Sie sind auf Seite 1von 71

EC901 M ICROECONOMIC T HEORY

(For teaching only - Do not cite)


Andres Carvajal 1
Fall term, 2008 - 09

1 E-mail

address: a.m.carvajal@warwick.ac.uk.

C ONSUMER T HEORY

Human beings are complicated objects, and human behavior is difficult to model.
Here we consider the problem of a decision-maker who has to choose a bundle of
commodities, subject to a budgetary constraint. The decision-maker could be a person, or a group of people (for instance a family); for simplicity we will treat the
decision-maker as a person. In order to make this problem tractable, we will abstract from the problems of what commodities are available (we take them as given),
and will model the person through two elements: what she wants, and what she can
do.

1.1

C ONSUMPTION S PACE AND P REFERENCES

We consider a situation in which a person is to choose a bundle of L commodities.


These commodities are perfectly divisible and can be consumed in any nonnegative
amount: the consumption set is the nonnegative orthant RL+ , so a bundle of commodities is x = (x1 , . . . , xL ), where each xl represents the number of units of commodity
l that make part of the bundle. We take the facts that these commodities exist as
exogenous.
The first element in our model of the person is what she wants. For us, the individuals preferences are subjective judgments about the relative desirability of bundles: given two bundles, preferences are defined by her answer to the question is the
first bundle at least as good as the second one? Formally, then, the decision-makers
preferences are a binary relation % defined on the consumption set: given a pair of
commodity bundles x and x0 , we write x % x0 if, according to the persons tastes,
x is at least as good as x0 .1 We also take the persons preferences as exogenous, in
the sense that we do not explain where they come from. Instead, we concentrate on
the problem of studying the individuals behavior given her preferences, under the
assumption that these preferences will not be affected by the persons choices.
We start by studying properties that the individuals preferences may (but need
not) satisfy. We first study properties of a binary relation under which it makes sense
to identify this relation with someones preferences.
D EFINITION . We say that a binary relation % is
1. complete if for any x and any x0 , either x % x0 or x0 % x;
1

L
Formally, % can be seen as a subset of RL
+ R+ .

2. reflexive if for any x, x % x;


3. transitive if x % x0 and x0 % x00 imply x % x00 ; and
4. rational if it is complete, reflexive and transitive.
Consumers with incomplete preferences may find instances in which they are unable to choose: they are simply unable to make a value judgments about the relative
(subjective) quality of two bundle. Reflexivity is consistent with our interpretation
of weak preference. Consumers with nontransitive preferences are open to full rent
extraction, as a person could find a cycle of bundles for which the person is willing
to pay a positive premium at each step. In economics, one usually assumes that the
decision-maker under consideration has rational preferences, although in some cases
(e.g. very complicated problems) it may be reasonable to consider that in individuals
preferences are incomplete; also, some cases of nontransitive preferences are sometimes observed in real life. In any case, from now on, we fix a rational binary relation
%, and define the following (induced) binary relations on the consumption set: (i)
the strict preference relation , by saying x  x0 if it is not true that x0 % x; and the
indifference relation , by saying x x0 if it is true that x % x0 and that x0 % x.
E XERCISE 1.1. Argue that  is transitive, but not reflexive, and that is reflexive and
transitive. Could these relations be complete? Could they be rational?
A second set of properties studies whether our consumer likes the commodities
available, in the sense that the more she consumes them the happier she is.
D EFINITION . We say that a binary relation % is
1. strictly monotone if x > x0 implies x  x0 ;
2. monotone if x  x0 implies x  x0 ;
3. locally nonsatiated if for every x and every  > 0, one can find x0 with ||x x0 || < 
and x0  x.
The first property captures the case when all goods, one by one, are good for the
individual: with strictly monotone preferences, getting more of any one commodity
improves the bundle. With monotone preferences, getting more of all commodities
improves the bundle. Local nonsatiation does not capture that the commodities are

good,2 but it implies that the individual will not have bliss points, in the sense that
any bundle can be improved, even with a small perturbation. In particular, the assumption of locally nonsatiated preferences rules out thick indifference curves.
E XERCISE 1.2. Argue that strict monotonicity implies monotonicity and that monotonicity
implies local nonsatiation. Does monotonicity imply strict monotonicity? Does local nonsatiation imply monotonicity or strict monotonicity? Does strict monotonicity imply local
nonsatiation?
A third group of properties studies whether our consumer likes to combine commodities in bundles.
D EFINITION . We say that a binary relation % is
1. convex if for any bundle x, any bundle x0 such that x % x0 , and any scalar 0 1,
it is true that x + (1 )x0 % x0 ;
2. strongly convex if for any bundle x, any bundle x0 6= x such that x % x0 , and any
scalar 0 < < 1, it is true that x + (1 )x0  x0 .
Convex preferences favor balanced bundles, in the sense that if the individual
has two bundles that have different composition but make her equally happy, then
she would not be worse off with a third bundle that just combined (took an average)
of them . Strongly convex preferences do too, in a strong sense: if the two original
bundles were different, then the combination is considered to be strictly better by
the consumer. The indifference map of a convex binary relation has the usual shape,
whereas if the relation is strongly convex, then its indifference curves cannot have
straight portions.
E XERCISE 1.3. Argue that strong convexity implies convexity. Does convexity imply strong
convexity?
It is most usual in economics to represent a decision-makers preferences by a
function that gives a higher value the more the person likes a bundle.
D EFINITION . We say that a binary relation % is
1. represented by function u : RL+ R if u(x) u(x0 ) occurs if, and only if, x % x0 ;
2

Except in the very weak sense that it cannot be that not getting anything of any commodity is the

best that can happen to the consumer

2. representable if there is some u : RL+ R that represents it.


The function u that represents % is called utility function. Notice that if a preference relation is representable, then there are infinitely many different utility functions
that represent it. All these representations will have the same contour sets (i.e. the
same ordinal information), but may give nontrivially different utility levels (i.e. different cardinal information). It is for this reason that interpersonal comparisons of
utility are problematic.
E XERCISE 1.4. Argue that representability implies rationality. Does rationality imply representability? What property must u satisfy if it represents a convex %? What property must u
satisfy if it represents a monotone %?

1.2

P REFERENCE M AXIMIZATION : M ARSHALLIAN DEMAND

We now study the consumers behavior when deciding a consumption bundle. The
second element in the formalization of this behavior is the definition of what bundles are available for the person to choose. Here, we adopt the competitive setting:
a price is given for each commodity, and the consumer has a nominal wealth that
she can spend in her bundle. Both of these variables are here considered to be exogenous; later, when studying general equilibrium, we will endogenize them. Formally, let us fix some rational preferences %, a vector of prices for all commodities
p = (p1 , . . . , pL )  0 and nominal income m 0. The preference maximization problem is to find x that (i) is affordable: p x m; and (ii) cannot be improved upon:
for every other x0 that is affordable (i.e. such that p x0 m), it is true that x % x0 .
In consumer theory, we will assume that a (competitive) consumer behaves as if she
solved the problem above.
The most immediate questions are whether this problem has solutions and, if so,
how many. The problem of existence of solution is somewhat technical, so well skew
it: suffice it to say that when preferences are such that they dont change abruptly
as one changes consumption (e.g. if they can be represented by a continuous utility
function), the problem has a solution. From now on, let us assume that a solution
exists. Notice that if preferences are strongly convex, the solution is unique. From
now on, let us assume that % is strongly convex and denote by x(p, m) the unique
solution to this problem. As we vary prices and income, the preference maximization
problem defines an optimal demand function x : RL++ R+ RL+ , which is known
as the Marshallian demand function.
4

1.2.1

P ROPERTIES OF THE M ARSHALLIAN D EMAND

P ROPOSITION 1.1. The following are properties of the Marshallian demand, under our assumptions on %:
1. Homogeneity of degree zero: for any (p, m) and any > 0, it is true that x(p, m) =
x(p, m);
2. Walrass law: if % is locally nonsatiated, then the consumer spends all her nominal
income: for any (p, m), p x(p, m) = m;
3. Weak Axiom of Revealed Preferences (WARP): for any (p, m) and any (p0 , m0 ), if p
x(p0 , m0 ) m and p0 x(p, m) m0 , then it must also be true that x(p, m) = x(p0 , m0 ).
E XERCISE 1.5. Let L = 2, and suppose an agent who spends all her money, but who may or
may not be rational.
1. Suppose that at p = (1, 1) she demands x = (1, 1). Suppose that at prices p0 = (2, 3)
and income m0 = 4 her demand, x0 = (x01 , x02 ), is unknown. For what values of x01
would she violate WARP?
2. Suppose that at p = (10, 10) she demands x = (1, 1). Suppose that at prices p0 = (10, 8)
and income m0 her demand is x0 = ( 65 , x02 ), where both m0 and x02 are unknown. For
what values of x02 would she violate WARP?
1.2.2

T HE I NDIRECT U TILITY F UNCTION

Under representability of preferences, we can simply write that x(p, m) solves maxx u(x) :
p x m. The value function, v(p, m) = u(x(p, m)) is known as the Indirect Utility
function.3
P ROPOSITION 1.2. The following are properties of the indirect utility function:
1. Homogeneity of degree zero: for any (p, m) and any > 0, v(p, m) = v(p, m);
2. If % is locally nonsatiated, then v is increasing in m and nonincreasing in p: (i) if
m > m0 , then, for any p, v(p, m) > v(p, m0 ); and if p p0 , then, for any m, v(p, m)
v(p0 , m).
3. Quasiconvexity: if (p, m) and (p0 , m0 ) are such that v(p, m) v(p0 , m0 ) and 0 1,
then it must be true that
v(p + (1 )p0 , m + (1 )m0 ) v(p, m).
3

We should simply write v(p, m) = maxx u(x) : p x m.

1.2.3

D IFFERENTIABLE C ONSUMER

Suppose furthermore that u is twice continuously differentiable4 and has interior contours.5 Suppose that u is strictly monotone and strongly quasiconcave,6 and consider
only m > 0.
P ROPOSITION 1.3. Under the assumptions stated above,
1. Marshallian demand is interior: for any (p, m), x(p, m)  0;
2. For given (p, m), bundle x(p, m) is the only x for which there exists > 0 such that
Du(x) = p and p x = m;
3. The Marshallian demand function x is differentiable;
4. The indirect utility function, v, is differentiable, and the marginal utility is given by
m v(p, m) = (p, m) =

1
u(x(p, m)).
p 1 x1

Moreover, the following properties are important restrictions in applied work:


P ROPOSITION 1.4. Under the assumptions stated above, the following are properties of the
Marshallian demands, at all (p, m):
1. Cournot aggregation: for any commodity l0 ,
X
pl pl0 xl (p, m) = 0;
xl0 (p, m) +
l

2. Engle aggregation:

pl m xl (p, m) = 1;

3. Euler aggregation: for any l0 ,

pl pl xl0 (p, m) + mm xl0 (p, m) = 0.

And the following property of the indirect utility function is very useful in theoretical work:
P ROPOSITION 1.5 (Roys identity). Under the assumptions stated above, for any commodity l0 we have that

pl0 v(p, m)
= xl0 (p, m).
m v(p, m)

The following notation will be used: for any function f (x, y), the partial derivative with respect

to x will be denoted by x f (x, y); the gradient of any function f (x) will be denoted by Df (x) and its
Hessian will be D2 f (x).
5
0
0
L
That is, that for every x RL
++ , it is true that {x |u(x ) u(x)} R++ .
6
In this setting, we are assuming that, whenever x  0, it is true that Du(x)  0 and that
T D2 u(x) < 0 for any 6= 0 such that Du(x) = 0.

Proof: A direct proof can be given by the Envelope theorem. Alternatively, recall that
v(p, m) = u(x(p, m)). Then, differentiating with respect to pl0 , by the chain rule,
X
pl0 v(p, m) =
xl u(x(p, m))pl0 xl (p, m).
l

By the first order conditions (Proposition 1.3), we can substitute xl u(x(p, m)) = (p, m)pl ,
to get
pl0 v(p, m) = (p, m)

pl pl0 xl (p, m) = (p, m)xl0 (p, m),

where the second equality comes from Cournot aggregation (Proposition 1.4). Similarly, but using Engle aggregation,7
m v(p, m) =

xl u(x(p, m))m xl (p, m)

= (p, m)

pl m xl (p, m)

= (p, m).
Q.E.D.

E XERCISE 1.6. Suppose that L = 2, and consider a consumer whose preferences are represented by
u(x) = ln(x1 ) + (1 ) ln(x2 ),
for 0 < < 1. Are the preferences of this consumer rational? Are they locally nonsatiated?
Are they convex? Find the Marshallian demand and the indirect utility functions? Verify
homogeneity of degree zero of both functions. Verify Walrass law. Verify that the indirect
utility is increasing in income and nonincreasing in prices. Verify differentiability. Verify the
conditions of aggregation of Cournot, Engel and Euler. Verify Roys identity.

1.3

E XPENDITURE M INIMIZATION : H ICKSIAN DEMAND

Fix rational, strongly convex preferences %. Suppose that a continuous utility function u represents %.
For prices p  0 and a feasible utility level , the expenditure minimization problem is to find x such that: (i) it gives utility at least equal to : u(x) ; and (ii) for
every other x0 such that u(x0 ) , it is true that p x0 p x. That is, to find x that
solves minx p x : u(x) .
7

The equations that follow prove the last statement in Proposition 1.3.

Under our assumptions, the problem is guaranteed to have a unique solution,


which we denote by h(p, ). Function h : RL++ u[RL+ ] RL+ is known as the Hicksian
demand function.8
1.3.1

P ROPERTIES OF THE H ICKSIAN DEMAND

P ROPOSITION 1.6. The following are properties of the Hicksian demand, under our assumptions on %:
1. Homogeneity of degree zero in p: for any (p, ) and any > 0, it is true that h(p, ) =
h(p, ).
2. No excess utility: for any (p, ), it is true that u(h(p, )) = .
3. The compensated law of demand: for any p and p0 , and any , it is true that
(p p0 ) (h(p, ) h(p0 , )) 0.
Proof: We only proof the compesated law of demand. Notice that, by definition,
p h(p, ) p h(p0 , ) and p0 h(p0 , ) p0 h(p, ). Immediately,
p h(p, ) + p0 h(p0 , ) p h(p0 , ) + p0 h(p, ).
Q.E.D.
Notice that there is no analogous of the last property that holds true for the Marshallian demand!
E XERCISE 1.7. For the consumer considered in Exercise 1.6, find the Hicksian demand function. Verify homogeneity of degree zero in p, no excess utility and the compensated law of
demand.
1.3.2

T HE E XPENDITURE F UNCTION

The value function of the expenditure minimization problem is the Expenditure Function: e : RL++ u[RL+ ] R+ , defined by e(p, ) = p h(p, ).9
P ROPOSITION 1.7. The following are properties of the expenditure function:
8
9

L
L
Here, u[RL
+ ] denotes the set of feasible utility levels: u[R+ ] = { R|x R+ : u(x) = }.
Or, e(p, ) = minx p x : u(x) .

1. Homogeneity of degree one in p: for any (p, ) and any > 0, e(p, ) = e(p, ).
2. If preferences are monotonic, then e is increasing in and nondecreasing in p: if > 0 ,
then, for any p, it is true that e(p, ) > e(p, 0 ); and if p p0 , then, for any , it is true
that e(p, ) e(p0 , ).
3. Concavity in p: for any p, p0 and and any 0 1, it is true that
e(p + (1 )p0 , ) e(p, ) + (1 )e(p0 , ).
Proof: We prove only concavity in p. Notice that, by definition, e(p, ) p h(p +
(1 )p0 , ) and e(p0 , ) p0 h(p + (1 )p0 , ). Then, taking the average,
e(p, ) + (1 )e(p0 , ) (p + (1 )p0 ) h(p + (1 )p0 , ).
Q.E.D.
Notice that the last property is cardinal, while the analogous result for in the indirect utility function is only ordinal!
D IFFERENTIABLE C ONSUMER

1.3.3

Suppose again that the utility function u representing the consumers preferences is
(twice continuously) differentiable and has interior indifference curves. Suppose also
that u is strictly monotone and strongly quasiconcave.
P ROPOSITION 1.8. Under the assumptions stated above, and considering only interior utility
levels > u(0), the following is true:
1. Hicksian demand is interior: for any (p, ), h(p, )  0.
2. For given (p, ), bundle h(p, ) is the only x for which there exists > 0 such that
p = Du(x) and u(x) = .
3. The Hicksian demand function, h, is differentiable;
4. The expenditure function, e, is differentiable, and the marginal expenditure is given by
e(p, ) = (p, ) = (x1 u(x(p, )))1 p1 .
Furthermore, the following restrictions are important in applied and theoretical
work:
9

P ROPOSITION 1.9. Under the assumptions stated above, and considering only interior utility
levels > u(0),the following is true:
2
1. Negative semidefiniteness: for any p and , matrix Dp,p
e(p, ) is negative semidefinite;

2. Shephards lemma: for any commodity l0 , pl0 e(p, ) = hl0 (p, );


3. Symmetry and negative-semidefiniteness of substitution effects: matrix Dp h(p, ) is
symmetric and negative semidefinite.
2
Proof: Negative semidefiniteness of Dp,p
e(p, ) is immediate from the fact that e is

concave (Proposition 1.8).


