Block I: Static Exchange Economies

Eduardo Dávila∗

October 30, 2024, 8:42 PM

Contents
1 The Economy 3
1.1 What is an Economy? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Describing an Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Exogenous Variables, Endogenous Variables, and Comparative Statics . . . . . . . . . . . . . 4

2 Edgeworth Box Economy: I = J = 2 4

2.1 Physical Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Planning Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Axiomatic (Nash) Bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Competitive Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Offer curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Excess demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.3 Walras’ Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Graphical Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Final Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Static Exchange Economies 16

3.1 Physical Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Planning Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Axiomatic (Nash) Bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Competitive Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.5 Application: Armington Model: I = J ≥ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Efficiency and Welfare 25

4.1 Planning Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.1 Lagrangian Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.2 Perturbation Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.3 Remarks on Pareto Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Welfare Assessments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
∗ eduardo.davila@yale.edu. Yale University.

1
 4.3 First Welfare Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Second Welfare Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.5 Application: Transfer Problem/Paradox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.6 Application: Optimal Taxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5 Microfounding Competition 43
5.1 Core Convergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Non-Cooperative Strategic Foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

6 Competitive Equilibrium 47
6.1 Existence and Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Excess Demand Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.3 Uniqueness and Multiplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.4 Convergence to Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.5 Comparative Statics of Equilibrium Allocations and Prices . . . . . . . . . . . . . . . . . . . . 52
6.6 Economies with Many Individuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.7 Final Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2
1 The Economy
1.1 What is an Economy?
The unit of analysis in general equilibrium is the economy. What is an economy? An economy is a model of
how economic activity — consumption, exchange, and production — takes place. When economic activity
takes place simultaneously at a single instant, we say that an economy is static. When economy activity
involves time or uncertainty, we say that an economic is dynamic or stochastic. We exclusively consider
static economies until we reach Block III, which deals with dynamic and stochastic economies.
Economies are populated by individuals, who consume goods (and services) and supply factors of
production. Production takes place by transforming factors and goods into other goods. To simplify the
exposition, I use the term goods to refer to both goods and services.1 Figure 1 illustrates this physical
circular flow of economic activity. It is common in elementary macroeconomics textbooks to include a set
of opposite arrows to represent the flow of income. Including those would only be justified after defining
notions such as prices, wages, or income.

Figure 1: Physical Flow of Economic Activity

1.2 Describing an Economy

Every economy can be described in terms of i) its physical structure, and ii) its economic structure. The
physical structure of an economy is given by

i) preferences over consumption and factor supplies for each individual in the economy,

ii) technologies that transform factors and goods into other goods at a point in time, over time, and across
states, and

iii) resource constraints for goods and factors, which ensure that good and factor supplies add up to their
uses.

The economic structure of an economy defines the remaining features that describe how economic activity
takes places. For instance, the economic structure of a static economy includes:

i) who has ownership over goods and factors,

ii) how an individual behaves given the behavior of others (e.g.: price-taking behavior with linear budget
constraints),
1 In
classical general equilibrium, it is common to refer to both goods (and services) and factors as commodities. This
terminology is somewhat arcane, so I will not use in this text.

3
 Figure 2: Describing an Economy

iii) how individuals interact (e.g.: equilibrium notion),

iv) how technologies are operated (e.g.: profit maximization by firms), etc.

In dynamic stochastic economies, the economic structure of the economy also includes, among possibly other
features:

i) the set of possible financial arrangements among individuals (e.g., are markets complete or incomplete),

ii) the ability or inability of individuals to keep promises (e.g., whether individuals can fully or partially
renege on their promises),

iii) the information structure (e.g., who knows what when and how learning takes place),

iv) restrictions on the ability to reallocate factors (e.g., search and matching technologies).

The reason to draw a sharp distinction between the physical and economic structures is that once we have
specified the former we will be able to characterize the set of Pareto efficient (or Pareto optimal) allocations
for a particular economy. Such allocations — often referred to as first-best allocations — provide the
benchmark against which we compare the economic outcomes generated by different economic structures.

1.3 Exogenous Variables, Endogenous Variables, and Comparative Statics

Variables in an economy — or in any other model — are either exogenous or endogenous.2 Exogenous
variables are primitives of the economy, and are predetermined. All parameters of a model are exogenous.
Endogenous variables are the variables that are are determined by the model.
It is important to be aware at all times of which variables are exogenous and which are endogenous. This
may be harder that it seems, since variables that are exogenous in some economies can be endogenous in
others. Also, variables treated as exogenous by some agents within the model can be endogenously determined
within the equilibrium of an economy. For instance, in competitive economies, individuals treat prices as
exogenous when choosing their demand, even though prices are endogenously determined in equilibrium.
Comparative statics study the response of endogenous variables when we change exogenous variables.
Many important results in economics are obtained through comparative statics. Comparative statics can be
used to generate counterfactual scenarios.

2 Edgeworth Box Economy: I = J = 2

The simplest and most foundational economy in general equilibrium is the Edgeworth box economy. The
Edgeworth box economy is a pure exchange economy (or an endowment economy), that is, an economy
2 At times, we may also consider exogenous and endogenous functions or functionals. The same logic applies to these.

4
without production. The only form of economic activity in this economy is the exchange and consumption
of goods. Edgeworth (1881), Pareto (1906), and Bowley (1924) are the classic references, while Humphrey
(1996) provides a readable history of the origins and uses of the Edgeworth box diagram.

2.1 Physical Structure

An Edgeworth box economy is populated by I = 2 individuals, who consume consume J = 2 goods. We
index individuals by i ∈ I = {1, 2}, and goods by j ∈ J = {1, 2}. In this economy, goods are not produced
and appear as endowments.
Individual i has preferences over the consumption of both goods given by the utility function:
 
(Preferences) V i = ui cij j∈J
, (1)

where cij ≥ 0 denotes the consumption of good j by individual i. Unless noted, we assume that preferences
are continuous, strictly convex, and strongly monotone.
The resource constraint for good j is given by
X
(Resource Constraints) cij = ȳ j , (2)
i

where ȳ j > 0 denotes the total endowment of good j in the economy. Typically, we also assign endowments
to each individual, so that
X
(Endowment Assignment) ȳ j = ȳ ij , (3)
i

where ȳ ij ≥ 0 denotes the endowment of good j that belongs to individual i. However, as we will see
shortly, this is not necessary to characterize the set of efficient allocations. The restriction to non-negative
consumption is justified by the fact that variables that appear in resource constraints need to be non-negative,
since they describe the physical assignment of resources to uses. Models in which consumption of a good
can be negative can be reformulated by suitably defining production technologies, as explained in Block II.

Edgeworth Box Economy: Physical Structure

Individual preferences for each individual and resource constraints for each good define the physical
structure of an Edgeworth box economy:

V 1 = u1 c11 , c12


Preferences Individual 1
V 2 = u2 c21 , c22


Preferences Individual 2
c11 + c21 = y 1 Resource Constraint Good 1
12 22 2
c +c =y Resource Constraint Good 2

An allocation is defined as a non-negative assignment of the consumption of each good to each individual.

5
We express it in vector form as

c̊ = c11 , c21 , c12 , c22 ,


(Allocation)


where c̊ ≥ 0.3 We also denote individual consumption allocations in vector form as c1 = c11 , c12 and

c2 = c21 , c22 . We say that an allocation is feasible when both resource constraints hold. Given these


definitions, we can compactly write preferences as V i = ui ci .
Similarly, we express endowments in vector form as

ȳ = ȳ 11 , ȳ 21 , ȳ 12 , ȳ 22 ,
˚

(Endowments)

where ˚
ȳ ≥ 0. We also denote individual endowments in vector form as ȳ 1 = ȳ 11 , ȳ 12 and ȳ 2 = ȳ 21 , ȳ 22 .
 

The autarky allocation is the one in which c̊ = ˚

ȳ. Before proceeding, I would like to make three remarks:
Remark 1. (Utilities vs. Preferences) Although we have specified individual preferences using utility

functions, we could have specified instead preference relations or choice rules over cij j∈J . See e.g. Mas-
Colell, Whinston and Green (1995) or Kreps (2013) for conditions (essentially, continuity of preferences)
under which preference orderings can be represented by a utility function.
Remark 2. (Resource Constraints with Equality) We could have written resource constraints with an
P ij j
inequality: i c ≤ ȳ , and then define non-wasteful allocations as those in which the resource constraints
hold with equality. Using the version with equality involves little loss of generality since one could always
introduce a free-disposal technology for each good — see Section 2 of the static production notes.
Remark 3. (Individuals vs. Types of Individuals) At times, we will take literally the assumption that there
are I = 2 individuals. However, when working with a competitive market structure, we should think instead
that there are not two individuals but two types of individuals, with a continuum of individuals of each of
the two types.
Remark 4. (Self-Interest Individuals) We have assume that individual i’s preferences are exclusively a
function of individual i’s consumption, that is, that individuals are self-interested. This is a pervasive
assumption in economics, although it can be easily relaxed, as we do in Block II.
Figure 3 shows the box diagram that gives Edgeworth box economies their name. The origin for individual
1 is in the southwest corner of the box, while the origin for individual 2 is in the northeast corner of
the box. Any point inside the box corresponds to a feasible allocation. Under standard assumptions on
preferences, individual 1 prefers allocations toward the northeast, while individual 2 prefers allocations
toward the southwest. Individual preferences can be represented via indifference curves, as illustrated in
Figure 4.

2.2 Planning Problem

Once we have defined the physical environment, we proceed to sequentially study different economic
structures that have clear predictions over which allocations emerge endogenously. First, we characterize
Pareto efficient allocations by solving planning problems. Second, we characterize Pareto efficient and
3 The need for denoting allocations in vector form by c̊ rather than c, and for the slightly different ordering of the elements

of c̊ and ci will become clear in Block II. The same applies to ˚ȳ.

6
 y2 Origin
y 21 c21 i=2

˚
ȳ
y 12 y 22

c̊
c12 c22

y1
Origin y 11 c11
i=1

Figure 3: Box Diagram

individually rational allocations by solving (axiomatic) bargaining problems. Finally, we define and
characterize competitive equilibria.
We introduce several definitions:

i) An allocation is Pareto efficient (or Pareto optimal) if there is no other feasible allocation such that
all individuals are (weakly) better off, with at least one individual being strictly better off.

ii) The Pareto set is the set of Pareto efficient allocations.

iii) An allocation is individually rational for individual i if that individual prefers it to his/her endowment.

iv) The contract curve is the set of allocations that are Pareto efficient and individually rational for all
individuals.4

The left panel of Figure 4 shows the Pareto set and the contract curve in the box diagram for an economy
with Cobb-Douglas preferences. For these preferences, all Pareto efficient allocations are interior, so the
Pareto set is fully characterized by the set of allocations in which indifference curves are tangent. Pareto
efficient allocations in which indifference curves are not tangent can only occur at the boundary of the
Edgeworth box. This may occur, for instance, when both goods are perfect substitutes for at least one of
the individuals.
We can formally characterize the Pareto set by solving the planning problem (at times also called Pareto
problem) for different values of the Pareto weight α. Note that all the “planners” we consider in the text are
always benevolent, in the sense that they seek to maximize the well-being of the individuals in the economy.
It is important to highlight that solving planning problems in practice is incredibly challenging, in particular
because the informational requirements may be enormous — see the discussion in Block II on the Socialist
Calculation Debate.
4 Someauthors use the term contract curve to refer to the set of Pareto efficiency allocations. It is useful to draw a distinction
between the two.

7
 3.5
3
3
2.5
2.5
2
2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

Figure 4: Pareto Set, Contract Curve, and Pareto Frontier

Note: This figure illustrates the Pareto set, contract curve, and utility possibility set, and Pareto frontier for an economy

αi
1−αi 1 3
with Cobb-Douglas preferences given by V i = ci1 ci2 , where α1 = 2
, α2 = 5
, y 11 = 0.5, y 12 = 3, y 21 = 2.5, and
y 22 = 0.5.

Edgeworth Box Economy: Planning Problem

The Pareto set corresponds to the set of allocations c̊ = c11 , c12 , c21 , c22 characterized by solving
the planning problem:

max αu1 c11 , c12 + (1 − α) u2 c21 , c22 ,

for α ∈ [0, 1] ,
c̊

subject to

c11 + c21 = ȳ 1 Resource Constraint Good 1

c12 + c22 = y 2 Resource Constraint Good 2

The solution to the planning problem is characterized — in addition to the resource constraints — by
the following two conditions:
∂u1 ∂u1
∂c11 ∂c12 1−α
∂u2
= ∂u2
= ,
∂c21 ∂c22
α

where the first equality ensures that indifference curves are tangent — as we will see, this condition ensures
exchange efficiency — and the second equality simply selects a particular Pareto efficient allocation depending
on the weight assigned to each individual. By varying the Pareto weight α between 0 and 1 we can trace the
Pareto set in the box diagram.
Note that the planning problem can also be formulated as solving

max u1 c11 , c12 subject to u2 c21 , c22 ≥ ū

c̊

8
and resource constraints. In this case, varying the level of ū is equivalent to varying α, exploiting the duality
between constraints and objectives.
Given preferences and resource constraints, allocations in the Pareto frontier have exhausted all gains
from trade. Hence, from a positive perspective, it seems natural to argue that these allocations are good
candidates to emerge endogenously once individuals can exchange goods and communicate with each other.


Pareto Frontier. Each allocation in the Edgeworth box is associated with a pair of utilities V = V 1 , V 2 .
It is thus natural to represent the possible combinations of individual utilities associated with the set of
feasible allocations: this is the utility possibility set, which we define here along with the Pareto frontier

i) The utility possibility set corresponds to the set of attainable utility levels in an economy.

ii) The Pareto frontier (or utility possibility frontier) corresponds to the set individual utilities associated
with allocations in the Pareto set.

The right panel of Figure 4 illustrates the utility possibility set and the Pareto frontier. It also shows
the subset of the Pareto frontier that corresponds to allocations in the contract curve
 and the
 utility
 1 2 i i ij
levels associated with the endowment allocations: V̄ = V̄ , V̄ , where V̄ = u ȳ j∈J . Since
individual utilities are only defined up to increasing transformations, there are many preference-preserving
representations of the utility possibility and Pareto sets. If the utility functions are concave (and it is possible
to find a concave utility representation under minimal assumptions), then the utility possibility set is convex
— see 16.E of Mas-Colell, Whinston and Green (1995).

2.3 Axiomatic (Nash) Bargaining

It should be evident that if individuals can unilaterally decide to consume their endowment, they will only
choose allocations that are individually rational. However, solutions to the Pareto problem need not be
individually rational. In particular, when α → 0 or α → 1, the planning problem selects allocations that
make one of the individuals worse off relative to consuming their endowment.
The set of efficient allocations that are also individual rational (i.e., the contract curve) can be
characterized by solving a Nash bargaining problem. Nash bargaining is an axiomatic bargaining procedure
that selects allocations that satisfy a set of reasonable properties or axioms.

