The Fundamental Welfare Theorems

The so-called Fundamental Welfare Theorems of Economics tell us about the relation between
market equilibrium and Pareto efficiency.
The First Welfare Theorem: Every Walrasian equilibrium allocation is Pareto efficient.
The Second Welfare Theorem: Every Pareto efficient allocation can be supported as a
Walrasian equilibrium.
The theorems are certainly not true in the unconditional form in which weve stated them
here. A better way to think of them is this: Under certain conditions, a market equilibrium
is efficient and Under certain conditions, an efficient allocation can be supported as a
market equilibrium. Nevertheless, these two theorems really are fundamental benchmarks
in microeconomics. Were going to give conditions under which theyre true. One (very
stringent) set of conditions will enable us to prove the theorems with the calculus (KuhnTucker, Lagrangian, gradient) methods weve used very fruitfully in our analysis of Pareto
efficiency. Additionally, for each theorem well provide a much weaker set of conditions under
which the theorem remains true.
Assume to begin with, then, that the consumers preferences are very nice viz., that
theyre representable by utility functions ui that satisfy the following condition:
ui is continuously differentiable, strictly quasiconcave, and xi Rl+ : uik (xi ) > 0
n

b , (b
The First Welfare Theorem: If p
xi )1
i

E = ((u ,
x

n
))1

()

is a Walrasian equilibrium for an economy

in which each u satisfies (), then (b

xi )n1 is a Pareto allocation for E.

Proof:
n

b , (b
bi maximizes ui subject to xi = 0
Because p
xi )1 is a Walrasian equilibrium for E, each x

b xi 5 p
b
and to the budget constraint p
xi . Therefore, for each i N there is a i = 0 that
satisfies the first-order marginal conditions for is maximization problem:
k : uik 5 i pbk , with equality if x
bik > 0,

(1)

bi . In fact, because each uik > 0,

where of course the partial derivatives uik are evaluated at x
we have i > 0 and pbk > 0 for each i and each k. The market-clearing condition in the
definition of Walrasian equilibrium therefore yields

n
X

x
bik =

i=1

n
X

xik for each k (because each pbk > 0).

(2)

i=1

1
, because i > 0, and then we can rewrite (1) as
i
k : 0i uik 5 pbk , with equality if x
bik > 0,

For each i we can define 0i =

(3)

for each i N . But (2) and (3) are exactly the first-order conditions that characterize
the solutions of the Pareto maximization problem (P-max). Therefore (b
xi )n1 is a solution of
(P-max), and since each uik > 0, a solution of (P-max) is a Pareto allocation. 
In the proof, we could have instead expressed the first-order marginal conditions for the
individual consumers maximization problems in terms of marginal rates of substitution:
i
k, k 0 : M RSkk
0 =

pbk
pbk0

(4)

(written for the interior case, to avoid a lot of inequalities), which yields the (Equal MRS)
condition
j
i
i, j, k, k 0 : M RSkk
0 = M RSkk 0 .

(5)

The conditions (5) and (2) then guarantee that (b

xi )n1 is a Pareto allocation.
Now lets see if we can prove the First Welfare Theorem without any of the assumptions in
(). It turns out that we cant quite do that; however, the only thing we need to assume is
that each consumers preference %i is locally nonsatiated.
n

b , (b
xi )1
The First Welfare Theorem: If p
i

E = ((%i ,
x

n
))1

is a Walrasian equilibrium for an economy

in which each %i is locally nonsatiated, then (b

xi )n1 is a Pareto allocation for

E.
Proof:
Suppose (b
xi )n1 is not a Pareto allocation i.e., some allocation (e
xi )n1 is a Pareto improvement on (b
xi )n1 :
(a)

Pn

ei 5
1 x

Pn
1

xi

(b1)

e i %i x
bi
i N : x

(b2)

ei i0 x
bi .
i0 N : x

n

b , (b
bi maximizes %i on the budget set
Because p
xi )1 is a Walrasian equilibrium for E, each x

b xi 5 p
b
B(b
p,
xi ) := { xi Rl+ | p
xi }. Therefore, (b2) implies that
0

bx
ei > p
bx
bi ,
p
2

(6)

and since each %i is locally nonsatiated, (b1) implies that

bx
ei = p
bx
bi for each i.
p

(7)

((7) follows from a duality theorem that well state and prove below in the final three
pages of this set of notes.) Summing the inequalities in (6) and (7) yields
n
X

bx
e >
p

i=1

i.e.,
b
p

n
X

bx
bi ,
p

(8)

i=1
n
X

e >p
b
x

n
X

i=1

bi .
x

(9)

i=1

b Rl+ , it follows from (9) that there is at least one k for which
Since p
pbk > 0 and

n
X
i=1

x
eik

>

n
X

x
bik .