For Shephards lemma, notice first that, by no excess utility (Proposition 1.6),
u(h(p, )) = . Then, taking derivatives,
X

xl u(h(p, ))pl0 hl (p, ) = 0.

l
1
By first-order conditions (Proposition 1.8), we can substitute xl u(h(p, ) = (p,)
pl , to
P
get that l pl pl0 hl (p, ) = 0. Now, recall that e(p, ) = p h(p, ), so, differentiating,

pl0 e(p, ) = hl0 (p, ) +

pl pl0 hl (p, ) = hl0 (p, ).

l
2
e(p, ), so negative
Shephards lemma immediately implies that Dp h(p, ) = Dp,p

semidefiniteness follows from part 1, while symmetry follows from a well-known result in mathematics, Youngs Theorem.

Q.E.D.

Symmetry of Dp h(p, ) is an example of a result that was discovered after the


application of mathematics, but was not anticipated by intuitive arguments.
E XERCISE 1.8. For the same consumer as in Exercise 1.6, verify that e is increasing in p and
in . Verify that e is homogeneous of degree one and concave in p. Verify Shephards lemma.

1.4

D UALITY

For the purposes of this section, fix rational, strongly convex, locally nonsatiated preferences %, and let the utility function u represent %.
P ROPOSITION 1.10 (Duality Theorem). Fix prices p  0, nominal income m and a feasible
utility level . Under the assumptions above, the following is true:
10

1. Marsallian demand at income equal to minimized expenditure is the same as Hicksian


demand: x(p, e(p, )) = h(p, );
2. Hicksian demand at utility level equal to maximized utility is the same as Marshallian
demand: h(p, v(p, m)) = x(p, m);
3. Maximized utility at income equal to minimized expenditure is the same as required
utility: v(p, e(p, )) = ;
4. Minimized expenditure at utility level equal to maximized utility is the same as nominal
income: e(p, v(p, m)) = m.
Proof: For part 1, suppose that the equality does not hold. Since p h(p, ) = e(p, ), it
must be that u(x(p, e(p, ))) > u(h(p, )) = . But then, since p x(p, e(p, )) e(p, ),
we have that x(p, e(p, )) solves the expenditure minimization problem too, which
would violate no excess utility (Proposition 1.6).
For part 2, again suppose otherwise. Since u(x(p, m)) = v(p, m), it must be that
p h(p, v(p, m)) < p x(p, m) = m. But then, since u(h(p, v(p, m)) v(p, m), we have
that h(p, v(p, m)) solves the utility maximization problem too, which would violate
Walrass law (Proposition 1.1).
Part 3 is immediate from part 1, given no excess utility, and part 4 follows from
Q.E.D.

part 2, given Walrass law.

The Duality Theorem allows us to go from one problem to the other without needing to solve them both. Notice that the assumption that p  0 is crucial. For instance,
if p = 0, then any bundle with u(x) solves the expenditure minimization problem
(and at least one such bundle exists), whereas the utility maximization problem has
no solution, given that preferences are locally nonsatiated.
E XERCISE 1.9. For the same consumer as in Exercise 1.6, verify the duality equalities.
When the utility function u is twice continuously differentiable and has interior
indifference curves, one has the following crucial result.
P ROPOSITION 1.11 (Slutskys Identity). Suppose that u is strictly monotone and strongly
quasiconcave. Fix a feasible utility level and define a nominal income m = e(p, ). Then,
for each pair of commodities l and l0 , the following is true:
pl0 xl (p, m) = pl0 hl (p, ) xl0 (p, m)m xl (p, m).
11

Proof: From the Duality Theorem (Proposition 1.10), we know that xl (p, e(p, )) =
hl (p, ). Then, differentiating,
pl0 xl (p, e(p, )) + m xl (p, e(p, ))pl0 e(p, ) = pl0 hl (p, ).
By Shephards Lemma (Proposition 1.9) and duality, we can substitute pl0 e(p, ) =
hl0 (p, ) = xl0 (p, e(p, )). The result follows since m = e(p, ).

Q.E.D.

E XERCISE 1.10. For the same consumer as in exercise 1.6, verify the duality equalities and
Slutskys identity.
Letting the substitution matrix be S(p, m) = Dp h(p, v(p, m)), Slutskys Identity
writes in matrix terms as Dp x(p, m) = S(p, m) Dm x(p, m)x(p, m)T . That the substitution matrix is symmetric and negative semidefinite is a necessary condition implied
by rationality of the consumer. Another necessary condition is that the matrix should
have (exactly) L1 of its columns linearly independent. Remarkably, these conditions
are also sufficient for rationality!

1.5

A DDITIONAL E XERCISES

E XERCISE 1.11. Consider a standard consumer with preferences %, over nonnegative consumption of two commodities, represented by u(x) = x1 + ln(x2 ). Answer and solve:
1. Are these preferences rational? Are they convex? Are they strongly convex? Are they
monotone? Are they strictly monotone?
2. Find Marshallian demands and the indirect utility function. Verify Roys identity.
(Warning: be careful about the nonnegativity constraint!)
3. Find the expenditure function and the Hicksian demands.
E XERCISE 1.12. For a consumer in a two-commodity world, solve:
1. Suppose that the following information is known: when her income is m = 5 and prices
are p = (1, 1), her demand of commodity 1 is x1 = 3; when her income is m0 = 5
and prices are p0 = (, ), all that is known is that her consumption of commodity 2 is
x02 3. For what values of x2 , and x0 does this consumer satisfy WARP? When are
these observations consistent with maximization of strongly convex, locally nonsatiated
preferences?
12

2. Suppose that the following information is known: when her income is m = 5 and prices
are p = (1, 1), her demand of commodity 1 is x1 = 3; when her income is m0 = 5
and the price of commodity 1 is p01 = , all that is known is that her consumption of
commodity 2 is x02 3. For what values of x2 , , p02 and x0 does this consumer satisfy
WARP?
E XERCISE 1.13. Consider a standard consumer with preferences %, over nonnegative consumption of two commodities, represented by
1
1
1
1
u(x) = (x1 ) 2 + (x2 ) 2 .
2
2

Answer and solve:


1. Are these preferences rational? Are they convex? Are they strongly convex? Are they
monotone? Are they strictly monotone?
2. Find Marshallian demands and the indirect utility function. Verify Roys identity.
(Warning: be careful about the nonnegativity constraint!)
3. Find the expenditure function and the Hicksian demands.
E XERCISE 1.14. Suppose that L = 3, and consider a consumer whose preferences are represented by
u(x) = min{x1 , x2 + x3 }.
Let prices be p  0.
1. Are the preferences of this consumer rational? Are they locally nonsatiated? Are they
convex?
2. Find the Marshallian demand and the indirect utility function. Verify also that the
indirect utility is homogeneous of degree zero, increasing in income and nonincreasing
in prices.
3. Considering only p such that p2 < p3 , verify that the Marshallian demand and the
indirect utility are differentiable functions, verify the condition of aggregation of Engel,
and verify Roys identity for commodity 1.
4. Find the Hicksian demand and the expenditure function. Verify that the expenditure
function is homogeneous of degree 1 in prices, and verify Shephards lemma for commodity 1.

13

A PPENDIX : W HY ?
Here, I briefly sketch the arguments why the various propositions stated above are true. Some
bits here are technical, and this appendix is to be seen as optional.
Marshallian demand is guaranteed to exist when preferences can be represented by a
continuous utility function and prices are strictly positive, thanks to Weierstrasss Theorem, since set {x RL
+ |p x m} is, then, closed and bounded.
Marshallian demand is guaranteed to be unique when preferences are strongly convex,
since if there were more than one solution, an average of two of these solutions would
be strictly better and still affordable.
In Proposition 1.1:
1. Homogeneity of degree-zero holds because the budget set does not change when
one multiplies all prices and nominal income by the same positive constant.
2. Walrass law follows since otherwise the consumer would be able to find  > 0
such that all x0 with ||x0 x(p, m)|| <  is affordable; by local nonsatiation, at least
one of these x0 would also be strictly superior to x(p, m).
3. For WARP, suppose otherwise; then, x(p, m) % x(p0 , m0 ) and x(p0 , m0 ) % x(p, m)
while, by strong convexity, 21 x(p0 , m0 ) + 21 x(p, m)  x(p, m); but this is impossible,
since p ( 21 x(p0 , m0 ) + 12 x(p, m)) m.
In Proposition 1.2:
1. Homogeneity of degree zero is immediate from the same property of Marshallian
demand.
2. That v is increasing in m follows from Walrass law: otherwise, x(p, m0 ) would be
optimal at (p, m) and p x(p, m) < m0 ; that v is nonincreasing in p is immediate as
p p0 implies that p0 x m is true whenever p x m.
3. For quasiconvexity, notice that
(p + (1 )p0 ) x m + (1 )m0
implies that either p x m or p0 x m0 , and hence that
v(p + (1 )p0 , m + (1 )m0 ) max{v(p, m), v(p0 , m0 )}.
In Proposition 1.3:
1. Interiority of demand follows from interiority of the contour sets, given m > 0.
2. The characterization of demand via first-order conditions follows from Kuhn-Tuckers

theorem, given that strong quasiconcavity guarantees the second-order conditions.


3. Differentiability follows from the Implicit Function Theorem: differentiate the system of first-order conditions, and notice that the Jacobian with respect to (x, ) is
nonsingular when u is strongly quasiconcave.

14

4. That v is differentiable is immediate from the previous result; the derivative can
be taken by the Envelope theorem, or it can be obtained as in the proof of Roys
identity.
In Proposition 1.4:
1. Cournot aggregation follows from Walrass law: differentiate with respect to prices.
2. And so does Engle aggregation: differentiate with respect to income.
3. Euler aggregation follows from homogeneity of degree zero of demand, via Eulers
theorem.
That the expenditure minimization problem has a solution again follows from Weierstrasss theorem: since is feasible, one can bound the feasible set of the problem to cost
0
less than some feasible bundle x0 ; the set {x RL
+ |u(x) and p x p x } is closed
and boundsd, given that u is continuous.
That the solution to the expenditure minimization problem is unique follows from
strong quasiconcavity: if there are multiple solutions, an average of any two of then
will give strictly more utility than ; this average bundle x must satisfy p x > 0 and
then, multiplying x by a number  < 1, close enough to 1, one gets a feasible and
cheaper bundle.
In Proposition 1.6:
1. Homogeneity of degree zero follows from the fact that multiplying all prices by a
positive constant only re-scales the objective function.
2. No excess utility follows by continuity: if u(h(p, )) > , then p h(p, ) > 0 and
then, multiplying h(p, ) by a number  < 1, close enough to 1, one gets a feasible
and cheaper bundle.

15

P RODUCER T HEORY

Firms are complicated. Unlike with consumers, we have to wonder why and how
they are created; what a firm can do is the result of decisions made within and without
the firm; many different people may work for a firm, and it isnt always the case
that they all agree when they make a decision; moreover, they may all have different
objectives and interests leading their decisions. While all these issues are interesting,
we will assume them away. We take take a given capacity to produce and study the
implications of the assumption that (everyone involved agrees that) the firm wants
to make as much money as possible.

T ECHNOLOGY

2.1

As before, let us assume that there exist L commodities. A firm is a set F RL . This
set says what is technologically feasible for the firm to produce: a production plan is
a bundle y = (y1 , . . . , yL ); in this plan, commodity l is used as an input if yl < 0 and
is produced as an output if yl > 0; a production plan y is feasible for the firm if and
only if y F.
Obviously, we want to consider a firm that is able to do something, so we assume that F 6= . For technical reasons, we also want to assume that the technology
does not change abruptly from feasible to unfeasible: we assume that F contains its
boundary (i.e., is closed).10
2.1.1

P ROPERTIES OF A TECHNOLOGY:

The following are properties that may be satisfied by a firm:


D EFINITION . Firm F is said to satisfy
1. no-free-lunch if y > 0 implies y
/ F;
2. possibility of inaction if 0 F;
3. free disposal if y F and y 0 y imply that y 0 F ;
4. non-increasing returns to scale if y F and 0 1 imply that y F;
5. non-decreasing returns to scale if y F and 1 imply that y F;
10

That is, we assume that if we take a sequence (yn )


n=1 of feasible production plans (i.e. yn F for

every n) that converges to some production plan y, then that limit plan is feasible too (i.e., y F).

16

6. constant returns to scale if if y F and 0 imply that y F;


7. free entry if y, y 0 F implies that y + y 0 F.
No-free-lunch says that the firm cannot get output without using inputs. Possibility of inaction means that one can just shut the firm down (there are no sunk costs).
Free disposal says that wasting (either inputs or outputs) is possible. Non-increasing
returns to scale says that one can shrink the firm, whereas non-decreasing returns
imply that any expansion is possible; with constant returns, both contractions and
expansions are possible. Free entry says that feasible production plans dont interfere
with one another.
E XERCISE 2.1. Argue that nonincreasing returns to scale implies possibility of inaction. Argue that if F satisfies nonincreasing returns to scale and free entry, then it is a convex set.
Argue that if if F satisfies free entry, then the following integer-constant returns to scale
holds: y F and N imply y F.11 Are the opposite assertions true?

P ROFIT M AXIMIZATION

2.2

Let us fix a firm F and prices p  0.


The profit maximization problem is to find y that (i) is feasible: y F; and (ii) and
cannot be improved upon, in the sense of profits at the given prices: every y 0 F
yields p y p y. Put another way, we want to find a solution to the problem
maxy p y : y F.
Let Y (p) be the set of all values of y that satisfy the two conditions (which may
be none, so Y (p) = is possible). As we vary prices, this set of optimal production
plans may change, so we are defining an optimal supply correspondence Y : RL++ F.
E XERCISE 2.2. Argue that if F satisfies nondecreasing returns to scale and there exists y F
such that p y > 0, then Y (p) = .
2.2.1

P ROPERTIES OF THE S UPPLY C ORRESPONDENCE

We only want to consider prices at which the firm is able to find a profit-maximizing
production plan, so let us define that set of prices D = {p  0|Y (p) 6= }.
P ROPOSITION 2.1. The following are properties of the supply correspondence:
11

Here, N denotes the set of Natural numbers.

17

1. Homogeneity of degree zero: for any p D and any > 0, it is true that p D and
Y (p) = Y (p);12
2. Convexity: If F is convex, then y, y 0 Y (p) and 0 1 imply that y+(1)y 0
Y (p);
3. Single-valuedness: suppose that F further satisfies the following property: whenever
y, y 0 F, y 6= y 0 and 0 < < 1, one can find y 00 F such that
y 00 > y + (1 )y;
then, Y (p) is singleton for any p D;
4. The law of supply: for any p and p0 , for any y Y (p) and any y 0 Y (p0 ), it is true that
(p p0 ) (y y 0 ) 0.
Proof: We prove the law of supply only: by definition, p y 0 p y and p0 y p0 y 0 .
Immediately, p y + p0 y 0 p y 0 + p0 y.

2.3

Q.E.D.

P ROFIT FUNCTION

For prices for which the profit maximization problem does have a solution, we define
the value function : D R by (p) = p y, for any y Y (p). This function is known
as the profit function.13
P ROPOSITION 2.2. The following are properties of the profit function:
1. Homogeneity of degree 1: for any p D and any > 0, it is true that (p) = (p);
2. Convexity: for any p, p0 D and any 0 1 such that p + (1 )p0 D, it is
true that
(p + (1 )p0 ) (p) + (1 )(p0 ).
Proof: We prove convexity only: by definition, for any y Y (p + (1 )p0 ), we have
that (p) p y and (p0 ) p0 y. Immediately,
p y + (1 )p0 y (p) + (1 )(p0 ).
Q.E.D.
12
13

Implicitly, we are saying also that D is a positive cone.


Formally, (p) = maxyF p y.

