Economics 2010b: Microeconomic Theory II

Michael Powell
Introduction; GE Model; Welfare Theorems; Pareto optimality

1 Introduction to General Equilibrium Theory1

In the …rst three weeks of this course, our goal is to develop a parsimonious model of the
overall economy to study the interaction of individual consumers and …rms in perfectly
competitive decentralized markets. The resulting framework has provided the workhorse
micro-foundations for much of modern macroeconomics, international trade, and …nancial
economics. In the last three and a half weeks, we will begin to study managed transactions
and in particular how individuals should design institutions such as contracts and property
rights allocations to governance structures, to achieve desirable outcomes.
The main ideas of general equilibrium theory have a long history, going back to Adam
Smith’s (1776) evocative descriptions of how competition channels individual self-interest in
the social interest and how a sense of “coherence among the vast numbers of individuals
and seemingly separate decisions” (Arrow, 1972) can arise in the economy without explicit
design. General equilibrium theory addresses how this aggregate “coherence”emerges from
individual interactions and can potentially lead to socially desirable allocations of goods and
services in the economy. The mechanism through which this coherence emerges is, of course,
the price mechanism. Individuals facing the same, suitably determined, prices will end up
making decisions that are well-coordinated at the economy-wide level.
What distinguishes general equilibrium theory from partial equilibrium theory, which you
have studied in Economics 2010a, is the idea that if we want to develop a theory of the price
system for the economy as a whole, we have to consider the equilibrium in all markets in the
economy simultaneously. As you can imagine, thinking about all markets simultaneously can
be a complicated endeavor, since markets are interdependent: the price of computer chips
1
These notes borrow liberally from Levin’s (2006) notes and Wolitzky’s (2016) notes.

1
will a¤ect the price of software, cars, appliances, and so on. General equilibrium theory was
the most active research area in economic theory for a good part of the 20th century and is
therefore a very rich topic. Our goal over the …rst three weeks of the course is to cover only
the basics.
In many ways, the …rst part of this class will be structured the way an applied theory
paper is structured. We will start by talking about the setup of a model: who are the players,
what do they do, what do they know, what are their preferences, what is the solution concept
we will be using? Then we will partially characterize its solution, focusing on its e¢ ciency
properties in particular. What do I mean by partially characterize? I mean that we will be
describing some properties of equilibria that we can talk about without actually solving the
full model. That should take us through the …rst week.
In the second week, we will do a little bit of heavier lifting and begin with a question
that is often easy to overlook but is very important to answer: does an equilibrium exist?
Existence proofs can sometimes seem like a bit of an esoteric detour, but a good existence
proof is especially useful if it helps build tools to answer other important questions about
the model. After we establish that an equilibrium exists, we will ask questions like: when
is there a unique equilibrium? Are equilibria stable? What are the testable implications of
equilibrium behavior?
The third week will focus on what I will blandly call “extensions,” but are really the
meat and potatoes of how general equilibrium theory gets used in practice. We will show
how we can introduce …rms, time, and uncertainty into the framework, and we will talk
about how and when the main results we identi…ed above apply in these settings as well.
And …nally, we will end by putting the model’s solution concept, competitive equilibrium, on
…rmer microfoundations. A lot of economics is buried in the solution concept, and developing
microfoundations for the solution concept is a useful way to ‡esh out some of the key insights.

2
2 Pure Exchange Economies

A general equilibrium model describes three basic activities that take place in the economy:
production, exchange, and consumption. For the …rst two weeks, we will set production aside
and focus on the minimal number of modeling ingredients necessary to give you a ‡avor of
the powerful results of general equilibrium theory.
We begin with a description of the model we will be using. As always, the exposition of
an economic model speci…es a complete description of the economic environment (players,
actions, and preferences) and the solution concept that will be used to derive prescriptions
and predictions.

2.1 The Model

Formally, a pure exchange economy is an economy in which there are no production op-
portunities. There are I consumers i 2 I = f1; : : : ; Ig who buy, sell, and consume L
commodities, l 2 L = f1; : : : ; Lg. A consumption bundle for consumer i is a vector
xi = (x1;i ; : : : ; xL;i ), where xi 2 Xi , which is consumer i’s consumption set and just de-
scribes her feasible consumption bundles. We will assume throughout that Xi contains the
0 vector and is a convex set. Consumer i has an endowment of the L commodities, which
is described by a vector ! i = (! 1;i ; : : : ; ! L;i ) and has preferences over consumption bundles,
which we assume can be represented by a utility function ui : Xi ! R. A pure exchange
economy is therefore a set E = (ui ; ! i )i2I , which fully describes the model’s primitives:
the set of players, their preferences, and their endowments.
Each consumer takes prices p = (p1 ; : : : ; pL ) as given and solve her consumer maxi-
mization problem:
max ui (xi ) s.t. p xi p !i,
xi 2Xi

where the right-hand side of the consumer’s budget constraint is her wealth, as measured by
the market value of her endowment at prices p. Consumer i’s feasible actions are therefore

3
xi 2 Bi (p) fxi 2 Xi : p xi p ! i g, where we refer to Bi (p) as her budget set at prices
p. Given prices p and endowment ! i , we will refer to consumer i’s optimal choices as her
Marshallian demand correspondence and denote it by xi (p; p ! i ). Simply put, all
consumers do in this model is to choose their favorite consumption bundles in their budget
sets.
We have now described the players and their actions, but no model description is complete
without a solution concept. Here, the solution concept will be a Walrasian equilibrium, which
will specify a set of prices and a consumption bundle for each consumer (we will refer to
a collection of consumption bundles for each consumer as an allocation) that satisfy two
properties: consumer optimization and market-clearing. Given prices p, each consumer
optimally chooses her consumption bundle, and total demand for each commodity equals
total supply.