Edgeworth Box Economy: Nash Bargaining Solution

The contract curve corresponds to the set of allocations c̊ = c11 , c12 , c21 , c22 characterized by solving
the Nash bargaining problem:

α 2 21 22


1−α
max u1 c11 , c12 − u1 ȳ 11 , ȳ 12 − u2 ȳ 21 , ȳ 22


u c ,c , where α ∈ [0, 1] ,
c̊

subject to

c11 + c21 = ȳ 11 + ȳ 21 Resource Constraint Good 1

c12 + c22 = ȳ 12 + ȳ 22 Resource Constraint Good 2

9


By subtracting ui ȳ i1 , ȳ i2 , the bargaining solution accounts for the each individual’s threat point, which
involves consuming their endowment. The Nash bargaining solution satisfies several reasonable properties.
In particular, it selects allocations that are i) Pareto efficient, ii) individually rational, and iii) satisfy
independence of irrelevant alternatives. There are other axiomatic solutions also with desirable properties,
for instance, the Kalai-Smorodinski solution (Kalai and Smorodinsky, 1975). Muthoo (1999) provides a
comprehensive treatment of axiomatic and strategic bargaining — see also Section 22.E of Mas-Colell,
Whinston and Green (1995).
Given preferences and resource constraints, allocations in the contract curve frontier have exhausted all
gains from trade and are robust to the threat that an individual could decide to separate from the other and
consume his endowment. Hence, from a positive perspective, it seems natural to argue that these allocations
are good candidates to emerge endogenously once individuals can exchange goods and communicate with
each other, and are able walk away and consume their endowment in isolation.

2.4 Competitive Equilibrium

After having first characterized the set of allocations that leave no scope for Pareto improvements, and once
we have argued that only a subset of those satisfy individual rationality, we are still left with many allocations
in the contract curve. We now introduce an economic structure based on competitive, price-taking behavior
that has sharp predictions for which allocations are compatible with such behavior: this is the competitive
or Walrasian model, after Walras (1874). Note that, a priori, it should not be obvious why assuming that
individuals take prices as given and face linear budget constraints is a desirable economic structure or a good
description of individual behavior. We will have to wait until Section 4 to study the normative properties
of this economic structure, showing that it yields desirable outcomes, and to Section 5, to characterize the
conditions under which we expect this type of economic structure to emerge in practice.
In the Walrasian model, individuals choose consumption to maximize their utility subject to a linear
budget constraint, taking prices as given. Formally, individual i chooses consumption ci to maximize

ui ci1 , ci2 ,


max
ic

subject to a linear budget constraint given by

p1 ci1 + p2 ci2 = p1 ȳ i1 + p2 ȳ i2 , (4)

where pj ≥ 0, with p = p1 , p2 . The solution to this problem yields a (Marshallian) demand function
of the form studied in classical demand theory — see e.g. Deaton and Muellbauer (1980) or Chapter 3 of
Mas-Colell, Whinston and Green (1995).5 We compactly denote individual i’s demand as a function of prices
and individual endowments by6

ci p, ȳ i .


(Marshallian Demand)

5 We could have alternatively formulated individual budgets sets as Bi (p) = ci1 , ci2 ≥ 0 : p1 ci1 + p2 ci2 ≤ p1 ȳ i1 + p2 ȳ i2 .
Local nonsatiation ensures that budget constraints hold with equality.

6 We could have alternatively define individual demands as ci p, p · ȳ i , highlighting the fact that individual i’s income is a

function of the value of the market value of endowments.

10
The idea behind a competitive equilibrium is simple. Each individual chooses how much to consume taking
prices as given, and an equilibrium is reached whenever there is a vector of prices that makes the consumption
decisions of all individuals mutually consistent with each other.

3.5 3.5

3 3

2.5 2.5

2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

Figure 5: Disequilibrium and Competitive Equilibrium

Note: This figure illustrates disequilibrium and competitive equilibrium scenarios for an economy with Cobb-Douglas

αi
1−αi 1 3
preferences given by V i = ci1 ci2 , where α1 = 2
, α2 = 5
, y 11 = 0.5, y 12 = 3, y 21 = 2.5, and y 22 = 0.5.

Edgeworth Box Economy: Competitive Equilibrium

A competitive equilibrium is an allocation c̊ = c11 , c12 , c21 , c22 and prices p = p1 , p2 such that

i) individuals choose consumption to maximize utility subject to their budget constraint taking
prices as given, that is, they solve

ci p, ȳ i = arg max ui ci1 , ci2 s.t. p1 ci1 + p2 ci2 = p1 ȳ i1 + p2 ȳ i2 ,

i c

ii) and markets clear, that is, resource constraints hold:

c11 + c21 = ȳ 11 + ȳ 21 Market Clearing Good 1

c12 + c22 = ȳ 12 + ȳ 22 Market Clearing Good 2

The left panel in Figure 5 illustrates a situation of disequilibrium. An economy is in disequilibrium for
some vector of prices when the market clearing conditions are not satisfied given individual demands. In
this figure, there is excess supply of good 2 and excess demand of good 1. The prices chosen to illustrate
this figure are not consistent with a competitive equilibrium.
The right panel in Figure 5 represents a competitive equilibrium in the box diagram. Given the budget
sets defined by the vector of prices p represented, the demands of both individuals are mutually compatible,
making the allocation c̊⋆ a competitive equilibrium.

11
 Since individual demands are homogeneous of degree zero in the price — that is, individual’s i budget is
the same if all prices are scaled up or down by the same amount: the prices p, 3p, or αp for any α > 0 yield
identical demands — then whenever prices p are a part of a competitive equilibrium, there are equilibria
with identical allocations with prices αp, for α > 0. Graphically, in the box diagram, we can see that all that
p2
matters is the slope of the budget constraints p1 , so only the ratio of prices is determined in a competitive
equilibrium. We have thus one (trivial) dimension of indeterminacy, which we typically use to normalize the
price of good 2 to p2 = 1. This normalization is often described as choosing good 2 as numeraire. There are
other possible normalizations. For instance, it is common to restrict prices to lie in the unit simplex, so that
P j
j p = 1.
Therefore, the competitive Walrasian model — at least in its basic form — is a theory of relative prices,
and has no predictions for the aggregate level of prices. The lack of predictions for the price level of the
Walrasian model — traditionally referred to as Hanh’s Problem (Hahn, 1965) — has spurred large literatures
on monetary theory. See Gale (1982, 1985) for earlier contributions closer to the classic general equilibrium
tradition, but also Woodford (2003), Williamson and Wright (2010), Starr (2013), or Cochrane (2023).

2.4.1 Offer curves

An alternative way of representing a competitive equilibrium in the box diagram is by drawing offer curves.
The offer curve of individual i represents the demand function in the box diagram for all different prices.
The offer curve is derived by considering how each individual’s demand function changes as relative prices
change. By construction, offer curves always go through the autarky allocation for each individual, and must
be tangent to the individual’s indifference curve at the endowment allocation. Whenever there is another
crossing — different from the endowment allocation — of the offer curves of both individuals, we have
found a competitive equilibrium. In a sense, the offer curves in the box diagram are the counterparts of the
Marshallian supply and demand diagram.

12
 3.5

2.5

1.5

0.5

0
0 0.5 1 1.5 2 2.5 3

Figure 6: Offer Curves

Note: This figure illustrates offer curves (bottom panel) for an economy with Cobb-Douglas preferences given by

αi
1−αi 1 3
V i = ci1 ci2 , where α1 = 2
, α2 = 5
, y 11 = 0.5, y 12 = 3, y 21 = 2.5, and y 22 = 0.5.

The bottom panel of Figure 7 shows offer curves for both individuals in an Edgeworth box economy with
isoelastic preferences.

2.4.2 Excess demand

In order to account for whether an individual is a net buyer or a net seller of a particular good, we define
individual i’s excess demand for good j as

z ij p, ȳ i = cij p, ȳ i − ȳ ij .



(Individual Excess Demand) (5)

For given prices p, when z ij p, ȳ i > 0, an individual i is a net buyer of good j, and when z ij p, ȳ i < 0,

an individual i is a net seller of good j.

Aggregate excess demand for good j, which corresponds to the sum of individual excess demands, is given
by
X
X ij
z j (p; ȳ) = z ij p, ȳ i = c p, ȳ i − ȳ ij .



(Aggregate Excess Demand) (6)
i i

For given prices p, when z j (p; ȳ) > 0, we say that there is aggregate excess demand of good j, and when
z j (pȳ) < 0, we say that there is aggregate excess supply. We can collect individual and aggregate excess
demands in J-dimensional vectors as follows:

X i
X i
z i p, ȳ i = ci p, ȳ i − ȳ i z p; ˚ z p, ȳ i = c p, ȳ i − ȳ i .





and ȳ =
i i

13
Note that in an Edgeworth box economy, we can define aggregate excess demand for each of the two goods.
Note also that prices that characterize the set of competitive equilibria can be found by finding the values
of p⋆ that solve the system of equations
z p⋆ ; ˚


ȳ = 0. (7)

And equilibrium allocations can be written in terms of primitives (endowments) as

ci⋆ ˚
ȳ = ci p⋆ , ȳ i ,

where p⋆ is defined by equation (7).7

Therefore, the aggregate excess demand map z p; ˚


ȳ summarizes the positive properties of competitive
equilibria, and will be the central object of study in Section 6. In economies with two goods (J = 2), like
the Edgeworth box economy, it is useful to plot individual and aggregate excess demand functions for one
of the goods, say good 1. The top panels of Figure 7 show individual and aggregate excess demands as a
function of p2 (or equivalently, p2 /p1 , for an economy with isoelastic preferences.
Individual excess demands inherit the properties of individual demands, which are studied in classical
demand theory. If individual preferences are continuous, strictly convex, and strongly monotone, individual
demands, individual excess demands, and aggregate excess demands are i) continuous and ii) homogeneous
of degree zero in prices. Note that without further restricting individual preferences, demands need not be
homogeneous in individual endowments.

2.4.3 Walras’ Law

Walras’ Law is the statement that

“if all but one markets in a competitive economy are in equilibrium, then the last market must
also be in equilibrium”.

Walras’ Law can be seen as a property of the aggregate excess demand function. Formally, aggregating all
individual budget constraints implies that
XX
X X j ij
pj cij − ȳ ij = p z p, ȳ i = p · z p; ˚



ȳ = 0,
i j i j

and in particular in the Edgeworth box economy:

p1 c11 + c21 − ȳ 11 − ȳ 21 + p2 c12 + c22 − ȳ 12 − ȳ 22 = 0.

Therefore, if c11 + c21 = ȳ 11 + ȳ 21 = 0, then it must be that c12 + c22 = ȳ 12 + ȳ 22 = 0. Hence, we can always
drop one of the aggregate excess demand conditions in equation (7). In practice, depending on the particular
application considered, it may more convenient to drop the condition associated to some specific goods.
An implication of Walras’ Law is that

“if there is aggregate excess demand in one market, there must be another market with aggregate
excess supply”.
7 With strongly monotone preferences, p⋆ > 0 in equilibrium since otherwise individuals would want to demand infinite
amounts of the goods with pj = 0.

14
Therefore, if c11 + c21 − ȳ 11 − ȳ 21 > 0, then it must be that c12 + c22 − ȳ 12 − ȳ 22 < 0. Intuitively, given that
individual’s total expenditures are constrained by their income, the value of excess demand for some goods
needs to be compensated with excess supply for another.
Note that the derivation of Walras’ law exclusively uses the linearity of budget constraints, but does not
require optimality regarding the choice of individual demands or price-taking behavior. This observation
implies that Walras’ law may also hold in models that are not perfectly competitive.

7
2
6
5
1
4
3
2 0

1
0 -1

0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

Figure 7: Excess Demands

Note: This figure illustrates excess demand functions as a function of p1 (good is chosen as numeraire: p2 = 1) for an

αi
1−αi 1 3
economy with Cobb-Douglas preferences given by V i = ci1 ci2 , where α1 = 2
, α2 = 5
, y 11 = 0.5, y 12 = 3,
y 21 = 2.5, and y 22 = 0.5.

2.5 Graphical Representation

There are four diagrams that systematically summarize an Edgeworth box economy for different preference
configurations:

i) [Box diagram] The top left diagram shows the box diagram with the endowments, the Pareto set, the
contract curve, and the competitive equilibrium. It displays indifference curves in autarky and at the
competitive equilibrium.

ii) [Box diagram with offer curves] The top right diagram also shows the box diagram but it only displays
the offer curves. The (non-autarky) allocations at which the offer curves intersect characterize the set
of competitive equilibria.

iii) [Utility diagram] The bottom left diagram shows the utility possibility frontier and the Pareto frontier,
it also displays the values of utilities at the autarky allocation and the competitive equilibrium.

iv) [Excess demand diagram] The bottom right diagram shows aggregate excess demand as a function of
p = p1 /p2 .

15
Figures 8 through 12 show the four diagrams for economies with different typical preference specifications:

αi
1−αi
i) Cobb-Douglas: Preferences are given by V i = ci1 ci2 .
  ϵ

ϵ−1

ϵ−1 ϵ−1
ii) Isoelastic (CES) Preferences: Preferences are given by V i = αi ci1 ϵ + 1 − αi ci2 ϵ .

 ϵ
 ϵ−1
iii1 i1

ϵ−1 i

i2 i2

ϵ−1
iii) Stone-Geary: Preferences are given by V = α c − c ϵ
+ 1−α c −c ϵ
.


iv) Perfect Substitutes: Preferences are given by V i = αi ci1 + 1 − αi ci2 .

v) Perfect Complements: Preferences are given by V i = min ci1 , ci2 .

2.6 Final Observations

Before moving on to study more complex economies, it is useful to make several remarks about the Edgeworth
box economy:

i) The box diagram is an incredibly powerful tool. It can illustrate — with very few exceptions — all
of the phenomena in general equilibrium and welfare economics in static exchange economies. It is
less useful to think about production and dynamic stochastic economies, yet the insights delivered by
thinking through the box diagram are often valid more generally.

ii) Many papers (including some of mine!) do not explicitly distinguish between physical and economic
structure. This is a bad idea, in particular if one is interested in understanding the efficiency and
welfare properties of an economy. I encourage you to always be clear about the physical structure of
your economy and to first characterize the solution to the planning problem.

iii) There are three types of (consumption) Edgeworth boxes. The one covered in this section corresponds
to a “static Edgeworth box”, in which two individuals consume two goods at a single instant. We will
explore dynamic and stochastic Edgeworth boxes in Block III.

iv) There are also factor Edgeworth boxes, which use the box diagram with two factors of production
instead of two goods. See Helpman and Krugman (1985) or Bhagwati, Panagariya and Srinivasan
(1998) for applications of factor Edgeworth boxes in international trade.

3 Static Exchange Economies

It is straightforward to generalize the Edgeworth box economy to the case with many individuals and goods.
We refer to these economies as static exchange economies.

3.1 Physical Structure

A static exchange economy is populated by I ≥ 1 individuals, who consume consume J ≥ 1 goods. We
index individuals by i ∈ I = {1, . . . , I}, and goods by j ∈ J = {1, . . . , J}. In this economy, goods are not
produced and appear as endowments.