18

2.4

D IFFERENTIABLE F IRM

As with consumers, we would like to have a setting in which we can use calculus to
deal with the optimization problem. So again, as there, we need to represent the firm
with a function.
Firm F is said to be represented by a function F : RL R, if y F occurs if, and
only if, F (y) 0. Function F is the transformation function.
So, for the remainder of this section, let us fix a representable firm, and let F be
the transformation function. Furthermore, let us suppose that
1. function F is twice continuously differentiable, nondecreasing and strongly convex;
2. the transformation frontier, which is the boundary of the technology, is the set of
production plans y for which F (y) = 0.14
Under the assumptions above, we can define, for any production plan y in the
transformation frontier, and for any pair of commodities, l, l0 , the Marginal Rate of
Transformation
M RTl,l0 (y) =

yl F (y)
.
yl0 F (y)

whenever the denominator is not zero. The meaning of the marginal rate of transformation depends on whether l and l0 are inputs or outputs:
1. If both commodities are outputs, it is the usual definition: the slope of the production possibilities frontier.
2. If l is an output and l0 is an input, it is the (negative) of the marginal product of
l0 (in the production of l).
3. If both commodities are inputs, it is the marginal rate of technical substitution.
Now, let us assume that for every p, Y (p) is singleton set.15 Then, we can further
14

Formally, the transformation frontier is F, the set of all production plans y with the property

that for any  > 0, one can find production plans y 0 and y 00 such that (i) ||y y 0 || < ; (ii) ||y y 00 || < ;
/ F. That is, a boundary point is arbitrarily close to points within and without
(iii) y 0 F; and (iv) y 00
the set. Since we are assuming that F is closed, we have that the transformation frontier is part of the
feasible set, F F. The assumption we are imposing is that F = F 1 (0).
15
It suffices, for instance, that F be increasing and that F 1 (0) be bounded. In this case, the problem
is guaranteed to have a solution, as the set F 1 (0) is closed (because F is continuous) and bounded,
and the function p y is continuous. To see that the solution has to be unique, notice that since F is
strongly convex and continuous, the result follows from single-valuedness in proposition 2.1.

19

take the only solution to the profit maximization problem, y(p), to define the supply
function, and we have the following property:
P ROPOSITION 2.3. Under the assumptions above,
1. for any p, y(p) is the only production plan y for which there exists > 0 such that
p = DF (y) and F (y) = 0;
2. function y is differentiable;
3. Hotellings lemma: function is differentiable, and for any commodity l, we have that
pl (p) = yl (p);
4. matrix D2 (p) is positive semi-definite;
5. matrix Dy(p) is symmetric and positive semi-definite.16
Remarkably, we obtain symmetry of Dy(p) (notice that this is not naturally true
in consumer theory), thanks to the fact that income effects have no analogous in a
firm.
E XERCISE 2.3. Suppose that L = 3. Suppose that if a firm uses y2 units of commodity 2 and
y3 units of commodity 3, then it obtains y2 y3 units of commodity 1. Assume that , > 0 and
+ < 1. Describe F. What properties does F satisfy? Derive the supply function, verify
the law of supply, derive the profit function, verify that is convex and verify Hotellings
lemma.
An associated problem, for the case when there is only one commodity which is
output and the remaining ones can be used only as inputs, fixes the level of production of the output and only finds the cheapest combination of inputs that delivers
at least that level of output. This problem is known as the cost minimization problem
and the resulting bundles are known as conditional demands for inputs. Formally, this
problem is equivalent to the expenditure minimization problem, and we can translate
most of the results we know from Hicksian demands and the expenditure function
to this setting. The exception to the latter is the fact that duality theory is less
16

This matrix is

p1 y1 (p) . . .

..
..

Dy(p) =
.
.
p1 yL (p) . . .

20

pL y1 (p)

..
.
.

pL yL (p)

rich in this setting, for an obvious reason: in consumers duality, both problems determine an optimal bundle of commodities; here, the profit maximization problem
determines a full production plan (of L commodities), whereas the cost minimization
problem fixes the level of one of those variables and only determines the remaining
(L 1) of them optimally. Thus, while profit maximization implies cost minimization
(at the optimally chosen level of output), the fact that a combination of inputs is optimal at some production level does not imply that the production plan with that level
of output and the chosen combination of inputs will maximize profits.17

2.5

A DDITIONAL E XERCISES

E XERCISE 2.4.

1. Consider a firm
F = {x R2 | 1 x1 0 and x2 x1 }.

Does this firm satisfy possibility of inaction? Free disposal? No-free-lunch? Free entry?
Nondecreasing returns to scale? Nonincreasing returns to scale? Constant returns to
scale? Convexity? Find the supply correspondence and the profit function, considering
prices p  0.
2. Consider now a different firm:
F = {x R3 | 1 x1 0, x2 x1 and x3 = 1}.
Does this firm satisfy possibility of inaction? Suppose that p2 = 3 and p1 = 1, and find
the supply and expenditure functions of this firm, for any value of p3 > 0.
3. Consider now a different firm:
F = {x R3 |x2 0, x3 0 and x1 = max{x2 , x3 }}.
Does this firm satisfy possibility of inaction? Free disposal? No-free-lunch? Free entry?
Nondecreasing returns to scale? Nonincreasing returns to scale? Constant returns to
scale? Convexity? Find the supply correspondence and the profit function, considering
prices p  0.

17

Unless, of course, the pre-fixed production level is optimal, but then we are very close to a tautol-

ogy!

21

E XERCISE 2.5. Answer and solve:


1. Given a production function X = f (K, L), define the firm F = {y R3 : y2 0, y3
0, y1 f (y2 , y3 )}. Argue that if f is homogeneous of degree 1, the firm satisfies
constant returns to scale. Argue that if f is homogeneous of degree d 1, the firm
satisfies nonincreasing returns to scale.
2. Consider the following firm:
F = {y R2 : y2 min{y1 , y1 }},
where < < 0 are technological parameters. What properties does this firm satisfy?
Find the optimal supply correspondence, and the profit function.
E XERCISE 2.6. Consider a Leontieff production function: X = min{K, L}, where X
represents output of one commodity, K and L represent input of capital and labor, respectively,
and > 0 and > 0 are technological coefficients. Answer and solve:
1. Write the technology set, F, representing the firm.
2. Does this firm satisfy nondecreasing returns to scale? nonincreasing returns to scale?
No free lunch? Possibility of inaction? Free entry?
3. Is optimal supply defined, for this firm, for all positive prices? For those prices for which
optimal supply is defined, determine the optimal supply and the profit function.
Suppose now that the production function is X = ln(1 + min{K, L}). Answer the same
questions as before and, also, for those prices for which optimal supply is defined, verify
Hotellings lemma.
E XERCISE 2.7. Consider the following firm:
F = {y R2 : (1 + y1 )2 + (y2 )2 1}.
Answer and solve:
1. Does this firm satisfy possibility of inaction? Nondecreasing returns to scale? Nonincreasing returns to scale? Constant returns to scale? No free lunch? Free entry? Free
disposal?
2. Find the optimal supply function (for strictly positive prices) and the profit function.
3. Verify Hotellings lemma.
22

A PPENDIX : W HY ?
Again, the more formal arguments for statements made in this section are given here.
Homogeneity of degree 1 in Proposition 2.2 follows directly from homogeneity of degree zero in the supply correspondence.
For Proposition 2.3:
1. That optimal supply is characterized by the first-order conditions is immediate
from Kuhn-Tuckers

Theorem.
2. Differentiability follows from the Inverse Function Theorem, by differentiation of the
system of first-order conditions, given that F is twice continuously differentiable
and strongly convex.
P
3. For Hotellings lemma,18 notice that pl (p) = l0 pl0 pl yl0 (p) + yl (p). By the first
order conditions, this implies
X
pl (p) =
yl0 F (y(p))pl yl0 (p) + yl (p) = pl F (y(p)) + yl (p) = yl (p),
l0

where the last equality follows from the fact that F (y(p)) = 0 for all p. Alternatively, a simpler proof can be obtained using the Envelope theorem.
4. Positive semidefiniteness of D2 (p) follows from convexity of in proposition
2.2.19
5. Symmetry and positive semidefiniteness of Dy(p) follows from Hotellings lemma:
Dy(p) = D2 (p).

18

That is differentiable is immediate, since so is y.


Strictly speaking, we now require to be differentiable twice, for which we would need to assume
that F is differentiable three times.
19

23

G ENERAL E QUILIBRIUM

D EFINITIONS

3.1

Suppose that there is a finite number, L, of commodities that can be consumed in


nonnegative amounts.
3.1.1

T HE E CONOMY

We assume a society populated by a finite number of individuals, which we denote


by i = 1, . . . , I. In this society, we will consider the case in which only exchange of
commodities takes place, and also the case when production is undertaken.
Consumer i is modeled by what she likes and what she has. For simplicity of
expression, we will assume here that our consumers have preferences that are representable by utility functions ui : RL+ R.20 In general equilibrium, we want to endogenize the individuals nominal income, so we will assume that they are endowed
with a bundle, wi RL+ , of commodities. Notice that, by the latter assumption, we
are introducing one institution in our society: private property.
When the economy has production, we will assume that it has an industry consisting of a finite number of firms, which we will denote by j = 1, . . . , J. Firm j is
a nonempty set F j RL , which represents its technology.21 In this case, we will
maintain the assumption of private property and will also assume that the economy
is closed: we will assume that each individual i owns a proportion si,j 0 of the
P
stock of firm j, and will impose the condition that Ii=1 si,j = 1 for every firm j.
An exchange economy is defined by a society, and by the full description of its members,
{{1, . . . , I}, (ui , wi )Ii=1 }.
Later, for simplicity, we will simply write {(ui , wi )Ii=1 }, leaving the society implicit,
unless we need to be explicit about it.
A production economy consists of a society and an industry, and by the full description of the members of both sets,
{{1, . . . , I}, {1, . . . , J}, (ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 }.
20

The standard properties of preferences may be invoked. Here, we will interchangeably say that

the individual has convex preferences or that her utility function is quasiconcave.
21
The standard properties of preferences may be invoked.

24

Later, for simplicity, we will simply write {(ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 }, leaving both
the society and the industry implicit.
3.1.2

C OMPETITIVE EQUILIBRIUM

We want to study situations where agents trade voluntarily and where they think
that their actions do not impact the aggregate conditions at which trade takes place.
We, then, add a second institution, competitive markets, which are exchange facilities
where an anonymous price is announced for each commodity, denoted pl , and where
all traders can trade at that given price.
Let p = (p1 , . . . , pL ) RL denote commodity prices, and use xi and y j to denote,
respectively, individual is consumption plan and firm js production plan.
In an exchange economy with competitive markets, consumers take all prices as
independent of their demands, and the only constraint that individual i recognizes
is that she cannot spend more than her nominal wealth, which is the nominal value
of her endowment. If it is a production economy, individual is nominal wealth is
given by the nominal value of her endowments and the dividends she receives from
the firms. In the latter case, firms too take all prices as fixed, and only recognize their
own technology as a constraint.
D EFINITION . In an exchange economy {(ui , wi )Ii=1 }, a competitive equilibrium is a pair
consisting of a vector of prices and a profile of consumption plans, (p, (xi )Ii=1 ), such that
1. for each consumer i, bundle xi solves the problem maxx ui (x) : p x p wi ;
2. all markets clear:

PI

i=1

xi =

PI

i=1

wi .

(Later on, for simplicity of notation, we will write as x the allocation (xi )Ii=1 .)
The definition of equilibrium takes preferences and endowments as given fundamentals, and determines values for all endogenous variables of the problem; in the
case of an exchange economy, the endogenous variables are all the prices and the
consumption decisions of all individuals. Equilibrium is then the requirement that
all these variables be feasible and that no agent regret the decision she is making at
the time she is making it. Importantly, notice that in the interpretation of the definition of competitive equilibrium, there are endogenous variables that are not decided
by any one particular agent: while prices are endogenous to the whole economy, each
decision-maker thinks that she cannot affect them. Notice also that the definition of
equilibrium does not say what occurs in the economy when it is not in equilibrium.
25

Finally, notice the assumptions implicit in the definition: (i) it is assumed, as an institution, the existence of complete competitive markets; (ii) it is assumed, as a rule
of behavior, that all agents are price takers; (iii) each individuals behavior affects her
well-being only; and (iv) no unit of a commodity can be consumed by more than one
consumer. Many results crucially depend on these assumptions.22
The following property is well known, and simplifies the treatment of competitive
equilibrium.
P ROPOSITION (Walrass law). Fix an exchange economy {(ui , wi )Ii=1 } in which all consumers have locally nonsatiated preferences, and at least one of them has strongly monotone
preferences. Suppose that (p, (xi )Ii=1 ) satisfies that:
1. for each individual i, xi solves maxx ui (x) : p x p wi ;
2. for all commodities but one, say for all l {1, ..., L 1}, it is true that
PI
i
i=1 wl .

PI

i=1

xil =

1
Then, all prices are positive, p  0, and it is true that (p, (xi )Ii=1 ), ( p11 p, (xi )Ii=1 ), ( ||p||
p, (xi )Ii=1 )

and ( P1 pl p, (xi )Ii=1 ) are all competitive equilibria.


l

Proof: Since one individuals preferences are strictly monotone, it follows from condition 1 that all prices must be strictly positive. Since all consumers are locally nonsatiated, condition 1 also implies, by the version of Walrass covered in Consumers
P
P
theory, that Ii=1 p (xi wi ) = 0. But then, by condition 2, pL Ii=1 (xiL wLi ) = 0,
P
which implies that Ii=1 (xiL wLi ) = 0, since pL > 0. This means that (p, (xi )Ii=1 ) is
a competitive equilibrium for the economy. That the other pairs are equilibria too
follows from homogeneity of degree zero of Marshallian demand.

Q.E.D.

The result says that when looking for general equilibria of an economy with strongly
monotone consumers, it suffices to guarantee that all of the markets but one clear.
This says that the L L system of market-clearing equations is underdetermined (as
a function of prices), and is in fact an L(L1) system. So, one can drop one variable
by letting, for instance, p1 = 1 and solving a (L 1) (L 1) system.
E XERCISE 3.1. Consider an exchange economy with two consumers and two commodities:
u1 (x1 , x2 ) = x1 + x2 , w1 = (1, 1),
u2 (x1 , x2 ) = x1 + x2 , w2 = (1, 1).
22

When there are two consumers and two commodities, a graphical representation of the economy,

its equilibria and other concepts is obtained via Edgeworth boxes.

26

Find all competitive equilibria.


D EFINITION . In a production economy {(ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 }, a competitive equilibrium is a triple consisting of a vector of prices, a profile of consumption plans and a profile
of production plans, (p, (xi )Ii=1 , (y j )Jj=1 ), such that
1. for each consumer i, bundle xi solves the problem
i

max u (x) : p x p w +
x

J
X

si,j p y j ;

j=1

2. for each firm j, bundle y j solves the problem maxy p y : y F j ;


3. all markets clear:

PI

i=1

xi =

PI

i=1

wi +

PJ

j=1

yj .

Later on, again for simplicity of notation, we will write as y the profile (y j )Jj=1 of
production plans. For production economies, a version of Walrass law also holds.
3.1.3

PARETO E FFICIENCY

Competitive equilibrium is the canonical noncooperative (some people say individualistic) solution. The simplest form of cooperative solution is the concept of Pareto
efficiency.
In an exchange economy, an allocation is a profile of consumption plans, x =
P
P
(xi )Ii=1 , such that Ii=1 xi = Ii=1 wi .23 In a production economy, an allocation is a pair
consisting of a profile of consumption plans and a profile of production plans, (x, y) =
P
P
P
((xi )Ii=1 , (y j )Jj=1 ) such that y j F j , for each firm j, and Ii=1 xi = Ii=1 wi + Jj=1 y j .
D EFINITION . In an exchange economy {(ui , wi )Ii=1 }, an allocation x is Pareto efficient if
there does not exist an alternative allocation x that
1. no consumer finds worse: for every i, ui (
xi ) ui (xi ); and
2. at least one consumer prefers: for some i, ui (
xi ) > ui (xi ).
E XERCISE 3.2. For the same economy as in Exercise 3.1, find all Pareto efficient allocations.
What can you say about the competitive equilibrium allocations?
D EFINITION . In a production economy {(ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 }, an allocation (x, y)
is Pareto efficient if there does not exist an alternative allocation (
x, y) that
23

Sometimes, when the latter condition is imposed people say that x is a feasible allocation, while

the term allocation is used for any profile of consumption plans.

27

1. no consumer finds worse: for every i, ui (


xi ) ui (xi ); and
2. at least one consumer prefers: for some i, ui (
xi ) > ui (xi ).
It is important to notice that: (i) Pareto efficiency does away with the institutions
of competitive markets (and hence prices) and private property; (ii) it does not replace
the latter institutions by alternative mechanisms; and (iii) in production economies,
only the welfare of consumers, and not the profits of the firms, matters.
E XERCISE 3.3. Argue the following: given a production economy, an allocation (x, y) is
Pareto efficient if, and only if, for each individual i0 the allocation solves the following problem:

i i

x ) ui (xi ),
for all individual i 6= i0 ;

u (
0

max ui (
xi ) :
(
x,
y)

yj F j ,

PI xi = PI
i=1

3.1.4

i=1

for all firm j;

wi +

PJ

j=1

yj .