(10)

i=1

P
P
xik , and (10)
bik = ni=1
Since pbk > 0, the market-clearing equilibrium condition yields ni=1 x
P
P
xik i.e., (e
xi )n1 does not satisfy (a). Our assumption that
eik > ni=1
therefore yields ni=1 x
(e
xi )n1 is a Pareto improvement has led to a contradiction; therefore there are no Pareto
improvements on (b
xi )n1 , and its therefore a Pareto allocation. 

The Second Welfare Theorem: Let (b

xi )n1 be a Pareto allocation for an economy in which
the utility functions u1 , . . . , un all satisfy () and in which the total endowment of goods is

b Rl+ such that

x Rl++ . Then there is a price-list p
for every (
xi )n1 that satisfies
n
X

b
bx
bi ,
xi =
x and i : p
xi = p

(11)

i=1
n

n
b , (b
p
xi )1 is a Walrasian equilibrium of the economy E = ((ui ,
xi ))1 .
n

b , (b
i.e., p
xi )1 is a Walrasian equilibrium of the economy in which each consumer has the

utility function ui and the initial bundle

xi .
Proof:
Let (b
xi )n1 be a Pareto allocation for the given utility functions and endowment
x. We first
xi )n1 i.e., if each consumers initial bundle is
show that the conclusion holds for (
xi )n1 = (b
bi . Since (b
x
xi )n1 is a Pareto allocation, it is a solution of the Pareto maximization problem
(P-max), and it therefore satisfies the first-order marginal conditions
i, k : i , k = 0 : i uik 5 k , with equality if x
bik > 0.

(12)

Each i is the Lagrange multiplier for one of the utility-level constraints in (P-max), and
since uik > 0 for each i and each k, its clear that each constraints Lagrange multiplier must
be positive: relaxing any one of the constraints will allow u1 to be increased.
For each k, let pbk = k . Now (12) yields, for each i,
i : k : uik 5

1
pbk , with equality if x
bik > 0.
i

(13)

These are exactly the first-order marginal conditions for consumer is utility-maximization
b . Therefore, since each consumers initial bundle is x
bi , each conproblem at the price-list p
bi . And since pbk = k > 0 for each k, the constraint satisfaction
sumer is maximizing ui at x
P
bi =
first-order condition for (P-Max) yields ni=1 x
x, so the market-clearing condition for a
Walrasian equilibrium is satisfied as well. Thus, weve shown that all the conditions in the
n

b , (b
definition of Walrasian equilibrium are satisfied for p
xi )1 .
bi , but still satisfying (11), then
If each consumers initial bundle is some other
xi instead of x
the second equation in (11) ensures that each consumers budget constraint is the same as
before (the right-hand sides have the same value), and therefore the inequalities in (13) again
bi maximizes ui subject to the consumers budget constraint. And as
guarantee that each x
P
bi =
before, the constraint satisfaction first-order condition for (P-Max), ni=1 x
x, coincides
with the market-clearing condition for a Walrasian equilibrium. 

Just as with the First Welfare Theorem, the Second Theorem is true under weaker assumptions than those in () but in the case of the Second Theorem, not that much weaker:
we can dispense with the differentiability assumption, and we can weaken the convexity
assumption and the assumption that utility functions are strictly increasing.
The Second Welfare Theorem: Let (b
xi )n1 be a Pareto allocation for an economy in
which each ui is continuous, quasiconcave, and locally nonsatiated, and in which the total
b Rl+ such that
endowment of goods is
x Rl++ . Then there is a price-list p
for every (
xi )n1 that satisfies
n
X
b
bx
bi ,

xi =
x and i : p
xi = p

(14)

i=1
n

n
b , (b
p
xi )1 is a Walrasian equilibrium of the economy E = ((ui ,
xi ))1 .

The proof of the Second Theorem at this level of generality is not nearly as straightforward
as the proof of the First Theorem. The proof requires a significant investment in the theory
of convex sets, as well as some additional mathematical concepts. This is a case in which
(for this course) the cost of developing the proof outweighs its value.

