
1. X SSD Z if
_
w
a
F
X
(t) dt
_
w
a
F
Z
(t) dt for all w [a, b].
2. X SSD Y is E[v(X)] E[v(Y )] for every continuous, nondecreasing, concave function v.
28
Part (a)
Let Z be a random variable with E(Z) = 0 and let X be a random variable such that X = w Z for some
deterministic real number w. Then X + Z = w Z + Z = w, i.e., X + Z is deterministic. Then for any
w

< w,
_
w
a
F
X+Z
(t)dt = 0 <
_
w
a
F
X
(t)dt.
Then Y does not second-order stochastically dominate X + Z, so X + Z is not more risky than X.
Alternative solution: Suppose there are two states, s = 1, 2, with respective probabilities
1
= 1/4 and

2
= 3/4. Dene the random variables (here we will think of them as risky consumption plans) X and Z as
X =
_
3 if s = 1
1 if s = 2
Z =
_
3 if s = 1
1 if s = 2
In other words, Z = X. Note that E(X) = E(Z) = 0, so E(X + Z) = 0. Then E(X) = E(X + Z). Also
note that X +Z = (0, 0), i.e., X +Z gives 0 deterministically. Let v be a continuous, nondecreasing, strictly
concave function. Then
E[v(X)] = (1/4)v(3) + (3/4)v(1) < v(0) = E[v(X + Z)].
Then X does not second-order stochastically dominate X + Z, so X + Z is not more risky than X.
Part (b)
Proposition. If X and Z are independently distributed random variables and E(Z) = 0, then X + Z is
more risky than Z.
Part (c)
Proof. Let X and Z be independently distributed random variables such that E(Z) = 0. Then E(X) =
E(X + Z). Let v be a continuous, nondecreasing, concave function. Then
E[v(X + Z)] =
_
X
_
Z
v(x + z) dF
X
(x) dF
Z
(z) (since X and Z are independently distributed)
=
_
X
__
Z
v(x + z) dF
Z
(z)
_
dF
X
(x)

_
X
_
v
__
Z
(x + z) dF
Z
(z)
__
dF
X
(x) (Jensens inequality since v is concave)
=
_
X
v (x + E(Z)) dF
X
(x)
=
_
X
v(x)dF
X
(x) (since E(Z) = 0)
= E[v(X)]
Thus E[v(X +Z)] E[v(X)], so X second-order stochastically dominates X +Z. Therefore X +Z is more
risky than X.
Fall 2005, I.1 (Werner, utility representations)
Denition. Preference relation _ is continuous if it satises the following equivalent conditions:
(i) For all convergent sequences x
n
, y
n
, x
n
_ y
n
, n limx
n
_ limy
n
.
(ii) For all x R
n
+
, the sets y R
n
+
: y _ x and z R
n
+
: x _ x are closed.
29
Denition. Preference relation _ is strictly increasing (strongly monotone) if for all x, y R
n
+
such
that x y and x ,= y, x y.
Denition. Function u is a utility representation for preference relation _ if for all x, y R
n
+
,
u(x) u(y) x _ y.
Part (a)
Let x R
n
+
and dene the sets A
+
and A

as
A
+
= u R
+
: ue _ x
and
A

= u R
+
: x _ ue
where e = (1, 1, . . . , 1) R
n
+
. Since x R
+
, x 0e = (0, 0, . . . , 0). By strict monotonicity, x _ 0e.
Then 0e A

. Similarly, take M = max(x

1
, . . . , x
n
). Then Me x, so by strict monotonicity, Me _ x.
Then Me A
+
. Thus A
+
and A

are both nonempty. By continuity of _, A

+
and A

are closed. By
completeness of _, A
+
A

= R
+
. Since R
+
is connected, it cannot be the union of two nonempty, closed,
disjoint sets. Thus A
+
A