T HE C ORE

If one maintains the institution of private property and the self-interest of individuals,
one can refine the definition of Pareto efficiency to a coalitional solution for exchange
economies:
D EFINITION . An allocation x is in the core of exchange economy {(ui , wi )Ii=1 }, if there do
not exist a (nonempty) group of individuals, H {1, . . . , I}, and a sub-profile of consumption plans x = (
xi )iH such that
1. group H can implement the sub-profile x:

iH

xi =

iH

wi ;

2. no individual in group H finds herself worse off: for all i H, it is true that ui (
xi )
ui (xi ); and
3. at least one individual in group H finds herself better off: for some i H, ui (
xi ) >
ui (xi ).
E XERCISE 3.4. For the same economy as in Exercise 3.1, find all core allocations. What can
you say about the competitive equilibrium and the Pareto efficient allocations?
E XERCISE 3.5. Argue that any allocation in the core of an exchange economy is Pareto efficient. Is the opposite true?
The relation between the core and efficiency is studied in the previous exercise.
The relation between efficiency (and hence the core) and competitive equilibrium is
studied in subsection 3.3.
28

P OSITIVE A NALYSIS

3.2

Properties are said to be positive when they are necessary conditions that do not
involve any value judgement. Some of these properties are very important, but their
exposition is somewhat technical. For instance, one can use fixed point theorems to
show that any well behaved economy24 has a competitive equilibrium. Moreover, if
one assumes that preferences and technologies are sufficiently differentiable, one can
use calculus, transversality theory in particular, to show that, for almost all values of
the profile of endowments, there are only finitely many competitive equilibria and,
at least in a local sense, competitive equilibrium changes smoothly in response to
small perturbations in endowments.25 Here, we will skip these interesting, but more
technical, issues. For the sake of completeness, and appendix includes a proof of
existence of competitive equilibrium for exchange economies.

N ORMATIVE A NALYSIS

3.3

We now study whether, ethically, competitive equilibria are acceptable: we study


the relationship between the competitive equilibrium allocations, the set of Pareto
efficient allocations and the core of the economy.
3.3.1

T HE F IRST F UNDAMENTAL T HEOREM OF W ELFARE E CONOMICS

Pareto efficiency is a minimal criterion for social optimality.26 The first key result in
normative general equilibrium theory is that, under mild assumptions, equilibrium
allocations display this minimal property.
P ROPOSITION (The FFTWE for production economies). Given a production economy,
{(ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 },
let (p, x, y) be a competitive equilibrium. If all consumers have locally nonsatiated preferences,
then the equilibrium allocation (x, y) is Pareto efficient.
24

The key properties are that the economy be bounded, in the sense that arbitrarily large production

is unfeasible, convex, in the sense that its demand and supply correspondences are convex-valued, and
continuous, in the sense that these latter correspondences are also continuous.
25
Importantly, one must notice that the last result holds for almost all values of endowments,
but not for all of them: a result known as the Sonnenschein-Mantel-Debreu Theorem shows that there are
economies where the predictive power of competitive equilibrium is very low.
26
This is a personal value judgement: just my opinion.

29

Proof: Suppose that the statement is not true: suppose that there exists an alternative
allocation (
x, y) such that
1. for all i, ui (
xi ) ui (xi ); and
0

xi ) > ui (xi ).
2. for some i0 , ui (
By feasibility, we must also have that
3. for all j, yj F j ;
4.

PI

i=1

xi =

PI

i=1

wi +

PJ

j=1

yj ;

By 3, it must be that for all j, p y j p yj , since y j maximizes profits for firm j at


P
0
0
0
0
prices p. By 2, p xi > p wi + Jj=1 si ,j p y j , since xi maximizes utility for individual
P
i0 , given prices p. Suppose now that for some i, p xi < p wi + Jj=1 si,j p y j ; then,
by local nonsatiation and 1, there exists an alternative bundle x such that p x
P
x) > ui (xi ), which is impossible since xi maximizes
p wi + Jj=1 si,j p y j and ui (
P
utility for individual i, given prices p. Since Ii=1 si,j = 1 for all j, all this implies that
PI
PJ
PI
i
j
i
y

+
p

>
p

(
i=1 w ), which contradicts condition 4.
j=1
i=1
It must be, then, that such alternative allocation does not exist.

Q.E.D.

For exchange economies, we can, in fact, say more.


P ROPOSITION (The FFTWE for exchange economies). Let (p, (xi )Ii=1 ) be a competitive
equilibrium of exchange economy {(ui , wi )Ii=1 }. If all consumers have locally nonsatiated
preferences, then the equilibrium allocation (xi )Ii=1 is Pareto efficient and is a core allocation.
E XERCISE 3.6. Notice that the previous theorem makes two statements: that, under the given
hypotheses, the equilibrium allocation is Pareto efficient, and that it lies in the core of the
economy. Which of the two statements is stronger? Argue the stronger statement, and obtain
the weaker one by immediate implication.
Notice that these two theorems: (i) do require local nonsatiation; (ii) do not use
continuity or convexity, and take as given a competitive equilibrium (so they do not
imply its existence); and (iii) crucially require the implicit assumptions of competitive equilibrium (as we have so far defined it): markets are complete, all agents,
firms and producers, are price takers, and there are no external effects. On the other
hand, it is necessary to understand the implication of the theorem. If one accepts its
assumptions, the theorem implies that competitive markets deliver allocations with
30

the minimal property of social desirability, as Smith had suggested. But it does not
say more than that! It is clear that Pareto efficiency does not take into account any
distributional considerations, and hence many efficient allocations may be socially
objectionable. In that sense, the theorem should not be understood to imply that economic policy is unnecessary if competitive markets operate. What the theorem does
say is that any economic policy beyond the equilibrium outcome will make at least
one individual worse off; although this result may be socially desirable, what cannot
be expected is victimless policies.
E XERCISE 3.7. Suppose that there are two societies, I1 = {1, . . . , I1 } and I2 = {I1 +
1, . . . , I1 + I2 }, where all individuals are locally nonsatiated. Let the global society be denoted by I = I1 I2 . Argue that there can be no unanimous opposition to globalization in
any of the two societies: suppose that (p, x) is a competitive equilibrium of the global economy,
{I, (ui , wi )iI } and (
xi )iIk is an allocation of the autarkic economy {Ik , (ui , wi )iIk }, for
k = 1, 2; if there is an individual i Ik that would prefer autarky, ui (
xi ) > ui (xi ), then there
0

is also an individual i0 Ik who prefers globalization: ui (


xi ) < ui (xi ).
E XERCISE 3.8. Consider an exchange economy with two consumers and two commodities:
u1 (x1 , x2 ) = min{x1 , 2x2 } , w1 = (3, 1),
u2 (x1 , x2 ) = min{2x1 , x2 } , w2 = (1, 3).
Find all competitive equilibria, all Pareto efficient allocations and the core of this economy.
Verify the relations that exist between these solutions.

3.4

T HE S ECOND F UNDAMENTAL T HEOREM OF W ELFARE E CONOMICS

We now study the opposite problem: given an efficient allocation, can we say that,
for sure, it is an equilibrium allocation? Stated like this, the answer to the question is
obviously negative: there are efficient allocations that cannot be sustained as equilibrium. However, we now show that if redistribution policies are allowed, all efficient
allocations can be sustained by competitive trading.
P ROPOSITION (The SFTWE for exchange economies). Given an exchange economy {(ui , wi )Ii=1 },
let x be a Pareto efficient allocation. If all consumers have continuous, convex, locally nonP
satiated preferences and Ii=1 wi  0, then there exist prices p and a distribution of wealth
(w i )Ii=1 such that

31

1. distribution (w i )Ii=1 is a reallocation of the existing aggregate endowment:


PI
i
i=1 w ;

PI

i=1

w i =

2. (p, x) is competitive equilibrium for the economy after redistribution, {(ui , w i )Ii=1 }.
The simplest argument to see that the theorem is true is as follows. Redistribute
wealth so that each individual receives the allocation that would correspond her in
the efficient allocation: let w i = xi for each i. The first condition in the theorem is
immediate, since x is an allocation for the economy. The economy after redistribution,
{(ui , w i )Ii=1 }, must have a competitive equilibrium (p, x).27 By individual rationality,
ui (
xi ) ui (w i ) for all i, and, since (
xi )Ii=1 is efficient, it must be that a fortiori ui (
xi ) =
ui (w i ), so (p, (xi )Ii=1 ) is a competitive equilibrium for {(ui , w i )Ii=1 }. This proves the
second claim and completes the argument.
In the case of production economies, the argument is more complicated and requires the use of a result known as the Separating Hyperplane Theorem. We wont cover
that argument here, but if you really feel like studying it you can find it in the appendix of this note. The theorem itself is a bit more complicated.
P ROPOSITION (The SFTWE for production economies). Given a production economy,
{(ui , wi , (si,j )Jj=1 )Ii=1 , (F j )Jj=1 },
let (x, y) be a Pareto efficient allocation such that xi  0 for all i. If all preferences are
continuous, convex and locally nonsatiated and all technologies are convex and closed and
satisfy free disposal, then there exist prices p and a distribution of nominal wealth in the
economy (mi )Ii=1 that satisfy the following conditions:
1. the nominal wealth being distributed is indeed the nominal value of the aggregate wealth
P
P
P
of the economy at the Pareto efficient allocation: Ii=1 mi = p ( Ii=1 wi + Jj=1 y j );
2. given their nominal wealth, each consumer is individually rational at prices p: for all i,
xi solves the problem maxx ui (x) : p x mi ; and
3. each firm maximizes profits at prices p: for all j, y j solves the problem maxy p y : y
Fj.
27

This is the step where the argument is less formal: we have not studied existence results in detail,

yet we are arguing that an equilibrium must exist given the assumptions that we have made. While
the latter is true (see the appendix: an equilibrium is guaranteed to exist under these assumptions),
here we will have to take it for granted.

32

Notice that, unlike the first theorem, the second fundamental does imply existence of equilibrium, so the convexity assumption is crucial. The policy implication
is that policy-makers do not need to close competitive markets to attain social objectives (which, one assumes, are Pareto efficient). Quite the opposite: well chosen
redistribution policies and competitive markets, under the assumptions of the theorem, deliver the desired objectives! Of course, the problem of how much information
a policy-maker needs in order to figure out the correct redistribution is not addressed
by the theorem.
E XERCISE 3.9. Fix an exchange economy {(ui , wi )Ii=1 }. An allocation x is said to be envyfree if no agent would (strictly) prefer someone elses consumption: for every pair of individ0

uals i and i0 , it is true that ui (xi ) ui (xi ). Argue that income reallocation can ensure that
every competitive allocation is envy-free: there exists a distribution of endowments (w i )Ii=1
such that:
1. distribution (w i )Ii=1 is a reallocation of the existing aggregate endowment:
PI
i
i=1 w ;

PI

i=1

w i =

2. after redistribution, competitive allocations are envy free: if (p, x) is a competitive equilibrium of {(ui , w i )Ii=1 }, then x is envy-free.
E XERCISE 3.10. Suppose that there are two societies, I1 = {1, . . . , I1 } and I2 = {I1 +
1, . . . , I1 + I2 }. Let the global society be denoted by I = I1 I2 . For each society, k = 1, 2, let
(pk , (xi )iIk ) be a competitive equilibrium of the autarkic exchange economy {Ik , (ui , wi )iIk }.
1 +I2
Argue that there exists an global income reallocation (w i )Ii=1
such that:

1. income reallocation is balanced in each society: for each k = 1, 2,

iIk

w i =

iIk

wi ;

2. every competitive equilibrium of the global economy gives an allocation that everybody
prefers to the given autarkic allocation: for any equilibrium (p, (
xi )iI ) of the global
economy after redistribution, {I, (ui , w i )iI }, it is true that ui (
xi ) ui (xi ) for every
individual i.

3.5

A DDITIONAL E XERCISES

E XERCISE 3.11. Consider an exchange economy. An allocation x is said to be Weakly Pareto


efficient if there does not exist an allocation x such that ui (
xi ) > ui (xi ) for all i. Argue that:
1. any Pareto efficient allocation is also weakly Pareto efficient;
33

2. if all preferences are continuous and strongly monotone, any weakly Pareto efficient
allocation is also Pareto efficient.
E XERCISE 3.12. In a production economy, a profile of production plans y = (y j )Jj=1 is said
to be (i) feasible if it is true that y j F j for each firm j; and (ii) technically efficient if it
is feasible and there does not exist an alternative, feasible production plan y = (
y j )Jj=1 such
P
P
that j yj > j y j . Argue the following: if at least one individual has strictly monotone preferences and allocation (x, y) is Pareto efficient, then the profile of production
plans y must be technically efficient.
E XERCISE 3.13. Consider an exchange economy with two consumers and two commodities:
u1 (x1 , x2 ) = x1 , w1 = (1, 1),
u2 (x1 , x2 ) = x2 , w2 = (1, 1).
Find all competitive equilibria, all Pareto efficient allocations and the core of this economy.
Verify the relations that exist between these solutions.
E XERCISE 3.14. Consider an exchange economy with two commodities and two consumers.
Preferences are u1 (x) = min{x1 , x2 } and u2 (x) = x1 x2 . Endowments for individual 2 are
w2 = (0, 20).
1. Compute all competitive equilibria if w1 = (30, 0).
2. Compute all competitive equilibria if w1 = (5, 0).
Arent these results funny?
E XERCISE 3.15. Given an exchange economy ({1, ..., I}, (ui , wi )Ii=1 ), argue:
1. That if the endowment (wi )Ii=1 is itself an efficient allocation, then it lies in the core of
the economy.
2. That if all individuals have strongly convex preferences and the endowment (wi )Ii=1 is
an efficient allocation, then it is the only allocation in the core of the economy.
3. That if all individuals have locally nonsatiated and strongly convex preferences and the
endowment (wi )Ii=1 is an efficient allocation, all competitive equilibria of the economy
involve no (nontrivial) trade between agents.

34

E XERCISE 3.16. Let C and P be, respectively, the core and the set of efficient allocations of a
given exchange economy with two consumers. Argue that
C = {(xi )Ii=1 P|ui (xi ) ui (wi ) for both i}.
Argue that if there are three or more individuals, then the claim is not true.
E XERCISE 3.17. Consider an economy with one consumer, one firm and two commodities; the
individual has preferences over both commodities represented by u and has an endowment w
of commodity 1 only; the firm transforms commodity 1 into commodity 2, under a production
function f ; the individual owns the stock of the firm. Define competitive equilibrium for this
economy; define Pareto efficiency for this economy; state and prove the First Fundamental
Theorem of Welfare Economics for this economy.
E XERCISE 3.18. Consider an exchange economy in which each ui represents locally nonsatiated, strongly convex preferences. For each i, denote by hi the Hicksian demand function.
Define, (p, (xi )Ii=1 ) to be a pseudoequilibrium if:
1. For all i, xi = hi (p, ui (xi )) and p xi = p wi ;
2.

PI

i=1

xi =

PI

i=1

wi .

Considering only strictly positive prices, argue that (p, (xi )Ii=1 ) is a pseudoequilibrium if, and
only if, it is a competitive equilibrium.