De…nition 1. A Walrasian equilibrium for the pure exchange economy E is a vector

p ; (xi )i2I that satis…es:

1. Consumer optimization: for all consumers i 2 I,

xi 2 argmaxxi 2Bi (p ) ui (xi ) ,

2. Market-clearing: for all commodities l 2 L;

X X
xl;i = ! l;i .
i2I i2I

We have now fully speci…ed the model, but before we start to go into more detail dis-
cussing the properties of Walrasian equilibria, there are a couple more important de…nitions
to introduce. The …rst is the notion of a feasible allocation, which is just a collection of
consumption bundles for each consumer for which the total amount consumed for each com-
modity does not exceed the total endowment of that commodity.

4
 P P
De…nition 2. An allocation (xi )i2I 2 RL+ I is feasible if for all l 2 L, i2I xl;i i2I ! l;i .

The next de…nition is going to describe how we will be thinking about optimality in
the economy. For a lot of optimization problems you have seen in your other courses, the
appropriate notion of optimality is straightforward. For example, if a consumer has a well-
de…ned utility function, it is straightforward to think about what is optimal for her given
her budget set. Once we start thinking about environments with more than one consumer,
we would, in some sense, like to maximize multiple objective functions (i.e., each consumer’s
utility) simultaneously. In general, there are no allocations that simultaneously maximize the
utility of all consumers— consumers’objectives are typically in con‡ict with one another’s—
so the appropriate notion of optimality is not as straightforward. The notion will we use is
that of Pareto optimality, which means that all we are doing is ruling out allocations that
are dominated by other feasible allocations.

De…nition 3. Given an economy E, a feasible allocation (xi )i2I is Pareto optimal (or
Pareto e¢ cient) if there is no other feasible allocation (^
xi )i2I such that ui (^
xi ) ui (xi )
for all i 2 I with strict inequality for some i 2 I.

In words, all Pareto optimality rules out is allocations for which someone could be made
better o¤ without making anyone else worse o¤. This notion of optimality is therefore
silent on issues of distribution, since it may be Pareto optimal for one consumer to consume
everything in the economy and for everyone else to consume nothing.

2.2 Assumptions on Consumer Preferences and Endowments

Throughout the next few sections, we will invoke di¤erent sets of assumptions for di¤erent
results. I will collect these assumptions here and will be explicit in referring to them when
they are required for a result.

Assumption A1 (continuity): For all consumers i 2 I, ui is continuous.

Assumption A2 (monotonicity): For all consumers i 2 I, ui is increasing: ui (x0i ) >

5
ui (xi ) whenever x0l;i > xl;i for all l 2 L.

Assumption A3 (concavity): For all consumers i 2 I, ui is concave.

Assumption A4 (interior endowments): For all consumers i 2 I, ! l;i > 0 for all l 2 L.

The …rst three assumptions should be familiar from Economics 2010a. The results we will
be establishing in the upcoming sections will hold under weaker assumptions— for example,
(A2) can typically be relaxed to local nonsatiation,2 and (A3) can typically be relaxed to
quasiconcavity. The last assumption is a strong assumption that will prove to be su¢ cient
for ruling out some pathological cases in which a Walrasian equilibrium does not exist.

2.3 Graphical Examples

Many of the main ideas of general equilibrium theory can be understood in a two-consumer,
two-commodity pure exchange economy. We can get most of the results across graphically
in what is referred to as an Edgeworth box. Edgeworth boxes are informationally dense, so
let me introduce the constituent elements separately.
Figure 1(a) depicts the relevant information for consumer 1. On the horizontal axis is her
consumption of commodity 1 and on the vertical axis is her consumption of commodity 2. Her
endowment is ! 1 = (! 1;1 ; ! 2;1 ). At prices p = (p1 ; p2 ), she can a¤ord to buy any consumption
bundle in the set B1 (p). The slope of her budget line is p1 =p2 . Given her preferences,
which are represented by her indi¤erence curve, and given prices p and endowment ! 1 , she
optimally chooses to consume x1 (p) = x1;1 (p) ; x2;1 (p) . In other words, at prices p, she
would optimally like to sell ! 1;1 x1;1 (p) units of commodity 1 in exchange for x2;1 (p) ! 2;1
units of commodity 2. Figure 1(b) depicts the same information for consumer 2.
2
We say that consumer i’s preferences satisfy local non-satiation if for every xi 2 Xi and every " > 0,
there is an x0i 2 Xi such that jjx0i xi jj " and ui (x0i ) > ui (xi ).

