Question 2

Consider a pure exchange economy with two consumers I = {i1 , i
2 } and two goods L = {a, b}.
For each consumer i ∈ I, the Marshallian demand is di p, wi = argmax ui (x) : x ∈ RL

+
and p · x ≤ wi for any price vector p ∈ RL i
++ and wealth w = p · e .
i

We can choose the normalization pb = 1 (that is, pa will be the quantity of commodity b
needed to purchase one unit of commodity a), so that p = (pa , 1).
The agents’ utility functions are

ui1 (x) = ln (xa ) + xb and ui2 (x) = xa + 2xb ,

and initial endowments are given by ei1 = (1, 0) and ei2 = (0, 1).

(a) State the KKT conditions in this setting for each agent.
The KKT conditions are two sets of equations that, if the constraint qualification of the
problem is satisfied, help us to solve the problem in a easier way. There are two conditions;
following the Lecture Notes order,
m
X
∗
KKTa: ∇f (x ) + λ∗j ∇gj (x∗ ) = 0
j=1

KKTb: λ∗j gj (x∗ ) = 0 for all j ∈ {1, . . . , m},

In this setting, we have that the agent 1’s problem is
n o
max ln(xia1 ) + xib1
xa ,xb

s.t. g1i1 (x) = wi1 − (pa xia1 + pb xib1 ) = pa − (pa xia1 + xib1 ) ≥ 0
g2i1 (x) = xia1 ≥ 0
g3i1 (x) = xib1 ≥ 0
The KKT conditions are
!
1      
i
xa1
pa 1 0
KKTa1 : − λ1 + λ2 + λ3 =0
1 1 0 1

i1

i1
λ1 (pa − pa xa − xb ) = 0

KKTb1 : λ2 xia1 = 0
λ3 xib1 = 0



The agent 2’s problem is

n o
max xia2 + 2xib2
xa ,xb

s.t. g1i2 (x) = wi2 − (pa xia2 + pb xib2 ) = 1 − (pa xia2 + xib2 ) ≥ 0
g2i2 (x) = xia2 ≥ 0
g3i2 (x) = xib2 ≥ 0
The KKT conditions are
       
1 p 1 0
KKTa2 : − λ1 a + λ2 + λ3 =0
2 1 0 1

3
 i2

i2
λ1 (1 − pa xa − xb ) = 0

KKTb2 : λ2 xia2 = 0
λ3 xib2 = 0



(b) Why are they necessary conditions for the optimization? Why are they sufficient conditions?
For the KKT conditions to be necessary for the solution of the consumer’s problem,
Slater’s Condition must hold. It states that:
If the restrictions g1 , . . . , gm are pseudo-concave and there exists a point x+ such that
g(x+ ) >> 0, then solving KKT is necessary for solving the optimization problem.
All the restrictions of both agents are affine, so they are pseudo-concave. Intuitively, x+ is
a point where no constraints are binding. In the context of the agents’ problems, it is a point
in the feasible set of baskets where they actually consume both commodities, but doesn’t use
all of their wealth. Graphically, it can be seen as follows:

xib

g3 g1
x+

g2 xia

Figure 2: Feasible Baskets and x+ – Agents 1’s and 2’s Problems

As both agents have positive wealth, we know such a point exists. Therefore, Slater’s
Condition holds and x∗ solving KKT is necessary for it to solve the consumer’s problem.
For sufficiency, we look at Arrow-Uzawa’s Condition:
If the objective function is pseudo-concave and g1 , . . . , gm are quasi-concave, then solv-
ing KKT is sufficient for solving the optimization problem.
Notice that both objective functions are positive-weighted linear combination of concave
functions, so are also concave and, therefore, pseudo-concave. Furthermore, all the restric-
tions of both agents are affine, so they are concave and, therefore, quasi-concave. Thus,
Arrow-Uzawa’s Condition holds.
Therefore, x∗ solving the corresponding KKT equations is both a necessary and a sufficient
condition for it to solve the agents’ problems.

