Preferences and Utility

Simon Board
This Version: October 6, 2009
First Version: October, 2008.

These lectures examine the preferences of a single agent. In Section 1 we analyse how the
agent chooses among a number of competing alternatives, investigating when preferences can be
represented by a utility function. In Section 2 we discuss two attractive properties of preferences:
monotonicity and convexity. In Section 3 we analyse the agents indifference curves and ask how
she makes tradeoffs between different goods. Finally, in Section 4 we look at some examples of
preferences, applying the insights of the earlier theory.

The Foundation of Utility Functions

1.1

A Basic Representation Theorem

Suppose an agent chooses from a set of goods X = {a, b, c, . . .}. For example, one can think of
these goods as different TV sets or cars.
Given two goods, x and y, the agent weakly prefers x over y if x is at least as good as y. To
avoid us having to write weakly prefers repeatedly, we simply write x < y. We now put some
basic structure on the agents preferences by adopting two axioms.1
Completeness Axiom: For every pair x, y X, either x < y, y < x, or both.

Department of Economics, UCLA. http://www.econ.ucla.edu/sboard/. Please email suggestions and typos

to [email protected].
1
An axiom is a foundational assumption.

Eco11, Fall 2009

Simon Board

Transitivity Axiom: For every triple x, y, z X, if x < y and y < z then x < z.
An agent has complete preferences if she can compare any two objects. An agent has transitive
preferences if her preferences are internally consistent. Lets consider some examples.
First, suppose that, given any two cars, the agent prefers the faster one. These preferences are
complete: given any two cars x and y, then either x is faster, y is faster or they have the same
speed. These preferences are also transitive: if x is faster than y and y is faster than z, then x
is faster than z.
Second, suppose that, given any two cars, the agent prefers x to y if it is both faster and bigger.
These preferences are transitive: if x is faster and bigger than y and y is faster and bigger than
z, then x is faster and bigger than z. However, these preferences are not complete: an SUV
is bigger and slower than a BMW, so it is unclear which the agent prefers. The completeness
axiom says these preferences are unreasonable: after examining the SUV and BMW, the agent
will have a preference between the two.
Third, suppose that the agent prefers a BMW over a Prius because it is faster, an SUV over a
BMW because it is bigger, and a Prius over an SUV, because it is more environmentally friendly.
In this case, the agents preferences cycle and are therefore intransitive. The transitivity axiom
says these preferences are unreasonable: if environmental concerns are so important to the
agent, then she should also take them into account when choosing between the Prius and
BMW, and the BMW and the SUV.
While it is natural to think about preferences, it is often more convenient to associate different
numbers to different goods, and have the agent choose the good with the highest number. These
numbers are called utilities. In turn, a utility function tells us the utility associated with
each good x X, and is denoted by u(x) <. We say a utility function u(x) represents an
agents preferences if
u(x) u(y) if and only if x < y

(1.1)

This means than an agent makes the same choices whether she uses her preference relation, <,
or her utility function u(x).
Theorem 1 (Utility Representation Theorem). Suppose the agents preferences, <, are complete and transitive, and that X is finite. Then there exists a utility function u(x) : X <
which represents <.

Eco11, Fall 2009

Simon Board

Theorem 1 says that if an agent has complete and transitive preferences then we can associate
these preferences with a utility function. Intuitively, the two axioms allow us to rank the goods
under consideration. For example, if there are 10 goods, then we can say the best has a utility
u(x) = 9, the second best has u(x) = 8, the third best has u(x) = 7 and so on. For a formal
proof, see Section 1.2.

1.2

A Proof of Theorem 12

The idea behind the proof is simple. For any good x, let N BT (x) = {y X|x < y} be the
goods that are no better than x. The utility of x is simply given by the number of items in
N BT (x). That is,
u(x) = |N BT (x)|.

