Lecture Notes (EC740)

Hideo Konishi
October 26, 2014

[Note] Please let me know if you …nd typos. I will update the
notes later in class.

0. Math One-Minute Lessons

1. Introduction

Examples: taxation, labor supply, Robinson-Crusoe economy

Basic de…nitions of economy and equilibrium

2. Exchange economy

o¤er curves, Walras equilibrium, Pareto e¢ ciency

the …rst welfare theorem

3. Existence of Equilibrium

Debreu’s standard proof

(in)dispensability of convex preferences and/or convex consump-
tion set (location choice)

4. Miscellaneous Topics (Illustrations)

The core: WE Core, core convergence

uniqueness of equilibrium: gross substitutes and index theorem
Stability: WARP and Liapunov function

1
5. Production Economy

the …rst and second welfare theorems

comments on existence

6. Scope of Application and Market Failures

applications
– time
– location choice
– transportation cost
missing markets
– new products
– externalities, public goods
– asymmetric information: adverse selection, moral hazard
– absence of complete asset market
OLG model
ill-de…ned property right
distortions (tax, monopoly power)

7. Uncertainty: Complete and Incomplete Markets

Arrow-Debreu equilibrium, Radner equilibrium, Asset market

Complete and incomplete markets

8. 2 2 Production Economy

Shephard’s lemma,
factor price equalization theorem
the Stolper-Samuelson theorem

9. Introduction to the Theory of Second Best

Corlett and Hague

optimal taxation

2
 Math One-Minute (?) Lessons
1. [Closed sets] In each class (just the …rst part), I talk about math tools
we need in class.

Consider an n-dimensional Euclidean space. Let fx1 ; x2 ; x3 ; :::::::g =

fxk g1 n 1 n n
k=1 be a sequence in R where xk = (xk ; ::::; xk ) 2 R for all
k = 1; 2; :::. A sequence is convergent if there is x = (x ; :::; xn ) 2
1

Rn such that kxk xk ! 0 (for each > 0 there exists a natural

number K( ) such that for all k > K( ) we have kxk xk < . El-
ement x is called the limit of fxk g1 1
k=1 , and we write fxk gk=1 ! x.

Consider X Rn . Set X is nonempty if there is at least an

element in X: X =
6 ;.
A point x 2 Rn is a contact point of X if there is a sequence in
X with fxk g1
k=1 ! x.

Let X S Rn . Set X is closed in S if all contact points of X

in S is in X.
– In other words: Set X is closed in S if any convergent sequence
in X has a limit in X if S is closed. [Pictures].

2. [Open sets: supplement]

**Let X S Rn . Set CS (X) = fx 2 S : x 2

= Xg is comple-
ment of X S.
**Set X S Rn is open in S if CS (X) is closed in S. [Pictures]
**An element x 2 X is an interior point of X on S if x is not a
contact point of CS (X). Denote the set of interior points of X on
S by intS (X). [pictures]
F **Set X S Rn is open in S if every element of X is an interior
point of X on S.
F If set X S Rn is open in S, then for all x 2 X there is an
> 0 such that B (x) = fz 2 S : kz xk < g X \ S.

3. [Convexity and addition of sets]

3
 Consider set X Rn . Set X is convex if x + (1 )y 2 X for
all x; y 2 X and all 2 [0; 1].
Consider nonempty sets X; Y Rn . A set X + Y = fz 2 Rn :
z = x + y for some x 2 X and y 2 Y g is an addition of sets X
and Y .
Similarly, X Y = fz = x y for some x 2 X and y 2 Y g is an
subtraction of Y from X.
Consider set X Rn . For all 0, set X = fz 2 Rn : z = x
for some x 2 Xg is a multiplication of set X and .
F Consider nonempty sets X; Y Rn . If X and Y are convex, X +Y
is convex.
F Consider set X Rn . If X is convex, then X is convex for all
0.

4. [Compactness]

Consider set X Rn . Set X is bounded if there is > 0 such

that X B = fz 2 Rn : kz 0k g.
Consider set X Rn . Set X is compact if X is closed and
bounded.
Pick any sequence fxk g1 k=1 in a nonempty set X Rn . A sub-
sequence of fxk g1 k=1 is a sequence generated by dropping some
elements of fxk g1
k=1 and renaming the leftover elements of it. For
example, dropping x2 ; x4 ; x5 ; ::: from a sequence fxk g1
k=1 , we ob-
k0 1
tain a subsequence fx gk=1 :

x1 ; x/2 ; x3 ; x/4 ; x/5 ; x6 ; x7 ; :::

=)
x1 ; x3 ; ; x6 ; x7 ; :::
=)
fx10 ; x20 ; x30 ; x40 ; :::g

F Pick any sequence fxk g1k=1 in a nonempty and compact set X

Rn . Then, there is a convergent subsequence fxk0 g1 k 1
k=1 of fx gk=1
with fxk0 g1
k=1 ! x 2 X.

4
 F Consider nonempty sets X; Y Rn . If X and Y are compact,
X + Y is compact.
F Consider nonempty sets X; Y Rn . Even if X and Y are closed,
X + Y may not be closed.

5. [Continuity of functions]

Consider nonempty sets X Rn and Y Rm , and a function

f : X ! Y . Function f is continuous at x 2 X if for all
convergent sequence fxk g1 k
k=1 ! x in X, limk!1 f (x ) = f (x).
Function f is continuous in X if f is continuous at any x 2 X.
F **Function f : X ! Y is continuous if and only if f 1
(T ) is closed
in X for all closed set T Y .
F **[special case] Real-valued function f : X ! R is continuous if
and only if f 1 ([y; 1)) and f 1 (( 1; y]) are closed in X for all
y 2 R.
**Note that the above special case is exactly the same as the
"continuity axiom" for preference ordering: "both upper and lower
contour sets are closed."
F Consider a continuous function f : X ! Y . If S X is compact
then f (S) is compact.
F [special case: Weierstrass’s Theorem] Consider a continuous
real-valued function f : X ! R. If S X is nonempty and
compact, f has the maximum and minimum values in S.
Note that this Weierstrass’s theorem proves existence of demand
(optimal consumption bundle) when the budget set is compact.
Similarly, existence of …rm’s supply (optimal input-output choice)
if production set is compacti…ed.

6. [Continuity of correspondences]

A correspondence : X Y is a set-valued function: i.e.,

assigns a subset (x) of Y to each element x 2 X. Note that a
function f : X ! Y is a single-valued correspondence: i.e., for
each x 2 X a single element in Y is assigned by f (x). Thus,
a function is a special kind of correspondence (a single-valued
correspondence).

5
A correspondence is nonempty-valued, i¤ (x) 6= ; for any
x 2 X. (This can be assured by the Weierstrass’s theorem.)
A correspondence is upper hemi-continuous, i¤ for any con-
vergent sequence fxk g1 1
k=1 ! x, any sequence fyk gk=1 with yk 2
(xk ) for each k = 1; 2; :::, and any convergent subsequence of
fyk g1 0 1
k=1 , fyk gk=1 ! y, we have y 2 (x). Note that graph of
an upper hemi-continuous is closed in X Y . An upper hemi-
continuous correspondence is one of generalizations of a continu-
ous function for multi-valued function world. A continuous func-
tion f has the following property: for any convergent sequence
fxk g1 1
k=1 ! x, a sequence ff (xk )gk=1 ! f (x). Since we are work-
ing on correspondence, we need to select an element of (xk ) for
each k.
A correspondence is lower hemi-continuous, i¤ for any x 2 X,
and any y 2 (x), and any convergent sequence fxk g1 k=1 ! x,
1
there is a sequence fyk gk=1 such that yk 2 (xk ) for each k =
1; 2; :::, and fyk g1
k=1 ! y.

A correspondence is continuous, i¤ is upper and lower hemi-

continuous. A typical PLexample is a budget correspondence. Let
L
= fp0 2 RL+ : p
i=1 i
0
= 1g be the price simplex (the set
of relative price vector). Imagine B i : L B i (p) = fxi =
i i L i
(x1 ; ::::; xL ) 2 R+ : p x p ! g where ! = (! 1 ; :::; ! iL ) satis…es
i i i

! i` > 0 for all ` = 1; :::; L. This is a continuous correspondence.

– Assumption " ! i` > 0 for all ` = 1; :::; L" is important for
the continuity of budget correspondence. Assume L = 2,
X i R2+ , and ! i = (2; 0), and consider a sequence fpk g1 k=1 !
k k k k k
p = (0; 1) such that p = (p1 ; p2 ) 0 with p1 + p2 = 1
for all k = 1; 2; :::. Then, while x1 2 is satis…ed for all
x = (x1 ; x2 ) 2 B i (pk ) for all k = 1; 2; :::, B i (p) = f(x1 ; x2 ) 2
R2+ : x2 = 0g suddenly at p = (0; 1). Thus, lower hemi-
continuity is violated in this case. (Take (10; 0) 2 B i (p).
There is no convergent sequence fxk g1 k=1 ! (10; 0) such that
k k
x 2 B(p ) for all k = 1; 2; :::) In contrast, any convergent
sequence fxk g1 k=1 ! x in the budget correspondence (x 2
k

B(pk ) for all k = 1; 2; :::), we have x 2 B(p), and B is upper

hemi-continuous.

6
7. [Berge’s maximum theorem]

F [Berge’s maximum theorem] Suppose that X and Y are non-

empty subsets of Euclidean spaces, ' : Y X is a nonempty-
valued, compact-valued, and continuous correspondence, and u :
X Y ! R is a continuous function. Let : Y X be such that
(y) fx 2 '(y) : u(x; y) u(x0 ; y) for any x0 2 '(y)g. Then,
is a (nonempty-valued and) upper-hemicontinuous correspondence
(u.h.c.).
– In words, Berge’s maximum theorem says "The limit of opti-
mal choice is optimal."
– Note that convexity of sets is needed here (discrete choice is
allowed).
Berge’s maximum theorem guarantees u.h.c. of demand corre-
spondence and supply correspondence.
– Let Y = L , X be consumption set X i , ' be budget corre-
spondence (B(p)), and u(x; y) = ui (x) be a utility function (a
special case). Then i (p) is consumer i’s demand correspon-
dence.
– Let Y = L , ' be a constant mapping to production set
('(p) = Y j : trivially continuous correspondence), and u(x; y)
be j (p). Then j (p) is …rm j’s supply correspondence.
– Note that in these examples, we need to compactify X (choice
set). And for consumer demand, we need ! i 2 intX i , where
int( ) denotes the interior of set X i .
– Recall the previous budget correspondence example without
lower hemi-continuity. If consumer i’s utility is ui (x1 ; x2 ) =
x1 + x2 , then (pk ) = (2; 0) for large ks, but (p) is in…nite
consumption of commodity 1 (formally speaking, demand is
empty set at p). This discontinuity of demand occurs even if
commodity space is compacti…ed.
– This theorem is directly applicable to Nash equilibrium exis-
tence theorem in game theory, too.

8. [Convex-valuedness of correspondence]

7
 A correspondence : Y X is convex-valued, i¤ for any y 2 Y ,
any x; x 2 (y), and any t 2 (0; 1), tx + (1 t)x0 2 (y) holds.
0

A function u : X Y ! R is quasi-concave in X, if for any

y 2 Y , any x; x0 2 X, and any t 2 (0; 1), we have u(tx + (1
t)x0 ; y) minfu(x; y); u(x0 ; y)g. If the inequality is strict for all
x 6= x0 , then u : X Y ! R is said strictly quasi-concave in
X.
F [conveniently adjusted Berge’s theorem I] Suppose that X and
Y are nonempty and compact subsets of Euclidean spaces, ' :
Y X is a nonempty-valued, continuous, compact-valued, and
convex-valued correspondence, and u : X Y ! R is a continuous
and quasi-concave function in X. Let : Y X be such that
(y) fx 2 X : u(x; y) u(x0 ; y) for any x0 2 '(y)g. Then,
is a nonempty-valued upper-hemicontinuous and convex-valued
correspondence.
F [conveniently adjusted Berge’s theorem II: even stronger restric-
tion] Suppose that X and Y are nonempty and compact sub-
sets of Euclidean spaces, ' : Y X is a nonempty-valued and
continuous function, and u : X Y ! R is a continuous and
strictly quasi-concave function in X. Let : Y X be such that
0 0
(y) fx 2 X : u(x; y) u(x ; y) for any x 2 '(y)g. Then, is
a continuous function.
– Here, …nally you have assurance for continuity of demand and
supply (factor demand) functions, which is assumed in Mar-
vin’s class!

9. [Kakutani’s …xed point theorem]

F [Kakutani’s …xed point theorem] Suppose that X is a non-

empty, compact and convex subset of Rn , and that a mapping
: X X is a nonempty-valued, convex-valued, and upper-
hemicontinuous correspondence. Then, there exists a …xed point
of : i.e., there exists an x 2 X such that x 2 (x ).

10. [Separating hyperplane theorem]

Consider p 2 Rn nf0g and c 2 R. A hyperplane generated by p

and c is a set Hp;c = fz 2 Rn : p z = cg.

8
 A set S 2 Rn is separated by a hyperplane Hp;c if either S
fz 2 Rn : p z cg (half-space below) or S fz 2 Rn : p z cg
(half-space above).
F [Separating hyperplane theorem] Suppose that X; Y Rn
are convex and X \ Y = ;. Then, there exists p 2 Rn nf0g and
c 2 R such that Hp;c separates X and Y .