16
 3 3

2 2

1 1

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

3
6

2
4

1 2

0
0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

Figure 8: Edgeworth Box: Cobb-Douglas Preferences

1 3
Note: In this figure α1 = 2
, α2 = 5
, y 11 = 0.5, y 12 = 3, y 21 = 2.5, and y 22 = 0.5.

17
 2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

2.5

2 2

1.5
1
1
0
0.5

0
0 0.5 1 1.5 2 0 0.2 0.4 0.6 0.8 1

Figure 9: Edgeworth Box: Isoelastic (CES) Preferences

1 1
Note: In this figure ϵ = 1.2, α1 = 3
, α2 = 2
, y 11 = 0.75, y 12 = 1.5, y 21 = 2.25, and y 22 = 0.5.

18
 2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

15
4

3 10

2
5
1
0
0
2 3 4 5 6 0 0.5 1 1.5 2 2.5

Figure 10: Edgeworth Box: Stone-Geary Preferences

1 1
Note: In this figure ϵ = 1.2, α1 = 3
, α2 = 2
, c11 = −0.5, c12 = −7, c21 = 0, c22 = 0, y 11 = 0.75, y 12 = 1.5, y 21 = 2.25, and
y 22 = 0.5.

19
 2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

1.5
8

1 6
4
2
0.5
0
-2
0
1.4 1.6 1.8 2 2.2 0 0.5 1 1.5

Figure 11: Edgeworth Box: Perfect Substitutes

1 1
Note: In this figure α1 = 3
, α2 = 2
, y 11 = 0.75, y 12 = 1.5, y 21 = 2.25, and y 22 = 0.5.

20
 2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

2
1
1.5
0.5

1 0

-0.5
0.5
-1

0
0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3

Figure 12: Edgeworth Box: Perfect Complements

Note: In this figure y 11 = 0.75, y 12 = 1.5, y 21 = 2.25, and y 22 = 0.5.

21
 Preferences and resource constraints are introduced as in the Edgeworth box economy. Individual i has
preferences over the consumption of goods given by the utility function:
 
(Preferences) V i = ui cij j∈J
, (8)

where cij ≥ 0 denotes the consumption of good j by individual i. The resource constraint for good j is given
by
X
(Resource Constraints) cij = ȳ j , (9)
i

where ȳ j > 0 denotes the total endowment of good j in the economy, and where ȳ j = ȳ ij , where
P
i
ȳ ij ≥ 0 denotes the endowment of good j that belongs to individual i. As in the Edgeworth box economy,
the restriction to non-negative consumption is justified by the fact that variables that appear in resource
constraints need to be non-negative, since they describe the physical assignment of resources to uses.
Every definition from the Edgeworth box economy extends straightforwardly to the general case with I
individuals and J goods. Allocations are now represented in vector form by
 

 

(Allocation) c̊ = c11 , . . . , cI1 , . . . , c1j , . . . , cIj , . . . , c1J , . . . , cIJ ,
| {z }
 | {z } | {z }

good 1 good j good J


where c̊ ≥ 0. We also denote individual consumption allocations in vector form as ci = ci1 , . . . , ciJ .
An allocation is feasible when both resource constraints hold. We can compactly write preferences as


V i = ui ci .
Similarly, we express endowments in vector form as
 

 

(Endowments) ˚
ȳ = ȳ 11 , . . . , ȳ I1 , . . . , ȳ 1j , . . . , ȳ Ij , . . . , ȳ 1J , . . . , ȳ IJ ,
|
 {z } | {z } | {z }

good 1 good j good J

where ˚ȳ ≥ 0. We also denote individual endowments in vector form as ȳ i =

 i1 i2
ȳ , ȳ . The autarky
allocation is the one in which c̊ = ˚
ȳ.

3.2 Planning Problem

Solutions to the planning problem still characterize the Pareto set, as long as the utility possibility set is
convex. The planning problem can now be written as
X  
max αi ui cij j∈J
,
c̊
i

subject to
X
cij = ȳ j , ∀j ∈ J ,
i

22
 αi = 1 and αi ≥ 0. By varying the set of weights αi , it is possible to trace the Pareto frontier. We
P
where i
will explore this problem in detail in Section 4.

3.3 Axiomatic (Nash) Bargaining

Similarly, the axiomatic (Nash) bargaining problem can be written as
Y    αi
max ui cij j∈J
− ui ȳ ij j∈J
,
c̊
i

subject to
X
cij = ȳ j , ∀j ∈ J ,
i

αi = 1 and αi ≥ 0. As in the Edgeworth box economy, the Nash bargaining solution selects
P
where i
allocations that are i) Pareto efficient, ii) individually rational, and iii) satisfy independence of irrelevant
alternatives.

3.4 Competitive Equilibrium

Finally, a competitive equilibrium is an allocation c̊ and prices p = p1 , . . . , pJ such that

i) individuals choose consumption to maximize utility subject to their budget constraint taking prices as
given, that is, they solve
  X X
ci p, ȳ i = arg max i
cij pj cij = pj ȳ ij ,


i
u j∈J
s.t. ∀i ∈ I,
c
j j

ii) and markets clear, that is, resource constraints hold:

X X
cij = ȳ ij , ∀j ∈ J .
i i

The Lagrangian associated with the individual utility maximization problem, denoting the Lagrange
multipliers in the budget constraint and the non-negativity constraints by λi ≥ 0 and ν ij ≥ 0 respectively,
is given by  
  X X
L = ui cij j∈J
− λi  pj cij − pj ȳ ij  + ν ij cij ,
j j

with optimality conditions given by

dL ∂ui
= − λi pj + ν ij = 0. (10)
dcij ∂cij

The definitions of individual and aggregate excess demand in equations (5) and (6) remain valid for static
exchange economies. Also the characterization of competitive equilibrium by setting the aggregate excess
demand function to zero, z p⋆ ; ˚


ȳ = 0, in equation (7), and the characterization of in terms of primitives
(endowments) as ci⋆ ˚
ȳ = ci p⋆ , ȳ i remain valid. Unfortunately, out of the four diagrams from Section

23
2.5, only the excess demand diagram can be used to study general static exchange economies, and only
when J = 2. That said, the excess demand diagram becomes extremely useful to understand the positive
properties of competitive economies, as it will become clear in Section 6.

3.5 Application: Armington Model: I = J ≥ 2

A special interesting case of the static exchange economy is the Armington model (Armington, 1969;
Anderson, 1979), which provides an elementary theory of international trade based on product differentiation
by country of origin. The key idea is that goods “produced” in different countries are not perfect substitutes,
so countries prefer to consume a variety of goods from multiple other countries rather than solely rely on
the cheapest option. This is arguably the simplest model that generates a gravity equation (Anderson and
van Wincoop, 2003; Arkolakis, Costinot and Rodríguez-Clare, 2012). It contrasts with the Ricardian model,
which we’ll study in Block II, in which countries specialize in goods in which the have comparative advantage.
An Armington economy is populated by I > 1 individuals (typically interpreted as countries) who would
like to consume the J = I goods available. Critically, each country has an endowment of a particular good,
that is, country 1 is endowed with good 1, country i with good j = i, and so on, until country I is endowed
with good J. The Edgeworth box economy is a special case of the Armington economy when I = J = 2.8
Formally, country i has CES preferences
 σ
 σ−1
X 1 σ−1
Vi = aij σ cij σ 



,
j

where σ ≥ 0 denotes the elasticity of substitution and aij > 0, and faces the budget constraint
X
pij cij = pi ȳ i ,
j

where we allow for country specific prices — pij rather than pj — to allow for transportation costs, as
explained below. Country i’s optimality conditions imply that for any two goods j, ℓ ∈ J
−σ
cij aij pij

iℓ
= iℓ ,
c a piℓ

which in turn allows us to write country i’s expenditure on good j as

1−σ
pij

pij cij = aij pi ȳ i , (11)
Pi

P 1

1−σ  1−σ
where P i = ℓ piℓ has the interpretation of a price index. The Armington model typically
features iceberg shipping costs, so when country j ships cij units to country i, τ ij units are lost (“melt”) in
the process.9 This can be interpreted as a technology, as explained in Block II. Here, it simply implies that
8 Instead of assuming that each country starts with an endowment of their domestic good, the Armington model is typically

formulated assuming that each country starts with a fixed factor of production that can be exclusively used to produce a
domestic good using a constant returns to scale technology.
P Both
formulations are equivalent,P with pi ȳ i equal
Pto ijlabor income.
9 Resource constraints can therefore be written as 1 + τ ij cij = ȳ j or, equivalently, cij = ȳ j − τ cij
i i i

24


pij = 1 + τ ij pj , that is, the price paid at destination for a unit of good j’s consumption is equal to the
price at origin augmented by the iceberg cost.
Hence, equation (11) can be written as

−σ i
1−σ j
1−σ i i

pj cij = aij 1 + τ ij P p p ȳ ,
| {z } | {z } |{z}
bilateral/multilateral j- term i−term

which has the interpretation of a generalized gravity equation, in which expenditure of country i on good j

−σ i
1−σ
is a function of a bilateral/multilateral term, 1 + τ ij P , and two terms that are country specific,
j 1−σ

i i
p and p ȳ . See Allen and Arkolakis (2016) for a detailed exposition of gravity equations.

4 Efficiency and Welfare

Consistently with the welfare primacy approach followed in this text, we study efficiency and welfare in
static exchange economies first, in this section, leaving the analysis of the positive properties of Walrasian
equilibrium to Section 6.

4.1 Planning Problem

It seems nature to initially characterize and understand the set of Pareto efficient allocations since we have
shown that those will be the same for all economies with the same physical structure. Given the importance
of these allocations, we will characterize the solution to the planning using two different methods, both
yielding different insights. First, we use a conventional Lagrangian approach, followed by a perturbation
approach.

4.1.1 Lagrangian Method

The Lagrangian associated with the planning problem stated in Section 3.2 is given by
!
X   X X XX
i i ij j ij j
L= αu c j∈J
− η c − ȳ + κij cij ,
i j i i j

where we use η j ≥ 0 to denote the Lagrange multipliers in each of the J resource constraints, and κij ≥ 0 to
denote the Lagrange multipliers in each of the IJ non-negativity constraints in consumption. The optimality
conditions for this problem are given by

dL ∂ui
ij
= αi ij − η j + κij = 0. (12)
dc ∂c

The first term in equation (12) corresponds to the marginal benefit of increasing the consumption of good j
by individual i. The second term can be interpreted as the social shadow cost of providing that consumption,
i
∂u
which is the same across all individuals. The third term, when non-zero, implies that αi ∂c j
ij < η , so marginal

benefit of increasing cij is lower than its marginal cost, making it optimal to set cij = 0. The multiplier κij
can be interpreted as a shadow marginal benefit that the planner attaches to cij , as to be happy with setting
cij = 0 and not a negative amount.

25
 Focusing on interior allocations, there are two combinations of the optimality conditions in equation
(12) worth highlighting. First, we can combine the optimality conditions for any two goods consumed by
individual i, in this case, j and ℓ ∈ J . In this case, we find that

∂ui
∂cij ηj
∂ui
= . (13)
∂ciℓ
ηℓ

Equation (13) confirms and generalizes the result derived graphically in the Edgeworth box economy, in which
the slopes of the indifference curves of all individuals are equalized at interior Pareto efficient solutions. In
particular, note that the left-hand side of equation (13) corresponds to the marginal rate of substitution of
individual i between goods j and ℓ. The right-hand side of equation does not depend on i, which means
whenever two individuals i and n ∈ I consume any two goods j and ℓ, it must that their marginal rates of
substitution between the goods are equalized, that is,

∂ui ∂un
∂cij ∂cnj ηj
∂ui
= ∂un
= , ⇒ Efficiency (14)
∂ciℓ ∂cnℓ
ηℓ

where equation (13) holds ∀i, n ∈ I and ∀j, ℓ ∈ J such that cij , ciℓ , cnj , cnℓ > 0.
Second, we can combine the optimality conditions for any two individuals i and n who consume good j.
In this case, we find that
∂ui
∂cij αn
∂un
= . (15)
∂cnj
αi
αn
Equation (15) ensures that the ratio of marginal utilities is equal to ratio of the Pareto weights αi . This
equation captures that by choosing Pareto weights (or utility units) that favor a particular individual, this
individual will have overall higher consumption of good j — and of all goods, since equation (15) is valid for
any good j that both individuals consume, hence

∂ui ∂ui
∂cij ∂ciℓ αn
∂un
= ∂un
= , ⇒ Redistribution (16)
∂cnj ∂cnℓ
αi

where equation (16) holds ∀i, n ∈ I and ∀j such that cij , cnj > 0. Graphically, this condition has the
interpretation of how different Pareto weights choose different points in the Pareto frontier in the utility
diagram of the Edgeworth box economy.
Given our discussion, we will say that the conditions in (14) characterize efficiency since any allocation
that satisfies those is a Pareto efficient allocation. We will also say that the conditions in (16) characterize
redistribution since these conditions encode preferences of the planner for one individual versus another.
Importantly, while the conditions in (14) are invariant to all preference preserving transformations, the
conditions in (16) depend on the choices of i) Pareto weights, and ii) individual utility units. The use of
these terms will be consistent with the results of the welfare assessments analysis in Section 4.2.

26
4.1.2 Perturbation Solution

We now proceed to attack the planning problem using a direct perturbation approach.10 For this purpose
we define the value of the social objective, which we refer to as social welfare, by
X
W = αi V i ,
i

where V i denotes individual utility levels. The perturbation approach is based on looking for marginal
perturbations of allocations, and then studying its impact on W . We index all perturbations by an index θ.
This variable is simply a placeholder which will index changes in allocations. Alternatively, we could have
dW
used differentials, and simply write dW instead of dθ . In order to characterize the solution to this problem,
we must find allocations in which no feasible perturbation away from those allocations increase W , that is,
dW
they must feature dθ ≤ 0, or = 0 at an interior optimum, the case we will consider here.
In this case, the change in social welfare can be written as

dW X dV i dV i X ∂ui dcij
= αi , where = ,
dθ i
dθ dθ j
∂cij dθ

and where any feasible perturbations must satisfy that

X dcij dȳ j
= = 0. (17)
i
dθ dθ

Hence, we can write the change in social welfare induced by a perturbation as

dW X X ∂ui dcij
= αi ij . (18)
dθ i j
∂c dθ

Given equation (18) and the restriction that all feasible perturbations must satisfy equation (17), it should
dW
be evident that the condition that ensures that dθ = 0 for all feasible perturbations is that

∂ui ∂un
αi ij
= αn nj (19)
∂c ∂c

for any two individuals i and nm and ∀j such that cij , cnj > 0. Otherwise, it would always be possible to
construct a feasible perturbations that shifts one unit of consumption of good j from individual i to n (or
dcij dcnj
vice versa) — that is dθ = 1 and dθ = −1 — with

dW ∂ui ∂un
= αi ij − αn nj > 0.
dθ ∂c ∂c

But the conditions in equation (19) imply both the efficiency and redistribution conditions in (14) and (16),
so we have simply characterized the solution to the planning problem in a different way.
Both the planning and the perturbation approaches are useful in different circumstances. On the one
10 It should be evident that the perturbation approach is the mathematical foundation of the Lagrangian approach.