6
 Figures 1(a) and 1(b): consumer-optimization problems

The Edgeworth box represents both consumers’endowments and their optimal choices as
a function of the prices p, so it will incorporate all the information in Figures 1(a) and 1(b).
To build towards this goal, Figure 2(a) depicts all the non-wasteful allocations in the
economy: allocations (xi )i2f1;2g for which xl;1 + xl;2 = ! l;1 + ! l;2 for l 2 f1; 2g. The bottom-
left corner is the origin for consumer 1, and the upper-right corner is the origin for consumer
2. The length of the horizontal axis is equal to the total endowment of commodity 1, and the
length of the vertical axis is equal to the total endowment of commodity 2. The horizontal
axis, read from the left to the right, represents consumer 1’s consumption of commodity 1,
and read from the right to the left, represents consumer 2’s consumption of commodity 1.
The vertical axis, read from the bottom to the top, represents consumer 1’s consumption of
commodity 2, and read from the top to the bottom, represents consumer 2’s consumption
of commodity 2. The endowment ! is a point in the Edgeworth box, and it represents a
non-wasteful allocation, since ! l;1 + ! l;2 = ! l;1 + ! l;2 for l 2 f1; 2g. The allocation x also

7
represents a non-wasteful allocation, since xl;1 + xl;2 = ! l;1 + ! l;2 for l 2 f1; 2g.

Figure 2(a): non-wasteful allocations Figure 2(b): prices and budget sets

Figure 2(b) adds prices into the picture and shows that, given any price vector p, the
Edgeworth box can be partitioned into the budget sets for the two consumers. Given these
prices, consumer 1 can choose any consumption bundle to the bottom left of the diagonal
line, and consumer 2 can choose any consumption bundle to the upper right of the diagonal
line. Given these prices, Figure 3(a) shows that consumer 1 will optimally choose bundle
x1 (p; p ! 1 ), and Figure 3(a) shows how x1 (p; p ! 1 ) varies as the price ratio varies. Note
that, in terms of determining consumer 1’s optimal choice, the price ratio p1 =p2 is a su¢ cient
statistic for the price vector p. This is because Marshallian demand correspondences are
homogeneous of degree zero in prices (i.e., x1 (p; p ! 1 ) = x1 ( p; p ! 1 ) for all 2 R++ ).The
curve traced out in Figure 3(b) is referred to as consumer 1’s o¤er curve.

8
Figures 3(a) and 3(b): Consumer 1’s Marshallian demand for a …xed p and her o¤er curve

Recall from above that in a Walrasian equilibrium p ; (xi )i2f1;2g , consumer i opti-
mally chooses xi given equilibrium prices p . This means that in any Walrasian equilibrium,
both consumers’ optimal choices lie on their o¤er curves. Figure 4(a) depicts, for a given
price vector p, both consumers’ optimal choices. At this price vector, consumer 1 would
like to sell ! 1;1 x1;1 (p; p ! 1 ) units of commodity 1 in exchange for x2;1 (p; p ! 1 ) ! 2;1
units of commodity 2, and consumer 2 would like to buy x1;2 (p; p ! 2 ) ! 1;2 units of com-
modity 1 and sell ! 2;2 x2;2 (p; p ! 2 ) units of commodity 2. The associated allocation,
(x1 (p; p ! 1 ) ; x2 (p; p ! 2 )), is not a Walrasian equilibrium allocation, since consumer 1
would like to sell more units of commodity 1 than consumer 2 would like to buy, so the

9
market for commodity 1 does not clear: ! 1;1 x1;1 (p; p ! 1 ) > x1;2 (p; p ! 2 ) ! 1;2 .

Figures 4(a) and 4(b): disequilibrium and equilibrium allocations

It should be clear from the above argument, then, that any Walrasian equilibrium al-
location has to occur at a point where both consumers’o¤er curves intersect. Figure 4(b)
illustrates such a point. The upper-left point at which the two o¤er curves intersect is a
Walrasian equilibrium allocation, and the price vector that ensures both players optimally
choose the associated consumption bundles is a Walrasian equilibrium price vector. This
example also illustrates that, relative to the Walrasian equilibrium allocation, there are no
other feasible allocations that can make one of consumers better o¤ without hurting the
other consumer. The Walrasian equilibrium allocation is therefore Pareto optimal. Note
that the o¤er curves also intersect at the endowment point, but the endowment point is not
a Walrasian equilibrium allocation in this particular example— why?
The Edgeworth box is also useful for illustrating why there may be multiple Walrasian
equilibria and why a Walrasian equilibrium might fail to exist. Figure 5(a) illustrates a
situation in which there are multiple Walrasian equilibria. Here, the two consumers’ o¤er

10
curves intersect multiple times. Exercise 2 asks you to solve for the set of Walrasian equilibria
in another example in which there are multiple equilibria.

Figures 5(a) and 5(b): multiple Walrasian equilibria and no Walrasian equilibria

Figure 5(b) illustrates a situation in which there are no Walrasian equilibria. In the
example, consumer 1 is endowed with no units of commodity 1 and with all the units of
commodity 2. Consumer 2 is endowed with all the units of commodity 1 and with no units of
commodity 2. Consumer 2 cares only about her consumption of commodity 1, and consumer
2 cares about both her consumption of commodity 1 and her consumption of commodity
2. Moreover, consumer 1’s marginal utility of consuming the …rst unit of commodity 1 is
in…nite, and her marginal utility of consuming commodity 2 is strictly positive. For any
prices p with p1 > 0, the market for commodity 1 cannot clear, since consumer 2 will always
choose x1;2 (p; p ! 2 ) = ! 1;2 , and consumer 1 will always choose x1;1 (p; p ! 1 ) > 0, unless
p2 = 0. And if p2 = 0, then x2;1 (p; p ! 1 ) = +1, so the market for commodity 2 cannot
clear. This example illustrates why things can go awry when assumption (A4) is not satis…ed.
These examples tell us that the answers to the following two important questions is “no”:

11
(a) is there always a Walrasian equilibrium? (b) if there is a Walrasian equilibrium, is it
unique?

