Richard Katzwer

WWS 511c, F14

09/20/14

1 Utility Maximization with Two Goods

At its most basic, a utility maximization problem over goods x and y takes the form

max u (x, y)
x,y
s.t. p x x + py y  I
0  x, y

where p x , p y , I 0 are prices and income and u is utility.

In this course, we assumed “more is better”, that is you want as much x and y as you can afford. Because
of this assumption, without loss of generality we may write the constraint as an equality constraint,
p x x + p y y = I . That is, you spend your entire income. Furthermore, if I > 0 and we assume more is
better, we know that the non-negativity constraints x, y 0 will never both bind.
There are several ways to solve this:

Substitution We can use the budget constraint to make the constrained two-variable problem a one-
variable unconstrained problem. More formally, given income and prices, we may write
I px x ⇥ ⇤
y (x; I ) = , x 2 0, I /p x .
py

Then we can transform the utility function in x, y to a utility function in just x:

!
I px x
u (x, y) = u x, = ũ (x).
p x x+p y y I py

It follows that we may solve the problem

max ũ (x)
x
⇥ ⇤
s.t. x 2 0, I /p x

a univariate problem we know how to solve easily via differential calculus. It is easy to check for corner
solutions using this method!

The Method of Lagrange We reviewed this method in depth in the math review. The Lagrangian
takes the form f g
L x, y = u (x, y) p x x + py y I ,
which yields a system of three equations in three unknowns to solve

0 = u x (x, y) px
0 = u y (x, y) py
0 = p x x + py y I

Note that we ignored the constraints x, y 0. We will discuss “corner solutions” soon.

1
Straight to the Tangency Condition Notice that from the Lagrange FOCs we can write
u x (x, y) = p x , u y (x, y) = p y ,
which implies
MU x MUy
= .
px py
What does this mean? At an interior optimum, for every good, the marginal utility per dollar spent (price
of that good) is equated. Why does this hold. Suppose (wlog)
MU x MUy
>
px py
at an optimum. Then there exists some tiny amount x of x that we could buy. So we would go from
bundle !
I (x + x) p x
x, y ! x + x, .
py
Our approximate change in utility is then
x · px
u = MU x x MUy ·
py
!
u MU x MUy
= x > 0.
px px py
So we can’t be optimizing unless the tangency condition holds,

MU x MUy
= .
px py
Another way to write this is
MU x px
= ,
MUy py
that is, MRS of the indifference curve at the optimum equals the slope of the budget constraint through
that optimal bundle.

1.1 Example
Suppose that u (x, y) = x ↵ y where ↵, > 0. Substitution

I px x
Substitution From the budget constraint, y = py . So
!
↵ I px x
ũ (x) = x .
py
Then the FONC is
! ! 1 !
0 ↵ 1 I px x ↵ I px x px
ũ (x) = 0 = ↵x +x
py py py
I px x px
0 = ↵ x
py py
so
⇣ ⌘ ↵ I ⇣ ⌘ I
x ⇤ p x , py , I = , y ⇤ p x , py , I = .
↵+ px ↵+ py

2
The Method of Lagrange f g
L = x↵y p x x + py y I
which yields equations

px = ↵x ↵ 1 y
1
py = x↵y

Combining yields
px ↵y px ⇤
= =) y⇤ = x . (1)
py x ↵ py
Plugging into the budget constraint gives
px ⇤
p x x ⇤ + py x = I,
↵ py

which yields
⇣ ⌘ ↵ I ⇣ ⌘ I
x ⇤ p x , py , I = , y ⇤ p x , py , I = . Marshallian Demands
↵+ px ↵+ py

Straight to the Tangency Condition Go straight to (1)!

1.2 Beware of Corner Solutions

Everything above has assumed an interior solution, that is x ⇤ > 0 and y ⇤ > 0 at an optimum. But this is
hardly always the case.

MU x MUy
• There may be no feasible (x, y) for which px = py holds.

Example 1. Suppose u (x, y) = x + y and p x = 2 and p y = 1. Your MRS is constant and equals
p
one, but x is more expensive than y. You will only consume y! The condition MRS = pyx never
holds.
MU x MUy
• px = py may hold but in fact yield a minimum instead of a maximum.

Example 2. Suppose u (x, y) = x 2 + y 2 . Then we can solve

MU x 2x px
= =
MUy 2y py
py
and get y ⇤ = px x ⇤ . The budget constraint then yields

py ⇤
p x x ⇤ + py x = I
px
I px
x⇤ = ,
p x + p y2

But we have actually minimized u, not maximized it (check SOCs)!

3
How can we know that the interior solution is the correct? Concavity of the utility function is
important here. Put another way, convex indifference curves. Convexity means you generally prefer
mixes to extremes, you want a balance of all goods. Convexity in formal choice theory axioms is that if
(x, y) ⇡ ( x̃, ỹ), then
(↵x + (1 ↵) x̃, ↵y + (1 ↵)ỹ) (x, y), ( x̃, ỹ).
For us, with nice utility functions, it means convex indifference curves. Suppose we have a utility function
u (x, y). Fix a utility level ū and solve the equation ū = u (x, y) for y. This will give you an expression
y (x; ū) that is exactly the equation of the indifference curve with utility level ū.
Example 3. If u (x, y) = ↵ log x + log y, then

y (x; ū) = exp ū ↵ log x

is the ū-level indifference curve.