Unfortunately, the latter approach is typically taught in cookbook form, and many students are unaware of the connections.

27
hand, the planning problem is valuable because it provides Lagrange multipliers, which can be interpreted
as measure of value for each of the goods, and will map to prices in competitive economies. On the other
hand, the perturbation approach will be useful to systematically understand the origins of welfare gains and
losses, as we will learn in Section 4.2.

4.1.3 Remarks on Pareto Efficiency

Since we repeatedly use the notion of Pareto efficiency/optimality throughout this text, I’d like to make four
observations.

i) Pareto optimality is mute about distributional consequences. Allocations in the northeast and
southwest corners of the Edgeworth box displayed in Figure 4 are Pareto optimal, yet in both of
them one of the individuals is miserable, with no consumption of any of the goods. In general, as long
as one individual has strongly monotone preferences and consumes the totality of endowments, that
allocations will be Pareto efficient.

ii) Economists have little to say about which of the different Pareto efficient allocations is preferred.
While preferences for redistribution can be captured by social welfare functions, introduced in Section
4, normative statements that involve interpersonal comparisons must be qualified by the choice of social
welfare function and utility units.

iii) The Pareto criterion defines an incomplete order over allocations. Since typically any welfare
assessments involves winners and losers — in particular as the number of individuals in an economy
grows — the Pareto criterion is almost always inconclusive in practice. There exist other notions that
can be used to provide more complete order, for instance, the Kaldor-Hicks potential compensation
principle, introduced in Section 4.

iv) It is important to highlight that whether an allocation is Pareto efficiency or not depends on the
possible set of feasible allocations that are considered. Unless we are explicit about it, Pareto efficiency
in this text refers to the solution of a planning problem, although at times it is useful to define notions
of constrained Pareto efficiency, in which the set of feasible perturbations to consider is constrained.

4.2 Welfare Assessments

After characterizing the set of Pareto efficient allocations, and before diving into the welfare properties of
the competitive model, I will address the question of: given a particular physical structure, is there a way in
which we can systematically attribute the welfare gains of particular perturbations to specific sources? This
is the question of the origins of welfare gains that I have developed in recent work with Andreas Schaab —
see e.g. Dávila and Schaab (2024c) and Dávila and Schaab (2024b), among other work.
Many questions take the form of welfare assessments. For instance, policy counterfactuals, e.g. what is
the welfare impact of a tax change?; changes in primitives, e.g. what is the welfare impact of a change in
technology?; optimal policy exercises, e.g. what is the optimal tax system?, a question that can only be
answered by assessing the welfare consequences of different policies.

28
Welfarism and Social Welfare Functions. To provide an answer to these questions, we will study
welfare assessments for welfarist planners. Welfarist planners are those who assess welfare based on a social
welfare function given by

W = W V 1, . . . , V i, . . . , V I ,


(Social Welfare Function) (20)

∂W
where individual utilities V i are defined in equation (8), and where we assume that ∂V i > 0, ∀i. We refer
to the units of W as social utils. A welfarist planner finds a perturbation desirable (undesirable) if

dW X ∂W dV i
= > (<) 0.
dθ i
∂V i dθ

The critical restriction implied by the welfarist approach is that the social welfare function W (·) cannot
depend on any model outcomes besides individual utility levels. The main justification for using a welfarist
approach is that it concludes that every Pareto improving perturbation is desirable — that is, the welfarist
approach is Paretian. Kaplow and Shavell (2001) further show that the converse statement — welfarism is
the only approach that respects the Pareto criterion — is true under minimal assumptions. The use of social
welfare functions of the form described here dates to Bergson (1938) and Samuelson (1947).
Note that we have already used utilitarian social welfare functions, W = i αi V i , to solve planning
P

problems. By virtue of their linearity, utilitarian social welfare functions can be used to trace (convex)
Pareto frontiers, and that’s why they are the most used in practice. However, there are other reasonable
forms of social welfare functions, as described, for instance in Kaplow (2011). Intuitively, social welfare
functions are simply a way to encode social preferences over individuals.
The main challenge when working with social welfare functions is that individual utilities are ordinal, so
it is not a priori obvious how a welfarist planner makes tradeoffs in comparable units across individuals. We
will address this challenge next.

Addressing the Units Challenge. The change in social welfare induced by a particular perturbations,
dW
dθ , can be expressed as

i
dW X ∂W dV i X ∂W dV X ∂W dV i|λ
i dθ
= i dθ
= i
λ i
= i
λi (21)
dθ i
∂V i
∂V λ i
∂V dθ

utils of individual i


where λi > 0 is an individual normalizing factor — with units dim λi = units of welfare numeraire — that
allows us to express individual welfare gains or losses in units of a common welfare numeraire. The rationale
behind multiplying and dividing by λi is to be able to express individual  i  welfare gains in a common unit
dV i
for all individuals. In particular, while the units of dθ are dim dV dθ = utils of individual i
units of θ , the units of
i
 i

dV dV
dV i|λ
dθ = λi are dim
dθ dθ
λi = units of welfare numeraire
units of θ , ∀i, which are common to all individuals.
But what is exactly λi ? In static exchange economies, it is natural to choose a particular good as welfare
∂ui
numeraire. If good 1 is chosen as welfare numeraire, then λi = ∂ci1 . But we could also choose a bundle of
i i
∂u ∂u ∂ui 1 ∂ui 1 ∂ui 2
goods, so λi = ∂ci1 + ∂ci2 , or a proportional numeraire, so λi = i
∂ci1 c or λ = ∂ci1 c + ∂ci2 c . See DS25 for
more details.
dV i|λ
Therefore, the interpretation of dθ corresponds to individuals i’s willingness to pay for perturbation

29
in units of the common welfare numeraire (e.g. consumption of good 1, or of a bundle of goods). So by
normalizing by λi we have now cardinalized the welfare change for each individual i. It turns out to be
useful to cardinalize the chance in social welfare, which can be achieved by normalizing dW 1
P ∂W i
dθ by I i ∂V i λ ,
as follows
dW ∂W i
dW λ X dV i|λ ∂V i λ
= 1
P dθ∂W = ωi where ω i = 1
P ∂W
.
dθ I i ∂V i λ
i
i
dθ I i ∂V i λ
i

dW λ
This normalization implies that dθ is now expressed in units of taking one unit of welfare numeraire and
1
distributing it equally (with I shares) to all individuals. The weights ω i define normalized individual weights,
and have the property that they average to 1, that is

1X i
ω = 1.
I i

A normalized individual weight has the interpretation of the marginal social welfare gain of distributing 1
unit of welfare numeraire (consumption) to individual i that is equivalent to distributing ω i units to everyone.
For instance, transferring one unit of welfare numeraire from i = 1 to i = 2 implies a welfare gain of

dW λ
= ω2 − ω1 .
dθ

Alternatively, in relative terms, the relative welfare gain of giving a unit of welfare numeraire to i = 1 relative
ω1
to i = 2 is ω2 . For example, if ω 1 = 1.1, a planner values the same giving 1 units of welfare numeraire to
i = 1 and 1.1 units of welfare numeraire equally distributed to everyone. And if ω 2 = 0.9, this implies that
ω1 1.1
ω2 = 0.9 = 1.22, so we can say that planner likes individual i = 1 22% more than i = 2.
Intuitively, play the role ω i of social marginal rates of substitution across individuals. And importantly,
∂W
while Pareto weights αi or their general counterpart ∂V i are meaningless, because they depend on the
ordinal nature of individual utility functions, the normalized Pareto weights ω i have a meaningful cardinal
interpretation.

Efficiency/Redistribution Decomposition. Dávila and Schaab (2024c) show that there is a unique way
dW λ
to decompose the normalized welfare assessment dθ into an efficiency component — which corresponds to
Kaldor-Hicks efficiency — and its complement, called the redistribution component. Formally,

dW λ X dV i|λ X dV i|λ dV i|λ

 
= ωi = + CoviΣ ω i , , (22)
dθ i
dθ i
dθ dθ
| {z } | {z }
RD
ΞE (Efficiency) Ξ (Redistribution)

where CoviΣ [·, ·] = I · Covi [·, ·] denotes a cross-sectional covariance-sum among all individuals. The efficiency
component ΞE corresponds to Kaldor-Hicks efficiency, that is, it is the unweighted sum of individual
willingness-to-pay for the perturbation in units of the welfare numeraire. The notion of Kaldor-Hicks
efficiency (Kaldor, 1939; Hicks, 1939; Boadway and Bruce, 1984) is based on the compensation principle.
This notion says that a perturbation is desirable if the winners can hypothetically compensate the losers.
Hence, perturbations in which ΞE > 0 can be turned into Pareto improvements if transfers are feasible and
costless.

30
 While in practice the compensation of winners to losers may not available, Kaldor-Hicks efficiency can be
used as tool to think about Pareto efficiency. Two properties relate the efficiency component (equivalently,
Kaldor-Hicks efficiency) and Pareto efficiency.

i) First, the efficiency component is strictly positive (ΞE > 0) for (strict or weak) Pareto-improving
perturbations. Intuitively, since Pareto-improving perturbations feature no losers, the sum of
willingness-to-pay must be strictly positive.

ii) Pareto optimal allocations — defined as those solving the Pareto problem (Ljungqvist and Sargent,
2018) — must feature a weakly negative efficiency component (ΞE ≤ 0) for any feasible perturbation
given endowments. This property ensures that the efficiency component of our decomposition cannot
be positive by reallocating resources away from a (first-best) Pareto optimal allocation.11

dV i|λ
The redistribution component — equivalently be expressed as ΞRD = i
P
i ω −1 dθ — captures the
equity concerns embedded in a particular social welfare function. ΞRD is positive when the individuals
relatively favored in a perturbation are those relatively preferred by the planner, i.e., have higher normalized
individual weights ω i .
It is worth highlighting two properties of the efficiency/redistribution decomposition. First, the efficiency
component is invariant to i) the choice of social welfare function and ii) preference-preserving utility
transformations. Second, as long as one individual does relatively better than another, it is possible to
select individual weights ω i (by varying the social welfare function or the formulations of individual utilities)
so that ΞRD is positive or negative for a given perturbation. For instance, ΞRD can be negative for Pareto-
improving perturbations, even though ΞE + ΞRD > 0. Note also that equation (22) implies that welfare
assessments based on a social welfare function are equivalent to relying on Kaldor-Hicks efficiency with a
correction for redistribution.

Welfare Accounting: Exchange Efficiency in Static Exchange Economies The decomposition in

equation (22) has made no assumptions (beyond differentiability) about preferences, technologies, or resource
constraints. In fact, it applies equally to static and dynamic stochastic exchange and production economies.
Here we will specialize the results for the static exchange economy that is the subject of study in this Block I.
In this economies, we are only two possible perturbations: those that change the allocations of consumption,
and those that change the size of the endowments of goods. Formally, we can index aggregate endowments
by θ, as in ȳ j (θ), and allow for
dȳ j (θ)
⋛ 0.
dθ
dV i
dV i|λ
In a static exchange economy dθ = dθ
λi , can be written as

i
dV i|λ X ∂uij dcij X dcij
= ∂c
i
= M RScij (23)
dθ j
λ
|{z} dθ j
dθ
=M RScij

The interpretation of equation (23) is clear. The welfare change of a perturbation in individual i’s welfare is
dcij
given by the change in consumption of each of the goods, dθ , valued at the marginal rate of substitution
11 The critical feature of the Pareto problem is the presence of linear resource constraints, which allow for costless transfers.
In general, it is possible to find perturbations of constrained Pareto efficient allocations such that ΞE > 0.

31
between that good and the assumed welfare numeraire, which we denote here by

∂ui
∂cij
M RScij = .
λi

dV i|λ
As explained above, the units of dθ are common across all individuals, which allows to meaningfully
dV i|λ E dcij
aggregate dθ in Ξ . While we could stop at equation (23), it should be evident that dθ ’s could change
because consumption is reallocated among individuals or perhaps because there is a change in the amount
of aggregate endowment, which would allow for consumption to go up for all individuals.
Defining individual i’s consumption share of good j by

cij
χij
c = ,
cj

dcij
where cj denotes aggregate consumption of good j, cj = cij , we can express
P
i dθ as follows:

dcij dχij
c j dcj
= c + χij
c .
dθ dθ dθ

This expression simply states that the change individual i’s consumption of good j can be attributed to
dχij
a change in the consumption share, dθ ,
c
for a given amount of aggregate consumption, or to a change in
aggregate consumption, for a given consumption share, χij
c .
Therefore, the efficiency component of a welfare assessment in a static exchange economy is given by

X dV i|λ dcij dχij j

ij dc
XX XX XX
ΞE = = M RScij = M RScij c cj + χij
c M RSc
i
dθ i j
dθ j i
dθ j i
dθ
" #
X dχij
c
X dcj
= CoviΣ M RScij , cj + AM RScj ,
j
dθ j
dθ

where we define an aggregate marginal rate of substitution AM RScj = χij ij

P
i c M RSc — see Dávila and
Schaab (2024b). This object has the interpretation of increasing aggregate consumption by a unit that is
dcj dȳ j (θ)
distributed in proportion to consumption shares, using the resource constraint, we have that dθ = dθ ,
so we can express efficiency as

dχij dȳ j (θ)

X   X
ΞE = CoviΣ M RScij , c cj + AM RScj . (24)
j
dθ j
dθ
| {z } | {z }
Cross-Sectional Consumption Good Endowment
Exchange Efficiency Change

Equation (24) is the conclusion of the study of welfare assessments in general static exchange economies. In
this economies, all efficiency gains need to be due to i) reallocating consumption across individuals or ii) the
fact that aggregate endowments change. As one would expect, if M RScij are equalized across individuals,
then the starting allocation is efficient, so the first term is necessarily zero. But note that the efficiency
conditions in equation (14) imply the equalization of M RScij for any feasible numeraire.
In this case, AM RScj corresponds to the marginal social value of having more units of good j (M SV j )
in the economy holding consumption shares fixed — see Dávila and Schaab (2024a) for why this is well-

32
defined notion. In more general economies, the definition of the marginal social value of a good needs to be
augmented to include the use of the good as an input in production, as shown in Dávila and Schaab (2024b),
who show how to generalize equation (24) to general production economies without accumulation equations
and Davila and Schaab (2025), who do the same for general dynamic stochastic production economies with
accumulation equations.

Global Kaldor vs. Hicks Criteria and Scitovsky Paradox. The general global definitions of the
Kaldor and Hicks efficiency/compensation criteria are the following:

i) the Kaldor compensation criterion (Kaldor, 1939) states that an allocation A is preferred to B if
winners can hypothetically compensate losers after the change takes place

ii) the Hicks compensation criterion (Hicks, 1939) states that an allocation A is preferred to B if losers
cannot bribe the winners for the change not to take place.