(1.2)

If there are 10 goods, then the worst has a no better than set which is empty, so that u(x) = 0.
The second worst has a has a no better than set which has one element, so u(x) = 1. And so
on.
We now have to verify that this utility function represents the agents preferences. We do this
in two steps: first, we show that x < y implies u(x) u(y); second, we show that u(x) u(y)
implies x < y .
Step 1: Suppose x < y. Pick any z N BT (y);3 by the definition of N BT (y), we have y < z.
Since preferences are complete, we know that z is comparable to x. Transitivity then tells us
that x < z, so z N BT (x). We have therefore shown that every element of N BT (y) is also
an element of N BT (x); that is, N BT (y) N BT (x). As a result,
u(x) = |N BT (x)| |N BT (y)| = u(y)
as required.
Step 2: Suppose u(x) u(y). By completeness, we know that either x < y or y < x. Using
Step 1, it must then be the case that either N BT (y) N BT (x) or N BT (x) N BT (y), so
the no better than sets cannot partially overlap or be disjoint. By the definition of utilities
(1.2) we know that there are more elements in N BT (x) than in N BT (y), which implies that
2
3

More advanced.
z N BT (y) means that z is an element of N BT (y).

Eco11, Fall 2009

Simon Board

N BT (y) N BT (x). Completeness means that a good is weakly preferred to itself, so that
y N BT (y). Since N BT (y) N BT (x), we conclude y N BT (x). Using the definition of
the no better than set, this implies that x < y, as required.

1.3

Increasing Transformations

A number system is ordinal if we only care about the ranking of the numbers. It is cardinal
if we also care about the magnitude of the numbers. To illustrate, Usain Bolt and Richard
Johnson came 1st and 2nd in the 2008 Olympic final of the 100m sprint. The numbers 1 and
2 are ordinal: they tell us that Bolt beat Johnson, but do not tell us that he was 1% faster or
10% faster. The actual finishing times were 9.69 for Bolt and 9.89 for Johnson. These numbers
are cardinal: the ranking tells us who won, and the magnitudes tells us about the margin of
the win.
Theorem 1 is ordinal: when comparing two goods, all that matters is the ranking of the utilities;
the actual numbers themselves carry no significance. This is obvious from the construction:
when there are 10 goods, it is clearly arbitrary that we give utility 9 to the best good, 8 to the
second best, and so on. This idea can be formalised by the following result:
Theorem 2. Suppose u(x) represents the agents preferences, <, and f : < < is a strictly
increasing function. Then the new utility function v(x) = f (u(x)) also represents the agents
preferences <.

The proof of Theorem 2 is simply a rewriting of definitions. Suppose u(x) represents the agents
preferences, so that equation (1.1) holds. If x < y then u(x) u(y) and f (u(x)) f (u(y)), so
that v(x) v(y). Conversely, if v(x) v(y) then, since f () is strictly increasing, u(x) u(y)
and x < y. Hence
v(x) v(y)

if and only if x < y

and v(x) represents <.

Theorem 2 is important when solving problems. Suppose an agent has utility function
u(x) =

15
1/2
(x1

1/2

+ x2

+ 10)3

Solving the agents problem with this utility function may be be algebraically messy. Using

Eco11, Fall 2009

Simon Board

Theorem 2, we can rewrite the agents utility as

1/2

v(x) = x1

1/2

+ x2

Since u(x) and v(x) preserve the rankings of the goods, they represent the same preferences.
As a result, the agent will make the same choices with utility u(x) and v(x). This is useful
since it is much simpler to solve the agents choice problem using v(x) than u(x).
Theorem 2 is also useful for cocktail parties. For example, some people dislike the way I rank
movies of a 1-10 scale. They claim that a movie is a rich artistic experience, and cannot be
summarised by a number. However, Theorem 1 tells us that, if my preferences are complete and
transitive, then I can represent my preferences over movies by a number. Moreover, Theorem
2 tells us that I can rescale the numbers to put them on a 1-10 scale.

Choosing from Budget Sets

Theorem 1 assumes that the consumer chooses from a finite number of goods. While this is
realistic, it is more mathematically convenient to allow consumers to choose from a continuum
of goods. For example, if the agent has $10 and a hamburger costs $2, it is easier to allow the
consumer to any number between 0 and 5, rather than forcing her to choose an integer.
Suppose the choice set is given by X <n+ . A typical element is x = (x1 , . . . , xn ), where xi is
the number of the ith good the agent consumes. In order to prove a representation theorem for
this larger set of choices, we need one more (rather technical) axiom.
Continuity Axiom: Suppose x1 , x2 , x3 , . . . is a sequence of feasible choices, so that xi X
for each i, and suppose the sequence converges to x X. If xi < y for each i, then x < y.
Theorem 3 (Representation Theorem for Budget Sets). Suppose the agents preferences, <,
are complete, transitive and continuous, and that X <n+ . Then there exists a continuous
utility function u(x) : X < which represents <.