9
1 Introduction
1. Organization Matters: Hideo Konishi 482, 2-1209, [email protected],
O¢ ce Hours just stop by or by appointment
2. General equilibrium theory in our …rst year micro sequence has two
aspects. Putting Marvin’s material together to analyze the entire econ-
omy is one thing, and regarding market equilibrium as a mechanism to
think about other possible allocation mechanisms, we can start think-
ing about theory of social choice and mechanism design.
3. General equilibrium does not mean that the analysis is very compli-
cated. The important thing is that the model is closed (with relevant
sectors and economic agents).
4. This course will make use of mathematics. However, the important
thing is to understand de…nitions and logics. The course is designed to
be more or less self-contained in math tools.
5. General equilibrium theory deals with resource allocation problem in
the whole economy. It provides foundations for market economy (the
…rst and second welfare theorems and the core convergence theorem).
Also general equilibrium theory is an important language to talk about
economy as a whole. For example, if you want to study macroeconomic
phenomenon, you need general equilibrium theory to describe the econ-
omy. If you are interested in e¤ects of government policies, you need
think about the e¤ects of both governmental expenditure and …nance
(taxes), since the government is a big player that has a signi…cant im-
pacts in both expenditure and …nancing. International trade (market
interaction between countries) and public economics (economics of the
government) need general equilibrium for sure.
6. Why is general equilibrium important? There are pitfalls in partial
equilibrium analysis. A trivial example is Pepsi and Coke markets.
The two demand curves interact with each other. If Pepsi marginal
cost exogenously goes down (by a technological innovation), then Pepsi
market price decreases. Then, Coke demand decreases, and Coke mar-
ket price goes down. Then, the Pepsi demand decreases and the Pepsi
market price goes down, etc. Clearly, a partial equilibrium analysis is
not su¢ cient.

10
 7. Let me explain more substantial pitfalls of partial equilibrium analysis
by using a taxation analysis. Let’s …rst review undergraduate-level
partial equilibrium analysis. We use surplus analysis, so tax revenue
will be returned to consumers (in a lump sum manner).

8. producer tax:

(producer price) = (market price) (producer tax)

and
(demand) = (supply)
Thus, the traded quantity is less, and the total surplus

(after tax CS)+(after tax PS)+(tax revenue)+(deadweight loss) = (CS)+(PS)

9. consumption tax:

(consumer price) = (market price) + (consumer tax)

As long as ($ consumer tax) = ($ producer tax), the above two taxes

achieve the exactly the same allocation. Both taxes generate welfare
losses (the deadweight loss).

10. In partial equilibrium analysis, which commodity should be taxed?

From the …gure on the blackboard, it is clear that if one of supply
or demand curve is inelastic then the deadweight loss is small. Thus,
an obvious policy implication is that the government should impose
higher tax rates on inelastically supplied (demanded) goods. The in-
verse elasticity rule.

11. Consider ad valorem tax (tax rates in %: not tax rates in $). Suppose
that consumption of all goods face the same ad valorem tax rates,
and that the tax revenue is returned to consumers (assume identical
consumers). Then, what happens? Intuitively,

all prices are increased by the same percentage

all tax paid is returned to compensate the reduction in consump-
tion

11
12. M RS = M RT is satis…ed q` = (1 + t)p` for all `. For all ` and `0
q` (1 + t)p` p`
M RS = = = = M RT
q `0 (1 + t)p`0 p `0
Thus, the allocation will not be a¤ected. This is totally contradic-
tory with partial equilibrium analysis. It is important to think about
interactions among markets.

[NOTE] It is hard to tax all goods. It can be hard to observe

consumption level of a good if household is endowed with the
good. For example, think about leisure. Consumers sell some of
her leisure endowment (labor supply in hours). Taxing on leisure
consumption means consumption price of leisure is raised and the
tax base is 24 hours minus working hours. What is (pretax) con-
sumption price of leisure? It is wage rate since wage rate is the op-
portunity cost of leisure consumption. Commonly observed labor
income tax reduces consumption price of leisure, so it is actually
a leisure subsidy.
[NOTE] Actually, labor supply is an interesting problem that
bridges Marvin’s stu¤ and general equilibrium theory. In Mar-
vin’s class, consumer i’s (uncompensated: Marshallian) demand
function of good ` 2 f1; :::; Lg is written as
xi` = xi` (p; I i )
This is actual consumption amount of good ` when consumer i
has income I i . Note that all prices matter p = (p1 ; :::; p` ; :::; pL ) in
deciding xi` . Now, instead of income I i , consumer has a vector of
goods ! i = (! i1 ; :::; ! iL ). Then, what is her demand for commodity
`?
xi` = xi` (p; p ! i )
Just I i is replaced by p ! i . How much more does she purchase
good i in the market (net demand for good i)?
xi` ! i` = xi` (p; I i ) ! i`
In the case of ` being leisure, leisure consumption must be less
than ! i` = Li (leisure endowment), so xi` Li 0. This is negative
labor supply z i = Li xi` , where z i denotes consumer i’s labor
supply. Draw pictures.

12
 More generally, given arbitrary ! i , draw pictures.
[Note on notations] In Mas-Colell’s book, both i and ` are sub-
scripts of consumption vector (xi = (xi1 ; :::; xi` ; :::; xiL )). However,
I will use the above notations so that you can see what each vari-
able represents (Imagine x12 in an economy with I = 2 and L = 2.
Isn’t x12 clearer than x12 ? The same?)

13. Informal introduction of a one-consumer, one-producer economy (Robin-

son Crusoe economy: page 525).

one consumer (or many homogeneous consumers)

one …rm (or many homogeneous …rms)
commodity 1 is leisure; commodity 2 is a consumption good pro-
duced by a …rm
…rm uses labor input z and produce the consumption good using
production function q = f (z)
price of labor is w; price of the consumption good is p
…rm’s optimization problem

max pf (z) wz
z 0

given price vector (w; p), …rm’s labor demand is z(w; p), output
supply is q(w; p) and pro…t is

(w; p) = pq(w; p) wz(w; p)

with q(w; p) = f (z(w; p))

noting that …rm is ultimately owned by consumer, pro…t is a part
of consumer income

I=w L x1 + (w; p)

consumer’s optimization problem is

max u(x1 ; x2 ) s:t: px2 w L x1 + (w; p)

given price vector (w; p), consumer demand functions are x1 (w; p)
and x2 (w; p).

13
A (Walras or market) equilibrium price vector (w; p) requires
that demand equals supply in every market. (i) consumption mar-
ket: demand = supply

x2 (w; p) = q(w; p)

and (ii) labor market: supply = demand

L x1 (w; p) = z(w; p)

Consider the budget constraint (with equality) in which demand

functions and the de…nition of pro…t function are used

px2 (w; p) = w L x1 (w; p) + pq(w; p) wz(w; p)

Rearranging the above

p (x2 (w; p) q(w; p)) = w L x1 (w; p) z(w; p)

This equation implies that if either of consumption good or labor

market is cleared, then the other market is also cleared (the Walras
law: strict version). We only need to consider one of the two
markets. This fact is true for general economy. "If there are
L goods, then equilibrium conditions all but one market implies
equilibrium in the last good market." The Walras law
draw pictures of the Walras equilibrium.
only relative price matters (demand and supply are homogeneity
of degree zero) only L 1 prices matter (a numeraire)
e¢ ciency of equilibrium (the highest utility allocation under the
feasibility: in this case) "If there is an equilibrium, it is e¢ cient."
The …rst welfare theorem
equilibrium exists if utility function is quasi-concave (convex pref-
erence), production function is concave (convex technology), and
more "hidden conditions" are satis…ed. "Under appropriate con-
ditions including convex preferences and technology, there exists a
Walras equilibrium." Walras equilibrium existence theorem

14
 an e¢ cient allocation is a Walras equilibrium under convex pref-
erence and technology and more hidden assumptions. "Under ap-
propriate conditions including convex preferences and technolo-
gies, every (Pareto) e¢ cient allocation is supported by a Wal-
ras equilibrium with appropriate transfers. The second welfare
theorem

14. Informal presentation of economies (Chap 10.B): An economy is de-

scribed by

I consumers: indexed by i = 1; :::; I

J …rms: indexed by j = 1; :::; J
L goods: indexed by ` = 1; :::; L
market price vector p = (p1 ; :::; pL ) 2 RL+
each consumer i has preferences over consumption bundles xi =
(xi1 ; :::; xi` ; :::; xiL ) in her consumption set X i RL
each consumer i’s preference is represented by utility function ui :
Xi ! R
each consumer i has her endowment vector ! i 2 RL
each …rm can choose its production vector y j = (y1j ; :::; y`j ; :::; yLj )
from its production set Y j RL
ij
each consumer i has share of each …rm j, : i.e., consumer i can
earn her pro…t income from …rms

15. A Walras (market) equilibrium is a list (p; (xi )Ii=1 ; (y j )Jj=1 ) of an

economy (I; J; L; (X i ; ! i ; ( ij )Jj=1 ; ui )Ii=1 ; (Y j )Jj=1 ) is such that

(a) For every j, y j 2 Y j maximizes pro…ts in Y j : i.e., p y j p yj

for all y j 2 Y j
(b) For every i, xi 2 X i is maximal for i (or maximizes utility ui )
in the budget set
( )
X
J
ij
xi 2 X i : p xi p !i + p yj
j=1

15
 PI PI PJ
(c) i=1xi = i=1 !i + j=1 y j (equilibrium in market of every
good `)

16. [Note] You may wonder why p y j is the pro…t of …rm j. As you will see,
in general equilibrium theory, an input of …rm j is regarded as negative
output of …rm j. Thus, we have

X
L
p yj = p` y`j
`=1
X X
= p` y`j p `0 ( y`j0 )
| {z }
`2Outputs `0 2Inputs
input amount
= (Revenue) (Cost)

17. How do we …t one consumer one …rm economy to this general frame-
work?

I = 1 and L = 2 such that ` = 1; 2 represent leisure and consump-

tion good, respectively
X = (x1 ; x2 ) 2 R2+ : x2 L where L is leisure endowment
! = (! 1 ; ! 2 ) = (0; L)
Y = f( z; q) 2 R2 : q f (z)g: input is negative output: if y` > 0
then y` is the amount of output, while if y` < 0 then y` denotes
the amount of input (This is the key in production economy.)

18. We will consider general economies after we talk about a less-trivial

example than the one-consumer-one-…rm example: the Edgeworth (ex-
change) economy.

19. Informal introduction of a two person two good exchange economy

(Edgeworth economy).

i 2 f1; 2g: consumers

` 2 f1; 2g: goods
draw two consumers’budget lines and optimal consumption bun-
dles
put them together in the Edgeworth box

16
 describe o¤er curves in two consumers’consumption sets
describe the market equilibrium
the Walras law

2 Exchange Economy
1. An exchange economy is a list (I; (X i ; ! i ; ui )i2I ) such that I =
f1; :::; Ig is a set of households (an abuse of notation), X i RL+ is
i i i i
a consumption set for all i, ! 2 X is i’s endowment, and u : X ! R
is i’s utility function. Here L is the number of goods in the economy
(…nite).

2. Draw an Edgeworth box for the case of I = f1; 2g.

endowment
feasible allocations
budget sets (include infeasible allocations)
o¤er curves and net demand and supply: O¤er curve for i 2 I:
xi = xi (p; p ! i ).
market equilibrium

3. A feasible allocation isPa list x P= (xi )i2I = (x1 ; :::; xI ) such that
xi 2 X i for all i 2 I and i2I xi = i2I ! i , where xi = (xi1 ; :::; xiL ).

4. A market (Walrasian) equilibrium is a list (p; x) where x is a fea-

sible allocation and p 2 RL such that (i) p xi p ! i for all i 2 I and
(ii) for all i 2 I, ui (xi ) ui (~
xi ) holds for all x~i 2 X i with p x~i p ! i .

5. Draw an Edgeworth box with o¤er curves again. O¤er curve for i 2 I:
xi = xi (p; p ! i ). There are substitution e¤ect and income e¤ect: income
e¤ect depends on two things — (i) is it a normal good or an inferior
good, and (ii) does consumer i’s income increase or decrease by the

17
 price change. Recall labor supply. Slutsky equation:
dxi @xi @xi
= + ! i`
dp` @p` @W i
@xi @xi i @x
i
= xi` + ! `
@p` u=const @W i @W i
| {z }
Slutsky equation
i
@x @xi
= (xi` ! i` )
@p` u=const
| {z } @W i
net demand

where W i = p ! i . Whatever can happen dependent on the sign of

xi` ! i` (Hicks’s backward bending labor supply curve).

multiple equilibria
unique equilibrium
monotonicity of preferences
convex preferences
i i
6. Cobb-Douglass example: ui (xi1 ; xi2 ) = (xi1 ) 1 (xi2 ) 2 with i
1+
i
2 = 1 for
i = 1; 2. In Marvin’s class, you learned
i i
`W
xi` (p; W i ) = :
p`
Thus, it is easy to see
@xi` i
`W
i
@xi`
= < 0 and = 0:
@p` (p` )2 @p`0

However, in general equilibrium model, W i = W i (p). So variety of

things can happen:

Consider ! i = (! i1 ; 0): say, ` = 1 is leisure. Then, W i (p) = p1 ! i1

p i !i
and xi1 (p; W i ) = i1 ! i1 and xi2 (p; W i ) = 1 p12 1 . This implies a
perfectly inelastic leisure demand (labor supply) function.

dxi1 dxi2
= 0 and > 0:
dp1 dp1

18
 i (p !i )
However, in general, what we have is xi` (p; W i ) = `
p`
. Thus,

@xi` i
` (p` !i) i i
`!`
= +
@p` (p` )2 p`

and
P i
P
@xi` i
`
i
`0 2L p`0 ! `0
i i
`!` ` `0 6=` p`0 ! i`0 @xi`0 i i
`0 ! `
= + = 0 and = 0:
@p` (p` )2 p` (p` )2 @p` p`

Plot excess demand in the Edgeworth box. What do you …nd?