These definitions can be seen as the “aggregate” counterpart of compensating (Kaldor) and equivalent (Hicks)
variations. Importantly, for marginal perturbations, the Kaldor and Hicks criteria are identical — boiling
P dV i
down to whether i λdθi > 0 — hence my use of the Kaldor-Hicks term.12
It is well understood that the global formulations of the Kaldor and Hicks criteria can yield paradoxical
results — see e.g. Boadway and Bruce (1984). For instance, the Scitovsky paradox (Scitovsky, 1941)
highlights that it is possible to have an allocation A that Kaldor dominates allocation B, while at the same
time allocation B Kaldor dominates allocation A. These paradoxical results — as well as the fact that
equivalent and compensating variations are different — are all a manifestation of the some phenomenon: the
fact the marginal valuations (in particular λi ) vary with the allocation. This is in turn a consequence of the
presence income effects. In quasilinear/transferable utility economies (λi = 1) welfare aggregation is trivial
and these issues do not arise, as we describe next.

Quasilinear/Transferable Utility Economies. In a quasilinear economy, all individuals have prefer-

ences that are separable and linear in consumption of good 1. Formally, individual i has preferences given
by
V i = ci1 + Ψi ci2 , . . . , ciJ .

These preferences have a cardinal interpretation in which good 1 can be interpreted as “money”. Under this
interpretation, adopting an equal-weighted utilitarian social welfare function implies that ω i = 1 — since
∂W
∂V i = λi = 1 — so with these formulation of preferences and social welfare function maximizing welfare is
equivalent to maximizing efficiency. These economies are often called transferable utility economies, and are
useful because they do not feature income effects (or more precisely, all income effects emerge in good 1).
Quasilinear economies yield the Marshallian supply-and-demand diagram ubiquitous in introductory
economics, as explained in Block II. The ω i = 1 property thus justifies the use of consumer and producer
surplus as valid welfare metrics. The results presented in this subsection can be interpreted as generalizing
those notions to economies without preferences that are not quasilinear.
12 It is also the case that compensating variation, equivalent variation, and consumer surplus are identical for marginal

perturbations.

33
4.3 First Welfare Theorem
After having characterized the set of efficient allocations and the origins of welfare gains at any allocation
(efficient or inefficient), it is time to study the welfare properties of Walrasian equilibria. Formally, we are
trying to understand the relation between the allocations that emerge as part of a competitive equilibrium
and the Pareto efficient allocations. The results relating both are of major importance in economics, and are
referred to as the first and second welfare theorems. The first welfare tells us every allocation that emerges
from a competitive equilibrium is Pareto efficient.

Figure 13: Welfare Theorems

Theorem. (First Welfare Theorem). Every competitive equilibrium allocation is Pareto efficient.

Given its central importance in economics, I will provide three different proofs of the first welfare theorem.
The first proof, independently provided by Arrow (1951) and Debreu (1951) is the most powerful one if one
is interested in identifying the minimal set of assumptions required for the first welfare theorem to hold. In
particular, this proof highlights that local non-satiation of preferences is all that is needed for a competitive
equilibrium to be Pareto optimal. It is also remarkably simple.

34
 Proof #1 of First Welfare Theorem: Preferred Allocations Not Affordable

Consider a competitive equilibrium allocation given by c̊⋆ and p⋆ . Suppose there exist another feasible
allocation c̊ that Pareto dominates c̊⋆ , in which,

i) At least one individual is strictly better off: V i > V i⋆ for some i.

ii) No individual is worse off: V i ≥ V i⋆ , ∀i.

It must have been that the individual that is strictly better off could not have afforded the newly
considered allocation at the competitive equilibrium prices (otherwise he would have demanded that
allocation, and the original allocation could not have been an equilibrium), so
X X
pj⋆ cij > pj⋆ eij .
j j

Similarly, for all the other individuals, local non-satiation ensures that
X X
pj⋆ cij ≥ pj⋆ eij .
j j

But aggregating across all individuals implies that

!
XX XX X X X
j⋆ ij j⋆ ij j⋆ ij ij
p c > p e ⇒ p c − e > 0. (25)
i j i j j i i

cij = eij ,
P P
Therefore, since the resource constraints/market clearing conditions require that i i
equation (25) cannot hold, which means that there are no feasible allocations c̊ that Pareto dominate
c̊⋆ . Therefore, every competitive equilibrium is Pareto efficient.

Figure 14 illustrates the failure of the first welfare theorem in an Edgeworth box economy with preferences
that exhibit satiation.

Figure 14: Counterexample to First Welfare Theorem

Note: Figure 16.C.1 of MWG. Individual 2 is better off. Individual 1 is indifferent.

35
 Proof #2 of First Welfare Theorem: Equivalence between Competition and Planning

To simplify the exposition, we focus in the interior case in which cij > 0, although the results can
be extended to situations in which cij = 0 for some individuals. In this case, individual optimality
conditions — presented in equation (10) — are given by

∂ui
(Competitive Equilibrium) − λi pj = 0,
∂cij

where λi denotes the Lagrange multiplier in individual i’s budget constraint.

The optimality conditions of the planning problem — presented in equation (12) — are given by

∂ui 1
(Planning Solution) − i η j = 0,
∂cij α

where αi denotes the Pareto weight assigned by the planner to individual i and η j denotes the
Lagrange multiplier in good j’s resource constraint. Comparing both equations it becomes evident
that there are one-to-one mappings between λi and αi , and between η j and pj :

1
λi ↔ and pj ↔ η j . (26)
αi
1
Therefore, given a competitive equilibrium allocation, if we choose Pareto weights αi = λi , we
j j
automatically know that p = η is a solution of the planning problem, which is enough to conclude
that the competitive equilibrium is Pareto efficient.

The second proof of the first welfare theorem is constructive, and shows clearly the connections between
the competitive and planning problems, as well as the conditions that competition satisfies by virtue of
being. This proof is typically attributed to Lange (1942) — see also Samuelson (1947) and Dávila and
Schaab (2024b). This proof relies on differentiability in a way that the first proof does not.

36
 Proof #3 of First Welfare Theorem: Pecuniary Effects Cancel Out

Starting from a competitive equilibrium allocation, we can express the individual welfare gains
associated with a general perturbation by
i
dV i X ∂ui dcij X ∂uij dcij X dcij
i i
= ij dθ
= λ ∂c
i dθ
= λ pj , (27)
dθ j
∂c j
λ j
dθ

where now we use λi to denote the Lagrange multiplier in individual i’s budget constraint, and where
the last equality follows form individual optimality. In particular, we consider perturbations that
modify individual demands, but ensuring that budget constraints and market clearing conditions
remain satisfied.a Since a perturbation must satisfy individual budget constraints, it must be that

X dcij X dpj ij X dpj ij X dcij X dpj

pj pj ȳ ij − cij .


+ c = ȳ =⇒ =
j
dθ j
dθ j
dθ j
dθ j
dθ

This equation tells us that aggregate changes in the market value of individual i’s consumption,
P j dcij
j p dθ , need to equal the sum of the distribute pecuniary effects — using the language of Dávila
j
and Korinek (2018) — of the perturbation, j dp
P ij ij


dθ ȳ − c . These pecuniary effects are composed
by

i) net trading positions (net buying or net selling), ȳ ij − cij , and

dpj
ii) sensitivity of equilibrium prices, dθ .

dV i
We can thus express the normalized individual welfare gain dθ
λi as

dV i X dcij X
dθ
pj dpj ȳ ij − cij .


= =
λi j
dθ j

After aggregating across all individuals, it must be that

X dV i X dpj X
dθ
ȳ ij − cij = 0,


=
i
λi j
dθ i
| {z }
=0

where the last equality follows from market clearing, implying that distributive pecuniary effects
P dV i
cancel out in the aggregate. The final argument follows by contradiction. Since i λdθi = 0, if for
dV i dV i
some individual dθ
λi > 0, there must be another individual for whom dθ
λi < 0, so every perturbation
features losers, implying that the competitive equilibrium is Pareto efficient.




a Formally, we can assume that individuals submit perturbed demands of the form c̃i p, ȳ i = ci p, ȳ i +θhi p, ȳ i ,


where hi p, ȳ i ’s are arbitrary. This corresponds to a Frechet/Gateaux differential of individual demands.

Versions of the approach in the third proof are used in Geanakoplos and Polemarchakis (1986) and Dávila
and Korinek (2018) to study constrained inefficiency of competitive equilibrium in models with incomplete
markets. The advantage of the third proof over the first two is that it illustrates the role paid by pecuniary

37
effects/externalities in achieving efficiency. In particular, the last equality exploits market clearing, showing
dpj
that the pecuniary effects/externalities of any perturbation dθ “cancel out” or “add up to zero”. This will
not be true in economies with incomplete markets. The main drawback of this proof relative to the first one
is that it only provides a local argument.
This third proof highlights a central feature of competitive markets: agents exclusively interact through
(linear) prices, so any spillovers or externalities among individuals need to operate through prices. At times,
one may read that pecuniary externalities are not “real externalities”, a view that I do not subscribe. This
proof illustrates that if a perturbation increases (decreases) the prices of the goods that individual i purchases
(sales), individual i will be worse off, and vice versa, and this can be interpreted as a real externality. What
it is true is that at the competitive equilibrium of a static competitive economy, the sum of the pecuniary
externalities/effects of perturbation has to be zero, which implies that there are always winners and losers,
eliminating the possibility of finding Pareto improvements.

Assumptions Underlying the First Welfare Theorem. There are several assumptions that are critical
for the first welfare theorem to hold in a static exchange economy. These are

i) local non-satiation, as highlighted by the first proof;

ii) price-taking behavior or, equivalently, absence of market power;

iii) self-interested preferences or, equivalently, absence of consumption externalities;

iv) markets for each of the goods, or, equivalently, complete markets for goods.

In production economies, it is also necessary that all goods are rival, i.e. that there are no public
goods. In dynamic stochastic economies, the presence of incomplete markets, asymmetric information, belief
distortions, and a double infinity of individuals and goods or periods may break the first welfare theorem.
Continuity, differentiability, or convexity of preferences are not required for the first welfare theorem to hold,
although they may be critical assumptions to ensure that a competitive equilibrium exists.

Formalizing the Invisible Hand. The first welfare theorem shows that individuals pursuing their self-
interest in a competitive market will, without intending it, achieve an allocation of resources that cannot
be (Pareto) improved upon. Hence, this theorem provides a formal foundation for Adam Smith’s concept of
the invisible hand:

"Every individual (...) neither intends to promote the public interest, nor knows how much he is
promoting it. By directing that industry in such a manner as its produce may be of the greatest
value, he intends only his own gain, and he is (...) led by an invisible hand to promote an end
which was no part of his intention."

The Wealth of Nations, Book IV, Chapter II.

Smith highlights the arguably paradoxical reliance on the assumption of individual self-interest as follows:

"It is not from the benevolence of the butcher, the brewer, or the baker that we expect our dinner,
but from their regard to their own interest."

The Wealth of Nations, Book I, Chapter II.

38
 Figure 15: Counterexample to Second Welfare Theorem
Note: Figure 16.D.4 of MWG.

4.4 Second Welfare Theorem

The second welfare theorem is a converse result to the first. It concludes that any Pareto efficient allocation
can be supported as part of a competitive equilibrium after appropriately reshuffling endowments (or,
equivalently, wealth) across individuals.

Theorem. (Second Welfare Theorem) Every Pareto efficient allocation can be decentralized as a competitive
equilibrium (with transfers).

The second proof of the first welfare theorem presented earlier is also a proof of the second welfare
theorem. In fact, this is a constructive proof, in the sense that it characterizes the prices that support the
chosen allocation as part of a competitive equilibrium, via equation (26). I will not provided additional
proofs.
In its more general form, the second welfare theorem is an application of the Hahn–Banach separation
theorem, as presented for instance in Stokey, Lucas and Prescott (1989). The classical proof of the second
welfare theorem relies on showing that the the set of feasible allocations and the set of allocations that
improve over the Pareto efficient allocation that is to be supported are convex. Since the allocation that
is to be supported is Pareto efficient, it must be that both sets are disjoint, which justifies the use of
the separating hyperplane theorem, which has the interpretation of finding a set of supporting prices by
construction a separating hyperplane/linear functional of the form p · c̊.
Proving the second welfare theorem in the Edgeworth box is straightforward. Once an allocation in the
Pareto set is identified, the slope of the indifference curves at that allocation defines the set of supporting
prices. If preferences are convex, it should be evident that an competitive equilibrium can be supported by
reshuffling endowments so that individuals end up with an endowment consistent with the budget constraint
induced by the supporting prices.

Assumptions Underlying the Second Welfare Theorem. In contrast to the first welfare theorem,
the second welfare theorem relies on convexity (as does the existence result, which we discuss in Section 6.1),

39
in addition to non-satiation. This is not true for the first welfare theorem. The assumption of convexity is in
principle restrictive, but as we will discuss later, as the number of individuals grows, individual nonconvexities
wash out in the aggregate.
Figure 15 illustrates a scenario in which the second-welfare theorem does not hold. In the left panel,
preferences a non-convex, so the set of supporting prices that would support the candidate Pareto efficient
allocation is inconsistent with individual optimization. Mas-Colell, Whinston and Green (1995) also highlight
the need for non-satiation in establishing the second welfare theorem.
More important in practice is allowing the planner to reshuffle endowments freely. But if the planner
can transfer endowments costlessly, why not simply enforcing directly the desired allocation? In practice,
transfers are not costless, and need to be carried through some tax and transfer procedure, similar to the
one studied in Section 4.6.