35

A PPENDIX : E XISTENCE OF COMPETITIVE EQUILIBRIUM


The key mathematical result that we will be invoking in this appendix is the following:
T HEOREM (Kakutanis fixed-point theorem). Let RL and let : be a non-emptyvalued correspondence. If is compact and convex, and is convex-valued and upper hemicontinuous, then there exists such that ().
When is single-valued (i.e. a function), the result is referred to as Browers fixed-point
theorem and is easy enough to visualize in the case L = 1.
P
P ROPOSITION . Suppose that Ii=1 wi  0 and that each ui is strongly quasiconcave and strictly
monotone. Then, there exists a competitive equilibrium.
Proof: Denote by o and the relative interior and the boundary of the (L 1)-dimensional
unit simplex, , respectively. The aggregate excess demand function over strictly positive
prices, Z : o RL , is continuous and bounded below and satisfies that for all p o ,
p Z(p) = 0.
o

Let a sequence (pn )


n=1 in be such that pn p . Suppose it is not true that
maxl=1,...,L {Zl (pn )} . Then, for some x R it is true that for all n , there exists n n
such that maxl=1,...,L {Zl (pn )} x. Since Z is bounded below, there exists (pn(m) )
m=1 such
PI
i

that (Z(pn(m) ))m=1 is bounded. Since i=1 w  0, then for some i {1, . . . , I} we must
i

have p wi > 0. Fix one such i. Since (Z(pn(m) ))


m=1 is bounded, then (x (pn(m) ))m=1 is
bounded and, hence, has a convergent subsequence. For notational simplicity, assume that
L

(xi (pn(m) ))
RL
+
m=1 itself converges to x R+ . Let l {1, . . . , L} be such that p
l = 0. Let x
be defined as follows:
(
xl ,
if l 6= l;
x
l =
xl + 1 if l = l.
Since x
> x, ui (
x) > ui (x). By continuity,  > 0 such that for all x0 B (
x) RL
+ and all
00
i
0
i
00
i
x B (x), u (x ) > u (x ). Since x (pn(m) ) x, there exists some m1 N such that for all
m m1 , xi (pn(m) ) B (x). Fix l0 {1, . . . , L} such that pl0 > 0. Define (xn(m) )
m=1 as follows:

xl (pn(m) ) + 1, if l = l;
xl,n(m) =
xil0 (pn(m) ) 2 , if l = l0 ;

xi (p
otherwise.
l n(m) ),
Since pl0 ,n(m) pl0 > 0 and pl,n(m) pl > 0, there exists m2 N such that for all m m2 ,
pl0 ,k(m) ( 2 ) + pl,n(m) < 0. Now, let m max{m1 , m2 }. Then,

pn(m) xn(m) = pn(m) xi (pn(m) ) + pl0 ,n(m) pl0 ,n(m) ( )
2
< pn(m) xi (pn(m) )
pn(m) wi ,
and, nonetheless, xi (pn(m) ) B (x) and xn(m) B (
x), so ui (xn(m) ) > ui (xi (pn(m) )), which is
a contradiction. It follows that maxl{1,...,L} {Zl (pn )} .

36

Now, define correspondence : as follows:


(
argmax Z(p) , if p o ;
(p) =
{ |p = 0},
if p .
Notice that: (i) is nonempty- and convex-valued; (ii) if p o and Z(p) 6= (0, . . . , 0) then
(p) ; (iii) if p , then p
/ (p). That is upper hemicontinuous at p o follows
from the Theorem of the Maximum, given that Z is continuous. Now, let p , (pn )
n=1 in

such that pn p, and (n )n=1 in such that n (pn ) for each n. Since is compact, there
exist a subsequence (n(m) )
m=1 and a such that n(m) . Consider two cases: (1)

o
(pn(m) )n(m)=1 has no subsequences in o ; and (2) (pn(m) )
m=1 has a subsequence in . In (1),
since pn(m) p, for some m N we have that for all m m , pn(m) and pn(m) n(m) =
o

0, so p = 0. In (2), take the subsequence (pn(m(k)) )


k=1 in . Since pn(m(k)) p , by
the property above, there exists k N such that, for all k k and all l {1, . . . , L} such
that pl > 0, Zl (pn(m(k)) ) < maxl0 {1,...,L} {Zl0 (pn(m(k)) )}. It follows that for all k k and all
l {1, . . . , L} such that pl > 0, one has that l,n(m(k)) = 0 and, hence, pn(m(k)) n(m(k)) = 0.
Again, this implies that p = 0, and hence that is upper hemicontinuous at p .
Since is upper hemicontinuous, by property (i) and Kakutanis theorem there exists
some p such that p (p). By (iii), p o and hence, by (ii), Z(p) = 0.
Q.E.D.
Existence results under milder conditions can be given. For example, for production
economies:
T HEOREM . Given a standard production economy where each wi > 0, if each Y j satisfies free disposal
and possibility of inaction, and
{(x, y)

I
(RL
+)

J
Y

Y |

j=1

I
X
i=1

I
X
i=1

wi +

yj }

jJ

is compact, then there is a competitive equilibrium.

A PPENDIX : PROOF OF THE SFTWE FOR PRODUCTION ECONOMIES


The key mathematical result needed for this proof is the following.
T HEOREM (The separating hyperplane theorem). If Q, Q0 RA are disjoint and convex, then
there exist a vector RA , 6= 0, and a constant k such that (i) for every q Q, it is true that
q k; and (ii) for every q Q0 , it is true that q k.
Now, we can prove the SFTWE as follows. For each i, let U i = {x|ui (x) > ui (xi )} and
P
P
P
define the set U = Ii=1 U i .28 Define also the set F = { Ii=1 wi }+ Jj=1 F j . By quasiconcavity
of each function ui , each set U i is convex, so it follows that U is convex. By convexity of all F j
technologies, we also have that set F is convex. Since ((xi )Ii=1 , (y j )Jj=1 ) is efficient, it follows
that U and F are disjoint. Then, by the separating hyperplane theorem, one can find a vector
28

Notice that we are adding sets! This operation is defined as follows: for two sets A and B, we
define A + B = {x|x = xA + xB for some xA A and some xB B}.

37

p RL , p 6= 0, and some constant k such that p z k for all z U , and p z k for all z F .
By free-disposal, we have that p > 0.
For each consumer i, fix any x
i such that ui (
xi ) ui (xi ). By local nonsatiation, we can
i
find, for any natural number n, some bundle xn U i such that ||xin x
i || < 1/n. It follows
PI
from above, then, that p i=1 xin k for every n. Letting n , we conclude that p
PI
P
P
i k. (In particular, this implies that p Ii=1 xi k.) Since Ii=1 xi F , we have that,
i=1 x
moreover,
I
I
J
X
X
X
p
xi = p (
wi +
y j ) k,
i=1

i=1

j=1

P
so p Ii=1 xi = k.
As a consequence of the latter, we have that, for all individuals, ui (x) > ui (xi ) implies
p x p xi . Similarly, for all firms, y F j implies p y p y j , since
I
X

wi + y +

i=1

y j F,

j 0 6=j

which implies that


I
X
X 0
p(
wi + y +
y j ) k.
j 0 6=j

i=1

Define mi = p xi for each consumer. The first implication of the theorem follows by
construction, while the third part was argued above. Now, suppose that for some individual
i we have that for some bundle x, ui (x) > ui (xi ) and p x mi are both true. By our previous
result, p x = mi > 0, and, hence, by continuity of preferences, for  (0, 1) close enough to
1, we have ui (x) ui (xi ) and p (x) < mi = p xi , which contradicts our previous result.
This proves the second part of the theorem.

38

C HOICE UNDER U NCERTAINTY

In the first lectures of this course, we considered a situation in which a decision-maker


has to make a choice, the consequences of which we interpreted as certain. We now
consider a situation in which this is no longer the case.
A state of the world is a comprehensive description of the state of all the contingencies that can affect a decision-maker. In the words of Arrow,29 it is a description
of the world so complete that, if true and known, the consequences of every action
would be known. Let S be the set of states of the world. A random variable is a function whose domain is S. If the target set of a random variable is the set Y , we say that
it is a random variable over Y .
Let X 6= be a set of outcomes (or prospects),30 and let be the space of all probability distributions over X . For simplicity of presentation, let us take X to be finite, and
P
let us write X = {1, . . . , X}; then, is {p RX
+|
x px = 1}.
A lottery is a random variable over : it is a function L : S . That is, a lottery
is a device that assigns to each state of the world, s, a probability distribution over
the set of prospects, ps = L(s); under this device and given that state, the probability
P
s
of prospect x is psx , and X
x=1 px = 1.
A lottery fixes the probability of a prospect, given a state, but it does not determine the probability of that state. It is commonly interpreted that the probabilities of
states are exogenous to economic models, and are usually taken to be subjective to
the decision-maker, whereas the probabilities induced by lotteries, given a state, are
objective. A full theory that handles both types of probability is possible, but here,
for simplicity, we will deal with only one of the two types of uncertainty at a time.31

4.1

P REFERENCES OVER L OTTERIES

For simplicity, let us assume that there is only one state of the world, so that we can
ignore the set S and can refer to itself as the space of lotteries.
An individuals preferences, %, are a binary relation over . When we define
29
30

Arrow, K. (1971), Essays on the Theory of Risk Bearing, p. 45.


This could be an abstract set, or, if you would like more definiteness, an set X R, of monetary

values.
31
In fact, a richer theory where the decision maker is unsure of what subjective probabilities to
assign to states of the world is possible. Often, people reserve the term uncertainty for the latter
phenomenon, and use risk for the randomness that remains even when probabilities (subjective and
objective) are fixed. Here, we wont need this distinction.

39

preferences in this way, we are imposing the condition that the individual cares about
the risk (randomness) she faces, and not about the process that ultimately determines
that risk; this condition is known as consequentialism. As a binary relation, the
properties that we defined in the first lecture note may be brought to this setting.
Henceforth, we assume that % is rational.
4.1.1

P ROPERTIES OF P REFERENCES

Notice that is a convex set, and, therefore, the property of convexity of preferences
can be applied in this setting. Monotonicity, though, has to be redefined, as it is
impossible that p > p0 for two lotteries in !
We say that % satisfies monotonicity if given two lotteries, p and p0 such that p  p0 ,
the following statement is true: p + (1 )p0  p + (1 )p0 if, and only if, > . In
words, a decision-maker with monotonic preferences prefers more of a better lottery
to more of a worse lottery.
Another condition, for which we had no analogous before, imposes that the decisionmaker values the outcomes of the lotteries for themselves and then, independently,
the randomness induced over them by the lottery: we say that % satisfies independence
if given two lotteries p and p0 , the following statements are true:
1. if p % p0 , then for any number 0 1 and any lottery p00 we have that
p + (1 )p00 % p0 + (1 )p00 ;
2. if for some number 0 1 and some lottery p00 we have that
p + (1 )p00 % p0 + (1 )p00 ,
then p % p0 .
The latter property is controversial, and we will come back to it later. The following exercise relates the two properties.
E XERCISE 4.1. Argue that independence of % implies the following: for any pair of lotteries
p and p0 ,
1. if p p0 , then for any 0 1 and any lottery p00 , p + (1 )p00 p0 + (1 )p00 ;
2. if for some 0 1 and some lottery p00 we have that p+(1)p00 p0 +(1)p00 ,
then p p0 .
40

3. if p  p0 , then for any 0 < < 1 and any lottery p00 , p + (1 )p00  p0 + (1 )p00 ;
4. if for some 0 < < 1 and some lottery p00 we have that p+(1)p00  p0 +(1)p00 ,
then p  p0 ;
5. if p  p0 and 0 < < 1, then p  p + (1 )p0 and p + (1 )p0  p0 .
E XERCISE 4.2. Argue that independence of % implies its monotonicity.32
4.1.2

R EPRESENTABILITY

We again ask the question of when % can be represented by a utility function. It


turns out that, again, it depends on whether preferences have sudden jumps or
not, which is the same result that we had in consumers theory. Now, however, we
may want to have special properties on the utility function that represents %.
We say that % has an expected-utility representation if there exists a function u : X
R such that for any pair of lotteries p and p0 , we have that p % p0 if, and only if,
X

px u(x)

p0x u(x).

In this case, we can define the utility function over lotteries U (p) =

px u(x), and

it is immediate that U represents %.


A couple of words on jargon are in order. Notice that the expected utility name
is well chosen: by construction, U (p) = Ep (u). Now, sometimes different things are
given the same name: some people refer to u as Bernoulli utility function and to
U as von Neumann-Morgenstern utility function, while some other people refer to
u itself as the von Neumann-Morgenstern utility function, and some other people
use both names for u and leave U nameless. This can be problematic, as the two
functions are not the same thing: u measures utility over outcomes, while U does it
over lotteries. Here, we will refer to U as the utility function and to u as the utility
index.
E XERCISE 4.3. Argue that if % has an expected-utility representation then it satisfies independence and monotonicity.
A seminal result is decision theory is the following:
32

This exercise is a tiny bit more complicated than the others. Hint 1: suppose that you want to

write p + (1 )p0 as p + (1 )(p + (1 )p0 ), given that > ; what value must have? Hint
2: now, notice the last property of exercise 4.1.

41

P ROPOSITION (The von Neumann-Morgenstern Theorem). If % is continuous and satisfies independence, then it has an expected-utility representation (with continuous u).
E XERCISE 4.4. Argue the following: If U is an expected-utility representation of %, then it
satisfies the following linearity property: for any pair of lotteries p and p0 and any number
0 1,
U (p + (1 )p0 ) = U (p) + (1 )U (p0 ).
P ROPOSITION 4.1. Suppose that U and U are expected-utility representations of %. Let
u and u be their respective utility indices. There exist numbers and > 0 such that
u(x) = + u(x) for every x.
The previous proposition is important: only positive affine transformations of a
utility index preserve the expected-utility representation of %;33 this means that u
itself is a cardinal object.
4.1.3

D ISCUSSION :

IS INDEPENDENCE A GOOD ASSUMPTION ?

Independence is a strong assumption: it implies that the decision-maker is able to


evaluate outcomes without worrying about the randomness between them (which
is fine, as degenerate lotteries allow for that), and then evaluates the randomness in
a way which is perfectly consistent with the deterministic evaluation. The second
step is controversial: one can think of cases in which the winning and losing outcomes
are so related that the randomness cannot be assessed independently.
A canonical observation is the following: consider a space of monetary outcomes,
and suppose that the following lotteries are available:
x

p1 =

p3 =

p1x
x
p3x

0
0
0
0.89

1 5
1
1
0.11

p2 =

0
5

p4 =

p2x

0.01

p4x

0.9

0.1

0.89 0.1

It has been observed that when asked to compare these lotteries, many people respond that p1  p2 and p4  p3 (are these your preferences too?). But it turns out that
these preferences cannot have an expected-utility representation! To see this, notice
33

Dont get confused: any monotone transformation of U will represent % as well; but the transfor-

mation need not preserve the expected-utility property, which requires affinity.

42

that a representation would require numbers u(0), u(1) and u(5) such that U (p) = Ep u.
If one assumes that the decision-maker prefers more wealth to less, then without any
loss of generality, we can take u(0) = 0 and u(5) = 1, and the revealed choices say
that u(1) satisfies the following two inequalities:
Ep1 u = u(1) > 0.89u(1) + 0.1 = Ep2 u,
while
Ep3 u = 0.11u(1) < 0.1 = Ep4 u.
But, then, from the first equation u(1) > 10/11, while from the second equation u(1) <
10/11.
This observation, known as Allaiss Paradox, suggest that most people dont conform to the assumptions required for their preferences to have an expected utility
representation. A first interpretation of the puzzle is that indeed people dont satisfy
the independence condition. An alternative explanation is that people dont satisfy
the consequentialist principles, and are sensitive to how the final randomness is
obtained.

4.2

R ISK AVERSION

We now want to model the decision-makers taste or distaste for risk. We may be
tempted to say that if % is convex then she dislikes risk, but this would be a mistake:
a convex combination of lotteries does not reduce their riskiness! What we want to do,
instead, is to compare the agents ranking of a risky lottery and a degenerate lottery
that gives her the expected return of the risky lottery. For this to be possible, we
must abandon the simplifying assumption of finiteness of X , and we assume instead
that X is some interval in R, which we can interpret as the space of wealth levels,
X, of the individual. Still, a lottery is a probability distribution p over X , but we
restrict attention to lotteries for which an expected payoff is defined: there exists a
real number Ep X, such that

Z
xdp(x) = Ep X.
X

For simplicity of notation, we identify a wealth level x with the degenerate lottery that
gives that prize with probability 1.
In this setting, % is said to be
1. risk averse if for any p, Ep X % p;
43

2. strictly risk averse if for any p such that Ep X 6= p, Ep X  p;


3. risk neutral if for any p, Ep X p.
That is, a risk averse individual would always prefer the expected payoff of a lottery for sure to the lottery itself, and risk neutral if she is always indifferent between
the two lotteries so defined; she is strictly risk averse if she strongly prefers the riskless lottery, unless the other lottery is riskless itself.

4.3

E XPECTED U TILITY AND R ISK AVERSION

Let us assume that % has an expected-utility representation: we can find u : X R


such that p % p0 if, and only if, Ep u Ep0 u.
P ROPOSITION . Under the assumptions above:
1. % is risk averse if, and only if, u is concave;
2. % is strictly risk averse if, and only if, u is strongly concave.
3. % is risk neutral if, and only if, u is affine (for some numbers and , we have
thatu(x) = a + bx).
We can easily see that this is true, for a simplified case: lotteries that have only
two possible prizes. In this setting, if lottery p gives probability to prize x and 1
to x0 , risk aversion implies that
u(x + (1 )x0 ) u(x) + (1 )u(x0 ),
so it must be that the cardinal utility index u is concave. Under strict risk aversion, if
p is not degenerate, the latter inequality must always be strict, so u must be strongly
concave. Finally, under risk neutrality we must always have an equality, so u must be
both concave and convex, hence affine.
4.3.1

M EASURES OF R ISK AVERSION

Two measures of how averse to risk a person is are widely used. Given the result
above, it is not surprising that these measures are based on the curvature of the cardinal utility index. For the remainder of the section, we assume that u is differentiable
twice: it is by its second derivative that we will capture the curvature of u. We also
44

assume that u0 > 0.