If two consumers with i = 1; 2, then market equilibrium conditions
for goods 1 and 2 are
1
1 (p !1) 2
1 (p !2)
+ = ! 11 + ! 21 ;
p1 p1
1
2 (p !1) 2
2 (p !2)
+ = ! 12 + ! 22 ;
p2 p2
or the excess demand for goods 1 and 2 are zero:
1
1 (p !1) 2
(p ! 2 )
z1 (p) = ! 11 + 1 ! 21 = 0;
p1 p1
| {z } | {z }
1’s excess demand 2’s excess demand

1
2 (p !1) 2
2 (p !2)
z2 (p) = ! 12 + ! 22 = 0:
p2 p2
Let us calculate the value of excess demand:

p1 z1 (p) + p2 z2 (p)
= 11 (p ! 1 ) p1 ! 11 + 21 (p ! 2 ) p1 ! 21 + 1
2 (p !1) p2 ! 12 + 2
2 (p !2) p2 ! 22
= p ! 1 + p ! 2 p ! 1 p ! 2 = 0:

This is the Walras law. What it means is that if one market

is cleared (say good 1), then the other market is automatically
cleared. This law generalizes to L good economy. If L 1 markets
are cleared then the last market is cleared (see the Edgeworth
box).

19
 We can focus on one market, and demand function is homogeneous
of degree zero (even if all prices are doubled, demand does not
change — since the budget constraint is not a¤ected by that).
Let’s say p2 = 1, and focus on market 1.
1
1 (p !1) 2
1 (p !2)
z1 (p) = ! 11 + ! 21
p1 p1
1 1 1 2 2
1 (p1 ! 1 + ! 2 ) 1 (p1 ! 1 + ! 22 )
= ! 11 + ! 21
p1 p1
= 0

As is easily seen, equilibrium p1 is uniquely determined for all

values of 11 ; 21 2 (0; 1). That is, in a Cobb-Douglass economy,
equilibrium is unique.

7. In general, market equilibrium is not necessarily unique even if util-

ity functions are strictly monotonic (see pictorial examples). So, you
cannot expect uniqueness of equilibrium.

Utility function ui satis…es strict monotonicity if and only if for

all xi 2 X i and all x~i 2 X i with x~i` xi` for all ` = 1; :::; L with
xi > xi ), ui (~
at least one strict inequality (~ xi ) > ui (xi ) holds.

8. A Pareto e¢ cient allocation is a feasible allocation x such that there

is no other feasible allocation x~ with (i) ui (~
xi ) ui (xi ) for all i 2 I,
and (ii) ui (~
xi ) > ui (xi ) for some i 2 I.

This e¢ ciency is perhaps not very controversial. Without hurting

any consumer, there is no possibility of improving a consumer.
That is, this condition can be regarded as the minimum require-
ment for e¢ ciency.
Note that consumers’endowments do not play any role here.

9. The …rst welfare theorem in the Edgeworth box.

10. Utility function ui satis…es local nonsatiation if and only if for all
xi 2 X i and all > 0, there exists x~i 2 X i with jxi x~i j < such that
ui (~
xi ) > ui (xi ).

20
11. Utility function ui satis…es (weak) monotonicity if and only if for
all xi 2 X i and all x~i 2 X i with x~ii > xii for all i = 1; :::; L (~
xi xi ),
ui (~
xi ) > ui (xi ) holds.

Note that monotonicity implies local nonsatiation.

With local nonsatiation, market equilibrium prices for some goods
can be negative (bads). The following theorem allows negative
prices.

Theorem 1 Suppose that for all i 2 I, ui satis…es local nonsatiation.

Then, in every market equilibrium (p; x), the realized allocation x is
Pareto e¢ cient.

Proof. Suppose not. Then, there exists another feasible allocation

x~ such that (i) ui (~xi ) ui (xi ) for all i 2 I, and (ii) ui (~
xi ) > ui (xi )
for Psome i 2 I.PBy the de…nition of market equilibrium, we have
p ~i > p
i2I x
i
i2I x since (i) implies p x ~i p xi and (ii) implies
i i
p x~ > p x . However, by monotonicity P (or local
P nonsatiation), p xi =
p ! i holds for all i 2 I.PThus, p Pi2I x~i > p i2I ! i . By the de…nition
of feasibility, we have i2I x~i = i2I ! i . This is a contradiction.

12. Note that the proof of the …rst welfare theorem basically requires noth-
ing but local nonsatiation. It does not require:

quasi-concave utility (draw pictures in the Edgeworth box)

convex consumption set (location choice is allowed)

13. The …rst welfare theorem is applicable even if

production is included
time dimension is included
uncertainty is included
multiple locations are included

14. But also remember that it does not talk about existence of equilibrium.
The statement of the theorem is "If there is an equilibrium, it must be
Pareto-e¢ ceint."

21
15. Moreover, there are hidden assumptions:

there is a market for each commodity (in production economy and

with uncertainty it becomes more serious)
all agents are price takers
no externalities
there is no distortion
property rights (endowment here) are well de…ned (in production
economy it becomes more serious: labor managed …rms)
summing the values of consumption plans makes sense (cf. OLG
model, for example)
You need to check if these assumptions are satis…ed, since in var-
ious applied models, these assumptions can be violated in subtle
ways.

16. The Pareto (e¢ cient) set in the Edgeworth economy is found in the
following manner.

max
1 1
u1 (x11 ; x12 ) s:t: u2 (x21 ; x22 ) = u2
x1 ;x2

x11 + x21 = ! 1
x12 + x22 = ! 2

where ! ` is the aggregate endowment of good `. (Under convex pref-

erences) the necessary and su¢ cient conditions for Pareto e¢ ciency
are
u11 u21
=
u12 u22
and

x11 + x21 = ! 1
x12 + x22 = ! 2

From these conditions, the …rst welfare theorem is trivial, because mar-
ket equilibrium requires
u11 u21 p1
1
= 2
=
u2 u2 p2

22
 and feasibility constraints

x11 + x21 = ! 1
x12 + x22 = ! 2 :

Draw Edgeworth boxes describing them.

17. Under Cobb-Douglass utilities, we have

u11 1 1
1 x2
1 1
1 x2
2 2
1 x2
2 2
1 x2 u21
= = = = =
u12 1 1
2 x1 (1 1 1
1 )x1 (1 2 2
1 )x1
2 2
2 x1 u22

Substituting

x21 = ! 1 x11
x22 = ! 2 x12

into the MRS-equating condition, we can write x12 with x11 only.

x11 (1 1
1 ) 2
1
! 1 x11 1
(1 2
)
x12 =
1 1
!2:
x11 (1 1
1 ) 2
1
1+ ! 1 x11 1
1 (1 2
1 )

This characterization of Pareto e¢ cient allocation has no insight, but

you may try to set 11 = 21 . What do you …nd?

18. Leontief preference ui (xi1 ; xi2 ) = minfxi1 ; xi2 g and Edgeworth boxes.
Pareto set, market equilibrium.

19. Linear preference ui (xi1 ; xi2 ) = ai1 xi1 +ai2 xi2 and Edgeworth boxes. Pareto
set, market equilibrium.

3 Existence of Equilibrium
1. We will talk a bit about the existence of equilibrium. Before that let
us de…ne quasi-concave utility. Utility function ui is quasi-concave, if
and only if for all xi 2 X i , the better set B(xi ) fxi0 2 X i : ui (xi0 )
ui (xi )g is convex.

23
 Theorem 2 Suppose that ui is continuous and quasi-concave, and sat-
is…es weak monotonicity, and that X i RL+ is closed and convex,
! i 2 int(X i ) hold for all i 2 I. Then, there exists a market equi-
librium.

2. Nonexistence of equilibrium in two ways: (i) nonconvexity and (ii) no

cheaper point (draw Edgeworth boxes).

3. We will work on price simplex L so that all prices are nonnegative:

we assume weak monotonicity to be compatible.

4. We will use Kakutani’s P …xed point P

theorem to …nd market price p =
i i
(p1 ; :::; pL ) that satis…es
P i2Ii x P i2Ii ! : i.e., the aggregated excess
=
demand vector z = i2I x i2I ! = 0. So, we will consider a
mapping that maps each (p; z) to the space of (p; z) itself.

5. It is easy to imagine how to construct an excess demand mapping.

Aggregating each consumer’s demand correspondence up, and subtract
the total endowment. That gives us an excess demand correspondence.
Thus, in order to apply Kakutani’s …xed point theorem, the important
thing is individual demand correspondence satis…es all the conditions
(nonempty-valued, convex-valued, and upper hemicontinuous).

6. We will work on excess demand correspondence generated from indi-

vidual demand correspondences xi : L RL : z : L RL such that
for all p 2 L X
z(p) = xi (p; p ! i ) ! i
i2I

7. Thus, it is important to know if individual demand correspondence

xi : L RL is nonempty-valued, convex-valued, and upper hemicon-
tinuous. For this, we use the Weierstrass and Berge theorems (math
one-minute lessons: conveniently adjusted Berge’s theorem I). Since
utility function is continuous, the key is if budget correspondences are
compact-valued, convex-valued, and continuous.

8. Compact-valuedness of budget correspondences will be achieved by

truncating each consumer’s consumption set. What about continuity
of budget correspondences? It appears that the assumption is satis…ed.
However, there is a pitfall (even Arrow and Debreu made mistake here).

24
 This is about nonexistence of cheaper point. If there is no cheaper
point, the budget correspondence may fail to satisfy continuity. This is
the reason that type (ii) failure occurred in 2. With the cheaper point
assumption, we can show that individual demand is nonempty-valued
and upper hemicontinuous. Now, type (i) failure comes to play if pref-
erences are not convex (utility function is not quasi-concave). With
quasi-concavity, all conditions of Kakutani’s theorem are met on the
excess demand mapping.

9. Now, we need to think about how price is determined. Gale and Nikaido
had a brilliant idea to introduce a hypothetical auctioneer who deter-
mines market price from excess demand vector. In order to explain
their idea, let us state the Walras law.

10. [The Walras law] p z(p) 0 for all p 2 L .

Proof. Focus on consumer i 2 I. Her maximization problem is

max
i i
ui (xi ) s:t: p xi p !i
x 2X

That is, the value of consumer i’s optimal consumption bundle is less
than or equal to the one of her endowment. Summing them up, we
have X X
p xi (p) p !i
i2I i2I

This proves the Walras law.

The Walras law has two versions: strong and weak ones. The
above is weak one (inequality). If l.n.s. is assumed, then the
above inequality becomes equality. This is the strong version. The
strong version has an important implication. Due to the Walras
law, one freedom is lost. As a result, if L 1 markets clear, then
the last Lth market also clears.

11. This simple observation (the Walras law) plays an important role in
…nding the market equilibrium. Let us introduce a hypothetical player
— an auctioneer — who sets prices in order to maximize the value of
excess demand z.
(z) arg max p z
p2 L

25
 This arti…cial mapping together with the Walras law proves the exis-
tence of market equilibrium. Note that mapping is nonempty-valued
and u.h.c. since function p z is continuous and set L is compact.
Moreover, function p z is quasi-concave thus is convex-valued, too.
(See math one-minute lectures.) The reason why we use such an auc-
tioneer is revealed in 16 below.

This mapping has special nature. For each z, ` (z) > 0 can
occur only for `s such that z` z`0 for all `0 = 1; :::; L. That is,
(z) almost always takes the form of (z) = (0; :::; 0; 1; 0; :::; 0).
However, if z = 0 holds, (z) = , and a price vector p that is
consistent with z(p) = 0 will be a …xed point.