Figure 6: Pareto-optimal allocations and the

contract curve

Finally, Figure 6 shows that we can use the Edgeworth box to illustrate the entire set of
Pareto-optimal allocations. The Pareto set is the set of all feasible allocations for which
making one consumer better o¤ necessarily means making the other consumer worse o¤. It
also illustrates the contract curve, which is the set of Pareto-optimal allocations that both
players prefer to the endowment. If the two consumers were to negotiate a deal, given their
endowments as their outside options, they would likely reach a point on the contract curve.
Walrasian equilibrium allocations are typically a small subset of the contract curve, and in
particular, they lie on the Pareto set. This result is known as the …rst welfare theorem, and
we will now establish this result.
Exercise 1 (Adapted from MWG 15.B.2). Consider an Edgeworth box economy in
which the consumers have Cobb-Douglas utility functions u1 (x1;1 ; x2;1 ) = x1;1 x12;1 and
u2 (x1;2 ; x2;2 ) = x1;2 x12;2 , where ; 2 (0; 1). Consumer i’s endowments are (! 1;i ; ! 2;i ) >> 0

12
for i = 1; 2. Solve for the Walrasian equilibrium price ratio and allocation. How do these
change as you increase ! 1;1 ? Note: Feel free to avoid writing expressions out as much as
possible. For example, if you solve for price, feel free to leave the solutions for demand in
terms of the price variable instead of plugging in. For comparative statics, if you can …nd
the sign without having to write it out, that’s …ne.

Exercise 2 (Adapted from MWG 15.B.6). Compute the Walrasian equilibria for the
following Edgeworth box economy (there is more than one Walrasian equilibrium):
! 1=2
3
12
u1 (x1;1 ; x2;1 ) = x1;12 + x2;12 ; ! 1 = (1; 0) ;
37
! 1=2
3
12
u2 (x1;2 ; x2;2 ) = x1;22 + x2;22 ; ! 2 = (0; 1) .
37

Exercise 3 (Adapted from MWG 15.B.9). Suppose that in a pure exchange economy,
we have two consumers, Alphanse and Betatrix, and two commodities, Perrier and Brie.
Alphanse and Betatrix have the utility functions:
n o
u (xp; ; xb; ) = min fxp; ; xb; g and u (xp; ; xb; ) = min xp; ; (xb; )1=2 ,

(where xp; is Alphanse’s consumption of Perrier, and so on). Alphanse starts with an
endowment of 30 units of Perrier (and none of Brie); Betatrix starts with 20 units of Brie
(and none of Perrier). Neither can consume negative amounts of a commodity. If the two
consumers behave as price takers, what is the equilibrium? [Hint: consider the market-
clearing condition in the cases when both prices are positive, when only the price of Perrier
is positive, and when only the price of Brie is positive.]
Exercise 4. Consider an exchange economy with two consumers. The utility functions and
endowments are given by

x2;13
u1 (x1;1 ; x2;1 ) = x1;1 , ! 1 = (K; r)
3
x1;23
u2 (x1;2 ; x2;2 ) = x2;2 , ! 2 = (r; K) .
3
Assume that K is su¢ ciently large so that each consumer can achieve an interior solution
to her optimal consumption problem. Note that p = (1; 1) is an equilibrium price vector.
(a) For what values of r will there be multiple Walrasian equilibria in this economy? [Hint:
…rst solve for q = py =px by showing that rt4 t3 + t r = 0, where t = q 1=4 . Note this
expression factors as (t + 1) (t 1) (rt2 t + r) = 0.]
(b) For what value of r will p = (1; 3) be an equilibrium price vector?
(c) [Optional: algebra intensive] Assume that K = 10 and that r takes the value identi…ed
in part (b). Find all equilibrium prices and allocations.

13
(d) [Optional: algebra intensive] Rank the outcomes identi…ed in part (d) in terms of most
preferred to least preferred for each consumer.

3 First Welfare Theorem

At the Walrasian equilibria in the examples we just saw, there are no feasible allocations
that make both players better o¤: the Walrasian equilibrium allocation was Pareto optimal.
This, it turns out, is a general result and perhaps one of the most important results of GE.
This result is known as the …rst welfare theorem. Before stating and proving the …rst welfare
theorem, we will …rst establish an intermediate result known as Walras’s Law, which is a
direct implication of consumer optimization when consumers’preferences are monotonic (or
more generally, satisfy local non-satiation).

Lemma 1 (Walras’s Law). Given an economy E and prices p, if (A2) holds, then p
P P
i2I xi (p; p ! i ) = p i2I ! i .

Proof of Lemma 1. Since (A2) holds, each consumer will optimally choose to exhaust her
budget: p xi (p; p ! i ) = p ! i for all i 2 I. Summing this condition over consumers gives
us the expression in the Lemma.

Note that Walras’s Law holds for any set of allocations that are consumer-optimal—
the result does not require that the allocation (xi (p; p ! i ))i2I is a Walrasian equilibrium
allocation.

Exercise 5 (Adapted from MWG 15.B.1). Consider an Edgeworth box economy in

which the two consumers’ preferences satisfy local nonsatiation. Let xl;i (p; p ! i ) be con-
sumer i’s demand for commodity l at prices p = (p1 ; p2 ).
P P
(a) Show that p1 i2I (x1;l ! 1;l ) + p2 i2I (x2;l ! 2;l ) = 0 for all prices p 6= 0.
(b) Argue that if the market for commodity 1 clears at prices p >> 0, then so does the
market for commodity 2; hence p is a Walrasian equilibrium price vector.