A BSOLUTE RISK AVERSION :
Suppose that our decision-maker has an income subject to risk: it is determined
and Vp X = Ep (X X)
2 = . How much (premium)
by the lottery p. Let Ep X = X
instead of the lottery?
would she be willing to pay in order to secure an income of X
Let be the answer to this question: satisfies
) = U (p) = Ep u(X).
u(X
E XERCISE 4.5. Consider a consumer with expected-utility preferences and cardinal utility

index u(x) = x over nonnegative wealth levels. Suppose that she faces uncertain wealth,
distributed uniformly over the interval (0, 1). How much would she be willing to pay to
insure against this wealth uncertainty? How does your answer change if u(x) = x? What if
u(x) = x2 ?
Now, finding in general can be complicated, but we can study the last equality
by approximating its terms. For the right-hand side, notice that
+ u0 (X)(X

+ 1 u00 (X)(X

2 ) = u(X)
+ 1 u00 (X).

Ep u(X) Ep (u(X)
X)
X)
2
2
For the left-hand side,
) u(X)
u0 (X).

u(X
Equating, we get that

1 u00 (X)
,
2 u0 (X)

= u000 (X)
so it follows that the coefficient of absolute risk aversion, defined as A(X)
, is a
u (X)
To
good measure of the individuals aversion to risk, when her expected wealth is X.

see what type of risk is captured by this coefficient, notice that we could reinterpret
subject to shocks Z = X X,

things as if the individual had an expected income X,


where the shocks have mean EL Z = 0 and variance Vp Z = EL Z 2 = . Because of this,
is a good measure of the individuals attitude to
one usually understands that A(X)
additive risk.
E XERCISE 4.6. Suppose that u(x) = exp(x). What is the individuals coefficient of
What if u(x) = x1 , when 6= 1? What
absolute risk aversion? How does it depend on X?
1

if u(x) = ln x? What if u(x) = x? What if u(x) = x? What if u(x) = x2 ?

45

R ELATIVE RISK AVERSION :


Suppose now that, in the same situation as before, we ask what proportion of her
expected income the decision-maker would be willing to spend to secure her income?
Letting be that proportion, we have that
X)
= U (p) = Ep u(X).
u(X
Now, on the left-hand side we get
X)
u(X)
u0 (X)
X,

u(X
and, hence,

1 u00 (X)
1 u00 (X)

,
2 u0 (X)
2 u0 (X)

where

= 2 = Ep
X



2
X X
X X
= Vp
.

X
X
00

= u (0 X) X is a second
We thus get that the coefficient of relative risk aversion, R(X)
u (X)
The
measure of the individuals attitude to risk, when her expected wealth is X.
difference is that this measure is designed for multiplicative risk: suppose that the
but it is subject to proportional shocks z X,
with
individuals expected income is X,
the random variable
z=

X X
.
X

E XERCISE 4.7. For the cardinal utility indices defined in exercise 4.6, find the coefficients of
relative risk aversion and determine how they depend on the expected income.

4.4

S TOCHASTIC D OMINANCE

We now want to study monotonicity properties for lotteries. We need a new framework for this, as it would be a mistake to pretend that we can order lotteries using the
standard greater-than relation >. For simplicity, let us consider the case of lotteries
that pay in nonnegative units of some num`eraire (money), so that we represent them
by the probabilities they assign to any nonnegative number x: a lottery will be a c.d.f.
function F : R+ [0, 1].34
A first definition of when a lottery is larger than another is given by the following definition:
34

If a lottery F has a density function, we will denote this function by f : R+ R+.

46

D EFINITION . A lotery F is as large as lottery F in the sense of first-degree stochastic


dominance if F (x) F (x) for every every possible payoff level x. F is said to dominate F
in the sense of first-degree stochastic dominance if it is as large, and the above inequality
is strict at some payoff level.
For simplicity, we will write that F %F S F if F is as large as F in the sense of
first-degree stochastic dominance, and that F F S F if F dominates F in the same
sense. This is a good (i.e. intuitive) concept, under the premise that more wealth is
better than less: by definition, both F and F are nondecreasing and limx F (x) =
x F (x) = 1, so when one says that F %F S F , this means that, at any point, F
lim
leaves at least as much probability to be allocated to higher payoffs than F does. The
following proposition formalizes this; for simplicity of presentation, we consider here
only the case of a discrete random variable giving positive probability only to some
integer numbers, and defer the more general case to an appendix.
P ROPOSITION 4.2. Consider two lotteries, F and F , that give positive probability only to
payoffs in the set {0, 1, . . . , x}, for some positive integer x. Then,
1. if F F S F , then for any increasing utility index u, we have that EF [u(X)] >
EF [u(X)]; and
2. conversely, if F 6= F and it is not true that F F S F , then for some increasing utility
index u one has that EF [u(X)] < EF [u(X)].
Proof: For the first statement, by definition,
EF [u(X)] EF [u(x)] =

u(x)(F (x) F (x 1))

x=0

u(x)(F (x) F (x 1)),

x=1

with F (1) = F (1) = 0. Rearranging terms,35 the right-hand side of this expression
is
u(
x)(F (
x) F (
x))u(0)(F (1) F (1))+

(u(x)u(x1))(F (x1)F (x1)) < 0,

x=1
35

The following expression can be seen as an application of the following identity: for any functions

: {1, . . . , I} R and : {0, 1, . . . I} R, if one defines (i) = (i) (i 1) for each i = 1, . . . , I,


then

I
X

(i)(i) = (I)(I) (1)(0)

i=1

I
X
i=2

47

((i) (i 1))(i 1).

where the inequality follows since F (


x) = F (
x), F (1) = F (1), while u is increasing
and F F S F .
For the second statement, since F 6= F and it is not true that F F S F , it must
be that for some income level x , it is true that F (x ) > F (x ). Fix one such x ,
and construct the following (nondecreasing) function: v(x) = 0 for any x x , and
v(x) = 1 for all x > x . Then, for any c.d.f. F , by definition, E [v(X)] = 1 F (x ), so
F

it follows that EF [v(X)] > EF [v(X)]. Since this inequality is strict, we can modify v to
construct an increasing index u for which the inequality holds too.

Q.E.D.

First-degree stochastic dominance, however, can sometimes be too strong as a concept of deminance for lotteries. A second, weaker concept is given next.
D EFINITION . A lotery F is as large as lottery F in the sense of second-degree stochastic
dominance if
Z

Z
F (s)ds

F (s)ds

for every every possible payoff level x. F is said to dominate F in the sense of seconddegree stochastic dominance if it is as large, and the above inequality is strict at some
payoff level.
As before, we will use F %SS F and F SS F to denote stochastic dominance
in the second-degree sense. It is immediate that first-degree stochastic dominance
implies second-degree stochastic dominance, but the converse is not true. What the
second concept captures is the difference in the speeds at which different lotteries
accrue probability over low payoffs. The following proposition illustrates the importance of this concept; the proposition is stated without some technical assumptions,
which are deferred to the proof given in the appendix.
P ROPOSITION 4.3. Let F anf F be two continuous lotteries, with densities f and f. Then,
1. if F SS F , then for any increasing and strictly concave utility index u, we have that
EF [u(X)] > EF [u(X)]; and
2. conversely, if F 6= F and it is not true that F SS F , then for some increasing and
strictly concave utility index u one has that EF [u(X)] < EF [u(X)].
As before, we can illustrate this result in the discrete case considered in Proposition 4.2. In this case, since the domain of u is not convex, we replace the assumption
of concavity of the index by the condition that
(u(x) u(x 1)) (u(x 1) u(x 2)) < 0
48

for every x = 2, . . . , x, which is the discrete analogous of the condition the second
derivative of the function be negative. In this setting, suppose that F SS F , and
recall from the proof of Proposition 4.2 that
x
+1
X
EF [u(X)] EF [u(x)] =
(u(x) u(x 1))(F (x 1) F (x 1)).
x=1

Rewriting this expression, as before, we get that its right-hand side equals36
x
+1
x2
X
X

[(u(x) 2u(x 1) + u(x 2))


(F (s) F (s))].
x=2

s=0

Since, by assumption, each u(x)2u(x1)+u(x2) < 0 and each

Px2
s=0 (F (s)F (s))

0, with strict inequality somewhere, it follows that EF [u(X)] > EF [u(X)].


In order to understand exactly what second-order dominance captures, the following result is useful.
P ROPOSITION 4.4. Consider two continuous lotteries F and F with densities f and f, that
assign all the probability mass over the interval [0, x], and suppose that EF [X] = EF [X]. If
F SS F , then VF [X] = V [X].
F

The proof of the result is deferred to the appendix. Intuitively, under the premises
of the Proposition, lottery F takes probability mass from the center of the distribution (i.e. near the mean) and allocates it to both of its extremes; for a risk-averse
decision-maker, this makes the lottery worse. Under those premises, F is said to be a
mean-preserving spread of F .

A DDITIONAL E XERCISES
E XERCISE 4.8. Consider a firm
F = {x R3 | 1 x1 0, x2 x1 and x3 = 1}.
Suppose that p2 = 3 and p1 = 1, and find the supply and expenditure functions of this
firm, for any value of p3 > 0. Suppose that p3 is random, and can be 1 or 3 with equal
probability. Suppose that the firm is offered a future contract that allows it to secure the
price of commodity 3 at p. If the firms objective is expected profit, what is the largest p at
which the firm would be willing to take the future contract? If the owner of the firm has
The term (u(
x + 1) u(
x))(F (
x) F (
x)) (u(1) u(0))(F (0) F (0)), which also appears in the
expression, is zero since F (
x) = F (
x) and F (0) = F (0).
36

49

expected-utility preferences with cardinal utility index u(x) =

x, what is the largest p that

she would accept?.


E XERCISE 4.9. Consider a decision-maker facing risk. Her preferences over lotteries admit an
expected-utility representation with cardinal utility index u(c) = 1/c, and she expects to
have an income of w > 0. Suppose that she discovers that her income is subject to a random
shock, so her actual income will be w+X, where X is a random variable following the uniform
distribution over the interval [
x, x], with 0 < x < w. In this setting, answer:
1. What is the decision makers expected utility in the absence of any shocks to her income?
What is her expected utility in the presence of the shock to her income? What is her
expected income? Does this make sense? Show that limx0 E[u(X)] = u(w).
2. How much would she be willing to pay to insure against the random shock? What is her
coefficient of absolute risk aversion evaluated at her expected income? Let these values
be (w, x) and A(w, x), respectively.
3. Show that limx0 (w, x) = 0 but, still, limx0 A(w, x) = 2/w 6= 0. Does this make
sense?

50

A PPENDIX : T HE VON N EUMANN -M ORGENSTERN T HEO REM


Here, we give an informal argument for why the von Neumann-Morgenstern theorem is true.
For simplicity, we concentrate only on a small subclass of lotteries, rather than on the whole
space .
We say that a lottery is simple if it gives positive probability to at most two outcomes in
37
X . For simplicity, then, we can denote a simple lottery as a triple consisting of a number
and two outcomes, L = (p, x, x0 ), with 0 p 1 and x, x0 X, and with the interpretation
that the lottery gives outcome x with probability p, and outcome x0 with probability 1 p. Let
L1 be the space of simple lotteries, L1 = [0, 1] X X .
A compound lottery is a device that gives other lottery or lotteries as prizes. We will concentrate on compound lotteries that give positive probability to at most two simple lotteries,38
and denote them by (p, L, L0 ), a number and two outcomes, L, L0 L1 . Let L2 be the space of
simple lotteries, L1 = [0, 1] L1 L2 .
For our argument, we consider only degenerate, simple and our simplified definition of
compound lotteries, so we take % as defined over L = X L1 L2 . In order to keep consistency
with the analysis above, we need consider an individual who cares about outcomes, and
not about how these outcomes are presented, so we impose the following consequentialist
assumptions on %: for all p, p0 [0, 1] and for all x, x0 X,
1. (p, x, x0 ) (1 p, x0 , x);
2. (1, x, x0 ) x;
3. (p, (p0 , x, x0 ), x0 ) (pp0 , x, x0 ).
For simplicity, suppose also that we can find x , x X such that for every outcome x X
we have that x % x and x % x.
E XERCISE . Argue that:
1. x  x0 and 0 p < p0 1 imply that (p, x, x0 )  (p0 , x, x0 );
2. L  L0 and 0 p < p0 1 imply that (p, L, L0 )  (p0 , L, L0 );
3. if x % x0 , then for any x00 and any 0 p 1 it is true that (p, x, x00 ) % (p, x0 , x00 );
4. if for some x00 and some 0 p 1 it is true that (p, x, x00 ) % (p, x0 , x00 ), then x % x0 ;
5. if L % L0 , then for any L00 and any 0 p 1 it is true that (p, L, L00 ) % (p, L0 , L00 );
6. if for some L00 and some 0 p 1 it is true that (p, L, L00 ) % (p, L0 , L00 ), then L % L0 .
37

The term simple is normally used for lotteries that pay in outcomes and not in other lotteries;
here, I am using it is that sense, but making it stronger to require that they pay in only one or two
outcomes.
38
As before, the term compound is normally used for lotteries that pay in other lotteries; here, I
am using it is that sense, but making it stronger to require that they pay in only one or two lotteries.

51

Suppose that % satisfies the following continuity assumption: for any x, x0 , x00 X such
that x % x0 % x00 , we can find a number 0 p 1 such that (p, x, x00 ) x0 . Then, since %
satisfies monotonicity, it is relatively easy to construct a utility function representing it over
the space of simple lotteries: by continuity, for any lottery in L, we can find p [0, 1] such
that L (p, x , x ); by monotonicity, such p [0, 1] has to be unique; then, just let U : L R
be defined by letting U (L) be the unique number p [0, 1] such that L (p, x , x ).
Since L includes degenerate lotteries, we can define u : X R by letting u(x) = U ((1, x, x)).
Now, we just want to show that the expected utility property is satisfied in the following
sense: for every simple lottery (p, x, x0 ), U ((p, x, x0 )) = pu(x) + (1 p)u(x0 ).
Notice that, by construction,
(p, x, x0 ) (U ((p, x, x0 )), x , x ),
whereas, by independence,
(p, x, x0 ) (p, (u(x), x , x ), (u(x0 ), x , x )).
By direct computation, it follows that
(p, x, x0 ) (pu(x) + (1 p)u(x0 ), x , x ),
which implies, by monotonicity, that U ((p, x, x0 )) = pu(x) + (1 p)u(x0 ).

A PPENDIX : S TOCHASTIC D OMINANCE


We now give a more formal presentation of the results in stochastic dominance. We start by
giving a version of Proposition 4.2 for continuous lotteries, and its proof.
P ROPOSITION . Consider two lotteries, F and F , with densities f and f. Then,
1. if F F S F , then for any increasing and bounded utility index u C1 , we have that EF [u(X)] >
EF [u(X)]; and
2. conversely, if F 6= F and it is not true that F F S F , then for some increasing and bounded
utility index u C1 one has that EF [u(X)] < EF [u(X)].
Proof: For the first statement, integrating by parts and since u is continuously differentiable,
Z
EF [u(X)] EF [u(x)] =
u(x)(f (x) f(x))dx
0
Z

= [u(x)(F (x) F (x))]0


u0 (x)(F (x) F (x))dx.
0

Since F (0) = F (0) = 0 (remember that these c.d.f. have density) and limx F (x) =
limx F (x) = 1, and since u is bounded, it follows that [u(x)(F (x) F (x))]
0 = 0. Since
R
u0 > 0 and F F S F , we have that39 0 u0 (x)(F (x)F (x))dx < 0, so EF [u(X)]EF [u(x)] > 0.
39

Remember that any c.d.f. is right-continuous.