We will not prove this result. The following example examines a case where the continuity
axiom does not hold and no utility representation exists.
Suppose there are two goods and the agent has lexicographic preferences: when faced with
two bundles the agent prefers the bundle with the most of x1 ; if the two bundles have the same
5

Eco11, Fall 2009

Simon Board

x1 then she prefers the bundle with the most of x2 . To verify that this does not satisfy the
continuity axiom, consider a sequence of bundles xi = (1 + 1i , 1) which converges to x = (1, 1)
as i , and let y = (1, 2). For each i, xi is preferred to y since xi contains more of good 1.
However, in the limit, the agent prefers y to x since they have the same quantity of good 1, but
y has more of good 2. One can also show that there exists no utility function that represents
lexicographic preferences, but this is a little tricky.

Properties of Preferences

In this Section we introduce two key properties of preferences: monotonicity and convexity.
Throughout, we suppose X <n+ .
First we need a couple of definitions. If the agent weakly prefers x to y (i.e. x < y) and weakly
prefers y to x (i.e. y < x) then she is indifferent between x and y and we write x y. In
terms of utilities, an agent is indifferent between x and y if and only if u(x) = u(y).
If the agent weakly prefers x to y (x < y) and is not indifferent between x and y, then she
strictly prefers x to y and we write x y. In terms of utilities, an agent strictly prefers x to
y if and only if u(x) > u(y).

2.1

Monotonicity

Preferences are monotone if for any two bundles x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ),
xi yi for each i o
xi > yi for some i

implies x y.

In words: preferences are monotone if more of any good makes the agent strictly better off.
While monotonicity is stated in terms of preferences, we can rewrite it in terms of utilities.
Preferences are monotone if for any two bundles x and y,
xi yi for each i o

implies u(x) > u(y).

xi > yi for some i

Eco11, Fall 2009

Simon Board

As we will see, the assumption of monotonicity is very useful. It implies that indifference
curves are thin and downwards sloping. It implies that an agent will always spend her budget.
A slightly stronger version of monotonicity also rules out inflexion points in the agents utility
function which is useful when we analyse the agents utility maximisation problem.

2.2

Convexity

Preferences are convex if whenever x < y then

tx + (1 t)y < y

for all t [0, 1]

Convexity says that the agent prefers averages to extremes: if the agent is indifferent between
x and y then she prefers the average tx + (1 t)y to either x or y.
We can write this assumption in terms of utilities. Preferences are convex if whenever u(x)
u(y) then
u(tx + (1 t)y) u(y)

for all t [0, 1]

(2.1)

Slightly confusingly, a utility function that satisfies (2.1) is called quasiconcave.

The assumption of convexity if important when we analyse the consumers utility maximisation problem. Along with monotonicity, it means that any solution to the agents firstorder
conditions solve the agents problem.

Indifference Curves

An agents indifference curve is the set of bundles which yield a constant level of utility.
That is,
Indifferent Curve = {x X|u(x) = const.}
An agent has a collection of indifference curves, each one corresponding to a different level of
utility. By varying this level, we can trace out the agents entire preferences.
To illustrate, suppose an agent has utility
u(x1 , x2 ) = x1 x2
7

Eco11, Fall 2009

Simon Board

10
8
6

x2
4
2
0

10

Figure 1: Indifference Curves. This figure shows two indifference curves. Each curve depicts the
bundles that yield constant utility.

Then the indifference curve satisfies the equation x1 x2 = k. Rearranging, we can solve for x2 ,
yielding
x2 =

k
x1

(3.1)

which is the equation of a hyperbola. This function is plotted in figure 1.

In Section 3.1 we derive five important properties of indifference curves. In Section 3.2 we
introduce the idea of the marginal rate of substitution. For simplicity, we assume there are only
two goods.