12. [The …xed point mapping] Let Z RL be the set of possible excess
demand vectors. The …xed point mapping : L Z L
Z
L L
is the Cartesian product of mappings : Z and z : Z
( = z):
p p (z )
2 = ;
z z z(p )
or 0 1 0 1
p1 1 (z )
B .. C B .. C
B . C B . C
B C B C
B pL C B L (z ) C
B C B
2 C
B z1 C B z1 (p ) C
B . C B .. C
@ .. A @ . A
zL zL (p )

13. One thing to note is that we need to compactify Z RL ( L is

already compact). What is feasible individual consumption set in
the exchange
P economy? No one’s consumption of good ` should ex-
ceed i2I ! i` = ! ` . That is, for each consumer i, her potential con-
Q
sumption vector must satisfy xi = (xi1 ; :::; xiL ) 2 L`=1 [0; ! ` ]. At the
…xed point, feasibility needs to be satis…ed, it is harmless to compact-
ify consumerQi’s consumption set X i by setting an upperbound: i.e.,
X i = X i \ L`=1 [0; ! ` ] = fx 2 X i : x !g. More precisely, in or-
der toQavoid boundary problems, it is more convenient to let X ^i =
L
X i \ `=1 [0; ! ` + ] = fx 2 X i : x ! + g for some small > 0. (If

26
 under some p an element outside of X i is chosen, then such p cannot
be aP
part of any
P …xed point,
P so iwe can ignore such possibility.) Since
z = i2I z i = i2I ! i i2I x ,
( )
X X
Z= X^i !i
i2I i2I

^ i s are compact. Thus, Z is also

Since X i s are closed and convex, X
nonempty, compact and convex.
14. Consumer i’s budget set under p is now restricted by B i (p) = fxi 2 X^i :
p xi p ! i g. This compacti…cation together with ! i 2 int(X i ) allows
us to apply Weierstrass’s theorem and Berge’s maximum theorem (B i
is continuous (! i 2 int(X i )), and is compact-valued (B i (p) compact)).
Of course, under some p, consumer chooses some consumption bundle
that is not her real demand (due to compacti…cation). However, under
such p, if consumer chooses her optimal bundle freely, it can never be a
market equilibrium since it violates the feasibility constraint. So, this
compacti…cation is harmless. (Draw pictures.)
15. With compacti…cation, now consumer i’s demand correspondence i is
nonempty-valued (by continuity), u.h.c. (by continuity), and convex-
valued (by quasi-concavity). These properties are preserved under sum-
mation, so the excess demand correspondence z satis…es these proper-
ties. We have seen satis…es these properties. Thus, we are ready to
apply Kakutani’s …xed point theorem to : L Z L
Z, and
we …nd a …xed point (p ; z ) 2 (p ; z ). The rest of task is to show
that this …xed point is actually a market equilibrium. Since consumers
are optimizing under p , what needs to be done is to show z = 0.
16. [Gale-Nikaido’s lemma] In the …xed point (p ; z ) 2 (p ; z ), we
have z 0.
Proof. By the Walras law, p z(p) 0 for all p 2 L . Thus, p z 0
holds. Now, by the auctioneer’s choice, we have
0 p z p z
for all p 2 L . Pick `, and let p` = 1 and pk = 0 for all k 6= `. Then,
we have z` 0. This is true for any ` = 1; :::; L. This proves that
z 0. Q.E.D.

27
17. With weak monotonicity, z = 0 (budget is exhausted). This proves
the existence of market equilibrium.

18. Any idea why I did not say local nonsatiation?

19. Demand function needs to satisfy these conditions + compacti…cation

(from feasibility)

Weierstrass’s theorem (continuous function has a maximum in a

compact set) gives nonempty-valuedness.
Berge’s maximum theorem (continuous function and continuous
budget correspondence) gives upper hemicontinuity. Continuous
budget requires ! i 0.
quasi-concavity guarantees convex valuedness.

20. Quasi-concavity of utility functions, disconnected consumption sets are

sometimes dispensable by assuming a continuum of consumers (convex-
i…cation).

Discuss nonconvex preferences.

Discuss location choice (nonconvex consumption set)

21. **Measure theoretical GE ( looks complicated, but very intuitive)

Space of consumers (A; A; ): is measure of consumers ( (S) is

a population measure of set S A)
preferences u : A ! U (u[a](x) is person a’s utility function: u[a]
itself is a utility function)
endowment ! : A ! RL+
consumption plan f : A ! RL+ (f (a) is person a’s demand): this
can be an allocation
– demand correspondence ' : L A RL+ ('(p; a) is person
a’s demand under price p: '(p; ) is correspondence)
R
aggregate demand or mean demand A f d
R
– if correspondence A 'd is a collection of integrals of all se-
lections f : A RL+ with f (a) 2 '(p; a) for a 2 A a.e.

28
 L
– we want to …nd a selection f : A !
R R+ of demand
R correspon-
L
dence '(p; ) : A R+ such that A f d = A !d
[Liapunov’s theorem] If space of consumers is atomless (for all
S AR with (S) > 0, there exists T S with 0 < (T ) < (S)),
then A '(p; a)d RL+ is convex.
By Liapunov’s theorem, convex-valuedness of demand correspon-
dence is assured in an atomless economy.
Aumann (1966, Econometrica) In atomless economies, there exists
a market equilibrium if U is the space of continuous and strictly
monotonic utilities (with measurability restrictions). Later, Schmei-
dler weakened the restriction to local nonsatiation.
Nonconvex consumption set (location- or job-choice) is a bit harder
to deal with, but it is possible to apply the same method.

4 Miscellaneous Topics in General Equilib-

rium
This lecture talks about miscellaneous topics brie‡y (though with some for-
malism). More formal treatments, see Mas-Colell, Whinston and Green.

4.1 The Core

1. Let S I be a (nonempty) subgroup of consumers. A feasible al-
i i i
location
P i
P S isi a list (x )i2S such that x 2 X for all i 2 S and
for
i2S x = i2S ! .

2. A core allocation is a feasible allocation x such that for all (nonempty)

subgroup of consumers S I, there is no feasible allocation for S,
i i i
x ) ui (xi ) for all i 2 S, and (ii) ui (~
x )i2S , such that (i) u (~
(~ xi ) > ui (xi )
for some i 2 S. The set of all core allocations is called the core.

3. Draw a picture for the …rst welfare theorem. Show that the market
equilibrium is in the core.

4. The same proof applies to a stronger statement (prove it):

29
 Theorem 3 Suppose that for all i 2 I, ui satis…es local nonsatiation.
Then, in every market equilibrium (p; x), the realized allocation x is in
the core.

5. For this theorem, the same comments in 6 and 7 apply.

6. Replica economy (core convergence: Debreu and Scarf 1963 IER: simple
case in pictures). An N -replica of an exchange economy (I; (X i ; ! i ; ui )i2I )
is an exchange economy with each type i 2 I having N 2 consumers.
An allocation for N -replica economy is x = (x11 ; :::; x1N ; :::; xi1 ; :::; xiN ; ::::; xI1 ; :::; xIN ) 2
RL I N .

7. Assume strictly convex preference (strictly quasi-concave utility). Then,

0
if x is in the core of N -replica economy, then xin = xin for all n; n0 2
f1; :::; N g (equal-treatment property). From each type PiN2 I,intake
i 1
the worst-o¤ consumer (call her i1), and assign x = N n=1 x for
each type i 2 I. This allocation (xi )i2I is feasible for a coalition
f11; :::; i1; :::; I1g. Due to strictly convex preference, ui (xi ) > ui (xi1 )
for all i. A contradiction.

8. [The Debreu-Scarf theorem]. If (x1 ; :::; xI ) is in the core for all

replicas N = 1; 2; ::::, then x is a market equilibrium under strictly
convex preferences.

9. Proof by picture for the Edgeworth box economy (homework).

10. An idealized continuum economy (core equivalence: Aumann 1964

Econometrica). No convexity is required.

4.2 Uniqueness
1. First, I mention that it is fruitless to try to show the uniqueness
of Walras equilibrium if there are multiple types of consumers. Of
course, using utility functions in a restrictive functional form (say,
Cobb-Douglass), we can sometimes show uniqueness of equilibrium.
However, in general, we cannot hope such a well-behaved economy. An
example in an Edgeworth economy su¢ ces (see …gure). This and next
subsection mainly follow Varian’s text book.

30
2. Moreover, there is a theorem by Sonnenschein, Mantel and Debreu. It
in e¤ect says that any continuous function with the Walras law prop-
erty (p z(p) = 0 for all p) can be constructed by aggregating some
L consumers’ excess demand functions. That is, an aggregate excess
demand function can take any form, as a result, it is hopeless to achieve
uniqueness of equilibrium in a strict sense.

The Sonnenschein-Mantel-Debreu theorem (the SMD theorem)

sounds like a completely destructive result. However, one should
note that their construction of an economy from a continuous de-
mand with the Walras law is quite arti…cial, and the resulting
economy can assign very di¤erent endowments to di¤erent con-
sumers. If you take a view that the real economy has more homo-
geneous consumers in preferences and endowments, then the SMD
theorem does not mean much in a practical sense.

3. Still it may be good to know what sort of conditions are needed to

assure uniqueness of equilibrium. A popular condition is:
Gross Substitutes: A pair of commodities ` and `0 with ` 6= `0 are
@z`0 (p)
gross substitutes, if @p `
> 0 holds for all price vector p 2 RL++ . (Note
that in this and the next section, we do not use the price simplex as
the set of price vectors.)

Theorem 4 Assume that aggregate excess demand function satis…es

z` (p) > 0 for all p with p` = 0. If all commodities are gross substitutes
for all pair of commodities, then for all equilibrium price vectors p
and p0 there exists > 0 such that p0 = p .

Proof. Suppose that there are two equilibria with p0 6= p for any
p0
> 0. Since p 0 holds, we can de…ne m = max p ` 6= 0. By
`
homogeneity of degree zero, we have z(mp ) = 0. Thus, p0 mp
0 0
with p 6= mp and p` = mp` for at least one commodity `. Note
z` (mp ) = z` (p0 ) = 0. Increase price of commodity `0 with p0`0 < mp`0
one by one to mp`0 . By gross substitution, z` (mp ) > z` (p0 ) = 0 holds.
This is a contradiction.

4. More general approach. First consider two commodity situation. By

the Walras law, we only need to check the excess demand in one market.

31
 See …gures. Notice that if @z@p1 (p)
1
< 0 at all equilibrium price vectors p ,
then the equilibrium is unique. This argument can be extended to many
commodity case, though this theory requires sophisticated mathematics
(di¤erential topology).

Theorem 5 Assume that aggregate excess demand function satis…es

z` (p) > 0 for all p with p` = 0. If (L 1) (L 1) Jacobian matrix
0 1
@z1 (p) @z1 (p)
@p1 @pL 1
B .. ... .. C
Dz(p) = B . . C
@ A
@zL 1 (p) @zL 1 (p)
@p1 @pL 1

has positive determinant at all equilibria, then there is only one equi-
librium.

Remark 6 OK, looks nice, but what does it mean? Let us use Slutsky
equation:
!
X X
Dz(p) = Dp xi (p ; ui ) u=const + DW xi (p; p !) (xi (p) ! i )
i2I i2I

The …rst term is the (negative) Slutsky matrix, so it has positive de-
terminant. Thus, you can see that the potential issue is always income
e¤ects. As long as substitution e¤ect is dominant for all price vector,
there is a unique equilibrium. However, as I mentioned before, it is too
much to ask.

4.3 Stability
1. Again, the interest is limited. Late 50s, many researchers were trying to
show that there is a stable equilibrium in general. Unfortunately, Scarf
(1960 IER) presented a three-person-three-good counter example, so
there was no hope in a strict sense. But Scarf’s example is quite arti…-
cial, and three people have very di¤erent preferences and endowments.
Thus, his example does not mean that there is no stable equilibrium in
a practical sense.

32
2. Here, we consider a special price adjustment mechanism (that is gener-
alizable), and discuss stability of equilibrium. We will impose weak ax-
iom of revealed preference to market demand (this is actually nonsense
to extend rational behavior for individuals to the aggregated market
behavior, but it works well: equilibrium is essentially unique in this
case, too: see Mas-Colell et al.). Let us consider the following ad-
justment rule p_ = z(p). That is, the price of commodity with excess
demand supply) goes up (down). We need to impose the Walras law
p z(p) = 0 for all p. This system has a very convenient property.
Check the Euclidean norm of price vector change over time (I omit
square root, since it is irrelevant).
!
d X X
(p` (t))2 = 2p` (t)p_` (t) = 2p z(p) = 0
dt `2L `2L

That is, p stays on L-dimensional sphere (not simplex). Now, let us

recall weak axiom of revealed preference. If p z(p) p z(p ) then
p z(p) > p z(p ) holds. (If under p, z(p) is chosen over z(p ), then
under p , z(p) is not a¤ordable.) What are the implications here? Since
z(p ) = 0, 0 = p z(p) p z(p ) = 0 holds for all p (the …rst equality by
the Walras law; the second equality follows by the equilibrium condition
z(p ) = 0). That is, the presumption for the weak axiom is always
satis…ed. Thus, assuming the weak axiom implies p z(p) > 0 for all
p (again by the Walras law p z(p) > p z(p ) = 0 holds). (Did you
notice the reason why equilibrium is unique under the weak axiom?)

Theorem 7 Assume weak axiom of revealed preference. Assume that

the adjustment rule is given by p_` = z` (p) for all ` = 1; :::; L, and that if
p is the equilibrium price vector, then p z(p) > 0 holds for all p 6= p .
Then, all paths following the above price adjustment converges to p .

Proof. (sketch) We use Liapunov function: a function V : S ! R is

a Liapunov function if (1) V (x) reaches its P
minimum value at x , and
(2) V_ (x(t)) < 0 for x(t) 6= x . Let V (p) = `2L (p` p` )2 . Then, we

33
 have
dV (p) X
= 2 (p` p` ) p_` (t)
dt `2L
X
= 2 (p` p` ) z` (p)
`2L
= 2p z(p) < 0

Thus, the value is monotonically decreasing under weak axiom of re-

vealed preference.

5 Production Economy
1. Unfortunately, in public economics, almost always production economy.
A production economy is a list (I; J; (X i ; ! i ; ( ij )j2J ; ui )i2I ; (Y j )j2J ),
where J is the set of …rms, ij is the pro…t claim that consumer i owns
P
for …rm j ( i2I ij = 1 for all j 2 J), and Y j RL is the production
set available for …rm j.