52

For the second statement, we shall consider two c.d.f. that cross, so that none of them
dominates the other: fix x such that F (x) F (x) for all x x , with strict inequality somewhere, and F (x) F (x) for all x x . Define the index p , for each positive real number p,
by
(

p exp( xx
if x x ;
p ),
p (x) =
p + p1 (1 exp(p(x x ))), if x > x .
By construction, this function is differentiable and monotone, with
(

exp( xx
if x x ;
0
p ),
p (x) =

exp( xx
p )), if x x .
The function is also bounded, with lim x p (x) = p. Now, recalling the equation above,
R
we have that EF [u(X)] EF [u(x)] = 0 u0 (x)(F (x) F (x))dx, and the right-hand side of
this expression is, by direct substitution,
Z

x x
)(F (x) F (x))dx
exp(
p

exp(
x

x x
)(F (x) F (x))dx,
p

an expression that is ambiguous, in general, by our assumptions. However, since for and

xx

x x the term exp( xx


p ) is increasing in p, while for any x x the term exp( p ) is
decreasing in p, it follows that for p large enough the first term dominates and the whole expression is negative.
Q.E.D.
The result for second-order dominance requires some technical assumptions too:
P ROPOSITION . Let F anf F be two continuous lotteries, with densities f and f. Suppose that both
lotteries have finite mean. Then,
1. if F SS F , then for any increasing, bounded and strictly concave utility index u C1 , we
have that EF [u(X)] > EF [u(X)]; and
2. conversely, if F 6= F and it is not true that F SS F , then for some increasing, bounded and
strictly concave utility index u C1 one has that EF [u(X)] < EF [u(X)].
R
Proof: For the first statement, let us recall again that EF [u(X)]EF [u(x)] = 0 u0 (x)(F (x)
F (x))dx. By integration by parts again, the right-hand side of the expression is
Z x
Z
Z x
0

00

(u (x)
(F (s) F (s))ds)0 +
u (x)
(F (s) F (s))dsdx.
0

R
By concavity and boundedness, limx u0 (x) = 0, while 0 (F (s) F (s))ds R, since both
lotteries have mean, so the first term on this expression vanishes. The second term is positive,
since u00 < 0 and F SS F .
The proof of the second statement is similar to its analogous in the extension of Proposition 4.2 above, using the expression we just obtained, and considering the utility indices
p (x) =

1
exp(p(x x )),
p2

53

Q.E.D.

for p > 0. Details are omitted.


Proof of Proposition 4.4: Integrating by parts,
VF [X] VF [X] = (x2 (F (x) F (x)))x0

2x(F (x) F (x))dx.

The first summand in the right-hand side of the expression is zero, since EF [X] = EF [X],40
Integration by parts of the second term gives
Z
2(x

(F (s) F (s))ds)x0 + 2

Z
0

x
Z x

(F (s) F (s))dsdx.

R x
The first term in this latter expression is simply 2
x 0 (F (s) F (s))ds which, again, is null
since both lotteries have the same mean. The second term is is negative, since F SS F . Q.E.D

40

Simply notice that EF [X] =

R x
0

F (x)dx.

54

G ENERAL E QUILIBRIUM UNDER U NCERTAINTY

The general equilibrium model studied in the first lecture considered an abstract
economy where time and uncertainty were not taken into account, at least not explicitly. If one introduces these phenomena, the results of the abstract model have to
be reconsidered.

5.1

A N E XCHANGE E CONOMY WITH U NCERTAINTY

We consider only exchange economies with uncertainty.41 As before, we assume that


there are L commodities and I individuals i = 1, . . . , I. But suppose now that the
economy evolves over two periods, the present and the future, and there is a number
of contingencies that can occur in the future: there is a finite set, S = {1, . . . , S}, of
possible states of the world describing the future.
In order to define the economy, we need, again, to describe the preferences and
wealth of all individuals. For their wealth, we want to maintain the institution of
private property, so we will assume that each individual is endowed with real wealth
(in bundles of commodities). The endowment of individual i in the present date is
w0i , a bundle of commodities in RL+ . Now, we want to allow for the possibility that the
future wealth of individuals be contingent on the state of the world: the future wealth
of each individual is a random variable defined (from S) over RL+ . For simplicity of
notation, we just let wsi be individual is private endowment of commodities, when
the realized state of the world is s = 1, . . . , S; the random variable (w1i , . . . , wSi ) is
known by individual i, but she does not know what state s will realize in the future
date, so she faces risk. Denote wi = (w0i , w1i , . . . , wSi ).
Let us assume that individual i will, in this setting, make a consumption plan that
accounts for that risk: she will choose a consumption bundle xi0 , that she will consume, for sure, in the present, and will plan to consume a bundle xis in the future
date, if state s is the one that realizes. Denote by xi = (xi0 , xi1 , . . . , xiS ) the consumption
plan of individual i. It follows that the space of consumption (plans) of each conL(S+1)

sumer is R+

. We assume that each individual has rational preferences that allow

her to compare pairs of consumption plans; for simplicity of notation, we assume


41

Economies with production are interesting, but more difficult: if different shareholders of a firm

have different individual assessments of the risk faced by the firm, it is not obvious what one would
mean by saying that the firm maximizes profit.

55

L(S+1)

that function U i : R+

R represents individual is preferences.42 The economy

is defined by {{1, ..., I}, (U i , wi )Ii=1 }; for simplicity, we will ignore the society and will
refer to {(U i , wi )i } as the economy.

5.2

F INANCIAL M ARKETS

The equilibrium concept that we apply to the dynamic situation depends crucially on
the institutional assumptions we make about trade. Suppose, for instance, that in the
present date there are competitive, functioning markets for the L contemporaneous
commodities, and also for the L commodities contingent on each one of the possible
future S states of the world. In such case, one can simply extend the abstract model
to the present setting by a reinterpretation of the concept commodity,43 and all the
results obtained there, including, importantly, the fundamental theorems of welfare
economics, immediately extend. Of course, such an institutional setting is only a
theoretical benchmark, and one ought to consider a more realistic one if the model is
to be of interest.
Let as assume that in the present date, and in each contingency of the future, there
are markets for immediate trade of all the commodities. But let us also assume that,
in the present, individuals can also trade financial assets, which are contracts that
promise delivery of some future obligation, possibly with contingency in the realized
state of the world. Let us assume, for simplicity, that commodity 1 is the num`eraire
of this economy, namely the object in which all individuals (and we) keep nominal
accounts.
An asset is a contract that promises to pay a certain return, in units of the num`eraire,
and this return may depend on the state of the world: it is a random variable over
R. In order to keep notation simple (and consistent), define an asset as a vector
r = (r1 , . . . , rS )T RS , where rs is the return of the asset when state of the world
s occurs.44 We assume that in the present date, simultaneously with the markets for
commodities, individuals can trade A assets, which we denote by r1 , . . . , rA . The financial market is the S A matrix R = (r1 , . . . , rA ), whose a-th column is asset ra . The
s-th row of matrix R, which we denote by rs , is the profile of asset returns in that state
42

We do this for notational simplicity only, and give U i an ordinal interpretation only. In particular,

we do not assume that individual preferences have expected-utility representation.


43
Indeed, Arrows definition of commodity is given by a full description of the features that define
the object, but must also specify where, when and under what circumstances the object is available.
44
Notice that we are adopting the convention of denoting assets as column vectors.

56

of the world. Let us denote by i RA the portfolio purchased by individual i; we


allow for short sales of assets, so we do not constraint i to RA
+.
In the present trade, commodity prices are p0 and asset prices are q. In the future
commodity trade, in state s, prices are ps .45 It follows from our choice of num`eraire
that the price of commodity 1 must be always be 1: in the present, p0,1 = 1, and in
future contingency s, ps,1 = 1.
D EFINITION . In an economy {(U i , wi )i }, given a financial market R, a financial-markets
equilibrium is a four-tuple consisting of commodity prices, asset prices, individual consumption plans and individual portfolios, (p, q, (xi , i )i ) such that
1. for each individual i, the pair (xi , i ) of her consumption plans and her portfolio solves
the problem
(
max U i (x) :

p0 x0 + q p0 w0i ,

ps xs = ps wsi + rs , at all s = 1, . . . , S;

x,

2. all commodity markets clear in all states:


3. all asset markets clear:

and

xi =

wi ;

i = 0.

Importantly, even if this is not explicit in the notation, one usually imposes the
condition that in each individuals optimization problem, the constraint that xis 0 is
satisfied too. Intuitively, this means that all individuals avoid bankruptcy in all possible future states of the world, which can be seen as a (controversial) institutional assumption. Similarly, we are assuming that all financial contracts are honored, another
institutional assumption. It is also worthwhile noticing that the returns of portfolios
were written as rs i thanks to the assumption that Ps,1 = 1 at all s; if this convention
is not held, the nominal return of portfolios should be ps,1 rs si .46 It is also important to
notice that the definition of equilibrium is from the ex-ante point of view, and that, for
that reason, prices have different interpretation: while p0 and q are being observed by
all individuals, future prices ps , for s = 1, . . . , S, are just conditional forecasts of future commodity prices; the definition imposes the condition that all individuals agree
on these conditional forecasts (but see the next exercise).
45

Notice that here ps is a vector of prices and not a probability. In this setting, we do not need to

specify probabilities for the states of the world. Confusion should not arise.
46
In fact, we are already imposing a simplifying assumption when we say that all assets pay in
units of a given commodity only; more generally, one could assume that asset returns are in bundles
of commodities, an assumption that complicates the concept of equilibrium nontrivially.

57

E XERCISE 5.1. Given an economy {(U i , wi )i }, suppose that each individual has additively
separable preferences: there exist state-contingent functions uis : RL+ R such that
U i (x) =

S
X

uis (xs ).

s=0

Argue that if (p, q, (xi , i )i ) is a financial-markets equilibrium (for R), then the following is
true: for each future state of the world s 1, the pair (ps , (xis )i ) is a competitive equilibrium
of the exchange economy {(uis , wsi + rs i (1, 0, . . . , 0))i }.47

5.3

M ARKET C OMPLETENESS

The most critical difference between equilibrium in financial markets and the concept
of competitive equilibrium in the abstract economy is that in the new setting each
individual faces a series of budget constraints, whereas in the abstract case, where all
trade takes place simultaneously, individuals face one budget constraint only.48 Here,
if an individual wants to transfer wealth from the present to the future, or viceversa,
or from one future state of the world to other, she has to buy a portfolio that delivers
that transfer. But this difference is far from trivial: depending on the financial market
R, there may be transfers of revenue across states of the world which are simply
impossible. The key concept is whether markets are complete or incomplete. The space
of revenue transfers that are possible given the market R is the column span of R,
hRi = { RS |R = for some RA }.
When any revenue transfer is possible given R, we say that R is complete: hRi = RS .
When that is not the case, we say that R is incomplete.
E XERCISE 5.2. Argue that if R is complete, then there must exist at least S assets which are
nonredundant: one can find at least S linearly independent columns in the matrix R. Argue
that if R is incomplete, then it contains fewer than S nonredundant assets. Argue that one
can never find more than S nonredundant assets in R. Conclude that if one assumes that
R contains only nonredundant assets, then R is complete if, and only if, A = S; and R is
incomplete if, and only if, A < S. (Hint: how good is your linear algebra?)
47
48

i
Ignore the fact that ws,1
+ rs i could be negative.
If it were possible to write contracts for the delivery of all commodities, contingent on all possible

states of the world, as in the theoretical benchmark, each individual would face only one constraint.

58

5.4

C ONSTRAINED I NEFFICIENCY

When markets are complete, the ability of all individuals to make any transfer of revenue across states of the world implies that equilibria in financial markets be equivalent to equilibria in the theoretical benchmark: an allocation of consumption plans
(xi )i is part of an equilibrium in financial markets if, and only if, it is also part of a
competitive equilibrium in which individuals simultaneously trade promises of delivery for all commodities in all states.49 It follows that key properties, like Pareto
efficiency (the First Fundamental Theorem of Welfare Economics), apply when financial markets are incomplete.
When markets are incomplete, however, this relation breaks down, and one needs
to study all properties, positive and normative, of equilibrium. But not all properties
survive: critically, the First Fundamental Theorem of Welfare Economics fails, and
one can show that is most economies with incomplete markets every competitive
equilibrium allocation is Pareto inefficient. Moreover, one can show that in a large
subset of economies, if markets are sufficiently incomplete and there are sufficiently
many commodities being traded, competitive markets dont exploit well even the
existing arbitrage opportunities: a policy that induced (or forced) the individuals to
construct different portfolios could make all of the strictly better off. In this case, the
argument to defend the market mechanism cannot be that markets do things well,
but that, unless proved otherwise, an attempt to do things better than the markets
may be unrealistically complicated. The rest of this subsection is devoted to a more
formal presentation of this argument
D EFINITION . An allocation of commodities x is constrained-inefficient if there exist commodity prices p, an alternative commodity allocation x, date-zero revenue transfers ( i )Ii=1
and an asset allocation (i )Ii=1 such that:
1. revenue transfers are balanced:
2. the asset allocation is feasible

PI

i=1

PI

i=1

i = 0;

i = 0;

3. for every individual i, in the present


xi0 argmaxx ui0 (x) : p0 x p0 w0i + i ,
49

The relation existing between the prices of these equilibria and the determination of portfolios is

also well understood.

59

and, for every future state of the world, s,


xis argmaxx uis (x) : ps x ps wsi + rs i ;
4. the alternative commodity allocation is feasible:

PI

i=1 (w

xi ) = 0; and

5. for every individual i, U i (


xi ) > U i (xi ).
This is, an allocation is constrained inefficient if a reallocation of wealth, by revenue at date zero and the existing assets at date one, and competitive trade in the
commodity markets can make all individuals ex-ante better off. Conditions 1 and 2
imply that the reallocation is balanced, 3 implies that individuals are individuallyrational in the commodity markets, which clear by condition 4. Intuitively, this says
that if people simply traded the existing assets as in allocation y and made the revenue transfers (say, if some authority imposed this), and then let competitive markets operate to allocate commodities, then allocation x would result and everyone
would be strictly better off than at allocation x. The following result is immediate.
P ROPOSITION . If an allocation is constrained inefficient, then it is inefficient.
The striking result (Geanakoplos-Polemarchakis) is that when markets are (sufficiently) incomplete then, in a very general set of economies, all competitive equilibrium allocations are constrained suboptimal. The argument for this theorem is
somewhat involved (it requires some advanced mathematics), so here well only give
an example to illustrate the point:
E XAMPLE :50 Consider an economy populated by two types of individuals, i = 1, 2;
each type consists of a continuum of individuals of unit mass. One commodity is
exchanged and consumed at date 0, and quantities of the commodity are x, while
two commodities, l = 1, 2, are exchanged and consumed at date 1, and quantities of
the commodities are x1 and x2 . The intertemporal utility function of an individual of
type 2 is
U 2 (x, x1 , x2 ) = x + (1 ) ln x1 + ln x2 ,
where 0 < < 1, and his endowment at date 1 consists of b units of commodity 2
only. The intertemporal utility function of an individual of type 1 is
U 1 (x, x1 , x2 ) = x + ln x1 + (1 ) ln x2 ,
50

This example is from Carvajal and Polemarchakis (2006).

60

and his endowment at date 1 consists only of commodity 1; but, importantly, it is


subject to idiosyncratic shocks: it is a , with equal probability. At date 1, equal
proportions of individuals of type 1 have endowments a + , and a , and, as a
consequence, there is no aggregate risk. With quasi-linear preferences, as we have
assumed, it is not necessary to specify the endowments of individuals at date 0.
At date 0, the consumption good is numeraire, while q is the price of a risk-free
bond of that matures at date 1. At date 1, commodity 1 is numeraire, while the price
of commodity 2 is p. With holdings of the bond for individuals of type 1 and for
individuals of type 2, the equilibrium price at date 1 is (see exercise 5.3)
p() =

(1 )a + (1 2)
,
(1 )b

which depends non-trivially on asset holdings as long as 6= 1/2. At date 1, the


marginal utility of revenue for individuals of type 2 is 2 = 1/(pb ), while, for
individuals of type 1, it varies with the realization of the idiosyncratic shock (the
personal state of each individual) and is
1 () =

1
1
or 1 () =
,
a++
a+

with equal probability.


The optimization of individuals of type 2 at date 0 requires that
q=

1
(1 )
=
,
pb
(1 )a y

while optimization of individuals of type 1 at date 0 requires that


1
1
1
1
a+
q=( )
+( )
=
;
2 a++
2 a+
(a + )2 2
as a consequence, at equilibrium,
=

a +

p
a2 + 42 (1 )
.
2

A policy intervention is a pair (dx, d) of transfers of revenue and bonds to individuals of type 1. The welfare effects of a policy are
1
1
dU 1 = dx + qd (( )1 ()x12 () + ( )1 ()x12 ())p0 d
2
2
and
dU 2 = dx qd 2 (x22 b)p0 d.
61

It follows that Pareto improving interventions exist if the matrix


1

q (( 21 )1 ()x12 () + ( 21 )1 ()x12 ())p0

q 2 (x22 b)p0

is nonsingular, which is the case, for  6= 0: singularity of the matrix would occur if
and only if
1
1 1
()x12 () + 1 ()x12 () = 2 (x22 b),
2
2
which is equivalent to
1
1
(pb )
=
(
b),
p
pb
p
or = 0, which occurs only in the absence of idiosyncratic shocks, with  = 0.
E XERCISE 5.3. In the context of the example,
1. derive the first-order conditions of the individual intertemporal problems;
2. argue that
(1 )a + (1 2)
(1 )b
(hint: since there is no aggregate risk, p and x2 are independent of the random shock,
p() =

and aggregate demand is simply 12 (x1 () + x1 ()) + x2 );


3. argue that 2 = 1/(pb ), 1 () = 1/(a +  + ) and 1 () = 1/(a  + );
4. from question 1, argue that
q=
and

1
(1 )
=
pb
(1 )a y

1
1
1
1
a+
;
q=( )
+( )
=
2 a++
2 a+
(a + )2 2

5. argue that, at equilibrium,


=

a +

p
a2 + 42 (1 )
;
2

6. argue that the welfare effects of a policy are


1
1
dU 1 = dx + qd (( )1 ()x12 () + ( )1 ()x12 ())p0 d
2
2
and
dU 2 = dx qd 2 (x22 b)p0 d
(hint: use Roys identity).
62

S OLUTIONS TO SOME EXERCISES


E XERCISE 1.11:
1. These preferences are rational (they are representable), strongly convex (just draw indifference curves) and strictly monotone.
2. Marshallian demands are:
(
x1 (p, m) =

0,
m
p1

and

(
x2 (p, m) =

The indirect utility function is


(
v(p, m) =

if m < p1 ;
1, otherwise;
m
p2 ,
p1
p2 ,

if m < p1 ;
otherwise.

ln(m) ln(p2 ),
if m < p1 ;
m

1
+
ln(p
)

ln(p
),
otherwise;
1
2
p1

from where Roys identity is immediate.