We can now prove a version of the …rst welfare theorem.

14
Theorem 1 (First Welfare Theorem). Suppose p ; (xi )i2I is a Walrasian equilibrium
for the economy E. Then if (A2) holds, the allocation (xi )i2I is Pareto optimal.

Proof of Theorem 1. In order to get a contradiction, suppose the Walrasian equilibrium

allocation (xi )i2I is not Pareto optimal. Then there is some other feasible allocation (^
xi )i2I
for which ui (^
xi ) xi0 ) > ui0 (xi0 ) for some i0 . Since (xi )i2I
ui (xi ) for all i 2 I and ui0 (^
is a Walrasian equilibrium allocation, and consumers’preferences satisfy (A2), by revealed
preference, it has to be the case that p x^i p xi for all i 2 I and p x^i0 > p xi 0 .
Summing up over these conditions,
! ! !
X X X
p x^i >p xi =p !i ,
i2I i2I i2I

where the equality holds by Lemma 1. Since equilibrium prices p are nonnegative (why
are they nonnegative?), this inequality implies that there is some commodity l such that
P P
^i;l > i2I ! i;l , and therefore (^
i2I x xi )i2I is not a feasible allocation.

The …rst welfare theorem is a remarkable result because (a) its conclusion is both intel-
lectually important and powerful, (b) its explicit assumptions are quite weak, and (c) it has
a simple proof, in the sense that it involves only a couple steps, and each step is completely
transparent. Let me comment a bit more on each of these three points.
The …rst welfare theorem provides a formal statement of a version of Adam Smith’s
argument that the “invisible hand”of decentralized markets leads sel…sh consumers to make
decisions that lead to socially e¢ cient outcomes. Despite there being no explicit coordination
among consumers, the resulting equilibrium allocation is Pareto optimal.
Second, the only explicit assumption we made in order to prove the …rst welfare theorem
was that consumers have monotonic preferences— and even this assumption can be relaxed,
as exercise 6 below asks you to show. But in the background, there are several strong
and important assumptions. First, we assumed that all consumers face the same prices as
each other for all commodities. Second, we assumed that all consumers are price takers—

15
they take prices as given and understand that their consumption decisions do not a¤ect
these prices. Third, there are markets for each commodity, and all consumers can freely
participate in each market. Fourth, we assumed that each consumer cares only about her
own consumption and not about the consumption of anyone else in the economy— we have
therefore ruled out externalities. Finally, we assumed that there are a …nite number of
commodities and consumers. Exercise 6 asks you to show that when there are an in…nite
number of commodities and consumers, Walrasian equilibrium allocations need not be Pareto
optimal.

Exercise 6 (Adapted from MWG 16.C.3). In this exercise, you are asked to establish
the …rst welfare theorem under a set of assumptions compatible with satiation. First, we
will de…ne the appropriate notion of equilibrium. Given an economy E, an allocation (xi )i2I
and a price vector p = (p1 ; : : : ; pL ) constitutes a pricePequilibrium Pwith transfers if there
is an assignment of wealth levels (w1 ; : : : ; wI ) with i2I wi =Pp ! i such that: (i)
i2I P
consumers optimize: xi (p; wi ) = xi and (ii) markets clear: i2I xi = i2I ! i . Suppose
that every Xi is nonempty and convex and that every ui is strictly convex. Prove the
following:
(a) For every consumer i, there is at most one consumption bundle at which she is locally
satiated. Such a bundle, if it exists, uniquely maximizes ui on Xi .
(b) Any price equilibrium with transfers is a Pareto optimum.

Exercise 7. This exercise illustrates that the importance of the assumption that there are
a …nite number of commodities for the …rst welfare theorem. Consider an economy in which
there is one physical good, available at in…nitely many dates: t = 1; 2; : : : , so there are
e¤ectively an in…nite number of commodities: the physical good at date 1, the physical good
at date 2, and so on. One consumer (or “generation”) is born at each date t = 1; 2; : : : , and
lives and consumes at dates t and t + 1 (“young”and “old”). We will refer to the consumer
born in date t as consumer t. There is also one old consumer alive at date t = 1 (call her
consumer 0). Each consumer is endowed with one unit of the good when she is young, and
no storage is possible. Consumption in each period is non-negative, and each consumer t’s
preferences over consumption is given by ut (xt;t ; xt+1;t ) = u (xt;t ) + u (xt+1;t ), where u is
smooth, increasing, and strictly concave, with u0 (0) < 1.
(a) Show that there is a Walrasian equilibrium in which each consumer consumes her en-
dowment and gets utility u (1) + u (0).
(b) Show that the above Walrasian equilibrium is unique.
(c) Show that the above Walrasian equilibrium allocation is not Pareto optimal. In other
words, construct a feasible allocation that is strictly better for each consumer.