3.1

Properties of Indifference Curves

We now describe five important properties of indifference curves. Throughout, we assume that
preferences satisfy completeness, transitivity and continuity, so a utility function exists. We
also assume monotonicity.
1. Indifference curves are thin. We say an indifference curve is thick if it contains two points x
and y such that xi > yi for all i. This is illustrated in figure 2. Monotonicity says that y must

Eco11, Fall 2009

Simon Board

Figure 2: A Thick Indifference Curve. This figure shows a thick indifference curve containing points
x and y.

be strictly preferred to x and therefore rules out thick indifference curves.

2. Indifference curves never cross. Suppose, by contradiction, that two indifference curves
cross, as shown in figure 3. Since they lie on the same indifference curve, the agent is indifferent
between points A and D, and indifferent between points B and C. In addition, by monotonicity,
the agent strictly prefers A to B, and strictly prefers C to D. Putting all this together,
ABCDA
We conclude that A is strictly preferred to itself, which is false. Intuitively, two indifference
curves describe the bundles that yield two different utility levels. By monotonicity, one indifference curve must always lie to the northeast of the other.
3. Indifference curves are strictly downward sloping. If an indifference curve is not strictly
downward sloping, then we can find points x and y on the same indifference curve such that
yi xi for all i, and yi > xi for some i, as shown in figure 4. This contradicts monotonicity,
which says the agent strictly prefers y to x.
4. Indifference curves are continuous, with no gaps. We cannot have gaps in the indifference
curve, as shown in figure 5. This follows from preferences being continuous which, by Theorem
3, implies that the utility function is continuous.

Eco11, Fall 2009

Simon Board

Figure 3: Indifference Curves that Cross.

Figure 4: An Upward Sloping Indifference Curve.

10

Eco11, Fall 2009

Simon Board

Figure 5: An Indifference Curve with a Gap.

5. If preferences are convex then indifference curves are convex to the origin. Suppose x and y
lie on an indifference curve. By convexity, tx + (1 t)y lies on a higher indifference curve, for
t [0, 1]. By monotonicity, this higher indifference curve lies to the northeast of the original
indifference curve. Hence the indifference curve is convex, as shown in figure 6.

3.2

Marginal Rate of Substitution

The slope of the indifference curve measures the rate at which the agent is willing to substitute
one good for another. This slope is called the marginal rate of substitution or MRS.
Mathematically,
M RS =

dx2

dx1 u(x1 ,x2 )=const.

(3.2)

We can rephrase this definition in words: the MRS equals the number of x2 the agent is willing
to give up in order to obtain one more x1 . This is shown in figure 7.
The MRS can be related to the agents utility function. First, we need to introduce the idea of
marginal utility
M Ui (x1 , x2 ) =

u(x1 , x2 )
xi

which equals the gain in utility from one extra unit of good i.
11

Eco11, Fall 2009

Simon Board

Figure 6: Convex Indifference Curves.

Figure 7: Marginal Rate of Substitution.

12

Eco11, Fall 2009

Simon Board

Let us consider the effect of a small change in the agents bundle. Totally differentiating the
utility u(x1 , x2 ) we obtain
du =

u(x1 , x2 )
u(x1 , x2 )
dx1 +
dx2
x1
x2

(3.3)

Equation (3.3) says that the agents utility increases by her marginal utility from good 1 times
the increase in good 1 plus the marginal utility from good 1 times the increase in good 2. Along
an indifference curve du = 0, so equation (3.3) becomes
u(x1 , x2 )
u(x1 , x2 )
dx1 +
dx2 = 0
x1
x2
Rearranging,

dx2
u(x1 , x2 )/x1
=
dx1
u(x1 , x2 )/x2

Equation (3.2) therefore implies that

M RS =

M U1
M U2

(3.4)