4.5 Application: Transfer Problem/Paradox

The transfer problem in general equilibrium refers to the question of how a transfer of resources from one
country to another affects allocations and prices and, ultimately, the welfare of both countries. This classic
problem in international economics and general equilibrium was originally posed by Keynes (1929) and
Ohlin (1929) in the debate about German reparations following World War I. The transfer paradox is the
paradoxical outcome where a transfer of resources from a country to another could lead to a worsening of
the recipient’s welfare and an improvement of the donor’s welfare. Fortunately, the static exchange economy
we have analyzed allows us to provide a direct answer to the question of who gains and losses after a transfer
of resources amount individuals/countries.
Formally, let’s consider the welfare gains associated with a perturbation of individual endowments — as
usual indexed by θ — in a competitive environment. In this case, individual i’s budget constraint implies
that
X dcij X dpj ij
X j dȳ ij
pj + c − ȳ ij − p = 0. (28)
j
dθ j
dθ j
dθ

Intuitively, this equation captures that the value of the change in consumption (first term), the distributive
pecuniary effects (second term) and the gain from increases the level of endowments (third term), need to
add up to zero. Therefore, we can combine equations (27) and (28) to express individual welfare gains as

dV i X dcij X dpj X dȳ ij

dθ
pj ȳ ij − cij + pj


= = .
λi j
dθ j
dθ j
dθ
| {z } | {z }
Distributive Pecuniary Direct Transfer
Effects Effects

The transfer effects can be easily signed. If country i receives a transfer of good j, this increases welfare in
proportion to pj . Similarly, if country i transfers away good j, this generates a welfare loss proportional to
pj . The distributive pecuniary effects benefit those countries who are net sellers of good j, ȳ ij − cij > 0, if
dpj
good j’s price increases, dθ > 0, or countries who are net buyers of good j, ȳ ij − cij < 0, if good j’s price
j
dp
decreases, dθ < 0. Note that both distributive pecuniary effects and direct transfers effects cancel out in
the aggregate, that is
X X dpj XX dȳ ij
ȳ ij − cij = 0 and pj


= 0,
i j
dθ i j
dθ

40
 dȳ ij


ȳ ij − cij = 0 and the zero-sum nature of the transfer,
P P
because of market clearing, i i dθ = 0.
i
P dV
Therefore, it is also the case that i λi = 0.
dθ

In general, transfers of resources will have an impact on relative prices, thus affecting the economies
involved in complex ways.13 For the transfer paradox to occur, it simply has to be that a country for whom
dȳ ij
dθ < 0 experiences large beneficial distributive pecuniary effects. This could occur, for instance, when the
recipient country exerts a lot of upward pressure on the goods of the donor country starting from a situation
in which the donor’s exports are large.
Samuelson (1952) showed that the transfer paradox could only occur for unstable equilibrium in I = 2
economies. However, the transfer paradox can occur at unique and stable Walrasian equilibria in economies
with I > 3 individuals. Polemarchakis (1983) provides a fully worked out application along the lines of the
exposition of this section. See also Chichilnisky (1983) and Geanakoplos and Heal (1983).

4.6 Application: Optimal Taxation

The second welfare theorem assumes that it is possible to costlessly transfer resources among individuals, that
is, it assumes that “lump-sum” transfers and taxes are available. In practice, taxes need to be conditioned on
economic acts, for instance the act of consuming goods. Conditioning taxes on actions taken by individuals
modifies and distorts the behavior. Hence, a natural question in that case is how to achieve a particular
social objective (e.g. redistribution), typically minimizing the efficiency losses imposed: this is the realm of
Public Economics/Finance, in particular the theory of Optimal Taxation.
In static exchange economies, it is natural to study consumption taxes. Formally, the budget constraint
of an individual i who faces consumption taxes τ ij ≥ 0 can be written as

Rebate Tax Payment

X X X X z
z}|{ X }| {
ij
pj cij = pj ȳ ij + T i ⇒ pj cij = j ij i ij j ij


1+τ p ȳ + T − τ p c , (29)
j j j j j
| {z }
Net Transfer

where rebates must add up to zero in the aggregate

X XX
Ti = τ ij pj cij . (30)
i i j

At times, to eliminate redistributive forces, it useful to assume that rebates are perfectly targeted to those
(types of) individuals who paid taxes, so T i = j τ ij pj cij . Importantly, individual always take Ti as given,
P

so this substitution can only be made after an individual has optimality chosen her demand.14
In this case, the Lagrangian associated with individual i’s optimization takes the form
 
  X X
Li = ui cij − λi  1 + τ ij pj cij − pj ȳ ij − T i  ,


j∈J
j j

j
13 Inan open economy context, dp dθ
captures both changes in exchange rates and terms-of-trade movements.
14 Formally, given taxes τ ij , a competitive equilibrium with taxes for this economy is defined as allocations and prices such
that i) individuals maximize utility subject to the budget constraint in equation (29), ii) markets clear, and iii) net transfers
add up to zero in the aggregate, so equation (30) is satisfied.

41
with an optimality condition given by
∂ui
= λi 1 + τ ij pj .


∂cij

As above, we can express individual i’s welfare gains as

i
dV i X ∂ui dcij X ∂uij dcij X
j dcij
i ∂c i ij
= = λ = λ 1 + τ p ,
dθ j
∂cij dθ j
λi dθ j
dθ

where the last equality uses individual i’s optimality condition.

If we want to study the efficiency consequences of a perturbation, we can write15

dV i
E
X XX dcij X X ij j dcij
Ξ = dθ
= pj + τ p
i
λi i j
dθ i j
dθ
X X dcij XX dcij X X dcij
= pj + τ ij pj = pj τ ij ,
j i
dθ i j
dθ j i
dθ
| {z }
ȳ j
= ddθ =0

so
ΞE = 0 if τ ij = τ j , ∀i . (31)

Equation (31) yields a uniform consumption tax result. Formally, as long as τ ij = τ j , ∀i, the economy will
be efficient. This results follows from the fact that
X X dcij
ΞE = pj τ j =0
j i
dθ
| {z }
j
= ddθ
ȳ
=0

whenever τ ij = τ j , ∀i. Intuitively, in a static exchange economy, distortions can only emerge if any two
individuals exchange goods at different rates — in that case, there would be cross-sectional exchange efficiency
gains as defined in (24) by reallocating consumption across individuals. By ensuring that taxes for each
good are identical across all individuals, we have ensured that the outcome of the economy will be Pareto
efficient. Hence, different welfarist planners may decide to set different values of τ j since these have different
consequences for redistribution. But as long as equation (31) is satisfied, all of those allocations will be
Pareto efficient.
This uniform consumption tax result is clearly very special, and crucially relies on the absence of
production. We will revisit optimal taxation in economies with production in Block II.
15 Note that we can also write
 P 
XX dcij X X dpj
X d T i − j τ j pj cij
pj + cij − ȳ ij
= ,
dθ dθ dθ
i j i j i
| {z } | {z }
=0 =0

which ensures that the change in the value of individual consumption, the distributive pecuniary effects, and the net transfers
add up to zero one aggregated.

42
5 Microfounding Competition
Up to this point, whenever we have considered a competitive economic structure we have simply assumed
that individuals choose consumption taking prices as given and respecting linear budget constraints. It
should be evident that when the number of individuals is small (say when I = 2) there is no reason to
believe agents will take prices as given or that prices will be posted.
The question of this section is then to characterize whether the set of competitive allocations and prices
emerge endogenously in particular situations in which individuals behave strategically. This task is taken
up by the literature on strategic foundations of competitive equilibria. The most prominent result in this
literature is based on the cooperative game-theoretic notion of the core, which we study first. While one
could wait to present this material until describing the positive properties of competitive equilibrium, it is
useful to present the results on the core first, given the formal similarities with the normative results of the
model.

5.1 Core Convergence Theorem

The essential question of this section is how competition leads to an equilibrium where prices are taken as
given, despite individual incentives to behave strategically. As usual, we assume that individuals are self-
interested and act to maximize their utility through exchange. This sets the stage for a broader investigation:
how competition leads to an equilibrium characterized by price-taking behavior, as if determined by an
"invisible hand." The notion that competition nullifies any individual bargaining power, creating consistent
pricing and allocation outcomes across large populations has historical roots in Edgeworth (1881). He writes

“As the number of participants increases indefinitely, the final settlements tend to converge
towards a unique result, in which no participant has an incentive to vary his terms. Thus,
in the limit of an infinitely large number of bargainers, the monopoly element vanishes, and the
results of competition are obtained.”

He also writes:

“The range of indeterminateness of the final settlement decreases as the number of bargainers
increases.”

The formalization of Edgeworth (1881)’s large limit hypothesis is taken up by Debreu and Scarf (1963), whose
approach I developed here, and is the same as in Mas-Colell, Whinston and Green (1995). To formalize this
result, we will rely on the cooperative game-theoretic notions of “blocking coalition” and the “core”:

i) A coalition of individuals N ⊆ I blocks the feasible allocation c̊⋆ if there is another allocation c̊

1. in which every member of the coalition is better off

un (cn ) ≥ un (cn⋆ ) , ∀n ∈ N

2. that is feasible using the endowments of the coalition

X X
cnj ≤ ȳ nj , ∀j ∈ J
n∈N n∈N

43
 ii) The core is the set of allocations that cannot be blocked

In Edgeworth box economies, the core and the contract curve coincide. Intuitively, allocations outside of
Pareto set dominated by coalitions of the totality of individuals (I = 2), while allocations that are not
individually rational are dominated by coalitions of I = 1.
The core equivalence theorem — sometimes also called core convergence theorem or Edgeworth’s limit
theorem — states that as the number of individuals grows toward infinity, the core collapses to the set
of competitive equilibria. Therefore, a competitive equilibrium emerges as a valid solution to trading
environments with many individuals.
Formally, the core equivalence theorem consists of two results:

1. Competitive equilibrium allocations are in the core

2. As the number of individuals in an economy increases, the core shrinks to the set of CE

The logic of the first result is very similar to the logic behind Proof 1 of the first welfare theorem.
Proof that competitive equilibrium allocations are in the core

Consider a competitive equilibrium given by c̊⋆ and p⋆ . Consider a coalition N ⊆ I and an allocation
c̊ such that

i) At least one individual is strictly better off: un (cn ) > un (cn⋆ ) for some n ∈ N

ii) No individual is worse off: un (cn ) ≥ un (cn⋆ ), ∀n ∈ N

If this allocation was feasible, it would block the competitive equilibrium. At prices p⋆ :
X X
pj⋆ cnj > pj⋆ ȳ nj
j j

for one individual and

X X
pj⋆ cnj ≥ pj⋆ ȳ nj
j j

for the others. But aggregating among coalition members n ∈ N :

!
XX XX X X X
pj⋆ cnj > pj⋆ ȳ nj ⇒ pj⋆ cnj − ȳ nj > 0.
n∈N j n∈N j j n∈N n∈N

cnj = ȳ nj , the proposed

P P
But since the “coalition resource constraint” requires that n∈N n∈N
blocking allocation is not feasible. Therefore, competitive equilibria cannot be blocked and are in the
core.

Can thus all allocations in the core be supported as a competitive equilibrium (without transfers)? The
answer is obviously negative. For instance a Cobb-Douglas Edgeworth box economy will feature a unique
equilibrium, while its core/contract curve has many allocations. We can thus only hope to find a limiting
result as the number of individuals in the economy grows. We do so by considering replica economies,
introduced in Debreu and Scarf (1963).

44
 We study a replica economy populated by i ∈ I = {1, . . . , I} types of individuals, with m ∈ M =
{1, . . . , M } individuals per type. Hence, the total number of individuals in this economy is I × M . The
interesting limit we explore is one in which we fix number of types of individuals I, but increase the number
of individuals per type, that is, we take the limit M → ∞. We then establish the result that as the number
of individuals in an economy increases, the core shrinks to the set of competitive equilibria.
In order to establish this result, we will proceed assuming assuming every individual of type i must have
the same allocation in a competitive equilibrium — this is an equal treatment property. Its proof, which we
skip, is conceptually straightforward, and it is based on building coalitions of the worse treated individuals
of each type.

45
 Proof that as the number of individuals increases the core shrinks to the set of CE

Our proof hinges on showing that if c̊⋆ (IJ-dimensional) is a feasible type-allocation, but not a
competitive equilibrium, then it can be blocked. First, note that if c̊⋆ not Pareto Optimal, coalition
of everyone can block it, so it is without loss to assume that the candidate allocation c̊⋆ is Pareto
optimal.
Since c̊⋆ is Pareto Optimal, it can be supported as a CE with transfers (we denote prices as p⋆ ),
via the second welfare theorem. But if c̊⋆ is not a competitive equilibrium, then at least one type of
individuals (assume type-1) is favored relative to CE:
X
pj⋆ c1j⋆ − ȳ 1j > 0.


(32)
j

In this case, we can build a coalition (N , with n ∈ N = {1, . . . , N }) of all types with M −1 individuals
of type 1 to exclude one type-1 individual, where the total number of individuals N in the coalition
is
N = M − 1 + M (I − 1) .
| {z } | {z }
type 1 others types

This coalition can propose a new allocation that distributes symmetrically the net trades (c1 − ȳ 1 )
of the excluded type-1 individual:

1 1
∆cn = cn′ − cn = c − ȳ 1


N

In general, for large changes, the fact that the excluded individual has some negative entries of the
vector c1 − ȳ 1 makes things hard, but as M (and implicitly N ) grows, we can resort to perturbation
dcn
arguments of the sort covered earlier, so that limM →∞ ∆cn = dθ — using now the standard
placeholder notation θ. In this limiting case, we can show that everyone in the coalition will gain.
Formally, exploiting market clearing, for any individual n in the coalition we have formed

dV n ∂un
X
∂cnj dcnj X dcnj
dθ
= = pj ,
λn j
λn dθ j
dθ

but since type-1 was favored — see equation (32) — every member of the coalition is better off at
the margin:
X dcnj 1 X j⋆ 1j⋆
pj = p∆cn = p c − ȳ 1j > 0.


j
dθ N j

Intuitively, the excluded type-1 was consuming too much overall, with the value of his net consumption
being strictly positive at the prices that would support a competitive equilibrium. The other
individuals in the economy can thus get together, exclude him from the economy, and share that
extra value. Therefore, as M, N → ∞, the only allocations that cannot be blocked are competitive
equilibria.

The coalition that we have formed has Bertrand competition flavor, and captures the ruthless nature
of competition: whenever the value of the consumption of an individual in the economy is larger than his

46
contribution in the form of endowments, other individuals can cooperate to exclude him, erasing in this way
this undue value.
The Core Equivalence Theorem has been established under increasingly general conditions. Aumann
(1964) proves core convergence directly with a continuum of agents, without having to rely on growing
replica economies. Anderson (1978) does the same using nonstandard analysis. The notion of the core are
also widely used in discrete economies and matching theory. Shapley and Scarf (1974) apply the concept
of the core to the housing market, introducing top trading cycles, in which indivisibilities are critical. See
Afacan, Hu and Li (2024) for a recent survey. Other significant contributions include Gale and Shapley
(1962) and Shapley and Shubik (1971). See Roth and Sotomayor (1992) for a formalization of these ideas
with an eye to practical applications, and Roth (2015) for a popular account of how these ideas can be
applied to relevant questions, including organ donation, college admissions, and job placements.

5.2 Non-Cooperative Strategic Foundations

For any non-competitive economic structure, it is always possible to ask whether a particular limit of
such model delivers the competitive outcome. For instance, Bertrand equilibria of pricing games yield
a competitive outcome, while Cournot equilibria converge to the competitive outcome as the number of
participants grows. Shapley and Shubik (1977) analyze market games and how they converge to a competitive
equilibrium. Monopolistic competition models yield a competitive outcome as individual preferences are
perfectly elastic, but not as fixed costs go to zero (Córdoba and Liu, 2023). See, for instance, Gale (2000)
for strategic and game-theoretic foundation to competitive equilibrium applying dynamic matching and
bargaining models.

