4

Utility Functions

Varian, H. 2010. Intermediate Microeconomics, W.W. Norton.

Representing preferences with a single
function
The goal of this lecture is to develop our study of consumer preferences a
step further.

We want to find a simple mathematical representation that takes

preferences over bundles and assigns each bundle a single number, where
higher numbers represent more preferred bundles (higher indifference
curves).

We call this representation of preferences a utility function.

Formally, consider a utility function u(X ). This function will assign

numbers to bundles such that:

(x1, x2) > (y1, y2) if and only if u(x1, x2) > u(y1, y2)
Properties of the utility function
Note that the only important feature of the numbers a utility function
assigns to bundles is that it preserves the ranks.

Suppose we have three bundles X , Y , and Z , where

X > Y > Z.
All of the following utility functions represent the same preferences:

u1(X ) = 5 u1(Y ) = 3 u1(Z ) = 1

u2(X ) = 1000 u2(Y ) = −10 u2(Z ) = −20
u3(X ) = 0.01 u3(Y ) = 0.001 u3(Z ) = 0.0001
Properties of the utility function
Since only the ranking of the bundles matters, if we can find one function
to represent a particular set of preferences, we can find an infinite number of
other functions that do the same thing.

Any transformation of a set of numbers that preserves the order of the

numbers is called a monotonic transformation.

Typically, a transformation is achieved by applying some function f (u(X ))

that transforms each utility number, u, into another number f (u).

The transformation is monotonic if:

u1 > u2 implies f (u1) > f (u2)

𝑢
E.g. f (u) = or f (u) = u + 5
2
Monotonic transformations

Another way of saying that a transformation is (positive) monotonic is

to say that it has a positive slope.

In other words, a function f (u) will represent the same preferences as u

as long as:

𝑑ƒ(𝑢)
>0
𝑑𝑢
Monotonic transformations
Proposition: If f (·) is any monotonic transformation, and u(x1, x2) is a
utility function that represents a particular set of preferences, then f
(u(x1, x2)) is another utility function that represent the same preferences.

Proof:
•If u(x1, x2) represents a set of preferences, then by definition u(x1, x2)
> u(y1, y2) iff (x1, x2) > (y1, y2)
• And if f (u) is a monotonic transformation, then (also by definition)
u(x1, x2) > u(y1, y2) iff
f (u(x1, x2)) > f (u(y1, y2))
•Therefore, f (u(x1, x2)) > f (u(y1, y2)) iff
(x1, x2) > (y1, y2).
Thus, f (u) represents the same preferences as u.
Utility and indifference curves

Question: What is the relationship between a utility function and

indifference curves?

Answer: A utility function will assign the same number to all bundles
that lie on a single indifference curve!

Higher indifference curves will be assigned a larger number.

And any monotonic transformation of a utility function just relabels

indifference curves with different numbers while preserving the original
ordering.
Utilitarian ethics
The concept of utility was originally conceived by the English
philosopher Jeremy Bentham in the early 1800s as a way of evaluating
how well a society was governed.

His vision of society was that it should produce “the greatest happiness
for the greatest number”.

His concept (and others’since him) of utility gave importance to the

magnitude of utility, because his goal was to make interpersonal
comparisons of utility.

Why might interpersonal utility comparisons (aka cardinal utility) be

problematic?
Cardinal utility
Why might interpersonal utility comparisons (aka cardinal utility) be
problematic?

But how do we know if personA likes ice cream twice as much as person
B?

More importantly, how can you be sure that you like any bundle twice
as much as any other?

This is where Bentham’s philosophical project falters: we can’t really

compare happiness (or utility) across people. All we can say for sure is
restricted to choices that one individual makes.
Utility theory in economics
Adam Smith, another famous philosopher (but Scottish, not English)
was troubled by the following problem when he wrote his famous work
“The Wealth of Nations” in 1776:

Why are diamonds so much more expensive than water? Water is

necessary for survival, but diamonds are just pretty stones?

This is the famous Water-Diamond paradox, or the paradox of value. As

we’ll see, utility theory finally resolved this issue.
Keep it in mind as we go along.

Preview: The reason has to do with the notion of marginal utility; that
is, the additional utility gained from another unit of either good. We’ll
come back to this point later!
What kinds of preferences lead to a
utility function?
One kind of preferences that cannot be represented by a utility function
is intransitive preferences.

Suppose someone has intransitive preferences so that

X > Y > Z > X.
Then, a utility function representing those preferences would have to
output utility numbers such that
u(X ) > u(Y ) > u(Z ) > u(X ). But this is impossible!

It turns out, that given as long as preferences satisfy our Axioms

including the “Reasonable Restrictions” from Ch. 3, we can guarantee
that a utility function exists to represent those preferences.
Axioms
A1 Completeness
For all (x1, x2) and (y1, y2) in X either (x1, x2) < (y1, y2) or (y1, y2) <
(x1, x2), or both.