16
 I want to conclude this section with a couple comments on the simplicity of the proof
of the …rst welfare theorem. First, the theorem itself presents a partial characterization of
equilibrium allocations. To prove the statement, we did not need to solve explicitly for a
Walrasian equilibrium and show that it is Pareto optimal. Instead, we described properties
that all Walrasian equilibrium allocations must satisfy. Second, the statement itself is a
conditional statement. It is a statement of the form “if (xi )i2I is a Walrasian equilibrium,
then (xi )i2I is Pareto optimal.”This conditional statement dodges the question of whether
there is in fact a Walrasian equilibrium— we showed above that there does not always exist
a Walrasian equilibrium, and we will spend some time next week providing conditions under
which a Walrasian equilibrium in fact exists. Finally, the proof is a proof by contradiction,
and it e¤ectively takes the form of “if this Walrasian equilibrium allocation was not Pareto-
optimal, then stu¤ doesn’t add up.”While elegant, the proof itself provides little insight into
why the …rst welfare theorem holds. We will spend a little more time discussing the “why”
in week 3.

4 Second Welfare Theorem

The …rst welfare theorem establishes that Walrasian equilibrium allocations are Pareto op-
timal. The second welfare theorem in some sense establishes a converse. It says that, under
some assumptions, any Pareto optimal allocation can be “decentralized” as a Walrasian
equilibrium allocation, given the correct prices and endowments.

Theorem 2 (Second Welfare Theorem). Let E be an economy that satis…es (A1) (A4)
and Xi = RL+ . If (! i )i2I is Pareto optimal, then there exists a price vector p 2 RL+ such that
p; (! i )i2I is a Walrasian equilibrium for E.

Before proving the second welfare theorem, we will state a version of an important the-
orem in convex analysis, which is used in the key step of the proof of the second welfare
theorem.

17
Lemma 2 (Separating Hyperplane Theorem). If W Rn is an open convex set, and
z 62 W is a point not in W, then there exists a vector p 6= 0 and such that p x p z for
all x 2 cl (W).

Figure 7(a) illustrates this version of the separating hyperplane theorem in two dimen-
sions. The set W is open and convex, and z 62 W. The point z is on a line (which is a
hyperplane in a two-dimensional space) characterized by the equation p x = p z (i.e., p is
the normal vector to the line). All the points to the upper right of that line satisfy p x > p z,
and all the points to the lower left of that line satisfy p x < p z. And in particular, W is
fully to the upper right of this line. In the case illustrated in Figure 7(a), there are of course
many other separating hyperplanes satisfying p x p z for all x 2 W corresponding to
di¤erently sloped lines going through z but not intersecting W. Figure 7(b) shows why the
assumption that W is a convex set is important for this result. If z 62 W, but z 2 conv (W),3
then there is no vector p 6= 0 for which p x p z for all x 2 W. Exercise 8 asks you to
prove a stronger version of the separating hyperplane theorem, which shows that any two
disjoint convex sets can be separated by a hyperplane.
3
The set conv (W) is de…ned to be the smallest convex set containing W. In two dimensions, you can
visualize conv (W) by taking W and putting a rubber band around it.

18
 Figures 7(a) and 7(b): separating hyperplane theorem and convexity.

The idea of the second welfare theorem is to show that, if the endowment (! i )i2I is Pareto
optimal, we can always …nd a price vector that separates the set of allocations preferred by
all consumers in the economy from (! i )i2I and therefore show that p; (! i )i2I is a Walrasian
equilibrium.

Proof of Theorem 2. By the statement of the theorem, (! i )i2I is Pareto optimal. Let us
de…ne the set of aggregate consumption bundles that can be allocated in such a way among
consumers to make them all strictly better o¤ than under (! i )i2I . To do so, de…ne the set
of consumption bundles that consumer i prefers to ! i :

Ai = a 2 RL : a + ! i 0 and ui (a + ! i ) > ui (! i ) .

Since ui is concave, the set Ai is convex. The Minkowski sum of the sets Ai is therefore also

19
a convex set.4 That is, if we de…ne
( )
X X
A= Ai = a 2 RL : 9a1 2 A1 ; : : : ; 9aI 2 AI with a = ai ,
i2I i2I

then A is a convex set. The set A does not contain the 0 vector because (! i )i2I is Pareto
optimal. To see why this is the case, note that if 0 2 A, then there would exist (ai )i2I with
P
i2I ai = 0 and ui (ai + ! i ) > ui (! i ) for all i. That is, we could essentially just reallocate

the endowment (! i )i2I among the I consumers and make them all strictly better o¤, but
that would contradict the assumption that (! i )i2I is Pareto optimal.
Next, by Lemma 2, there is some price vector p 6= 0 such that p a 0 for all a 2 cl (A).
Moreover, each of the prices pl 0. To see why, suppose pl < 0 for some l. Take some a
for which al is arbitrarily large and all other al0 are arbitrarily small but positive. By the
monotonicity of consumer preferences, a 2 A, but if a is chosen this way, then p a < 0. We
therefore have that p > 0 (i.e., pl 0 for all l 2 L with at least one inequality strict).
We will now show that p ; (! i )i2I is a Walrasian equilibrium. To do so, we need to
show that at p , consumers optimally consume their endowments and that markets clear.
The second condition is immediate. It remains to show that at this p , consumers optimally
consume their endowments. To do so, suppose there is some x^i 2 RL+ for which ui (^
xi ) >
ui (! i ). We will show that this x^i 62 Bi (p ). By the de…nition of A, the allocation (xi )i2I
(! i )i2I with xi = x^i and xj = ! j for all j 6= i, is in cl (A). By the de…nition of p , we
P
necessarily have that p (^ xi ! i ) + p j6=i (! j !j ) 0, which implies that p x^i
p ! i > 0, where this last inequality holds because of Assumption (A4) that all consumers
have positive endowments of all commodities.
We are not yet done, because we have to show that this last inequality is strict. This is
where continuity of preferences (Assumption (A1)) comes into the picture. Since ui (^
xi ) >
4
The Minkowski sum of two sets A and B is just the set of vectors x that can be written as the sum of
vectors x = a + b for which a 2 A and b 2 B. The closest visual analog to thinking about the Minkowski
sum of sets in two dimensions is the way the clone stamp tool in Photoshop works if you are familiar with
it.