6 Competitive Equilibrium
In this last section, we analyze the positive predictions of the competitive model. This section is thus
useful to determine which phenomena can be captured by the model. First, we present issues related to
the existence of equilibria and their computation. We then study the question of multiplicity of equilibria,
globally and locally. Next, we present the excess demand theorem, and study the question of convergence to
an equilibrium. Finally, we show how to conduct comparative statics and study how nonconvexities can be
smoothed out in large economies

6.1 Existence and Computation

The "existence problem" studies the conditions under which there exists a solution to the conditions that
characterize a competitive equilibrium. In other words, when can we be sure that there are allocations and
prices that satisfy the definition of competitive equilibrium, that is, that individual optimize given prices and
market clear? Proving the existence of a competitive equilibrium was a major focus of early mathematical
economics, until the formal proofs of Arrow and Debreu (1954) and McKenzie (1954) — see Düppe and
Weintraub (2014) for a detailed detailed historical and sociological account of the search for the existence
theorem.
In line with the spirit of this book, I devote minimal attention to the existence question. In practice,
we are interested in understanding the predictions of a model in a particularly scenario. And to find those

47
predictions, it necessary to solve the model analytically or computationally, which is the ultimate assurance
than an equilibrium exists. If an equilibrium does not exist, it cannot be solved in closed form or in
the computer. There are also excellent available treatments of the existence question at different levels of
difficulty, including Debreu (1959), Arrow and Hahn (1971), Mas-Colell, Whinston and Green (1995), Bewley
(2007), Starr (2011), and Kreps (2013).

Equation Counting. As explained earlier, the set of prices that define a competitive equilibria can be
found by finding the values of the J-dimensional vector p⋆ that solve the system of equations that makes
aggregate excess demands equal to zero, that is, solving

z p⋆ ; ˚


ȳ = 0. (33)

Once equilibrium prices are determined, equilibrium allocations are also immediately characterized in terms
of primitives (endowments). Equation (33) is initially a system of J equations (aggregate excess demand for
each good) in J unknowns (prices for each good). For instance, in the Edgeworth box economy, equation
(33) can be written — dropping the dependence on endowments — as

z 1 p1⋆ , p2⋆ = 0

z 2 p1⋆ , p2⋆ = 0.

However, we can use two facts explained in Sections 2 and 3 to reduce the dimensionality of the system:

i) First, we can use the fact that aggregate excess demand functions are homogeneous of degree zero to
normalize one of the prices. Without loss of generality, we set good J as numeraire: pJ = 1.

ii) Second, we can use Walras’ law to conclude that as long as aggregate excess demand functions for
J − 1 goods are equal zero, the aggregate excess demand for the remaining good will also be zero.

Putting these observations together, characterizing the set of equilibrium prices boils down to solving a
nonlinear system of J − 1 equations in J − 1 unknowns, which we define by

z̃ p⋆ ; ˚


ȳ = 0, (34)

where z̃ p⋆ ; ˚
ȳ corresponds to the first J − 1 elements of z p⋆ ; ˚



ȳ . In the Edgeworth box economy, equation
(34) simply boils down — dropping the dependence on endowments — to

z̃ p⋆ ; ˚
ȳ = z 1 p1⋆ , 1 = 0.



(35)

If the system in equation (34) was linear, it would be straightforward to determine whether a unique, many,
or no solutions exist. Unfortunately, there is no systematic way to characterize the solutions of general
nonlinear systems of equations. In fact, seemingly innocuous nonlinear equations have no (real) solutions,
e.g. x2 + 1 = 0 or x2 + x + 1 = 0. However, despite this caveat, a rule-of-thumb that will typically (or more
formally, generically, as illustrated below) work is to count whether the number of equations is equal to the
number of unknowns: this is the “equation counting” approach to existence, which leverages the fact that
equation (34) features a system of J − 1 equations in J − 1 unknowns.

48
Existence with J = 2 via Excess Demand Diagram. Throughout this section, we will rely heavily
on the excess demand diagram, which is valid for general static exchange economies with I > 1 as long as
J = 2. We will always take good 2 as numeraire, and plot aggregate access demand for good 1 as a function
of p1 — defined in equation (35). Figure 7 illustrates the excess demand diagram.
In the J = 2 case, an existence proof can be established via the intermediate value theorem, or its
corollary, Bolzano’s theorem), which states that if a continuous function has values of opposite sign in an
interval, then it must have a root in that interval. Since aggregate excess demands are continuous, it suffices


to show that z 1 p1⋆ takes positive and negative values. First, as p1 → 0, strong monotonicity of preferences


implies that z 1 p1⋆ → ∞. Similarly, as p1 → ∞ (equivalently, p2 → 0), the same argument combined with


Walras’s law ensures that z 1 p1⋆ converges to a negative number. Moreover, since we have assumed that
cij ≥ 0, excess demands are bounded below by the aggregate endowments of goods.
This proof, despite its simplicity, illustrates how other more general existence proofs works. First, it is
important establishing continuity, and then provide continuous on the behavior of the system of study at its
limits or boundaries, ensuring that either a solution occurs in the interior or at boundary.

Existence via Fixed Point Theorems. The classical existence proofs — originally introduced by Arrow
and Debreu (1954) and McKenzie (1954) — are based on fixed point theorems. By constructing a suitably
defined continuous and convex mapping between an non-empty, compact, and convex set (of prices) into
itself, a fixed point theorem such as Brouwer or Kakutani’s will ensure that there is fixed point of the
mapping. Intuitively, a fixed point occurs when starting with a candidate vector of prices, the output of the
considered mapping is equal to that candidate vector, which is an equilibrium vector of prices.
The main downside of proofs based on fixed-point theorems is that they are not constructive. That is,
they do not tells us how to compute competitive equilibria, either analytically or numerically. The question
of how to compute an equilibrium is studied, for instance, in Scarf (1973).

Non-Existence. What does it mean that an competitive equilibrium does not exist? It simply means
that there are no allocations and prices that satisfy our definition. In practice, there are two threats to the
existence of equilibrium:

i) Non-Convexities

ii) Lack of strong monotonicity in preferences

FIGURE NONEXISTENCE

Negishi Approach. In general, planning problems are easier to compute than competitive equilibria.
This is because planning problems do not involve prices, and can be simply derived as a the solution of an
optimization problem. Competitive equilibria, on the other hand, require to solve many individual demand
problems, and then compute market clearing prices. Given this observation, it is seems natural to use
planning problems as a way to compute competitive equilibria: this is the Negishi (1960) approach. While
Negishi (1960)’s original contribution is more focused on providing another existence result, the approach he
proposed is mostly used in practice for its computational advantage — this is very common in neoclassical
macro models).

49
 Intuitively, when the welfare theorems hold, there is an equivalence between competitive equilibria and
Pareto efficient allocations, so we can solve planning problems and then find equilibrium prices that support
these allocations. As shown above, we can read off the prices as Lagrange multipliers from the appropriate
constraints of the social planners problem. Importantly, the Negishi approach, does not apply — at least in
its basic form — to inefficient economies in which the welfare theorems do not hold, for instance, economies
with taxes, market power, etc. In that case, the planning problem cannot be used to compute competitive
equilibria, invalidating the approach.

6.2 Excess Demand Theorem

The aggregate excess demand theorem — initially developed by Sonnenschein (1973), Mantel (1974), Debreu
(1974), see also Mas-Colell (1977) — illustrates that aggregate excess demand functions in a general
equilibrium model can take almost any shape, given only continuity, homogeneity of degree zero, and Walras’s
law. Intuitively, while individual excess demand satisfy restrictions imposed by individual optimization, such
properties are lost once we are able to add them up. It is the aggregation step that is critical, so these results
have more bite in economies as I → ∞. Importantly, this result is focused on the zeros and the overall shape
of the aggregate excess demand functions, but it imposes no restrictions on preferences and endowments.
These result are often interpreted as saying that anything goes (true), and that the competitive model
has no testable predictions (false). The correct way to interpret these results is to acknowledge the large
flexibility of the competitive model to capture a wide variety of phenomena in the aggregate, even without
the need to resort to imperfect competition or imperfect information. Obviously, the competitive model is
testable if we are allowed to conduct comparative static exercises on endowments or other primitives, as
highlighted by Brown and Matzkin (1996). Kübler and Polemarchakis (2024) provides a recent surveys the
theory of the recoverability of unobserved fundamentals from observable variables in general equilibrium.

FIGURE SMD

6.3 Uniqueness and Multiplicity

While an ideal theory of behavior has a unique prediction, a corollary of the excess demand theorem is that
multiplicity of equilibria is a priori pervasive. In fact, it is always possible to construct economies with many
equilibria.
That said, in practice, it is common to make assumptions on preferences (e.g. Cobb Douglas or CES)
such that there is a unique equilibrium. More generally, economies that satisfy the gross substitute property
will feature a unique equilibrium. In differential form, the gross substitutes property states that at every p,
it must be that ∂z j /∂pℓ > 0 for j ̸= ℓ. At an intuitive level, to ensure uniqueness, we must ensure that
demands slope down sufficiently.

FIGURE Multiplicity

Toda and Walsh (2024) provide a recent survey of the uniqueness results, highlighting the possibility that
multiplicity may arise with standard preference specifications.

50
Local Uniqueness. Even if there are multiple equilibria, it is worth to consider the question of whether
equilibria are locally unique or locally isolated, if we cannot find another normalized equilibrium price vector
arbitrarily close to it.

FIGURE Local Uniqueness

∂ z̃
We say that an equilibrium is regular if the Jacobian matrix ∂p has full rank (J − 1). If every equilibrium
is regular, the economy is regular. The following properties hold:

1. Any regular (normalized) equilibrium price vector is locally unique/isolated.

2. If the economy is regular, the number of normalized equilibria is finite.

3. Regular economies have an odd number of equilibria.

4. Equilibria in which the aggregate excess demand “slopes down” have to exist

The last two results follow from the Index theorem, which states that for a regular economy
X
index p = +1,
{p:z(p)=0,pJ =1}

∂ z̃ ∂ z̃
where index p = (−1)J−1 sgn ∂p , where |Dẑ(p)| is the determinant of the (J − 1) × (J − 1) matrix ∂p .
Intuitively, in the J = 2 case, index p is positive for equilibrium with downward-sloping demand and negative
for those with upward-sloping demand.
The final question here is whether we expect economies to be regular or not. The answer is, generically,
yes. That is, it is always possible to perturb a non-regular economy to find another economy that is regular.
This logic is related to the “counting equations” approach, in the sense that a non-linear system of equations
with the same number of equations and unknowns generically has a solution. Mas-Colell (1985) provides an
authoritative analysis of these issues.

FIGURE Generic

6.4 Convergence to Equilibria

Instead of focusing on what occurs at equilibrium, we could also consider what occurs as the economy
converges to equilibrium. We say that an equilibrium is stable if the economy converges to it (given an
initial price vector and a tatonnment rule). Convergence can be global, if this is true for any initial point,
or local, if it is the case for sufficiently close points. We say that an economy is stable, if for any initial
position, the economy converges to an equilibrium.
Unfortunately, our definition of equilibrium does not specify how an equilibrium is reached. A natural
procedure to study convergence is to postulate that prices adjust proportional to their excess demand: if
there is excess demand (supply) for a good j, its price will go up (down). This is a price tâtonnement
procedure, formally given by
dpj
= z j (p (t) ; ȳ) ,
dt
where we index prices by a “time” variable t ∈ [0, ∞], and we initially the system at some initial price vector
p (0). In the J = 2 excess demand diagram, we can see that this tâtonnement procedure always converges

51
to at least one of the equilibria. This procedure is typically referred to as playing the role of the “Walrasian
auctioneer”. But it suffers from a number of problems, for instance, why should all individual trade goods
at the same price out of equilibrium?
Even more unfortunately, the convergence/tâtonnement result is the one result in which the J = 2
analysis yields misleading results. In the general J > 2 case, anything can happen in terms of convergence
to an equilibrium. The seminal illustration of this phenomena is given by Scarf (1960). This paper shows
that the tâtonnement process does not necessarily lead to equilibrium. Instead, prices can oscillate without
ever converging. Formally, consider a static exchange economy with I = 3 individuals and J = 3 goods, with
preferences and endowments given by


1. V 1 = min c11 , c12 and ȳ 1 = (1, 0, 0)


2. V 2 = min c22 , c23 and ȳ 2 = (0, 1, 0)


3. V 3 = min c31 , c33 and ȳ 3 = (0, 0, 1)

In this economy, demand and excess demand for individual 1 is given by

     
p1 ȳ 1 p1 −p2
p1 +p2 p1 +p2 − 1 p1 +p2
p1 ȳ 1
c1 p, ȳ i =   ⇒ z 1 p, ȳ i = 

 
 p1
  p1

= ,
 p1 +p2   p1 +p2   p1 +p2 
0 0 0

while the demand for 2 and 3 can be derived similarly. Therefore, aggregate excess demand can be written
as:  
−p2 p2
p1 +p2 + p3 +p1
−p3

 
z p; ˚
ȳ = 
 p2 +p3 + p1
p1 +p2
.

−p1 p2
p3 +p1 + p3 +p3
j
If we assume a tâtonnement adjustment function dp j
dt = z (p (t) ; ȳ), ∀j, then Walras’ law implies that
P j

2
j p is constant, since its time derivative is zero:

X X dpj
pj z j (p (t) ; ȳ) = pj = 0.
j j
dt

It should be evident that p1 = p2 = p3 is the unique equilibrium, which solves z p; ˚

ȳ = 0. Scarf (1960)
shows that only solution to the tâtonnement ordinary differential equation is p1 p2 p3 = constant, that is,

2
prices will not adjust. If we choose, without loss of generality, that j pj = 3. Then, unless we start
P

from p1 p2 p3 = 1, the economy is unstable. We have thus established that this economy has a unique unstable
competitive equilibrium, a phenomenon that cannot arise when J = 2.

6.5 Comparative Statics of Equilibrium Allocations and Prices

When we discussed welfare assessments in Section (4.2), we took as an input how allocations changed as a
dcij
function of the perturbation parameter, that is, dθ . In competitive models, it is possible to characterize
those changes in allocations by conducting comparative statics on the equilibrium prices and allocations.16
16 The set of all possible equilibria in an economy is called the equilibrium manifold, for given combinations of parameters
like preferences, endowments, or technology. By conducting comparative statics of the equilibrium manifold we capture how

52
These changes are also critical to provide testable predictions of the equilibrium in response to changes in
parameters.
Formally, using the implicit function theorem, we can write

−1
∂ z̃ dp⋆ ∂ z̃ d˚
ȳ ⋆ dp⋆ ∂ z̃ d˚
ȳ ⋆

⋆ ∂ z̃
z̃ p ; ˚


ȳ = 0 ⇒ + =0⇒ = ,
∂p dθ ∂˚ȳ dθ dθ ∂p ∂˚ȳ dθ

∂ z̃ ∂ z̃
where ∂p and ∂˚ȳ
are Jacobian matrices. For this comparative statics exercise to be well-defined, we need
∂ z̃
∂p to have full rank.
dp⋆
Once we have characterized dθ , it is straightforward to characterize how individual allocations change,
using again the implicit function theorem:

dci p⋆ , ȳ i


∂ci dp⋆ ∂ci dȳ i
= + ,
dθ ⋆
∂p dθ ∂ ȳ i dθ

∂ci ∂ci
where ∂p⋆ and ∂ ȳ i denote another set of Jacobian matrices. In general, it is hard to obtain analytical
comparative static characterizations, but at times progress can be made under functional form assumptions.
More generally, we can parametrize also preferences or other exogenous elements of more general competitive
models, for instance, taxes.