A2 Transitivity
For all (x1, x2), (y1, y2) and (z1, z2) in X if (x1, x2) < (y1, y2) and (y1,
y2) < (z1, z2), then (x1, x2) < (z1, z2).

A3 Reflexivity
For all (x1, x2) in X , (x1, x2) < (x1, x2)
The axioms continue
A4 Monotonicity
If (x1, x2) ≥ (y1, y2) and either x1> y1 or x2> y2, then (x1, x2) > (y1, y2)

A5 Strict Convexity
For some weight t between 0 and 1:
(t x1 + (1 − t) y1, t x2 + (1 − t) y2) > (x1, x2) ∼ (y1, y2)

A6 Continuity
If (x1, x2) > (y1, y1) then all points “close to” (x1, x2) > all points “close to” (y1, y2).
No jumps in preferences.

Proposition: For any set of preferences that satisfy A1-A6, there exists a
continuous utility function that represents those preferences.
We won’t be going over the proof. This is just another reason that we imposed
all those restrictions on preferences in Ch. 3.
 Constructing a utility function

Indifference
Curves

4
x
2

3
2

x1

Just assign any “reasonable” preferences numeric labels in increasing

order as you move away from the origin.
Graphic representation

If you have a utility function u(x1, x2) and you want to create a graphical
representation, it’s easy to draw indifference curves.

Just find all the bundles (x1, x2) such that u(x1, x2) equals some constant,
k.

Then do it again. This set of numbers, that is, all (x1, x2) such that u(x1,
x2) = k is called a level set.
Drawing indifference curves
Suppose we have a utility function u(x1, x2) = x1x2
𝑘
Each indifference curve is defined by the formula x2=
𝑥1

k=4

k=3

Indifference
Curves
k=2
x
2

k=1

x1
A second example
Now suppose we have another utility function v 𝑥&, 𝑥 ( = 𝑥&( 𝑥 ((

Then we know that:

v (𝑥&, 𝑥( ) = 𝑥&( 𝑥(( = (𝑥&, 𝑥( )2 = u(𝑥&, 𝑥( )2

Since we must have at least 0 of each good, u(𝑥&, 𝑥 ( ) cannot be

negative.

Thus, v (·) is just a monotonic transformation of u(·), and both

functions have the same shaped indifference curves!

Only the labels change.

From Indifference Curves to a Utility
Representation
Going this direction is more complicated. There are two
methods.

1. Mathematically, we can try to find a function that is

constant along the given indifference curves and assigns a
higher value to higher indifference curves.

2. Or, given choice behavior of an individual, we can try to

figure out what the consumer is trying to maximize, i.e.
what combinations of goods describes the choice behavior.
Perfect substitutes

Remember the example comparing Coca-Cola and Pepsi (or red and blue peeps)?

Since all that matters is the total number of sodas (or peeps) we can use a very simple
utility function to represent these preferences:

u(x1,x2) = x1 + x2

How can we be sure that this utility function represents preferences for perfect
substitutes?

1. The function is constant along each indifference curve.

2. The function assigns a higher number to more preferred bundles.
Perfect substitutes
We can apply any monotonic transformation (squaring, adding 5), and
it will represent the same preferences.

What happens if the rate of substitution is greater than or less than 1

to 1?
Suppose the consumer requires 2 units of x2 to make up for losing 1
unit of x1.

u(x1, x2) = 2x1 + x2

equivalently
u(x1, x2) = x1 + &x2
(
What will be the slope of this indifference curve?
Answer: −2
Perfect substitutes

More generally, preferences for perfect substitutes can be represented

by any utility function of the form:

u(x1, x2) = ax1 + bx2

Where a and b are positive and measure the “value” of goods 1 and 2
to the consumer.
The slope of the indifference curve is −a/b and indicates the rate at
which a is traded off against b.
Perfect complements
Peanut butter and jelly. In this case, all you care is how many sandwiches
you can make.

This is based on the minimum of the units of peanut butter and the units of
jelly that you have.
u(x1, x2) = min {x1, x2}

Think about it this way: how many sandwiches can you make with 3 units
of peanut butter and 3 units of jelly?

Answer: min {3, 3} = 3

Now suppose you add one more unit of jelly, does that move you to a new
indifference curve?

Answer: No, min {3, 3} = min {3, 4} = 3.

Perfect complements
What happens if you like to have twice as much peanut butter as jelly on each
sandwich?

u(x1,x2) = min {1x1, x2}

2

Think about this, if we have more than twice as much peanut butter than jelly, then
we will have some peanut butter left over, so we’ll still only get 1 x sandwiches.