The intuition behind equation (3.4) is as follows. Using the definition of MRS, one unit of x1
is worth MRS units of x2 . That is, M U1 = M RS M U2 . Rewriting this equation we obtain
(3.4).
We can relate MRS to our earlier concepts of monotonicity and convexity. Monotonicity says
that the indifference curve is downward sloping. Using equation (3.2), this means that MRS is
positive.
Under the assumption of monotonicity, convexity says that the indifference curve is convex. This
means that the MRS decreasing in x1 along the indifference curve. Formally, an indifference
curve defines an implicit relationship between x1 and x2 ,
u(x1 , x2 (x1 )) = k
Convexity then implies that M RS(x1 , x2 (x1 )) is decreasing in x1 . This is illustrated in the next
section.
Finally, we can relate the MRS to the ordinal nature of the utility representation. In Theorem 2
we showed that the choices made under u(x) and v(x) = f (u(x)) are the same, where f : < <
is strictly increasing. One way to understand this result is through the MRS. Under utility
function u(x) the MRS is given by equation (3.4). Under utility function v(x), the MRS is
13

Eco11, Fall 2009

Simon Board

given by
M RS v =

v/x1
f 0 (u)u/x1
u/x1
= 0
=
= M RS u
v/x2
f (u)u/x2
u/x2

where the second equality uses the chain rule. This means that the agent faces the same
tradeoffs under the two utility functions, has identical indifference curves, and therefore makes
the same decisions.

3.3

Example: Symmetric Cobb Douglas

Suppose u(x1 , x2 ) = x1 x2 . We calculate the marginal rate of substitution two ways.

First, we can use equation (3.2) to derive MRS. As in equation (3.1), the equation of an
indifference curve is
x2 =

k
x1

(3.5)

Differentiating,
M RS =

k
dx2
= 2
dx1
x1

(3.6)

We can now verify preferences are convex. Differentiating (3.6) with respect to x1 ,
d
k
M RS = 2 3
dx1
x1
which is negative, as required.
Alternatively, we can use equation (3.4) to derive MRS. Differentiating the utility function
M RS =

M U1
x2
=
M U2
x1

(3.7)

We now want to express MRS purely in terms of x1 . Using (3.5) to substitute for x2 , equation
(3.7) becomes (3.6).

14

Eco11, Fall 2009

4
4.1

Simon Board

Examples of Preferences
Cobb Douglas

The CobbDouglas utility function is given by

u(x1 , x2 ) = x1 x2

for > 0, > 0

A special case if the symmetric CobbDouglas, when = . Using Theorem 2, we can then
normalise the symmetric CobbDouglas to = = 1.
The CobbDouglas indifference curve has equation x1 x2 = k. Rearranging,
/

x2 = k 1/ x1

These indifference curves look like those in figure 1.

The marginal utilities are
M U1 = x1
x2
1
M U1 = x1 x1
2
As a result the MRS is,
M RS =

4.2

M U1
x2
=
M U2
x1

Perfect Complements

Suppose an agent always consumes a hamburger patty with two slices of bread. If she has 5
patties and 15 slices of bread, then the last 5 slices are worthless. Similarly, if she has 7 patties
and 10 slices of bread, then the last 2 patties are worthless. In this case, the agents preferences
can be represented by the utility function
u(x1 , x2 ) = min{2x1 , x2 }
where x1 are patties and x2 are slices of bread. Note the 2 goes in front of the number of patties
because, intuitively speaking, each patty is twice as valuable as a piece of bread.

15

Eco11, Fall 2009

Simon Board

Figure 8: Perfect Complements. These indifference curves are Lshaped with the kink where x1 =
x2 .

In general, preferences are perfect complements when they can be represented by a utility
function of the form
u(x1 , x2 ) = min{x1 , x2 }
The resulting indifference curves are Lshaped, as shown in figure 8, with the kink along the
line x1 = x2 . Note that the indifference curve is not strictly decreasing along the bottom of
the L. This is because these preferences do not quite obey the monotonicity condition: when
the agent has 7 patties and 10 slices of bread, an extra patty does not strictly increase her
utility.
The MRS in this example is a little odd. When x1 > x2 ,
M RS =

M U1
0
= =0
M U2

M RS =

M U1

= =
M U2
0

When x1 < x2 ,

At the kink, when x1 = x2 , then MRS is not defined because the indifference curve is not
differentiable.

16

Eco11, Fall 2009

Simon Board

Figure 9: Perfect Substitutes. These indifference curves are linear with slope /.