,= . Then u(x) A
+
A

, i.e., u(x)e x.
Suppose that u(x) is not unique. Then v(x) R
+
such that v(x) ,= u(x) and v(x)e x. Since
u(x)e x and v(x)e x, transitivity of _ implies that u(x)e v(x)e. But since v(x) ,= u(x), either
u(x) > v(x) or u(x) < v(x). If u(x) > v(x), then u(x)e v(x)e, so strict monotonicity of _ implies that
u(x)e v(x)e. This is a contradiction. Similarly, if u(x) < v(x), then by the same reasoning we have
u(x)e v(x)e which is also a contradiction. Therefore it must be that u(x) is the unique element of R
+
such that u(x)e x. This implies that u(x) is a well-dened function on R
n
+
.
To see that u(x) is a utility representation of _, take x, y R
n
+
. Suppose that x _ y. Then u(x)e x _
y u(y)e, so transitivity of _ implies that u(x)e _ u(y)e. By strict monotonicity of _, u(x) u(y). Now
suppose that u(x) u(y). Then by strict monotonicity, u(x)e _ u(y)e. Then we have x u(x)e _ u(y)e y,
so transitivity of _ implies that x _ y. Therefore x _ y u(x) u(y), so u is a utility representation
of _.
Part (b)
Let n = 2 and dene the preference relation _ as
x y x
1
= y
1
and
x y x
1
< y
1
.
Take two sequences x
n
and y
n
such that x
n
x, y
n
y, and x
n
_ y
n
, n. Then x
n1
y
n1
, n. Since
x
n
x and y
n
y, x
n1
x
1
and y
n1
y
n
. This implies that x
1
y
1
, so x _ y. Then _ is continuous.
However, take x = (2, 2) and y = (1, 1). Clearly, x > y but x y. Then _ is not (strictly) monotone.
Let x = (1, 1) and let y = (2, 2). Then x y. Note that e = (1, 1), the unit vector in R
2
+
. Then
1e = (1, 1) x u(x) = 1 and 2e = (2, 2) y u(y) = 2. But then we have x y and u(x) < u(y), so u
is not a utility representation of _.
Part (c)
Let n = 2 and let _ be the lexicographic preference relation on R
2
+
dened as
x y x = y
and
x y x
1
> y
1
OR (x
1
= y
1
AND x
2
> y
2
).
30
Note that for all n > 0, x
n
= (1
1
n
, 1) (1, 0) = y but x = (1, 1) (1, 0) = y. Therefore _ is not
continuous.
Suppose that _ has a utility representation u. Then for each x R
+
, we can nd a rational number
r(x) such that u(x, 2) > r(x) > u(x, 1). Let x, x

R
+
such that x > x

. Then
u(x, 2) > r(x) > u(x, 1) > u(x

, 2) > r(x

) > u(x

, 1)
so r(x) > r(x

). But this implies that r : Q

+
R
+
is one-to-one. This is a contradiction since Q
+
is countable while R
+
is uncountable. Therefore _ has no utility representation. This implies that u(x)
specied in the problem cannot be a utility representation for _.
21 Fall 2005, III.1 (Aldo, Nash equilibrium)
Part (a)
Note that for a correspondence f : S S with compact range and closed values, the closed graph condition
is equivalent to upper hemicontinuity.
Theorem (Kakutanis xed point theorem). Suppose S R
n
is nonempty, compact and convex, and the
correspondence f : S S is nonempty-valued, closed-valued, convex-valued and upper hemicontinuous. Then
f has a xed point in S.
Part (b)
Denition (Nash equilibrium). Let S
i
denote the set of player is mixed strategies, i.e., S
i
= (A
i
). A
Nash equilibrium is a mixed strategy prole s S =
iI
S
i
such that s
i
is a best response to s
i
for all
i I, i.e.,
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i
.
Part (c)
First, note that by denition, S
i
= (A
i
), so S
i
is clearly nonempty, compact and convex. Dene player
is best response correspondence BR
i
as
BR
i
(s) = t
i
S
i
: u
i
(t
i
, s
i
) u
i
(r
i
, s
i
), r
i
S
i
.
Note that BR
i
: S S
i
. The rst step of the proof is to show that BR
i
is nonempty-valued, closed-
valued, compact-valued and u.h.c.
To see that BR
i
is nonempty-valued, note that we can write it as
BR
i
(s) = argmax
t
i
S
i
u
i
(t
i
, s
i
).
Since S
i
is compact and u
i
is continuous, BR
i
(s) ,= , i.e., BR
i
is nonempty-valued.
To see that it is closed-valued, let t
i
n
BR
i
(s) such that t
i
n
t
i
S
i
. We want to show that
t
i
BR
i
(s). Since t
i
n
BR
i
(s),
u
i
(t
i
n
, s
i
) u
i
(r
i
, s
i
), r
i
S
i
, n.
Continuity of u
i
implies that
u
i
(t
i
, s
i
) u
i
(r
i
, s
i
), r
i
S
i
.
Then t
i
BR
i
(s), so BR
i
is closed-valued.
To see that it is convex-valued, let t
i
, r
i
BR
i
(s). Then it must be that
u
i
(t
i
, s
i
) = u
i
(r
i
, s
i
) u
i
(q
i
, s
i
), q
i
S
i
.
31
By denition of a mixed strategy, we can write this as