20
ui (! i ), this implies that for just less than 1, ui ( x^i ) > ui (! i ), which in turn implies that
p x^i p ! i > 0. This cannot be the case if p x^i = p ! i , so we must therefore have that
p x^i > p ! i and hence x^i 62 Bi (p )— that is, any allocation preferred by consumer i to her
endowment is una¤ordable, and hence her optimal consumption bundle is her endowment.

The second welfare theorem does not show that every Pareto optimal allocation is a Wal-
rasian equilibrium given a particular endowment. Instead, it says that if we were to start
from a particular endowment (! i )i2I , and an allocation (xi )i2I is Pareto optimal, then we
could reallocate consumers’endowments in such a way that (xi )i2I is a Walrasian equilib-
rium allocation. The version of the theorem that we just proved carries out this exercise
using a particularly stark reallocation of endowments (i.e., it just sets (! i )i2I = (xi )i2I ).
There are versions of the theorem that involve carrying out lump-sum transfers of wealth
rather than directly moving around endowments. As you might expect, decentralizing a par-
ticular Pareto-optimal allocation in practice potentially requires large-scale redistribution of
wealth. I view the result more as establishing an equivalence between Walrasian equilib-
ria and Pareto-optimal allocations rather than as a practical guide for …guring out how to
achieve a particular distribution of consumption in society.
It is worth a reminder that convexity of consumers’preferences was critical in establish-
ing the result that A was a convex set, which in turn is required for using the separating
hyperplane theorem. Figure 8 shows an example where the conclusion of the second welfare

21
theorem fails if consumers’preferences are not convex.

Figure 8: Non-convex preferences

In this …gure, the endowment is a Pareto-optimal allocation, since consumer 1’s and consumer
2’s better-than sets are separated. But there are no prices that can make it optimal for
consumer 1 to consume ! 1 .
Nevertheless, a version of the second welfare theorem continues to hold when consumers
do not have convex preferences if you replicate the economy a large number of times. Think
of the 2-consumer economy as being a metaphor for a large economy with two types of
consumers: type-1 consumers have preferences u1 and endowments ! 1 , and type-2 consumers
have preferences u2 and endowments ! 2 . If we replicate the economy a large number of times,
so that there are N type-1 consumers and N type-2 consumers, where N is large, then we
can support ! as a Walrasian equilibrium allocation, at least on average. This result follows
from an application of the Shapley-Folkman lemma, which roughly says that the Minkowski
average of sets converges to the convex hull of that set. You don’t need to know the math
behind this result, but it is a useful result to be aware of. Figure 9 illustrates a replication

22
economy for the economy described in Figure 8. It shows that there may be a p for which
there is an x1 2 x1 (p; p ! 1 ) and an x01 2 x1 (p; p ! 1 ), so that if we allocate a fraction of
type-1 consumers to consume x1 and a fraction 1 of type-1 consumers to consume x01 ,
on average they are consuming ! 1 : x1 + (1 ) x01 = ! 1 .

Figure 9: Replication economy

This …gure illustrates the idea that large numbers “convexi…es” the economy. There
is a recurring theme throughout general equilibrium theory that many of the pathologies
that arise seem to “go away” in su¢ ciently large economies. Nonconvexities seem esoteric,
since we usually think of consumers’preferences as having diminishing marginal utility and
preferences for variety. Nonconvexities become especially relevant when we think of …rms,
though. When there are …xed costs, for example, the …rm analogue of consumers’“better-
than”sets are not convex, since the set of production levels better than “not even breaking
even”can include both “shut down”and “produce, but at a much larger scale.”
Finally, we made use of Assumption (A4) in a somewhat opaque way in the proof. What
Assumption (A4) rules out is cases like the one illustrated in Figure 5(b) in which there were
no Walrasian equilibria. The failure of equilibrium existence illustrated in Figure 5(b) arises

23
because of a sort of “division by zero”problem: supporting the endowment as an equilibrium
allocation would have required consumer 2 to buy only a …nite amount of a commodity with
a zero price when she has zero wealth.
Exercise 8. This question is intended to guide you through a proof of the separating
hyperplane theorem. This is more of an exercise in math than in economics, so feel free to
skip to the next step if you get stuck.
(a) Prove that if y 2 RN and C RN is closed, then there exists a point z 2 C such that
jjz yjj jjx yjj for all x 2 C. That is, there exists a point in C that is closest to y.
(You may assume that jj jj is the Euclidean norm.) Hint: use the Weierstrass extreme value
theorem— if f is a real-valued and continuous function on domain S, and S is compact and
non-empty, then there exists x such that f (x) f (y) for all y 2 S.
(b) Suppose further that C RN is convex, and note from above that if y 62 C, then there
exists z 2 C that is closest to y. Let x 2 C with x 6= z.
(i) Show that (y z) z (y z) x. Hint: consider jjy (z + t (x z))jj for t 2 [0; 1],
the distance between y and a convex combination of x and z.
(ii) Use the above result to show that for all x 2 C, (y z) y > (y z) x.
(iii) Explain how this is a special case of the separating hyperplane theorem, which states
that for any disjoint convex sets A; B RN , there exists nonzero p 2 RN such that p u p v
for any u 2 A and v 2 B.
(iv) Use the result of (ii) to deduce the separating hyperplane theorem. Hint: consider
y = 0 and C = A B = fu v : u 2 A; v 2 Bg.