3. By duality, the expenditure function is
(
exp()p2 ,
if exp()p2 < p1 ;
e(p, ) =
(1 + ln(p1 ) + ln(p2 ))p1 , otherwise;
from where, by Shephards lemma
(
0,
if exp()p2 < p1 ;
h1 (p, ) =
ln(p1 ) + ln(p2 ), otherwise;
and

(
h2 (p, ) =

exp(), if exp()p2 < p1 ;


p1
otherwise.
p2 ,

E XERCISE 1.12:
1. This consumer cannot satisfy WARP: her budget doesnt change when prices and nominal income are all multiplied by , so WARP would require x = x0 ; but x2 = 2 6= x02 3.
Since WARP is a necessary condition for maximization of strongly convex, locally nonsatiated preferences, this consumer cannot be maximizing preferences that satisfy those
conditions.
2. As long as p02 6= , the consumer satisfies WARP. If p02 6= , the new budget is a rotation
of the old budget with pivot on (5, 0); in the new budget, with p02 < , the only violation
of WARP would be x0 = (5, 0), which is impossible given that x02 3. When p02 = , the
situation is as in part 1.

63

E XERCISE 1.13:
1. These preferences are rational, strongly convex and strictly monotone.
2. Marshallian demands are
mp2
mp1
and x2 (p, m) =
.
(p1 + p2 )p1
(p1 + p2 )p2

x1 (p, m) =

The indirect utility function is




1
v(p, m) =
2

m(p1 + p2 )
p1 p2

1

Roys identity is immediate.


3. By duality,
e(p, ) =

4 2 p1 p2
.
p1 + p2

Then, by Shephards lemma,



h1 (p, ) =

2p2
p1 + p2

2


and h2 (p, ) =

2p1
p1 + p2

2
.

(This can be verified by actually solving the expenditure minimization problem.)


E XERCISE 1.14: Let prices be p  0.
1. These preferences are rational, monotone, hence locally nonsatiated and convex.
2. The Marshallian demands are
x1 (p, m) =

m
,
p1 + min{p2 , p3 }

0,

if p2 > p3 ;
if p2 < p3 ;
x2 (p, m) =

m
x
[0, p1 +p2 ], if p2 = p3 ;
m
p1 +p2 ,

and
x3 (p, m) =

m
p1 +p3

0,
m
p1 +p2

if p2 > p3 ;
if p2 < p3 ;
x
, if p2 = p3 .

The indirect utility function is


v(p, m) =

m
.
p1 + min{p2 , p3 }

The properties are immediate.

64

3. Differentiability is immediate; for Engel aggregation


p1 m x1 (p, m) + p2 m x2 (p, m) + p3 m x2 (p, m) =

p1
p2
+
= 1;
p1 + p2 p1 + p2

for Roys identity,


m

(p1 +p2 )2
p1 v(p, m)
m
= x1 (p, m).
=
=
1
m v(p, m)
p1 + p2
p +p
1

4. Hicksian demands are h1 (p, ) = ,

if p2 > p3 ;
0,
h2 (p, ) =
,
if p2 < p3 ;

u
[0, ], if p2 = p3 ;
and

if p2 > p3 ;
,
h3 (p, ) =
0,
if p2 < p3 ;

u
, if p2 = p3 .

The expenditure function is e(p, ) = (p1 + min{p2 , p3 }), which is homogeneous of


degree 1 in p. For Shephard, p1 e(p, ) = = h1 (p, ).
E XERCISE 2.4:
1. This firms satisfies convexity, possibility of inaction,
returns to scale only. The supply correspondence is

{(1, 1)},
Y (p) =
{(0, 0)},

{(
y , y)|0 y 1},

no-free-lunch and nonincreasing


if p2 > p1 ;
if p2 < p1 ;
if p1 = p2 .

The profit function is (p) = max{p2 p1 , 0}.


2. The new firm violates possibility of inaction. The supply correspondence is

if p2 > p1 ;
{(1, 1, 1)},
Y (p) =
{(0, 0, 1)},
if p2 < p1 ;

{(
y , y, 1)|0 y 1}, if p1 = p2 ;
and the profit function is (p) = max{p2 p1 , 0} p3 .
3. This firm satisfies possibility of inaction, no-free-lunch, nonincreasing returns to scale,
nondecreasing returns to scale and constant returns to scale. It violates free disposal,
free entry and convexity. The supply correspondence is

,
if p1 > min{p2 , p3 };

if p1 = p2 < p3 ;
{(x, x, 0)|x R+ },
Y (p) =
{(x, 0, x)|x R+ },
if p1 = p3 < p2 ;

{(x, x, 0)|x R+ } {(x, 0, x)|x R+ }, if p1 = p2 = p3 ;

(0, 0, 0),
if p1 < min{p2 , p3 }.
The profit function is not defined when p1 > min{p2 , p3 }; for every other p, it is (p) = 0.

65

E XERCISE 2.5: In these cases:


1. Suppose that if f is homogeneous of degree 1; let y F and 0; by construction, y2
0 and y3 0, so y2 0 and y3 0; Also, y1 f (y2 , y3 ), so, by homogeneity of
degree 1, f (y2 , y3 ) = f (y2 , y3 ) y1 ; it follows that y F, and hence that
F satisfies constant returns to scale. If f is homogeneous of degree d 1, the analysis is
similar, but, as long as 1, f (y2 , y3 ) = d f (y2 , y3 ) f (y2 , y3 ) y1 ;
this implies that F satisfies nonincreasing returns to scale.
2. Firm
F = {y R2 : y2 min{y1 , y1 }}
is the cone below the lines of slope and , going through the origin. This set is convex, satisfies free disposal, no-free-lunch, possibility of inaction, free entry and constant
returns to scale. For the profit maximization problem:
If pp21 > , there is no solution: let y = (1, ); since < < 0, y F; by direct
computation, p y > 0; since F satisfies constant returns to scale, there exists no
profit maximizing plan.
If pp21 < , there is no solution: let y = ( 1 , 1), and argue as above.
If pp12 = , there are infinitely many solutions: for any 0, y = (, ) is
optimal; maximized profits are 0.
If pp21 = , there are infinitely many solutions: for any 0, y = ( , ) is
optimal; maximized profits are 0.
< pp21 < , there is only one solution: y = (0, 0), with maximized profits equal
to 0.
E XERCISE 2.6: Under X = min{K, L}, with > 0 and > 0:
1. F = {x R3 : x2 0, x3 0, x1 min{x2 , x3 }}.
2. It satisfies constant returns to scale, no free lunch, possibility of inaction and free entry.
3. Let w, r > 0 be, respectively, the prices of labor and capital. The cheapest way to produce X units of output is K = X/ and L = X/, so the profit maximization problem
is simply
X
X
max pX r w ,
X

r
w
which has a solution if, and only if, p r + w
. If p = + , any X 0, K = X/ and
L = X/ is optimal; if p < r + w
, only X = 0, K = 0 and L = 0 is optimal. Maximized
profits are 0 in any case.

Suppose now that the production function is X = ln(1 + min{K, L}). Now, the cheapest way to produce X units is K = exp(X)1
and L = exp(X)1
, so the profit maximization

problem is simply
exp(X) 1
exp(X) 1
max pX r
w
.
X

66

Taking p >

only, this gives



X(p, w, r) = ln

and

1
K(p, w, r) =

1
L(p, w, r) =

p
w + r


,


p
1 ,
w + r


p
1 .
w + r

It follows that the profit function is



(p, w, r) = p ln

p
w + r


p+

w + r
,

from where Hotellings lemma is immediate.


E XERCISE 2.7: Firm
F = {y R2 : (1 + y1 )2 + (y2 )2 1}
is a closed circle of radius 1, centered at (1, 0). Then:
1. The firm satisfies no-free-lunch, possibility of inaction and nonincreasing returns to
scale, but violates nondecreasing returns to scale, constant returns to scale, free entry
and free disposal.
2. For strictly positive prices, the optimal supply functions are
y1 (p) =

p1
(p21

+ p22 ) 2

and
y2 (p) =

p2
(p21

+ p22 ) 2

The profit function is


(p) =

p21 + p22
1

(p21 + p22 ) 2

p1 .

3. Hotellings lemma follows immediately.


E XERCISE 3.11: Here:
1. That Pareto efficiency implies weak Pareto efficiency is immediate: if it is impossible to
make someone better-off without making anybody else worse-off, then it is impossible
to make everybody better-off.
2. Suppose that allocation x is not Pareto efficient. Let x
be such that ui (
xi ) ui (xi ) for all
0
0
0
0
i
i
i
i
0
i, and u (
x ) > u (x ) for some i . Now, for 0 <  < 1, define the following allocation:
0
0
i
i
x
= 
x , and for every i 6= i0 ,
x
i = x
i +

67

1  i0
x
.
I 1

If  is close enough to 1, by continuity we have that ui (


xi ) > ui (xi ). By monotonicity,
0
it cannot be that x
i = 0, so it follows that x
i > x
i for every i 6= i0 . By strong monotonicity, this implies that ui (
xi ) > ui (
xi ) ui (xi ), for all i, so x is not weakly Pareto efficient
either.
E XERCISE 3.12: Let i0 have strictly monotone preferences, and suppose that y is not techniP
P
cally efficient: let y be such that yj F j , for all j, and j yj > j y j . Construct x
= (
xi )i
P
P
P
P
0
0
as follows: x
i = xi +
yj
y j , and x
i = xi for every i 6= i0 . Since i x
i = i xi +
P j P j P i jP j j j
j
j y = i w + j y and y F for all j, it follows that (
x, y) is an allocation for
jy
0
i
the economy. Since u is strictly monotone, (
x, y) is Pareto superior to (x, y).
E XERCISE 3.13: The only equilibrium is ((
p, p), (2, 0), (0, 2)), for some p > 0; the only Pareto
efficient allocation is ((2, 0), (0, 2)), which is the only element in the core; it follows that the
core is a subset of the Pareto set, and that all competitive equilibrium allocations lie in the core.
E XERCISE 3.14: Here:
1. If w1 = (30, 0), there is only one equilibrium: p = (1, 2), x1 = (10, 10) and x2 = (20, 10).
2. If w1 = (5, 0), there are infinitely many equilibria. Take any 5 x 20 and p1 > 0;
p = (p1 , 0), x1 = (5, x) and x2 = (0, 20 x) is a competitive equilibrium.
What is funny about these results is that as individual 1 gets poorer, individual 2 is made
worse-off at equilibrium.
E XERCISE 3.15: Given an exchange economy ({1, ..., I}, (ui , wi )Ii=1 ):
1. Suppose not: the endowment (wi )Ii=1 is itself an efficient allocation, but it is not in the
P
core of the economy. Then, there exists H {1, ..., I} and (xi )iH such that iH xi =
P
i
i i
i
i
i0 i0
i0
i0
0
iH w , u (x ) u (w ) for all i H, and u (x ) > u (w ) for some i H. Define,
for all i,
(
xi , if i H;
x
i =
wi , otherwise.
P
P
It is immediate that: i{1,...,I} x
i = i{1,...,I} wi , ui (
xi ) ui (wi ) for all i {1, ..., I},
0
0
0
0
and ui (
xi ) > ui (wi ), so (wi )Ii=1 is not efficient, which is a contradiction.
2. Suppose not. From part 1, we know that (wi )Ii=1 is a core allocation, so it must be that
0
0
there exists (xi )Ii=1 , also in the core, such that for some i0 , xi 6= wi . By definition of
core, for all i, ui (xi ) ui (wi ). Now define, for all i, x
i = 21 (xi + wi ). Immediately,
X
i

x
i =

1X i 1X i 1X i 1X i X i
1X i
(x + wi ) =
x +
w =
w +
w =
w.
2
2
2
2
2
i

By strong convexity, for all i, ui (


xi ) min{ui (xi ), ui (wi )} = ui (wi ), whereas ui (
xi ) >
0
0
0
0
0
0
i
i
i
i
i
i
i
I
min{u (x ), u (w )} = u (w ). This contradicts the fact that (w )i=1 is an efficient
allocation.

68

3. If all individuals have locally nonsatiated preferences, equilibrium allocations lie in the
core. If they have strongly convex preferences and the endowment (wi )Ii=1 is an efficient allocation, the only core allocation, by 2, is (wi )Ii=1 . It follows that all competitive
equilibria of the economy have consumers consuming their own endowments.
E XERCISE 3.16: When there are only two consumers in the economy, only three non-empty
coalitions exist: the whole society, and each individual by herself. If an allocation is efficient,
the whole society does not object. For each individual in isolation, all she can do is consume
her endowment; if an allocation gives her at least the same utility as her endowment, she does
not object. When there are three consumers or more, the latter is not true: two-person coalitions are not captured by efficiency and can do more than each of their members in isolation.
E XERCISE 3.17: Here:
1. A competitive equilibrium is (p, x, y) such that
(a) y solves the problem maxy p2 y2 p1 y1 : y2 = f (
y1 );
(b) x solves the problem maxx u(
x) : p1 x
1 + p2 x
2 p 1 w + p 2 y2 p 1 y1 ;
(c) x1 + y1 = w and x2 = y2 .
Allocation (x, y) is Pareto efficient if there does not exist (
x, y) such that
(a) y2 = f (
y1 );
(b) x
1 + y1 = w and x
2 = y2 .
(c) u(
x) > u(x).
The statement of the first fundamental is: If (p, x, y) is a competitive equilibrium, then
(x, y) is Pareto efficient. For a proof, suppose not and let (
x, y) be Pareto superior; by
the second condition of equilibrium and the third condition of efficiency, p1 x
1 + p2 x
2 >
p1 w + p2 y2 p1 y1 ; by the first conditions of both definitions, p2 y2 p1 y1 p2 y2 p1 y1 ;
it follows that p1 x
1 + p2 x
2 > p1 w + p2 y2 p1 y1 , which contradicts the second condition
on the definition of efficiency.
E XERCISE 3.18: Since each ui represents locally nonsatiated, strongly convex preferences, and
considering only strictly positive prices, this follows by Duality in Consumers Theory: since
market-clearing is given, it suffices that we show individual rationality; but by duality,
hi (p, ui (xi )) = xi (p, ei (p, ui (xi ))) = xi (p, p hi (p, ui (xi ))) = xi (p, p xi ) = xi (p, p wi ).
E XERCISE 4.8: The supply correspondence is

if p2 > p1 ;
{(1, 1, 1)},
Y (p) =
{(0, 0, 1)},
if p2 < p1 ;

{(
y , y, 1)| 1 y 0}, if p1 = p2 ;
and the profit function is (p) = max{p2 p1 , 0} p3 . If the objective is expected profits, the

maximum is p = 2. With u(x) = x, the maximum is p = 5/2.

69

E XERCISE 4.9: In this setting,


1. The expected utility in the absence of shocks is simply u(w) = 1/w. Under the shocks,
1
w+
x
her expected utility is E[u(X)] = 2
x ln( w
x ). Her expected income is w + E[X] =
w. This all makes sense: this individual is risk-averse. By direct computation (using
LHopitals rule), limx0 E[u(X)] = 1/w = u(w).
2. By direct computation,
(w, x
) = w
whereas, by definition, A(w, x
) =

2
w,

2
x
,
w+
ln( wxx )

which is independent of x
.

3. That limx0 (w, x


) = 0 follows by direct computation, as in part 1. That limx0 A(w, x
) =
2/w 6= 0 is immediate. This makes sense, since
1
(w, x
) A(w, x
)V(X; x
)
2
and limx0 V(X; x
) = 0.

70

Das könnte Ihnen auch gefallen