a
i
A
i
t
i
(a
i
)u
i
(a
i
, s
i
) =

a
i
A
i
r
i
(a
i
)u
i
(a
i
, s
i
)

a
i
A
i
q
i
(a
i
)u
i
(a
i
, s
i
), q
i
S
i
.
Let [0, 1]. Then

a
i
A
i
t
i
(a
i
)u
i
(a
i
, s
i
) +

a
i
A
i
(1 )r
i
(a
i
)u
i
(a
i
, s
i
)

a
i
A
i
q
i
(a
i
)u
i
(a
i
, s
i
), q
i
S
i
or simply

a
i
A
i
[t
i
(a
i
) + (1 )r
i
(a
i
)]u
i
(a
i
, s
i
)

a
i
A
i
q
i
(a
i
)u
i
(a
i
, s
i
), q
i
S
i
Then
u
i
(t
i
+ (1 )r
i
, s
i
) u
i
(q
i
, s
i
), q
i
S
i
.
This implies that t
i
+ (1 )r
i
BR
i
(s), so BR
i
is convex-valued.
Finally, to see that BR
i
is u.h.c., note that since BR
i
is closed-valued and has compact range (S), it
suces to show that BR
i
has a closed graph. Let s
n
S such that s
j
n s S. Let t
i
n
S
i
such that
t
i
n
BR
i
(s
n
), n and t
i
n
t
i
S
i
. We just need to show that t
i
BR
i
(s). Since t
i
n
BR
i
(s), n,
u
i
(t
i
n
, s
i
n
) u
i
(r
i
, s
i
n
), r
i
S
i
, n.
By continuity of u
i
, we have
u
i
(t
i
, s
i
) = u
i
_
lim
n
t
i
n
, lim
n
s
i
n
_
= lim
n
[u
i
(t
i
n
, s
i
n
)] lim
n
[u
i
(r
i
, s
i
n
)] = u
i
_
r
i
, lim
n
s
i
n
_
= u
i
(r
i
, s
i
), r
i
S
i
.
Then t
i
BR
i
(s), so BR
i
has a closed graph and is therefore u.h.c.
Dene the correspondence BR : S S as
BR(s) =
iI
BR
i
(s).
Since BR
i
is nonempty-valued, closed-valued, convex-valued and u.h.c. for all i I, BR also has
those properties since it is just the Cartesian product of the #I correspondences BR
i
. Since S is nonempty,
compact and convex, S and BR satisfy the assumptions of Kakutanis xed point theorem. Therefore BR
has a xed point s S. By denition of BR,
u
i
( s
i
, s
i
) u
i
(t
i
, s
i
), i I, t
i
S
i
.
Thus s is a Nash equilibrium. Since s exists, G has a Nash equilibrium.
32