2. Explain Y j . y j = (y1j ; :::; yLj ): input is negative, output is positive.

Production set Y j is nonempty, closed and convex. Free disposal
Y j + RL Y j , inaction 0 2 Yj . Draw CRS, DRS and IRS.
i
3. If CRS, js do not matter (in the equilibrium, zero pro…t).

4. A feasible allocation for a production economy is a list (x; y)

where (i) x = (x1 ; :::; xI ) is such that xi 2 X i for all i 2 I,P (ii) yi =
(y 1 ; :::; y J ) is such that y j 2 Y j for all j 2 J, and (iii) i2I x` =
P i
P j
i2I ! ` + j2J y` for all ` = 1; :::; L.

5. A market equilibrium for a production economy (p; x; y) is a list

of price p 2 L and a feasible allocation (x; y) such that
P
(a) for all i 2 I, (i) p xi p ! i + j2J ij p y j , and (ii) ui (xi ) ui (~
xi )
P
holds for all x~i 2 X i with p x~i p ! i + j2J ij p y j
(b) for all j 2 J and all y~j 2 Y j , p y j p y~j holds

34
6. Instead of market equilibrium, Mas-Colell’s book uses "quasi-equilibrium"
which replaces a:(ii) with P
xi ) > ui (xi ) implies p x~i p ! i + j2J ij p y j ".
"a:(ii0 ) ui (~
This condition is of course a weaker one than a:(ii), and a quasi-
equilibrium is not necessarily a market equilibrium (example). The rea-
son why this concept is introduced is to avoid the problem of "! i 2 X i
while ! i 2
= int(X i )." This is an important yet a technical issue, so I do
not go into details.

7. A feasible allocation (in production economy) (x; y) is Pareto e¢ -

cient if and only if there is no other feasible allocation (~
x; y~) such that
i i i i
u (x ) u (~ x ) holds for all i 2 I with at least one strict inequality for
some i0 2 I.

Note that …rms’ pro…ts etc. do not matter. Just consumers’

utilities.

8. Robinson Crusoe economy again.

9. De…ne a feasible allocation, a Pareto e¢ cient allocation, and state and

prove the …rst welfare theorem (homework). You can …nd them any-
where, but try by yourself …rst. (Hint: the …rst welfare theorem: con-
tradiction with pro…t maximization.)

Theorem 8 Suppose that for all i 2 I, ui satis…es local nonsatiation.

Then, in every market equilibrium (p; x; y), the realized allocation x is
Pareto e¢ cient.

Proof. Suppose not. Then, there exists another feasible allocation

x; y~) such that (i) ui (~
(~ xi ) ui (xi ) for all i 2 I, and (ii) ui (~
xi ) >
ui (xi ) forPsome i 2 I.PBy the de…nition of market equilibrium, we
have p ~i > p
i2I x
i
i2I x since (i) implies p x ~i p xi and (ii)
i i
implies p x~ > p x . However, by monotonicity (or local nonsatiation),

35
 P ij
p xi = p ! i + j2J p y j holds for all i 2 I. Thus, we have
X X
p x~i > p xi
i2I i2I
X XX
ij
= p !i + p yj
i2I i2I j2J
X X
= p !i + p yj :
i2I j2J
P P P
By the de…nition of feasibility, we have i2I x~i = i2I ! i + j2J y~j ,
P P P
thus p ~i = p
i2I x
i
i2I ! + p ~j . From the above inequality,
j2J y
we conclude
X X X X
p !i + p y~j > p !i + p yj ;
i2I j2J i2I j2J
P P
or p ~j > p
j2J y
j
j2J y . This contradicts with …rms’pro…t maxi-
mization (at least one …rm is not maximizing pro…t at y j 2 Y j . Hence,
there cannot be a feasible allocation that Pareto-dominates a market
equilibrium allocation.

10. Remarks on the core: the core may be empty if DRS. Moreover, it
can be empty if IRS (Scarf). Often, we assume CRS. If you wish, you
can de…ne the core, and prove that a market equilibrium is in the core
under CRS. You can prove the core equivalence under the CRS and
atomless consumer assumptions.

11. If CRS is assumed, then i does not matter, and no

P reason to have
j
multiple …rms. Aggregate production set is Y = j2J Y . For all
y 2 Y , and all t > 0, ty 2 Y (CRS).

12. The core under publicly available technology (production set) Y . Let
S I be a (nonempty) subgroup of consumers. A feasible allocation P
i i i i
for
P S is a list (x )i2S such that x 2 X for all i 2 S and i2S x 2
i
i2S ! + Y . As in the exchange economy, a core allocation is a
feasible allocation x such that for all (nonempty) subgroup of consumers
S I, there is no feasible allocation for S, (~ xi )i2S , such that (i) ui (~
xi )
ui (xi ) for all i 2 S, and (ii) ui (~xi ) > ui (xi ) for some i 2 S. The set of
all core allocations is called the core.

36
 13. Prove that market equilibrium is in the core under CRS technology.

14. Existence of equilibrium. This is trickier than in exchange economy,

but is manageable in the exactly the same way. Excess demand corre-
spondence involves supply correspondence by …rms, but the treatment
is parallel. There are two main di¤erences. First one is feasibility set.
Note that even if Y j s are well behaved, the feasible set may expand to
entire RL space without conditions. (Draw a picture. Basically …rm 1
uses …rm 2’s output while …rm 2 uses …rm 1’s output in production.
By combining these two activities, two …rm together may be able to
produce in…nite amount of both goods.) We need condition such as
"Y is convex and Y \ RL+ is bounded." With this condition, Gale and
Mas-Colell (1975, Journal of Mathematical Economics) prove bound-
edness of feasibility set. Second, we typically assume free disposal
which requires "for all j 2 J, Y j + RL Y j " and local nonsatiation.
With free disposal, market price never becomes negative, so there is no
loss in weakening weak monotonicity to local nonsatiation at no cost.1
The formal proof is omitted, but it is not hard if you understand the
proof of existence theorem in exchange economy (each …rm’s supply
correspondence satis…es all nice conditions after compacti…cation).

Theorem 9 Suppose that ui is continuous and quasi-concave and sat-

is…es local nonsatiation, X i RL+ is closed and convex, ! i 2 int(X i )
hold for all i 2 I. And suppose that Y j RL is nonempty, closed and
convex with f0g 2 Y j , and satis…es free disposal (Y j +RL Y j ) for all
j 2 J. Moreover, assume that Y \ RL+ is bounded. Then, there exists a
market equilibrium.

15. We de…ne a market equilibrium with transfers. A market equilib-

rium for a production economy (p; x; y) with transfers W 2 RI
is a list of price p 2 L and a feasible allocation (x; y) such that

(a) for all i 2 I, (i) pxi W i , and (ii) ui (xi ) ui (~

xi ) holds for all
i i i i
x~ 2 X with p~ x W
1
Allowing negative price involves a bit more work (price simplex cannot be used any
more). For this, see Bergstrom (1976 Journal of Mathematical Economics "How to discard
‘free disposal’at no cost?").

37
 (b) for all j 2 J and all y~j 2 Y j , py j p~y j holds
P i
P i
P j
(c) i2I W = p i2I ! + j2J p y

16. Now, the second welfare theorem.

Theorem 10 Let (x; y) be a Pareto-e¢ cient allocation. Suppose that

X i is nonempty and convex, and ui is continuous and quasi-concave,
and satis…es local nonsatiation for all i 2 I, and that xi 2 int(X i ) holds
for all i 2 I. Suppose that Y j RL is nonempty, closed and convex
with f0g 2 Y , and satis…es free disposal (Y j + RL Y j ) for all j 2 J.
j

Then, there exist p 2 L and (W i )i2I which make (p; x; y; W ) be a

market equilibrium with transfers.

17. The proof of the second welfare theorem is as follows:

(a) Let V i (xP i

xi 2 X i : ui (~
) = f~ xi ) > ui (xi )g for all i 2 I, and
V (x) = i2I V i (xi ). Since V i (xi ) is convex by quasi-concavity of
ui , V (x) is a convex set (math one minute lesson).
P
(b) Since Y j is convex, Y = j2J Y j is convex (the same as above).
P
(c) Since (x; y) is a Pareto e¢ cient allocation, V (x)\Y +f i2I ! i g =
;.
P (Otherwise, P there is a Pareto-improving allocation (~
P x; y~) with
i j i
~ = j2J y~ + i2I ! .)
i2I x
(d) By the separating hyperplane theorem (math lesson), there are
p 2 RL P and W 2 R such that p x WPfor all x 2 V (x), and
p (y + i2I ! i ) W for all y 2 Y + f i2I ! i g. W.l.o.g., set
W 0.
P j
P i
(e) Since Y is closed, p j2J y + i2I ! = W . Thus, y j 2 Y j is
pro…t maximizing production vector for all j 2 J.
P
(f) Similarly, there is a vector (W i )i2I such that (i) i2I W i = W ,
and (ii) p x~i W i for all x~i 2 V i (xi ). By local nonsatiation,
p xi = W i .
(g) Suppose that p x~i = W i holds for some x~i 2 V i (xi ). Since
xi 2 int(X i ), there is > 0 such that xi ( ; :::; ) 2 X i . Since X i
is convex and V i (xi ) is open, there is t 2 (0; 1) such that t(xi
( ; :::; ))+(1 t)~ xi 2 V i (xi ) with p (t(xi ( ; :::; )) + (1 t)~ xi ) <
W i . This is a contradiction with f.(ii). Thus, x~i 2 = V i (xi ) for all
i i
p x~ = W . Q.E.D.

38
18. Importance of convexity: quasi-concave utility, convex technology, and
convex consumption set. We can dispense convexity by assuming a
continuum of consumers with additional assumptions.

19. In the classical economics, this theorem justi…es market equilibrium op-
timistically: as long as a proper redistribution (by lump sum transfers)
is done, then any Pareto e¢ cient allocation is supportable. However,
it requires tremendous amount of information.

20. In public economics and international trade, one important implication

of the second welfare theorem is that a social welfare maximizing al-
location can be supported by a market equilibrium with transfers as
long as social welfare function is increasing in each consumer’s utility
— the Bergson=Samuelson social welfare function: W : RI ! R such
that W (u1 ; :::; ui ; :::; uI ) is strictly increasing in all arguments.
P A sim-
i i
pler social welfare function is utilitalian WP (u) = i2I u . If each
i i i
ui (xP
i
) is concave
P function, then maximizing i2I u (x ) with respect
i i j j
to i2I x = i2I ! and y 2 Y for all j 2 J is a well-behaved
problem (under appropriate assumptions: what are they?).

21. [First-Order conditions for Pareto e¢ ciency] First, we denote

production sets by functions. Consider F j : RL ! R such that Y j =
fy 2 RL : F j (y) 0g. We assume F j (0) 0 and F j is twice continu-
ously di¤erentiable with
!
@F j @F j @F j
j ; :::; j ; :::; j = F1j ; :::; F`j ; :::; FLj 0:
@y1 @y` @yL

That is, F j (y j ) > 0 means that y j is infeasible and F j (y j ) = 0 means

production possibility frontier of …rm j. Similarly, for all i 2 I, X i =
RL+ and ui is twice continuously di¤erentiable with ui (0) = 0 and

@ui @ui @ui

; :::; ; :::; = ui1 ; :::; ui` ; :::; uiL 0:
@xi1 @xi` @xiL

The Kuhn-Tucker problem is

max u1 (x1 )
(x;y)2RLI
+ RLI

39
 subject to
ui (xi ) ui for all i = 2; :::; I
P i
P i
P j
i2I x` i2I ! ` + j2J y` for all ` = 1; :::; L
F j (y j ) 0 for all j = 1; :::; J
The Lagrangian function is
L(x; y; ; ; )
!
X X X X X X
= i i i
(u (x ) ui ) + ` ! i` + y`j xi` j
F j (y j );
i2I `2L i2I j2J i2I j2J

1
where = 1 and u1 = 0. If interior solutions, we have for all i; i0 ; `; `0 ; j;
and j 0
0
i ui` ui` i0
M RS`` 0 =
i
= i0 = M RS``0
u `0 u `0
0
j F`j F`j j0
M RT S`` 0 = j = j 0 = M RT S`` 0
F`0 F`0
i ui` F`j j
M RS`` 0 = = j = M RT S``0 :
ui`0 F`0
22. Now, recall consumer’s optimization problem
X
i i
max
i
u (x ) s:t: p` xi` Wi
x
`2L

and …rm’s optimization problem

X
max
j
p` y`j s:t: F j (y j ) 0:
y
`2L

Assuming interior solutions, we have for all i; `; `0 , and j

i ui` p`
M RS`` 0 =
i
=
u `0 p `0
and
F`j p`
j
j =
M RT S`` 0 =
:
F`0 p `0
Clearly, the …rst order conditions coincide for both cases (heuristic
argument for the second welfare theorem).

40
6 G.E. Theory: Scope of Applications and
Market Failures
1. The …rst and second welfare theorems are strong supports for mar-
ket mechanism. As we will see, we can allow for uncertainty, and we
can introduce time dimension (macro economics). We can also allow
location/job choice as well (nonconvex consumption sets with a large
number of consumers). However, we need to keep limitations of the
…rst welfare theorem in our mind.