And remember, we can always impose monotonic transformations, so min {x1,

2x2} represents the same preferences.
More generally, perfect complements can be represented by:

u(x1,x2) = min {ax1,bx2}

With a, b > 0 representing the proportions of each good.

Quasilinear preferences
Suppose someone has indifference curves that are vertical translations of
one another.

x
2

Indifference
Curves

We can write this form of utility function as u(x1, x2) = v (x1) + x2

Quasilinear preferences
To draw an indifference curve, set u(x1, x2) = v (x1) + x2 = k

Then, x2 = k − v (x1)

Thus, utility is linear in good 2, but it doesn’t have to be linear in good

1.

And as a result, perfect substitutes are a special case of

quasilinear preferences.

Economists like quasilinear preferences because they make a lot of the

math simpler.

Examples: u(x1, x2) = ln(x1) + x2

Cobb-Douglas preferences
This is another commonly used utility function, because it is
also relatively easy to manipulate mathematically.

u(x1, x2) = x1 x2

c = 1/2 and d = 1/2 c = 1/3 and d = 2/3

x
2

x1 x1
Cobb-Douglas preferences
These are basically the simplest of the “well-behaved” preferences we
discussed in chapter 3.

For that reason, we’re going to be using this kind of utility function a
lot.

As with other utility functions, we can represent the exact same

preferences with any monotonic transformation of the Cobb-Douglas
utility function.
Cobb-Douglas preferences
Here are two useful examples:

1. v (u(x1, x2)) = ln (xc xd ) = c ln x1 + d ln x2

• These indifference curves will look exactly the same as the Cobb-
Douglas indifference curves!

1 𝑐 𝑑
2. v (u(x1, x2)) = 𝑥$𝑐 , 𝑥2𝑑 𝑐# = 𝑥$𝑐#𝑑 𝑥2
𝑐#𝑑
𝑑

𝑐 𝑐' 𝑑
Then we define a = (so = 1) and we can rewrite:
𝑐'𝑑 𝑐'𝑑

v (u(x1, x2)) = 𝑥$𝑎 𝑥2$) 𝑎

This means we can always use a monotonic transformation to ensure that the
exponents in a Cobb-Douglas utility function sum to 1!
Marginal utility
How does a consumer’s utility as we increase the quantity of good 1
in her bundle?

The rate of change of the utility function as we add a little more of

good 1 is called the marginal utility with respect to good 1.

∆𝑈 𝑢 𝑥& + ∆𝑥&, 𝑥( − 𝑢(𝑥&, 𝑥( )

𝑀𝑈& = =
∆𝑥& ∆𝑥&

This shows how utility changes as you change the quantity of good 1,
holding the quantity of good 2 fixed.
Marginal utility
Let’s make this definition a little more precise.

When we talk about marginal, what we’re actually referring to is a

derivative.

In this case a partial derivative, of the utility function with respect to

good 1.

∆𝑈 𝑢 𝑥 1 . ∆ 𝑥 1 ,𝑥2 /𝑢(𝑥 1 ,𝑥 2 )
𝑀𝑈& = = lim ∆𝑥1 → - = 0𝑢(𝑥1,𝑥2 )
∆𝑥1 ∆𝑥1 0𝑥 1

This shows how utility changes as you change the quantity of good 1,
holding the quantity of good 2 fixed.
Marginal utility
We can do the same thing for the marginal utility w.r.t. good 2.

∆𝑈 𝑢 𝑥 1 . 𝑥 2 , .∆𝑥 2 /𝑢(𝑥 1 ,𝑥 2 )
𝑀𝑈( = = lim = 0𝑢(𝑥1,𝑥2 )
∆𝑥2 ∆𝑥 2 →- ∆𝑥2 0𝑥 2

Here we’re holding good 1 constant.

Taking monotonic transformations will change the magnitude of the

marginal utility, which means that we can’t calculate marginal
utilities by looking at choice.

But remember, all we care about is the ordering of bundles implied

by any utility function!
Marginal utility and the MRS

Remember, the MRS (marginal rate of substitution) measures the

slope of an indifference curve at any particular bundle of goods.

It tells you the rate at which a consumer will trade off good 1 and
good 2 at current consumption levels.

Note that this implies an easy method of computing the MRS.

Marginal utility and the MRS

Consider a change in consumption that keeps utility constant.

𝑀𝑈&∆𝑥&+ 𝑀𝑈( ∆𝑥( = ∆U = 0

Then, we can solve for the slope of the indifference curve.