5 Characterizing Pareto-Optimal Allocations

The welfare theorems provide a tight connection between the set of Pareto optimal allocations
and the set of Walrasian equilibrium allocations. This section will provide a short note on
how to …nd Pareto optimal allocations in particularly well-behaved environments. De…ne
the utility possibility set

U = (u1 ; : : : ; uI ) 2 RI : there is a feasible allocation (xi )i2I with ui (xi ) ui for all i .

If the sets Xi are convex sets and consumers’preferences are concave, then U is a convex set.
When this is the case, the problem of …nding Pareto-optimal allocations can be reduced to

24
the problem of solving Pareto problems of the form

max u
u2U

for some non-zero vector of Pareto weights 0. The objective function of this problem
is sometimes called a linear Bergson-Samuelson social welfare function. We will say that u
is a Pareto-optimal utility vector if there is a Pareto-optimal allocation (xi )i2I for which
ui (xi ) = ui for all i 2 I. The next theorem establishes the result.

Theorem 3. If u is a solution to the Pareto problem described above for some vector of
Pareto weights >> 0, then u is a Pareto-optimal utility vector. Conversely, if the utility
possibility set U is convex, then any Pareto-optimal utility vector u is a solution to the
Pareto problem for some non-zero vector 0.

Proof of Theorem 3. The …rst part is immediate: if u is not Pareto optimal, then any
Pareto-dominating utility vector would give a higher value in the Pareto problem for any
Pareto weight vector >> 0.
The second part of the theorem makes use of the supporting hyperplane theorem, which
says that a convex set can be separated from any point outside its interior (see Section M.G
of the mathematical appendix of MWG). If u is a Pareto-optimal utility vector, then it lies
on the boundary of U, so by the supporting hyperplane theorem, there exists 6= 0 such that
u u for all u 2 U. Further, the Pareto weights satisfy 0, since if i < 0 for some
i, then u < u~, where for some K > 0, u~ = u1 ; : : : ; ui 1 ; ui K; ui+1 ; : : : ; uI 2 U.
This contradicts the claim that u u for all u 2 U, so it must be the case that
0.

The theorem shows that when the utility possibility set is a convex set, the problem
of …nding Pareto-optimal allocations boils down to solving a class of Pareto problems. If
we further assume that consumers’utility functions are di¤erentiable, then Pareto-optimal
allocations can be characterized by taking …rst-order conditions. For example, suppose utility

25
functions are di¤erentiable with rui (xi ) >> 0 for all xi , and we have an interior solution,
we can …nd Pareto-optimal allocations by solving the problem:

X
max i ui (xi )
(xi )i2I
i2I

subject to feasibility for each commodity:

X X
xl;i ! l;i for all l 2 L.
i2I i2I

Then one can use the Kuhn-Tucker theorem to verify that any Pareto-optimal allocation
(xi )i2I with xi >> 0 for all i 2 I must satisfy

@ui =@xl;i @ui0 =@xl;i0 l

= = for all i; i0 ; l; l0
@ui =@xl0 ;i @ui0 =@xl0 ;i0 l0

for some l; l0 > 0. This condition says that the marginal rate of substitution between any
two commodities must be equalized across consumers in any Pareto-optimal allocation. If
this condition failed, there would be a Pareto-improving exchange of commodities l and l0
between consumers i and i0 . The values l corresponds to the Lagrange multiplier on the
P P
commodity-l feasibility constraint i2I xl;i = i2I ! l;i .
As an illustration of the second welfare theorem, given a Pareto-optimal allocation (xi )i2I
that satis…es the optimality conditions above, if you set pl = l for all l 2 L, then p; (xi )i2I
is a Walrasian equilibrium of the economy E = (ui )i2I ; (xi )i2I . This point is illustrated
in Figure 4(b). In that …gure, at the Walrasian equilibrium allocation, consumers’marginal
rates of substitution across the two commodities were equalized. Moreover, these marginal
rates of substitution were also equal to the price ratio that corresponded to the Walrasian
equilibrium (and given that price ratio, moving the endowment along the boundary of con-
sumers’budget sets does not change their ultimate consumption choices, so the same price
vector would also be an equilibrium price vector if we just set consumers’endowments equal

26
to their Walrasian equilibrium allocations).
Exercise 9 (Adapted from MWG 16.C.4). Suppose that for each consumer, there is
a “happiness function” depending on her own consumption only, given by u (xi ). Every
consumer’s utility depends positively on her own and everyone else’s “happiness”according
to the utility function

Ui (x1 ; : : : ; xl ) = Ui (u1 (x1 ) ; : : : ; ul (xl )) .

Show that if x = (x1 ; : : : ; xl ) is Pareto optimal relative to the Ui ( )’s, then x = (x1 ; : : : ; xl ) is
also a Pareto optimum relative to the ui ’s. Does this mean a community of altruists can use
competitive markets to attain Pareto optima? Does your argument depend on the concavity
of the ui ’s or the Ui ’s?

27