Location choice. Consider a life in NYC and one in Boston. Prac-

tically, a consumer needs to choose her location between two cities.
Having co¤ee in NYC is not the same as having co¤ee in Boston.
So, co¤ee in NYC and co¤ee in Boston are di¤erent commodities.
Leisure (labor) in NYC and leisure in Boston are di¤erent. Thus,
prices of commodities and wage rates can be di¤erent between two
cities. That is, the number of commodities will be 2 L (if two lo-
cations only). Some commodities like co¤ee cannot be transported
(practically), but others such as lobsters can be transported from
Boston to NYC. The prices need to be the same? Not necessarily.
Transportation of lobster from Boston to NYC is a production
activity that produces NYC lobster from Boston lobster and la-
bor (and others). In this way, we can treat consumers’ location
choice in the general equilibrium framework. The only di¢ culty is
nonconvex consumption set. (Note that New Haven co¤ee is not
a convex combination of NYC co¤ee and Boston co¤ee.) With
a continuum of consumers, the problem can be solved easily (a
consumer chooses her location by comparing her utility in NYC
and the one in Boston).
Job choice. Imagine a life working in a consulting company and
the one working in a college. Say, consulting company’s salary
is higher, but need to work for 70 hours or more (you are not
necessarily able to choose working hours unlike standard labor-
leisure choice problem), while college professor’s salary is lower
with summer vacation months. These two lives are very di¤erent,
and utility functions in these two jobs are perhaps very di¤erent.
Time. Imagine a consumer who lives forever (not necessarily)

41
facing expected price vectors (p(t))1 1
t=0 with interest rates (r(t))t=0 ,
L
where p(t) 2 R+ . Then, this consumer is actually facing price
p(t))1
vector (~ t=0 where

Y
t 1
1
p~(t) = p(t):
t0 =0
1 + r(t)

p(t))1
With price vector (~ 1
t=0 , consumer maximizes her utility U ((x(t))t=0 ).
So, besides in…nite dimension problem, everything is the same,
and we can talk about general equilibrium over time (although
I do not touch producer side: there is special structure of pro-
duction technology due to time. In many applications including
macroeconomics, we assume the following utility function:
X
1
t
U ((x(t))1
t=0 ) = u(x(t));
t=0

where 2 (0; 1) is her discount factor. Suppose that A(0) is her

asset (the value of her endowment), her asset in period t + 1 can
be described by

A(t + 1) = (1 + r(t)) (p(t)!(t) + A(t) C(t)) ;

where C(t) is the consumption expenditure this consumer spends

in period t. Thus, consumer’s optimization problem can be de-
scribed as
X
1
t
max1 V (p(t); C(t)) s:t: A(t+1) = (1+r(t)) (p(t)!(t) + A(t) C(t)) for all t =
(C(t))t=0
t=0

where V (p(t); C(t)) is her indirect utility function:

V (p; C) = max u(x) s:t: p x C:

x

You can use Roy’s identity to solve the problem. (See later sec-
tion: optimal taxation.) This is exactly the same as consumers’
behavior in the standard macroeconomics.

42
2. Although the statement and the proof of the …rst welfare theorem are
very simple, there are hidden assumptions such as complete market
(there is market for each commodity), property rights are well-de…ned,
there is no distortions such as taxes and monopoly power. There is
another subtle violation for the proof of the …rst welfare theorem due
to incomparable summations of values of consumption vectors among
all market participants (OLG model).

3. First, missing markets. There are many reasons for missing markets.

new products: If a producer developed a new product, how much

should producer produce? There is no market price for the com-
modity yet, so it is hard to imagine how the producer can be a
price taker.
externalities: In the proof of the …rst welfare theorem, it is essen-
tial for each consumer’s utility to depend only on her own con-
sumption vector (reread the proof). So, if there are externalities
in utility functions or production functions, then the …rst welfare
theorem would fail. In order to recover Pareto e¢ ciency, either
the government interventions (Pigouvian tax), or creating mar-
ket by adding property rights (Arrow). However, there are many
limitations.
public goods: Public goods are de…ned as commodities with no ri-
valry in consumption and consumption cannot be prevented (nonex-
0 0
clusion): ui (xi ; g) and ui (xi ; g) for consumers i and i0 . In this
kind of environment, the standard market would provide much less
public goods due to free-riding incentives by consumers. There are
remedies (Lindahl equilibrium in general equilibrium context; and
public good provision mechanisms in (mostly) partial equilibrium
context), but there are many issues.
asymmetric information: adverse selection, moral hazard
absence of complete asset market: We will discuss this later.

4. Overlapping Generation Model. You might have learned from Fabio

S. that under dynamic ine¢ ciency condition, the …rst welfare theorem
fails. This is caused by a subtle violation of the proof of the …rst welfare
theorem. I provide a heuristic argument. What is dynamic ine¢ ciency?

43
 The population growth rate exceeds the market interest rate. Recall
the …rst part of the proof:
"Suppose not. Then, there exists another feasible allocation x~ such
that (i) ui (~
xi ) ui (xi ) for all i 2 I, and (ii) ui (~
xi ) > ui (xi ) for some
i 2 I. By the de…nition of market equilibrium, we have
X X
p x~i > p xi
i2I i2I

since (i) implies p x~i p xi and (ii) implies p x~i > p xi ."
The problem is displayed inequality. With a higher population growth
rate than discounting (interest rates), the summations of both sides
blow up to in…nity (future weights become heavier and heavier). We
cannot compare in…nity with in…nity. This is the intuitive reason that
the …rst welfare theorem fails.

5. ill-de…ned property right: Imagine labor managed …rm. If you (a

worker) moves from a company to another company, your company
share is a¤ected by this move. In contrast, in Walras’s model, it does
not happen.

6. distortions (tax, monopoly power): with non-lump-sum taxes or monopoly

power, marginal rates of substitution and marginal rates of transforma-
tion are not equated. Then, it is obvious that the …rst welfare theorem
fails. Interesting properties of distorted economy is analyzed in the last
section.

7. asymmetric information. Adverse selection (hidden knowledge: a con-

sumer’s type is known by herself, while others cannot observe her type.
Ex. health insurance). Moral hazard (hidden action: an agent can
choose her action whithout being known by others. Her action a¤ects
the probability distribution of outcome. Ex: careless driving.)

7 Uncertainty: Complete and Incomplete Mar-

kets
1. Here, we only consider exchange economy. Introducing uncertainty to
general equilibrium is challenging conceptually more than technically.

44
 In order to describe uncertainty, we introduce state of the world.
Dependent on state of the world, endowments and/or preferences (and
technology in production economy) can change.

(a) natural disasters a¤ect consumers’endowments (typically this is

assumed)
(b) hot summer a¤ects consumers’preferences
(c) success of R&D a¤ects production technology

2. Let S = f1; :::; Sg be the set of states of the world. Consumer i has
state-dependent endowment

! i = (! is )s2S = (! i1s ; :::; ! i`s ; :::; ! iLs )s2S 2 X i

where consumer i’s consumption set is X i RLS . Consumer i cares

about her state-contingent commodity vector

xi = (xis )s2S = (xi1s ; :::; xi`s ; :::; xiLs )s2S 2 X i

Preferences over state-contingent commodity vectors can be described

by utility function U i : X i ! R. This can be general function, but
an easy example is expected Putility function with a probability vector
( s )s2S 2 S f i 2 RS+ : s2S is = 1g:
X
U i (xi ) = i i
s u (xs )
s2S

where ui (xis ) is Bernoulli utility function (or vNM utility function). If

probability vector is subjective and not common to all consumers, then
we can use ( is )s2S 2 S instead. Bernoulli utility function can be state
dependent (uis (xis )) as well. Note that concavity of ui (xis ) is associated
with risk aversion, and makes U i (xi ) a concave function in xi (thus
convex preferences in xi : upper contour set is convex in consumption
set).

3. Consider the following situation: before a state realizes, market opens.

For each state-commodity combination, there is a market. Purchasing
commodity `s by xi`s means that consumer i purchases the right of
obtaining commodity ` by xi`s if state s realizes. That is, consumer i

45
 can consume ! i`s + xi`s units of commodity ` if state s realizes. Arrow-
Debreu equilibrium assumes the existence of all such state-contingent
markets for all commodities. This market system is called a complete
market system.

4. Formally, a feasible allocation is a list x = (xi )i2I P = (x

1 I
Px ) 2i
; :::;
1 I i i i
X ::: X such that x 2 X for all i 2 I and i2I xs = i2I ! s
for all s = 1; :::; S, where xis = (xi1s ; :::; xiLs ). an Arrow-Debreu equi-
librium is a feasible allocation (x1 ; :::; xI ) 2 X 1 ::: X I and a
state-contingent price vector p = (p11 ; ::::; pLS ) 2 L S such that (i)
p xi p ! i for all i 2 I and (ii) for all i 2 I, U i (xi ) U i (~xi ) holds
i i i i
for all x~ 2 X with p x~ p !.

5. Note that the de…nition of Arrow-Debreu equilibrium is identical to

the one of market equilibrium if each state-contingent commodity is
regarded as a commodity. Thus, the …rst welfare theorem holds under
local nonsatiation, and the existence of equilibrium is also assured as
long as preferences are convex and the cheaper point assumption (! i 2
intX i ) is assumed. The second welfare theorem as well. So, although
Arrow-Debreu equilibrium is conceptually complicated, technically it
is no more complicated than market equilibrium.

6. Let us consider a simple two consumer two state example with one
physical commodity in each state (say money). You can regard this as
a large market with two types of consumers. Consumer i = 1; 2 has
endowments ! 1 = (! 11 ; ! 12 ) and ! 2 = (! 21 ; ! 22 ), Bernoulli utilities ui (xis )
that are common to all states s = 1; 2, and subjective probabilities
( i1 ; i2 ). Note that expected utility of consumer i is written as

U i (xi1 ; xi2 ) = i i i
1 u (x1 ) + i i i
2 u (x2 );

so at xi1 = xi2 = x (on the 45 degree line), we have

i @U i =@xi1 i
1
M RS21 = = :
@U i =@xi2 xi1 =xi2
i
2

Assuming that ui is concave function (risk aversion), indi¤erence curves

are convex to the origin. We can draw Edgeworth boxes for di¤erent
cases.

46
 (a) ! 1 = (1; 0), ! 2 = (0; 1), and 1 = 2 . The contract curve (the
Pareto set) is a straight 45 degree line.
(b) ! 1 = (1; 0), ! 2 = (0; 1), and 1 1
1= 2 < 2 2
1= 2. Then, Pareto set is
above the 45 degree line.
(c) ! 1 = (2; 0), ! 2 = (0; 1), and 1
= 2
. The Pareto set is between
the two 45 degree line.

7. So, the market equilibrium is again e¢ cient, but here there is a big as-
sumption: all L S markets exist. If S is a large number, it is a quite
demanding assumption. Do we really need so many markets to obtain
nice results? Arrow observed that if there is a contingent commodity,
then it may be su¢ cient to replicate the Arrow-Debreu equilibrium
when sequential trade is possible. The idea is as follows. There is a
commodity that is traded as state-contingent commodity in period 0
(before state realizes), and in each state of period 1, other commod-
ity markets open after the state-contingent commodity’s transfer has
been made. As long as consumers have rational expectations (state-
dependent perfect foresightedness), such an equilibrium replicates the
Arrow-Debreu equilibrium with only. This equilibrium concept is called
Radner’s rational expectation equilibrium.

8. Let us assume that commodity 1 is the contingent commodity. Con-

sumers face contingent commodity price vector q = (q1 ; :::; qS ) 2 RS
at t = 0 and have expectations (p1 ; :::; pS ) 2 RLS where ps 2 RL
is an expected spot price vector in period t = 1 with realized state
s 2 S. In period t = 0, consumer i chooses (net) trading plan z i =
(z1i ; :::; zSi ) 2 RS under price vector q as well as spot market consump-
tion plan xi = (xi1 ; :::; xiS ) 2 RLS + that may occur at period t = 1.
Consumer i’s optimization problem is
P i
i i i (i) s qs zs 0
max U (x1 ; :::; xS ) s:t:
(xi1 ;:::;xiS )2RLS
+
i
(ii) ps xs ps ! is + p1s zsi for all s
(z1i ;:::;zS
i )2RS

Note that zsi can take either sign. Actually, zsi < ! i1s can occur (i sells
commodity 1 in state s more than her endowment amount in state s).
This case is called short selling — if state s occurs, consumer i needs
to purchase commodity 1 to pay back after using up all commodity 1

47
 endowment.
A Radner (rational expectation) equilibrium is a list of contin-
gent commodity price vector q = (q1 ; :::; qS ) 2 RS , spot price vector
(p1 ; :::; pS ) 2 RLS , trading plan z i = (z1i ; :::; zSi ) 2 RS , and spot mar-
ket consumption plan xi = (xi1 ; :::; xiS ) 2 RLS + , such
i i
Pthat i(i) (z ; x )
solves
P the above
P maximization problem, and (ii) i2I z = 0 and
i i
i2I xs = i2I ! s for all s 2 S.
It is not hard to see that the set of Arrow-Debreu equilibrium alloca-
tions is equivalent to the one of Radner equilibrium allocations.

9. From (ii) of Radner equilibrium, pp1ss xis ps

p1s
! is zsi for all s. Sub-
stituting them into (i), we obtain
X ps ps
qs xi ! is
s
p1s s p1s
X
= s ps xis ps ! is 0;
s

where s = pq1ss 2 R. By setting p = (p1 ; :::; pS ) be such that ps = s ps

for all s, we can have Arrow-Debreu price vector. Converse follows the
same argument.