Recall that a function f is convex if f 00 > 0 everywhere. In the above example

d ↵
y (x; ū) = exp ū ↵ log x
dx x
d2 ↵2 ↵
y (x; ū) = exp ū ↵ log x + exp ū ↵ log x
dx 2 x2 x2
which is greater than zero for all x. This is the condition you should check when unsure about convexity,
d2
dx 2
y (x; ū) > 0.
Remark 4. It is often stated that convexity is the same as declining MRS. People will often write M Ux
M Uy is
declining in x, therefore the indifference curves are convex. This is incorrect! For example from above,

M Ux ↵y
=
M Uy x

which is in fact declining in x, but M Ux

M U y is a pointwise measure of slope of the indifference curve. You
have to see if MRS is declining along a fixed indifference curve, that is you must check if

M Ux
M Uy u (x,y)= ū

is declining. From the example above,

M Ux ↵y (x; ū) ↵ exp ū ↵ log x

= =
M Uy u (x,y)= ū x x

which is in fact declining in x.

MU x MUy
Condition 5. Indifference curves are convex and a solution to (x ⇤, y ⇤ ) solves px = py , then (x ⇤, y ⇤ )
is optimal bundle.
Condition 6. Indifference curves are convex and for any a 2 ( 1, 0) there exists x a 2 [0, pIx ] such that
d MU x MUy
dx
y (x; ū) = a, then there exists a bundle (x ⇤, y ⇤ ) that solves px = py and is optimal.

4
Notice that with Cobb-Douglas utility, u = x ↵ y ,

M Ux ↵y ↵ ū 1/ x ↵/ ↵ ū 1/
= = = .
M Uy u (x,y)= ū x x x 1+↵/
p 1/ p
For any slope of the budget line pyx 2 ( 1, 0) we can solve ↵xū1+↵/ = pyx , so we are guaranteed an interior
maximum regardless of prices or coefficients. The idea is that if every indifference curve has a slope than
runs the gamut from 1 to 0, we are guaranteed to find one where the tangency condition and budget
condition hold. This is an example of condition 6 in action.
We are not always so lucky though. Consider the utility function
u x, y = log x + y
which has the indifference curves y (x; ū) = ū log x. The MRS is given by
dy (x; ū) 1
= .
dx x
dy (x; ū)
We can make dx
arbitrarily large by making x arbitrarily small. Note though that we can’t make
dy (x; ū)
dx
arbitrarily small. If we are on the ū indifference curve, then y (x; ū) hits the x-axis at bundle
dy (x; ū) 1 px 1
exp ū, 0 . So along the ū indifference curve, dx
minimizes out at exp ū . So if p y > exp ū , we can use
px
the tangency condition to solve this problem for an interior solution. If exp1 ū py , the corner solution
I
x⇤ = px and y ⇤ = 0 is obtained. This is an example of condition 5 in action. [DRAW PICTURE]

1.2.1 Worked out example using Karush-Kuhn-Tucker

Note that if u x, y = log x + y, we know immediately that x > 0. The x 0 constraint cannot bind. If
it did, u = log 0 = 1. By consuming any x > 0 we can do better than 1. The maximization is
max log x + y
x,y
s.t. y 0
p x x + py y = I
The Lagrangian is f g
L = log x + y p x x + py y I + µy,
where is the Lagrange multiplier and µ 0 is the KKT multiplier. The system of equations yielded is
=
1
0 = px
x
0 = 1 py + µ
0 = µy
I = p x x + py y
Suppose that µ = 0 (that is, y 0 is non-binding). Then
1 px px
= =
x py + µ py
I = p x x + py y

5
 py
two equations, two unknowns. x = px and the budget constraint yields

py
I = px + py y = py + py y
px
I py
y = .
py

Is y 0? That is, is I p y ? If so, we have found a solution,

py I py
x⇤ = , y⇤ = .
px py

If I < p y , we have found y < 0 which cannot be correct! Therefore, µ , 0. Which immediately implies
y = 0. So x ⇤ = I p x 1 and y ⇤ = 0.
What is the significance of the condition I T y? Note that for an interior solution, we need the tangency
condition to hold,
px 1
= .
py x
Note that along our budget line, x ranges in [0, pIx ]. So we know the MRS on every indifference curve
I px
can get arbitrarily close to vertical (at x = 0), but becomes “flattest” at x = px , where M RS = I . If the
budget line is “flatter” than the indifference curve at this point, i.e.
px px
 (2)
py I
I
then we only consume x = px , y = 0, we are pushed to a corner solution. But (2) is exactly the condition
I p y from above!