𝑀𝑅𝑆 = ∆𝑥2 = - 𝑀𝑈1

∆𝑥1 𝑀𝑈2

As before, the MRS is negative, because to stay on the same

indifference curve, an increase in one good implies a decrease in the
other!
Marginal utility and the MRS

With calculus:

𝜕𝑢(𝑥&, 𝑥( ) 𝜕𝑢(𝑥&, 𝑥( )
𝑑𝑢 = 𝑑𝑥& + 𝑑𝑥(
𝜕𝑥& 𝜕𝑥 (

The first term measures the change in utility as we change x1, and the
second term measures the change in utility as we change x2.

Remember, we’re interested in changes such that utility is held

constant (i.e. we remain on a single indifference curve, du = 0).
Marginal utility and the MRS

𝑑𝑥 2
Now we can solve for 𝑑𝑥 1
, which is the MRS.
This is just:

𝑑𝑥( 𝜕𝑢(𝑥&, 𝑥( )/𝜕𝑥&

=
𝑑𝑥 & 𝜕𝑢(𝑥&, 𝑥( )/𝜕𝑥(

Which is directly analogous to our algebraic derivation a few slides

back.
Monotonic Transformations
Now suppose we take a monotonic transformation,
v (𝑥&, 𝑥( ) = f (u(𝑥&, 𝑥( ))

Using the chain rule, we can calculate the MRS for v (·):

MRS = - 0𝑣/0𝑥 1
= − 0ƒ ⁄0𝑢 0𝑢 ⁄0𝑥 1 = 0𝑢 ⁄0𝑥 1
0𝑣/0𝑥 2 0ƒ ⁄ 0𝑢0𝑢 ⁄ 0𝑥 2 0𝑢 ⁄ 0𝑥 2

This reiterates that the MRS is independent of the utility representation.

If two utility functions have the same MRS, then they represent the same
preferences.
Cobb-Douglas Preferences
Recall that Cobb-Douglas preferences are of the form:

𝑢 𝑥1, 𝑥2 = 𝑥 𝑐 𝑥 𝑑
& (

Then, the MRS is:

𝜕𝑢⁄𝛿𝜕
𝑀𝑅𝑆 = −
𝜕𝑢⁄𝜕𝑥2

𝑐𝑥 𝑐%1 𝑥&𝑑
= − 𝑐 𝑑%1
1
𝑑𝑥 1𝑥 &

= − 𝑐𝑥&
𝑑𝑥 1
Cobb-Douglas Preferences
We can take also compute the MRS by taking the log transformation
of the Cobb-Douglas utility function (assured that this will not alter
the underlying preferences) to get:

v 𝑥1, 𝑥2 = c ln 𝑥& + d ln 𝑥(

Here, the MRS is:

𝜕𝑢⁄𝜕𝑥&
𝑀𝑅𝑆 = −
𝜕𝑢⁄𝜕𝑥(
= − 𝑐/𝑥1
𝑑/𝑥 2
= − 𝑐𝑥 2
𝑑𝑥 1
Which is just the same as before.
In both cases, the MRS depends only on the c, d and the quantities of
the two goods currently in thparameterse bundle!
Perfect substitutes

Suppose we have the utility function 𝑢 𝑥&, 𝑥 ( = a𝑥& + 𝑏𝑥 ( .

What is the MRS?

𝜕𝑢⁄𝜕𝑥& 𝑎
𝑀𝑅𝑆 = − =−
𝜕𝑢⁄𝜕𝑥( 𝑏
Estimating utility functions
Preferences for commuting may depend on, among other things,
travel time, waiting time, explicit cost, etc...

Suppose:

U 𝑥&, , … , 𝑥𝑛 = 𝛽&𝑥& + ⋯ + 𝛽𝑛 𝑥𝑛

By observing commuter behavior, economists can infer the

average utility function in the population.

Then we can ask questions like, what happens to commuting

decisions when the price of gasoline goes up? or when the price
of transit passes goes up?
Estimating utility functions
A famous 1975 study by Thomas Domenich and Daniel
McFadden estimated just such a utility function (using powerful
statistical techniques that can find the β associated with each term
in the utility function) based on commuter data from 1967.

They got the following utility function, which fits their data very
well:

U 𝑇𝑊, 𝑇𝑇, 𝐶 = −0.147TW − 0.0411TT − 2.24C

TW = Time Walking, TT = Travel Time, and C = $ Cost.

So walking time is viewed as 3 times as costly as time in transit.

 Summary

1. Utility functions provide a convenient mathematical summary of

a preference ordering, so long as the preferences satisfy certain
axioms.
2. Utility numbers have no inherent meaning, so utility functions
that are positive monotonic transformations of one another will
all represent the same underlying preferences.
3. The marginal rate of substitution can be computed by taking
partial derivatives of the utility function:

𝜕𝑢⁄𝜕𝑥1
𝑀𝑅𝑆 = 𝑑𝑥2⁄𝑑𝑥1 = −
𝜕𝑢⁄𝜕𝑥2