10. A sequential market model with contingent commodities and rational

expectations in 7 is a good exercise, but it is nothing like the real
economy. However, in the real economy, there are other devices to
transfer wealth from states to states — - assets and securities. Consider
the same two-period setup as before. A unit of asset produces returns
dependent on realized states. We considered real assets of which returns
take form of commodity 1 in 8 (depending states). We generalize the
notion of real asset. A unit of asset is a vector r = (r1 ; :::; rS ) 2 RS ,
where rs is the amount of commodity 1 a buyer of the asset can receive
in case state s realizes. Note that rs can be negative for some state s.

11. Suppose that there are K di¤erent assets. Each unit of asset k =
1; :::; K is a vector rk = (r1k ; :::; rsk ; :::; rSk ). A vector of (net) trades
of assets (portfolio) and a price vector at t = 0 are z i = (z1i ; :::; zK i
)2
K K
R and q = (q1 ; :::; qK ) 2 R , respectively. Consumer i’s optimization

48
 problem is
P i
i (i) k2K qk zk 0 P
max U (xi1 ; :::; xiS ) s:t:
(xi1 ;:::;xiS )2RLS
+
(ii) ps xs ps ! is + k2K p1s zki rsk for all s
i

(z1i ;:::;zS
i )2RS

That is, asset k is no longer state contingent commodity. In state s,

from portfolio z i , consumer i receives
0 1
z1i
B C X i
rs1 ; ; rsK @ ... A = zk rsk
i
zK k2K

In a matrix form (using a return matrix R below), consumer i’s

budget constraint is written as
8 LS
9
>
> 0x 2 R : for some1 portfolio
0 z 2 RK with
10 q z 10, and >
>
>
< i i >
=
p1 (x1 ! 1 ) r11 r1K z1i
Bi (p; q; R) = B .. C B .. . . .. C B .. C = Rz i
>
> @ . A @ . . . A@ . A >
>
>
: i i i
>
;
pS (xS ! S ) rS1 rSK zK

The equilibrium concept is exactly the same as before.

A Radner (rational expectation) equilibrium is a list of asset price
vector q = (q1 ; :::; qK ) 2 RK , spot price vector (p1 ; :::; pS ) 2 RLS , trad-
ing plan z i = (z1i ; :::; zKi
) 2 RK , and spot market consumption plan
xi = (xi1 ; :::; xiS ) 2 RLS
+ , such
i i
P thati (i) (z ; x )Psolvesi the P above max-
i
imization problem, and (ii) i2I z = 0 and i2I sx = i2I ! s for
all s 2 S.

12. Is the Radner equilibrium in an asset market equivalent to the Arrow-

Debreu equilibrium? It might appear to be the case. However, it is not
necessarily true. It has something to do with the number of available
assets and correlation of asset returns. Imagine the following example.
There is only one physical commodity with three states s = 1; 2; 3.
There are two consumers i = 1; 2 with endowments ! 1 = (! 11 ; ! 12 ; ! 13 )
and ! 2 = (! 21 ; ! 22 ; ! 23 ). In this case, we can construct an Edgeworth
box with three dimensions (literally Edgeworth "box"). Suppose that
there are only two assets, r1 = (0; 1; 0) and r2 = (0; 0; 1). That is,

49
 whatever z i = (z1i ; z2i ) is chosen, consumer i must consume ! i1 if state 1
realizes. This cannot be an Arrow-Debreu equilibrium in general. The
problem is that the space of possible trades is just ! i + f0g R2 , and
consumer i’s budget set is B i (q1 ; q2 ) = f(xi1 ; xi2 ; xi3 ) 2 R3+ : xi1 = ! i1 and
q2 xi2 + q3 xi3 q2 ! i2 + q3 ! i3 g. Thus, although the consumption space is
three dimensional, the trading space is two dimensional, and the budget
line is only one dimensional line. What do we learn from this exam-
ple? In order to replicate the Arrow-Debreu equilibrium, we need full
dimensions (S-dimensions) need to be spanned by asset return vectors
(complete asset market). If this to happen, the Radner equilib-
rium is again equivalent to the Arrow-Debreu equilibrium. Otherwise,
Radner equilibrium is not necessarily e¢ cient (incomplete market).

13. Note that asset return vectors themselves are not important. The im-
portant thing is the space spanned by asset return vectors. This fact
has an implication to option plans. For example, consider the case
where only one asset is available r1 = (4; 3; 2; 1). With unique as-
set, consumers cannot buy or sell this asset (cannot be traded with
only one asset). However, consider strike options with strike prices
3:5, 2:5, and 1:5. Then, these options create assets r2 = (0:5; 0; 0; 0),
r3 = (1:5; 0:5; 0; 0), and r4 = (2:5; 1:5; 0:5; 0). These four asset return
vectors span the entire space. This shows that option plans can im-
prove e¢ ciency of the market by creating complete market. Also note
that trading volume can be unbounded in asset market. if there are
limitations, the entire space may not be spanned.

8 The 2 2 Production Model

1. Here, we will go back to Marvin’s framework of production theory. In
this section, we use the following notations: commodities (consumption
goods) are described by superscripts, while factors are described by
subscripts.

2. Let us start with the following situation: There are J …rms, and each
…rm j produces commodity j directly from L factors. Prices of com-
modities are described by p = (p1 ; :::; pJ ), and factor prices are de-
scribed by w = (w1 ; :::; wL ). production technology is described by a

50
 concave production function
q j = f j (z j )
Each …rm’s optimization problem is
max
j
pj f j (z j ) w zj ;
z

which determines output q j and factor demand vector z j . Clearly, pro…t

maximizing f.o.c.’s are: for all ` = 1; :::; L,
@f j (z j )
pj = w`
@z`j
Now, imagine a small open country in which commodities are interna-
tionally traded while factors are not tradable. That is, the world price
vector p is given, but factor price vector w is determined in domestic
market equilibrium within the country. Assume that there is an aggre-
gate factor endowment vector z = (z1 ; :::; zL ). In equilibrium, the sum
of factor demand and factor endowment are equated.
X j
z` = z`
j2J

Thus, w = (w1 ; :::; wL ) is determined by the above L equations. Given

j j j j
w , each …rm j maximizes P p jf (z ) w z and the factor demand
vectors satisfying z` = j2J z` .
3. Now, consider the central planner’s problem:
X X j
max pj f j (z j ) s.t. z` = z`
z=(z 1 ;:::;z J )
j2J j2J

The Lagrangian is
!
X X
L= pj f j (z j ) w^ z`j z` ;
j2J j2J

where w^ = (w
^1 ; :::; w^L ) is Lagrangian multiplier vector that is the same
as shadow price vector of factors. The f.o.c.s are: for all ` = 1; :::; L,
@f j (z j )
pj = w^` ;
@z`j

51
 and X
z` = z`j :
j2J

This shows that the pro…t-maximizing central planner’s problem and a

factor market equilibrium achieve the same allocation.

4. Let’s go back to equilibrium allocation with cost functions (Marvin’s

stu¤). Each …rm’s optimization problem is

max
j
pj f j (z j ) w z j = max
j
pj q j C j (w; q j ) ;
z q

where C j (w; q j ) is …rm j’s cost function. Firm j’s pro…t maximization
condition is
@C j (w; q j )
pj = :
@q j
Now, general equilibrium touch. Assume that there is an aggregate
factor endowment vector z = (z1 ; :::; zL ). In the equilibrium, the sum
of factor demand and factor endowment are equated.
X X @C j (w; q j )
z` = z`j =
j2J j2J
@w`

(Remember Shephard’s lemma.)

5. Assuming that technologies exhibit constant returns to scale, we can

rewrite cost function as

C j (w; q j ) = q j cj (w);

where cj (w) = C j (w; 1). Further assume J = L = 2: that is, there

are two outputs from two factors (typically K and L). By Shephard’s
lemma, we have: for j = 1; 2,

@cj (w) @cj (w)

aj1 (w) = ; and aj2 (w) = ;
@w1 @w2

where (aj1 (w); aj2 (w)) is …rm j’s cost-minimizing input combination to
produce one unit of output. As you remember we can draw an iso-
quant curve in the space of (z1j ; z2j ), but we can draw a dual of that:

52
 iso-cost curve in the space of (w1 ; w2 ). Firm j’s iso-cost curve is de…ned
in the following manner:

(w1 ; w2 ) 2 R2+ : cj (w) = c :

Note that the set (w1 ; w2 ) 2 R2+ : cj (w) c is convex, since cost
function is concave in w. By Shephard’s lemma, vector (aj1 (w); aj2 (w)) is
orthogonal to iso-cost curve at each point on the iso-cost curve. Clearly,
as w1 =w2 increases, aj1 (w)=aj2 (w) decreases.

6. Now, we consider an e¢ cient factor allocation problem. Recall the

Pareto set in the Edgeworth box with size (z1 ; z2 ) and iso-quant maps.
The set of e¢ cient factor allocations is exactly the same as the Pareto
set. From this Pareto set, we can derive the production possibility
frontier. Pick a point from the Pareto set (z 1 ; z 2 ) = ((z11 ; z21 ); (z12 ; z22 )).
Then, …nd corresponding (q 1 ; q 2 ) by

(z11 ; z21 ) = q 1 (a11 ; a12 )

(z12 ; z22 ) = q 2 (a21 ; a22 );

where (a11 ; a12 ) and (a21 ; a22 ) are found from iso-quant curve for unit pro-
duction. (Note that in exchange economy, if utility functions are ho-
mothetic, the Pareto set and utility possibility frontier can be derived
in the same way.)

7. [De…nition] We say that production of commodity j is relatively more

factor 1-intensive than production of commodity j 0 i¤
0
aj1 (w) aj1 (w)
> j0
aj2 (w) a2 (w)
for all factor prices w = (w1 ; w2 ). If factors are labor and capital,
then we say "production of commodity j is more labor-intensive than
production of commodity j 0 .

8. Given constant-returns-to-scale technology and interior solution, the

factor market equilibrium conditions are

p1 = c1 (w1 ; w2 )
p2 = c2 (w1 ; w2 )

53
 Thus, equilibrium factor prices are determined as the intersection point
of iso-cost curves in (w1 ; w2 )-space. Once w = (w1 ; w2 ) is determined,
we can see how much each factor is used in production of each com-
modity by evaluating (a11 (w ); a12 (w )) and (a21 (w ); a22 (w )) (these are
orthogonal to iso-cost curves) at the intersection point, and write them
in the (factor market) Edgeworth box. The intersection of two rays in
the box is the factor market equilibrium point:
(z11 ; z21 ) = q 1 (a11 (w ); a12 (w ))
(z12 ; z22 ) = q 2 (a21 (w ); a22 (w ))
From here, we can see q = (q 1 ; q 2 ). Since q maximizes the value
of the country under p = (p1 ; p2 ), p is orthogonal to the production
possibility frontier at q .
9. Note that given constant-returns-to-scale technology and interior solu-
tion, factor price w = (w1 ; w2 ) is solely determined by p = (p1 ; p2 ) as
the intersection point of two iso-cost curves. Thus, total endowment of
factors in a country z = (z1 ; z2 ) does not play any role in determining
the factor prices. This has an important implication in the theory of in-
ternational trade. As long as technology exhibits CRS and equilibrium
is interior solution (no specialization occurs), countries with the same
technologies have the same factor prices no matter what factor endow-
ment vectors they have. This is called a factor-price equalization
theorem.

9 Introduction to the Theory of Second Best

1. With distortions, the …rst welfare theorem does not hold. Moreover, it
is known that an economy with distortions is even harder to deal with.
For example, if there is a distortionary tax or a monopoly pricing in a
market, it is not necessarily good to keep all other markets undistorted.
Introducing more distortions in other markets can be welfare-improving
in general — - this is called the theory of second best (Lipsey and Lan-
caster 1956). In this section, we will give a brief introduction to the
theory of second best.
2. In this section, we will use di¤erent notations for notational simplicity.
Now, superscript i denotes good i = 0; 1; :::; L, while subscript j denotes

54
 a partial derivative with respect to price of good i. Good 0 denotes
leisure consumption.
3. First consumer theory you learned in Marvin’s class. Suppose that
consumer price vector is q = (q 0 ; q 1 ; q 2 ) while producer price is p =
(p0 ; p1 ; p2 ). These price vector can be di¤erent by commodity tax or
monopoly pricing. Then, production possibility frontier and indi¤er-
ence curve has intersections. You can measure economic distortions by
using the idea of compensating and/or equivalent variations. The cost
of distortions measured by producer price p is called excess burdens
(waste of resources).
4. Consider a one-consumer economy with leisure and two commodities
(1 and 2), in which all commodities are produced by using labor only
at …xed coe¢ cients p1 > 0 and p2 > 0, respectively. Taking leisure as
the numeraire, the commodity prices are …xed at p = (1; p1 ; p2 ), and
the production set is a hyperplane p y = 0. Now, we consider the
situation where the government needs to raise a certain …xed amount
of revenue T > 0 (measured by labor). No more lump sum transfers,
and the tax revenues are used for useful purposes.
X
2
t` p` x` = T:
`=0

5. One big question is if we can impose a uniform tax on all goods t > 0. If
it can be done, a uniform taxation is not distortionary, since consumer’s
budget constraint is
X
2 X
2
` `
qx = (1 + t)p` x` = ! 0 ;
`=0 `=0

where q ` > 0 is consumer price of commodity ` = 0; 1; 2, and ! 0 > 0 is

leisure endowment. Rewriting the above, we have
X
2
1 t
p` x` = !0 = !0 !0:
`=0
1+t 1+t
t
Thus, by setting T = 1+t ! 0 , we can see the equivalence between a
uniform consumption tax and a lump sum tax.