6.6 Economies with Many Individuals

As we have alluded to in different contexts, non-convexities at the individual level can still yield convex
behavior in the aggregate as the number of individuals in the economy grows, and in particular, when there
is a continuum of individuals.
These results is typically stated as the Shapley-Folkman-Starr theorem, a result that states that
the Minkowski sum (or average) of a large number of non-convex sets is approximately convex if the
dimensionality of the sets is lower than the number of sets being summed. This result shows that the
sum of a collection of non-convex sets behaves almost like a convex set. So results that require convexity,
most importantly, the existence result and the second welfare theorem, will become valid (or approximately
valid) in economies with non-convex preferences as long as there are many individuals.
In economics, it is common to consider models with a continuum of individuals, which corresponds to the
limit in which I → ∞. In this case, existence is guaranteed even when individual preferences are non-convex,
a result originally established in Aumann (1966).
An elementary illustration of the advantages of the continuum is the following. Consider an economy in
which all individuals have a discontinuous (bang-bang) demand of the form,

0, if p1 > p̄1i
ci1 = ,
1, if p1 ≤ p̄1i

where p̄1i is an individual specific threshold. Now let’s assume that the distribution of thresholds p̄i has a
continuous cumulative distribution function, which we denote by F (·). In this case, aggregate demand takes
the equilibrium outcomes (such as prices and allocations of goods) change as these underlying factors vary.

53
is given by
Z Z p1
c1 p1 = ci1 di = dF p̄i = F p1 .




0

But this expression is continuous, even though individual demands were not.
Many important general equilibrium models rely on having a continuum of agents, for example,
Dornbusch, Fischer and Samuelson (1977) or Geanakoplos (2010).

6.7 Final Observations

This is my summary of the main takeaways from the positive analysis of competitive equilibria.

1. Characterizing competitive equilibria amounts to solving a system of nonlinear equations. Under

reasonably assumptions (mostly continuity and convexity), this system will have a solution and
competitive equilibria will exist.

2. Although most models used in practice feature a unique equilibrium, we should not be surprised to
find models with multiple equilibria. Multiple equilibria are due to strong income effects.

3. Whenever a competitive economy has multiple equilibria, these cannot be Pareto ranked.

4. Generically, the number of equilibria is finite and odd. Equilibria are locally isolated.

5. By aggregating enough individual excess demands we can find aggregate excess demands with any
form, so the set of equilibrium prices can in general take any form. Once again, these phenomena are
due to strong income effects.

6. Similarly, the out-of-equilibrium convergence properties of the model can take any form. It is possible
to have a unique equilibrium to which it is not possible to converge under a reasonable tâtonnement
(adjustment) procedure.

7. Obviously, general equilibrium models are testable. Testing models may require to conduct comparative
static exercises, which are based on derivatives of the aggregate excess demand function.

8. As the number of individuals in an economy gets large, nonconvexities vanish in the aggregate, ensuring
for instance existence.

54
References
Afacan, Mustafa Oğuz, Gaoji Hu, and Jiangtao Li. 2024. “Housing Markets Since Shapley and Scarf.”
Journal of Mathematical Economics, 111: 102967.
Allen, Treb, and Costas Arkolakis. 2016. “Elements of Advanced International Trade.” Unpublished
graduate class notes.
Anderson, James E. 1979. “A theoretical foundation for the gravity equation.” The American Economic
Review, 69(1): 106–116.
Anderson, JE, and E van Wincoop. 2003. “Gravity with Gravitas: A Solution to the Border Puzzle.”
The American Economic Review, 93(1): 170–192.
Anderson, Robert M. 1978. “An Elementary Core Equivalence Theorem.” Econometrica, 46(6): 1483–
1487.
Arkolakis, Costas, Arnaud Costinot, and Andrés Rodríguez-Clare. 2012. “New Trade Models,
Same Old Gains?” American Economic Review, 102(1): 94–130.
Armington, Paul S. 1969. “A Theory of Demand for Products Distinguished by Place of Production.”
International Monetary Fund Staff Papers, 16(1): 159–178.
Arrow, Kenneth J. 1951. “An extension of the basic theorems of classical welfare economics.” Proceedings
of the Second Berkeley Symposium on Mathematical Statistics and Probability, 507–532.
Arrow, Kenneth J., and Gerard Debreu. 1954. “Existence of an Equilibrium for a Competitive
Economy.” Econometrica, 22(3): 265–290.
Arrow, Kenneth Joseph, and Frank Hahn. 1971. General Competitive Analysis. North Holland.
Aumann, Robert J. 1964. “Markets with a Continuum of Traders.” Econometrica, 32(1-2): 39–50.
Aumann, Robert J. 1966. “Existence of Competitive Equilibrium in Markets with a Continuum of
Traders.” Econometrica, 34(1): 1–17.
Bergson, Abram. 1938. “A Reformulation of Certain Aspects of Welfare Economics.” The Quarterly
Journal of Economics, 52(2): 310–334.
Bewley, Truman F. 2007. General Equilibrium, Overlapping Generations Models, and Optimal Growth
Theory. Harvard University Press.
Bhagwati, Jagdish N., Arvind Panagariya, and T. N. Srinivasan. 1998. Lectures on International
Trade. . 2nd ed., Cambridge, MA:MIT Press.
Boadway, Robin, and Neil Bruce. 1984. Welfare Economics. Basil Blackwell, Oxford.
Bowley, Arthur W. 1924. The Mathematical Groundwork of Economics. Oxford:Clarendon Press.
Reprinted, New York: Augustus M. Kelley, 1965.
Brown, Donald J., and Rosa L. Matzkin. 1996. “Testable Restrictions on the Equilibrium Manifold.”
Econometrica, 64(6): 1249–1262.
Chichilnisky, Graciela. 1983. “The Transfer Problem with Three Agents Once Again.” Journal of
Development Economics, 13(3): 237–248.
Cochrane, John H. 2023. The Fiscal Theory of the Price Level. Princeton, NJ:Princeton University Press.
Córdoba, Juan Carlos, and Xiying Liu. 2023. “A Unified Theory of Value: Oligopolistic Competition
and Optimum Product Diversity.” Iowa State University, Department of Economics.
Dávila, Eduardo, and Andreas Schaab. 2024a. “A Non-Envelope Theorem with Linearly Homogeneous
Constraints.” Working Paper.

55
Dávila, Eduardo, and Andreas Schaab. 2024b. “Welfare Accounting.” Working Paper.
Dávila, Eduardo, and Andreas Schaab. 2024c. “Welfare Assessments With Heterogeneous Individuals.”
Working Paper.
Dávila, Eduardo, and Anton Korinek. 2018. “Pecuniary Externalities in Economies with Financial
Frictions.” The Review of Economic Studies, 85(1): 352–395.
Deaton, Angus, and John Muellbauer. 1980. Economics and Consumer Behavior. Cambridge University
Press.
Debreu, Gerard. 1951. “The coefficient of resource utilization.” Econometrica, 19(3): 273–292.
Debreu, Gerard. 1959. Theory of Value: An Axiomatic Analysis of Economic Equilibrium. Vol. 17, Yale
University Press.
Debreu, Gérard. 1974. “Excess Demand Functions.” Journal of Mathematical Economics, 1(1): 15–21.
Debreu, Gérard, and Herbert Scarf. 1963. “A Limit Theorem on the Core of an Economy.” International
Economic Review, 4(3): 235–246.
Dornbusch, Rudiger, Stanley Fischer, and Paul Anthony Samuelson. 1977. “Comparative
advantage, trade, and payments in a Ricardian model with a continuum of goods.” The American Economic
Review, 67(5): 823–839.
Düppe, Till, and E. Roy Weintraub. 2014. Finding Equilibrium: Arrow, Debreu, McKenzie and the
Problem of Scientific Credit. Princeton, NJ:Princeton University Press.
Edgeworth, Francis Ysidro. 1881. Mathematical Psychics: An Essay on the Application of Mathematics
to the Moral Sciences. London:C. Kegan Paul & Co.
Gale, David, and Lloyd S Shapley. 1962. “College admissions and the stability of marriage.” American
mathematical monthly, 9–15.
Gale, Douglas. 1982. Money: In Equilibrium. Cambridge University Press.
Gale, Douglas. 1985. Money: In Disequilibrium. Cambridge University Press.
Gale, Douglas. 2000. Strategic Foundations of General Equilibrium: Dynamic Matching and Bargaining
Games. Cambridge, UK:Cambridge University Press.
Geanakoplos, J. 2010. “The leverage cycle.” NBER Macroeconomics Annual, 24(1): 1–66.
Geanakoplos, John, and Geoffrey Heal. 1983. “A Theory of the Regularity of Equilibrium and the
Transfer Paradox.” Journal of Economic Theory, 30(2): 131–141.
Geanakoplos, John, and Herakles Polemarchakis. 1986. “Existence, regularity, and constrained
suboptimality of competitive allocations when the asset market is incomplete.” Essays in honor of Kenneth
J. Arrow, 3: 65–95.
Hahn, Frank. 1965. “On Some Problems of Proving the Existence of an Equilibrium in a Monetary Econ-
omy.” In The Theory of Interest Rates. , ed. Frank Hahn and F.P.R. Brechling, 126–135. London:Macmillan.
Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade: Increasing
Returns, Imperfect Competition, and the International Economy. Cambridge, MA:MIT Press.
Hicks, John R. 1939. “The foundations of welfare economics.” The Economic Journal, 696–712.
Humphrey, Thomas M. 1996. “The Early History of the Box Diagram.” Federal Reserve Bank of Richmond
Economic Quarterly, 82(1): 37.
Kalai, Ehud, and Meir Smorodinsky. 1975. “Other solutions to Nash’s bargaining problem.” Economet-
rica: Journal of the Econometric Society, 513–518.

56
Kaldor, Nicholas. 1939. “Welfare propositions of economics and interpersonal comparisons of utility.”
Economic journal, 49(195): 549–552.
Kaplow, Louis. 2011. The Theory of Taxation and Public Economics. Princeton University Press.
Kaplow, Louis, and Steven Shavell. 2001. “Any Non-Welfarist Method of Policy Assessment Violates
the Pareto Principle.” Journal of Political Economy, 109(2): 281–286.
Keynes, John Maynard. 1929. “The German Transfer Problem.” Economic Journal, 39(153): 1–7.
Kreps, David M. 2013. Microeconomic Foundations I: Choice and Competitive Markets. Vol. 1, Princeton
University Press.
Kübler, Felix, and Herakles Polemarchakis. 2024. “Identification in General Equilibrium.” Journal of
Mathematical Economics, 113: 103017.
Lange, Oscar. 1942. “The foundations of welfare economics.” Econometrica, 10(3/4): 215–228.
Ljungqvist, Lars, and Thomas J. Sargent. 2018. Recursive Macroeconomic Theory. The MIT Press.
Mantel, Rolf R. 1974. “On the Characterization of Aggregate Excess Demand.” Journal of Economic
Theory, 7(3): 348–353.
Mas-Colell, Andreu. 1977. “The Equilibrium Price Set of an Exchange Economy.” Journal of Mathematical
Economics, 4(2): 117–126.
Mas-Colell, Andreu. 1985. The Theory of General Economic Equilibrium: A Differentiable Approach.
Cambridge:Cambridge University Press.
Mas-Colell, Andreu, Michael D. Whinston, and Jerry R. Green. 1995. Microeconomic theory.
Oxford University Press.
McKenzie, Lionel W. 1954. “On Equilibrium in Graham’s Model of World Trade and Other Competitive
Systems.” Econometrica, 22(2): 147–161.
Muthoo, Abhinay. 1999. Bargaining Theory With Applications. Cambridge University Press.
Negishi, Takashi. 1960. “Welfare Economics and Existence of an Equilibrium for a Competitive Economy.”
Metroeconomica, 12(2-3): 92–97.
Ohlin, Bertil. 1929. “The Reparation Problem: A Discussion.” Economic Journal, 39(154): 172–178.
Pareto, Vilfredo. 1906. Manuale d’economia politica. Società Editrice Libraria.
Polemarchakis, Heraklis M. 1983. “The Transfer Paradox.” International Economic Review, 24(3): 749–
760.
Roth, Alvin E. 2015. Who Gets What and Why: The New Economics of Matchmaking and Market Design.
Boston:Houghton Mifflin Harcourt.
Roth, Alvin E, and Marilda A Oliveira Sotomayor. 1992. Two-sided matching: A study in game-
theoretic modeling and analysis. Cambridge University Press.
Samuelson, Paul A. 1952. “The Transfer Problem and Transport Costs: The Terms of Trade When
Impediments Are Absorptive.” Economic Journal, 62(246): 278–304.
Samuelson, Paul Anthony. 1947. Foundations of Economic Analysis.
Scarf, Herbert E. 1960. “Some Examples of Global Instability of the Competitive Equilibrium.”
International Economic Review, 1(3): 157–172.
Scarf, Herbert E. 1973. The Computation of Economic Equilibria. Cowles Foundation Monograph Series,
New Haven, CT:Yale University Press.
Scitovsky, Tibor. 1941. “A note on welfare propositions in economics.” The Review of Economic Studies,
9(1): 77–88.

57
Shapley, Lloyd, and Herbert Scarf. 1974. “On cores and indivisibility.” Journal of mathematical
economics, 1(1): 23–37.
Shapley, Lloyd S, and Martin Shubik. 1971. “The assignment game I: The core.” International Journal
of game theory, 1(1): 111–130.
Shapley, Lloyd S., and Martin Shubik. 1977. “Trade Using One Commodity as a Means of Payment.”
Journal of Political Economy, 85(5): 937–968.
Sonnenschein, Hugo. 1973. “Do Walras’ Identity and Continuity Characterize the Class of Community
Excess Demand Functions?” Journal of Economic Theory, 6(4): 345–354.
Starr, R.M. 2013. Why Is There Money?: Walrasian General Equilibrium Foundations of Monetary Theory.
Edward Elgar.
Starr, Ross M. 2011. General Equilibrium Theory: An Introduction. Cambridge University Press.
Stokey, Nancy, Robert Lucas, and Edward Prescott. 1989. Recursive methods in economic dynamics.
Harvard University Press.
Toda, Alexis Akira, and Kieran James Walsh. 2024. “Recent Advances on Uniqueness of Competitive
Equilibrium.” Journal of Mathematical Economics, 113: 103008.
Walras, Léon. 1874. Éléments d’économie politique pure. Corbaz.
Williamson, Stephen D., and Randall Wright. 2010. “New Monetarist Economics: Models.” In
Handbook of Monetary Economics. Vol. 1, , ed. Benjamin M. Friedman and Michael Woodford, 25–96.
Amsterdam:Elsevier.
Woodford, Michael. 2003. Interest and Prices: Foundations of a Theory of Monetary Policy. Princeton
University Press.

58