Duality Theorems in Demand Theory:

Utility Maximization and Expenditure Minimization

These two Duality Theorems of demand theory tell us about the relation between utility maximization and expenditure minimization i.e., between Marshallian demand and Hicksian
(or compensated) demand. We would like to know that the utility-maximization hypothesis
ensures that any bundle a consumer chooses (if he is a price-taker) must minimize his expenditure over all the bundles that would have made him at least as well off. And conversely,
we would like to know that a bundle that minimizes expenditure to attain a given utility
level must maximize his utility among the bundles that dont cost more. The following two
examples show that the two ideas are not always the same.
Example 1: A consumer with a thick indifference curve, as in Figure 1, where there are
utility-maximizing bundles that do not minimize expenditure.
Example 2: A consumer whose wealth is so small that any reduction in it would leave
him with no affordable consumption bundles, as in Figure 2, where there are expenditureminimizing bundles that do not maximize utility.
The examples suggest assumptions that will rule out such situations.

Utility Maximization Implies Expenditure Minimization

b fails to minimize p x is because the preference
Its clear in Example 1 that the reason x
is not locally nonsatiated. If a consumers preference is locally nonsatiated, then preference
maximization does imply expenditure minimization.
First Duality Theorem: If % is a locally nonsatiated preference on a set X of consumption
b is %-maximal in the budget set {x X | p x 5 p x
b}, then x
b
bundles in R`+ , and if x
b}.
minimizes p x over the upper-contour set {x X | x % x
Proof:

(See Figure 3)

b does not minimize px on {x X | x % x

b}. Then there is a bundle x
e X such
Suppose x
e%x
b and p x
e < px
b. Let N be a neighborhood of x
e for which x N p x 5 p x
b
that x
b. Since % is locally nonsatiated,
i.e., N lies entirely below the p-hyperplane through x
e, and a fortiori x0  x
b. But then we have both
there is a bundle x0 N that satisfies x0  x
b and x0  x
b i.e., x
b does not maximize % on the budget set, a contradiction.
p x0 5 p x

6

Expenditure Minimization Implies Utility Maximization

b does not maximize utility among the bundles costing no
It seems clear that the reason x
more than it does is that there are no bundles in the consumers consumption set that cost
b. If there were such a bundle, we could move continuously along a line
strictly less than x
e costing no more than x
b but strictly better than x
b, and
from that bundle to any bundle x
b and that also is
we would have to encounter a bundle that also costs strictly less than x
b so that x
b would not maximize utility among the bundles costing
strictly better than x
b. Note, though, that in addition to having some bundle that costs less than
no more than x
b, this argument also requires both continuity of the preference and convexity of the set of
x
possible bundles. The proof of the Second Duality Theorem uses exactly this argument, so it
requires these stronger assumptions, which were not needed for the First Duality Theorem.
Can you produce counterexamples to show that the theorem does indeed require each of these
assumptions continuity and convexity in addition to the counterexample in Figure 2 in
b in the consumers set of possible consumptions?
which there is no cheaper bundle than x
Since we have to assume the preference is continuous, it will therefore be representable by
a continuous utility function, so we state and prove the theorem in terms of a continuous
utility function. We could instead do so with a continuous preference %.
Second Duality Theorem: Let u : X R be a continuous function on a convex set X of
b minimizes p x over the upper-contour set {x X | u(x) =
consumption bundles in R`+ . If x
b, then x
b maximizes u over the
u(b
x)}, and if X contains a bundle x that satisfies p x < p x
b}.
budget set {x X | p x 5 p x
Proof:

(See Figure 4)

b minimizes p x over the set {x X | u(x) = u(b

b does
Suppose that x
x)}, but that x
b} i.e., there is some x
e X such that
not maximize u over the set {x X | p x 5 p x
e 5 px
b but u(e
e such that
px
x) > u(b
x). Since u is continuous, there is a neighborhood N of x
each x N also satisfies u(x) > u(b
x). Weve furthermore assumed that there is a bundle
b. Since X is convex, the neighborhood N contains a convex
x0 X that satisfies p x0 < p x
e. Since p x
e 5 px
b and p x0 < p x
b, we have p x00 < p x
b.
combination, say x00 , of x0 and x
b does not minimize p x over the set
But since x00 N , we also have u(x00 ) > u(b
x). Thus, x
{x X | u(x) = u(b
x)}, a contradiction.

Figure 1

Figure 2

Figure 3

Figure 4