55
6. However, note that ! 0 is leisure endowment. In order to impose tax
on all commodities, the government need to check how much has been
"consumed" (not bought). In particular, leisure consumption taxation
is hard to implement. Do you wonder if labor income tax works as
leisure consumption taxation? Not at all. What is the price of leisure?
It is an opportunity cost of leisure consumption — wage rate. If labor
income tax is charged, then an after tax wage rate is lower than the
market wage rate. This is a subsidy for leisure consumption. If uniform
commodity tax is charged, consumer should face higher wage rate q 0 =
(1 + t)p0 . Moreover, tax base is also di¤erent, which can be seen as
follows: Let x and x be gross and net consumption vectors. Labor
supply is (! 0 x0 ), and leisure consumption is x0 , where commodity
0 is leisure here, and x0 is the gross leisure consumption. Commodity
taxes are paid for market transaction normally. So, net consumption
x` ! ` should be the tax base. Let x` x` ! ` for all ` = 0; 1; 2.

7. With constant returns to scale technology, there will be no pro…t in

the equilibrium. Note that ! 1 = ! 2 = 0. Thus, consumer’s budget
constraint is
X2 X
2
q x= q ` x` = q`!` = q0!0;
i=0 i=0
or
X
2
q x= q ` x` = 0:
i=0

Utility function u(x) is also translated to u(x) = u(x + !). From this
budget constraint, it is now clear that even if uniform commodity tax
is imposed, the tax revenue will be simply zero.

X
2 X
2
t ` ` t X ` `
2
` `
tp x = qx = q x = 0:
`=0 `=0
1+t 1 + t `=0

That is, given that the government needs to raise revenue T , it has to
introduce tax distortions to the economy.

8. Now, we start the analysis by Corlett and Hague (1953). Corlett and
Hague (1953) considered the following problem. Suppose that there
is uniform commodity tax when leisure is untaxed (this is equivalent

56
 to wage income tax system since relative price only matters). The
question they ask is — Is it better to introduce additional distortion?
Should some commodity bear higher tax rate than other commodities?
The smallest number of goods needed is 3 including leisure.
9. From 4, we know that consumers’ utility maximization is described
by the dual problem (expenditure minimization), and the consumers’
budget is zero in net term. Thus, expenditure to obtain utility u is zero
in equilibrium:
E(q; u) = 0:
The government’s budget constraint is
X
2
t` q ` x` (q; u) = T:
`=0

The resource constraint is

p x(q; u) + T = 0:
Clearly, by subtracting the government’s budget constraint from the
consumer’s budget constraint, we can generate the resource constraint.
This is the Walras law. Thus, the equilibrium is described by two
equations:
E(q; u) = 0:
p x(q; u) + T = 0:

10. Now, we consider a tax reform. We change t1 , but doing so alone will
not satisfy the resource constraint. Thus, we need to adjust the other
commodity 2’s tax rate. Totally di¤erentiating the system of equations,
we have
Eu x2 p2 du x1 p1
= dt1 ;
p xu p x2 p2 dt2 p x1 p1
@x `
where xk = (x0k ; x1k ; x2k ) and x`k = @q k for all `; k = 0; 1; 2. The wel-

fare e¤ect of changing t by adjusting t2 in order to keep the resource

1
du N
constraint is described by dt 1 = D where

x1 p1 x2 p2 Eu x2 p2
N= and D = :
p x1 p1 p x2 p2 p xu p x2 p2

57
11. What is the sign of D? Consider the e¤ect of changing t2 adjusting tax
revenue T . This can be seen from the following comparative statics:
Eu 0 du x2 p2
= dt2
p xu 1 dT p x2 p2
Thus, we have
dT D
= :
dt 2 Eu
This implies that D < 0 is more likely (non-La¤er situation). Note
that La¤er situation occurs at the …rst intersection (the one closer to
the origin) of the o¤er curve and the production possibility frontier.
du
12. How about N ? Given that D is negative, N < 0 () dt1
> 0.

p x1 p1 p x2 p2
N = x1 p1 x2 p2 :
x1 p1 x2 p2
p x1 p1
The term x1 p1
describes how much additional resource is needed ( @p x(q(t);u)
@t1
)
1 1 1
in order to raise one additional unit of tax revenue ( @(t @tp1x ) ). Thus,

du p x2 p2 p x1 p1
> 0 () N < 0 () >
dt1 x2 p2 x1 p1
p x2 p x1
() 2
> ;
x x1
that is, if raising tax of commodity 1 is less wasteful than raising tax
on commodity 2, this tax reform is welfare-improving.
13. We can manipulate the formula a bit.
P2 ` `
p xk `=0 p xk
=
xk k
P2 x ` k
`=0 p x`
=
xk
X q ` xk
2
1
`
=
`=0
1 + t` xk
X
2
1 k
= `;
`=0
1 + t`

58
 where xk` = x`k (symmetry of substitution matrix of Hicksian demand),
k q ` xk`
` = xk is compensated demand cross elasticity. Note that we have

X
2 X
2
q ` xk 1 X ` k
2
k `
` = = k q x` = 0:
`=0 `=0
xk x `=0

p xk k
Using the above, we can rewrite xk
by eliminating own price e¤ect k:

p xk X2
1 k
k
= `
x `=0
1 + t`
X
2
1 1 X
2
k k
= ` `
`=0
1 + t` 1 + tk `=0
X
2
1 1 k
= `
`=0
1 + t` 1 + tk
X2
tk t` k
= `
`=0
1 + tk 1 + t`
X2
k ` k
= `;
`=0

` ` t 1 1
where = 1+t ` . The third line to the fourth line: note 1+t` 1+tk
=
tk t` tk t`
(1+t` )(1+tk )
= 1+tk 1+t `:

14. Suppose that t0 = 0 (this is just for simplicity), and t1 = t2 = t

originally. Thus, 1 = 2 = and 0 = 0. Then, the welfare e¤ect of
raising t1 by adjusting t2 is

p x2 p x1 X2 X
2
2 ` 2 1 ` 1
= ` `
x2 x1 `=0 `=0
X
2 X
2
` 2 ` 1
= ` `
`=0 `=0
2 1
= 0 0
2 1
= 0 0 :

59
 We have the famous Corlett-Hague result (actually, t0 = 0 is not at all
necessary: just for presentation).
15. Theorem (Corlett-Hague 1953). Suppose that t1 = t2 and t0 = 0, and
dT
dt2
> 0. Then, raising t1 by adjusting t2 to keep the revenue intact
is welfare-improving if and only if 20 > 10 (commodity 1 is relatively
more complementary with leisure than commodity 2).
16. How do we interpret the above theorem? Note that leisure consumption
is not taxable (labor subsidy does not generate revenue). Thus, leisure
consumption is distorted under uniform commodity taxation t1 = t2 .
Raising tax on commodities relatively more complementary with leisure
works as an indirect tax on leisure. (Note: commodity 1 does not have
to be a complement of leisure. Just demands that commodity 1 is rel-
atively more complementary with leisure than commodity 2.) This is
the principle in the theory of second-best. If there is an unremovable
distortion, then it is not necessarily bene…cial to remove other distor-
tions. Some of other distortions may improve e¢ ciency of economy.
17. Now, optimal commodity taxation. This derivation is due to Mirrlees
(1976 J.Pub.E.).
max u subject to E(q; u) = 0, and
u;q
p x(q; u) + T = 0:
Note that constraints are in inequalities:
E(q; u) 0;
p x(q; u) + T 0:
It is because a higher expenditure is more distortionary (bad for utility),
and more resource is good for utility. The Lagrangian of the above is:
L = u + E(q; u) (p x(q; u) + T ) :
The …rst order conditions are
1 + Eu p xu = 0;
and
x` p x` = x` + x` = 0 for all ` = 0; 1; :::; L.

60
 0 1
x00 x0L
B C
Letting = = and letting S = @ ... x`k ... A be the Slutsky
xL0 xLL
matrix, we obtain the optimal commodity tax rule.
S = x;
or
X
L
k `
xk = x` ; (*)
k=0
`
where ` = q ` p` (= t` q ` ) is speci…c tax rate for commodity ` (t` = q`
is called ad valorem tax rate). The formula (*) is so-called the Ramsey
rule.
18. The standard interpretation is based on "small" taxes. Suppose that
T is very small (thus, k s are very small). Consider a tax reform
schedule (s) = s , where s 2 [0; 1]: i.e., (0) and (1) are before-
and after-tax vectors, respectively. Assuming that the Slutsky matrix
S does not change when tax rates change, the change in x` , x` , can
be approximated by the formula
X
L
` k `
x = xk = x` :
k=0

Note that is common to all commodities and leisure. Thus, if tax

rates are low, the formula says that we should impose taxes so that
all commodity consumption is reduced proportionally (net leisure con-
sumption as well).
19. Consider three good case (leisure, commodities 1 and 2). For simplicity,
assume 1 = 0 (just normalization). Then the Ramsey rule is

1 1 @x1
x1 + 2 = x1 ;
@q2
2 2
1 x1 + 2 x2 = x2 :
` t` `
Rewriting the above in elasticities and = 1+t`
= p` + ` , we have
1 1 2 1
1 + 2 = ;

61
 1 2 2 2
1 + 2 = :
Thus,
1 1 1
1 2
2 2 2 = :
1 2

By Cramer’s rule, we have

1 1 2
2 2
2 = 1 2 1 2 2 1
1 2 2 1 1 1
1 2 2
2 + 1 + 0
= 1 2 1 2 2 1 1 :
1 2 2 1 1 + 2 + 0

P
The last transformation used `0 + `1 + `2 = 0 (follows from Lk=0 q k x`k =
0). This implies that 1 R 2 holds if and only if 20 R 10 . That is, as in
Corlett-Hague, commodity 1 should face a higher tax rate if and only
if commodity 1 is more complementary with leisure.

20. Now, suppose that any commodity demand P is independent of any other
commodity prices. Note that we have Lk=0 qk x`k = 0. Thus, if x`k = 0
for all k = 1; :::; L with k 6= `, we need to have q 0 x`0 + q ` x`` = 0 for all
` = 1; :::; L. Supposing 0 = 0, we have
0 10 1 0 1
x00 x01 x0L 0 x0
B x1 x1 0 C B 1 C B x1 C
B 0 1 CB C B C
B .. .. . . .. C B .. C = B .. C :
@ . . . . A@ . A @ . A
L L L
x0 0 xL xL

This implies that for all ` = 1; :::; L, we have

`
` x`
= = = :
q` q ` x`` `
`

This is the inverse elasticity rule that is known in partial equilibrium

analysis.

21. Note that this inverse elasticity rule is not contradictory with the
Corlett-Hague rule,
P although they appear to be totally di¤erent con-
ditions. Recall Lk=0 `k = 0 (zero homogeneity of demand). With the

62
 above independence assumption, we have the following in three goods
case (though it works for L + 1 goods case).
1 1
0 + 1 = 0
2 2
0 + 2 = 0
`
Since we have ` < 0 for all ` (Slutsky matrix), we have
1 2 1 2
0 < 0 , 1 < 2 :

Thus, with the independence assumption, these two rules are actually
identical.

22. Finally, let us derive the Ramsey rule using more traditional way (using
indirect utility function). The properties of indirect utility function
v(q; W ) are (W is wealth or income):

(a) continuous in q and W

(b) nonincreasing in p, nondecreasing in W
(c) quasi-convex in q
(d) homogeneous of degree zero in (q; W )

23. Roy’s identity:

@v(q;W )
` @q `
x~ (q; W ) = @v(q;W )
;
@I

where x~` (q; W ) is the Marshallian demand function of commodity `.

The proof is as follows. We have the following identity:

u = v(q; E(q; u)):

Di¤erentiate both sides with q ` , and let u = v(q; W ). Then, we have

@v(q; W ) @v(q; W ) @E(q; v(q; W ))

0= + :
@q ` @W @q `
@E(q;v(q;W ))
By Shephard’s lemma, we have @q `
= x` (q; v(q; W )) = x~` (q; W ),
the desired result.

63
24. Slutsky equation:
x~`k (q; W ) = x`k (q; u) xk x~`W (q; W )
The proof is as follows. We have the following identity:
x~` (q; E(q; u)) = x` (q; u):
Di¤erentiating both sides with q k , and using Shephard’s lemma, we
have Ek (q; u) = xk (q; u), and we have desired result.
25. The optimal commodity tax problem is
max v(q; 0) subject to p~
x+T 0:
q

The Lagrangean is
L(q; ) = v(q; 0) (p~
x(q; 0) + T ) :

26. The f.o.c. with respect to q ` is

X
L
v` (q; W ) pk x~k` = 0:
k=0

By Roy’s identity, we have

X
L
`
0 = x vW pk x~k`
k=0
XL
= x` vW pk xk` x` x~kW
k=0
!
X
L X
L
`
= x vW pk x~kW pk xk`
k=0 k=0
!
XL X
L
k k
= x` vW q x~kW pk q k xk`
k=0 k=0
!
X
L X
L
k k k k
= x` vW + x~W + x`
k=0 k=0
!
X
L X
L
k k k k
= x` vW x~W + x` :
k=0 k=0

64
 1
PL k k
Let =1 vW k=0 x~W . Then, we have

X
L
k k
x` = x` ;
k=0

or
S = x:
This formula is again the Ramsey rule.

65