(c) Find agent i1 ’s demand set as a function of pa .

Notice that g2 can’t be a binding restriction, as the ln function isn’t defined when xia1 = 0.
We then have two cases to consider for possible values of xia1 and xia2 . The KKT Conditions are:

1
 xia1 − λ1 pa + λ2 = 0 (1)




1 − λ1 + λ3 = 0 (2)


i i
 λ1 (pa − pa xa1 − xb ) = 0 (3)
1

λ2 xia1 = 0




 (4)
λ xi1 = 0

(5)
3 b

Case 1: xia1 > 0 and xib1 = 0: By (4), λ2 = 0. By (1), λ1 = i

1
̸= 0. By (3), xia1 = 1.
xa1 pa

4
 1
Notice that, by (2), λ3 = λ1 − 1. Given that xia1 = 1 and λ1 = i , λ3 ≥ 0 ⇐⇒ pa ≤ 1.
xa1 pa
This means that the agent consumes only xa only if pa < 1.
Case 2: xia1 > 0 and xib1 > 0: By (4) and (5), λ2 = λ3 = 0. By (1), λ1 = i
1
̸= 0. By (2),
xa1 pa
λ1 = 1, so xia1 = 1
pa . By (3), pa − pa xia1 − xib1 = 0, so xib1 = pa − 1. Notice that xib1 = pa − 1 only
if pa ≥ 1; otherwise, the third restriction would be active and xib1 = 0.
Therefore, the agent’s demand set is
(
1
, pa > 1
xia1 = pa
1, pa ≤ 1
(
pa − 1, pa > 1
xib1 =
0, pa ≤ 1

(d) Find agent i2 ’s demand set as a function of pa .

The KKT conditions are


1 − λ1 pa + λ2 = 0 (1)

2 − λ1 + λ3 = 0 (2)



λ1 (1 − pa xia2 − xib2 ) = 0 (3)

λ2 xia2 = 0 (4)




λ3 xib2 = 0

(5)


Let’s divide in cases for the possible optimum values of xia2 and xib2 :
Case 1: xia2 = 0 and xib2 = 0: By (3), λ1 = 0. By (2), λ3 = −2: a contradiction, since
KKT multipliers are non-negative. We conclude that the optimum has positive consumption
of at least one of the goods.
Case 2: xia2 = 0 and xib2 > 0: By (5), λ3 = 0. By (2), λ1 = 2 ̸= 0. By (3), 1 − xib2 = 0, so
xib2 = 1.
Notice that, by (1), 1 − 2pa + λ2 = 0. As the restriction is active, it must be that λ2 > 0,
so pa > 12 . This means that the agent consumes only xb only if pa > 12 .
Case 3: xia2 > 0 and xib2 = 0: By (4), λ2 = 0. By (1), λ1 = p1a ̸= 0. By (3), 1 − pa xia2 = 0,
so xia2 = p1a .
Notice that, by (2), 2 − p1a + λ3 = 0. As the restriction is active, it must be that λ3 > 0,
so pa < 21 . This means that the agent consumes only xa only if pa < 12 ; it makes economic
sense, as agent 2 values more xb than xa .
Case 4: xia2 > 0 and xib2 > 0: By (4) and (5), λ2 = λ3 = 0. By (1), λ1 = p1a ̸= 0. By (3),
1 − pa xia2 − xib2 = 0, so any point satisfying the budget constraint with equality is optimal.
Notice that, by (2), λ1 = 2. By (1), λ1 = p1a . Therefore, the agent consumes both goods
only if pa = 12 .
Therefore, the agent’s demand set is

1
pa < 12

 pa ,

 0,
 pa < 12
xia2 = 0, pa > 12 and xib2 = 1, pa > 12
 i2 
 1−xb , p = 1
 
1 − pa xi2 , pa = 1
pa a 2 a 2