4.3

Perfect Substitutes

Suppose an agent is buying food for a party. She wants enough food for her guests and considers
3 hamburgers to be equivalent to one pizza. Since each pizza is three times as valuable as a
hamburger, her preferences can be represented by the utility function
u(xh , xp ) = x1 + 3x2
where x1 are hamburgers and x2 are pizzas.
In general, preferences are perfect substitutes when they can be represented by a utility function
of the form
u(x1 , x2 ) = x1 + x2
The resulting indifference curves are straight lines, as shown in figure 10. As a result, preferences
are only weakly convex. The marginal rate of substitution is
M RS =

M U1

=
M U2

That is, the MRS is independent of the number of goods consumed.

17

Eco11, Fall 2009

Simon Board

Figure 10: CES Preferences. In this picture > 0 since the indifference curve intersects with the
axes.

4.4

Constant Elasticity of Substitution (CES) Preferences

CES preferences have the form

u(x1 , x2 ) =

x1 x2
+

where 6= 0 and < 1.

This utility function can approximate the above examples. As 0 the limit of the above
utility function becomes
u(x1 , x2 ) = ln x1 + ln x2
which is the same as Cobb-Douglas with equal exponents. As 1, the preferences approximate perfect substitutes. As , the preferences approximate perfect complements.
The MRS is
M RS =

M U1
x1
x1
= 11 = 21 .
M U2
x2
x1

The last expression is convenient since 1 > 0. Substituting for x2 in this equation and
differentiating, one can show that MRS is decreasing in x1 , so the preferences are convex.

18

Eco11, Fall 2009

4.5

Simon Board

Additive Preferences

Additive preferences are represented by a utility function of the form

u(x1 , x2 ) = v1 (x1 ) + v2 (x2 )
The key property of additive preferences is that the marginal utility of xi only depends on the
amount of xi consumed. As a result, the marginal rate of substitution is
M RS =

M U1
v 0 (x1 )
= 10
M U2
v2 (x2 )

For example, suppose we have

u(x1 , x2 ) = x21 + x21
Differentiating, M Ui = 2xi , so the marginal utility of each good is increasing in the amount
of the good consumed. For example, one could imagine the agent becomes addicted to either
good.
As shown in Figure 11, these preferences are concave. One can see this formally by showing
the MRS is increasing in x1 along an indifference curve. Differentiating,
M RS =

2x1
x1
M U1
=
=
M U2
2x2
x2

(4.1)

The equation of an indifference curve is x21 +x22 = k. Rearranging, x2 = (kx21 )1/2 . Substituting
into (4.1),
M RS =

x1
(k x21 )1/2

which is increasing in x1 .

4.6

Bliss Points

Suppose preferences are represented by the utility function

1
1
u(x1 , x2 ) = (x1 10)2 (x2 10)2
2
2
Figure 12 plots the resulting indifference curves which are concentric circles around the bliss
point of (x1 , x2 ) = (10, 10). These preferences violate monotonicity; as a result the indifference
19

Eco11, Fall 2009

Simon Board

Figure 11: Addiction Preferences. These indifference curves are concave.

curves are sometimes upward sloping.

The marginal rate of substitution is
M RS =

10 x1
M U1
=
M U2
10 x2

Hence the MRS is positive in the northeast and southwest quadrants, and is negative in the
northwest and southeast quadrants. From figure 12 one can also see that preferences are convex.
This is also possible to see from the MRS, but is a little tricky since monotonicity does not
hold.

4.7

Quasilinear Preferences

An agent has quasilinear preferences if they can be represented by a utility function of the form
u(x1 , x2 ) = v(x1 ) + x2
Quasilinear preferences are linear in x2 , so the marginal utility is constant. These preferences
are often used to analyse goods which constitute a small part of an agents income; good x2
can then be thought of as general consumption.

20

Eco11, Fall 2009

Simon Board

Figure 12: Bliss Point. Utility is maximised at (10, 10). Indifference curves are circles around this
bliss point.

The marginal rate of substitution equals

M RS =

M U1
v 0 (x1 )
=
= v 0 (x1 )
M U2
1

Observe that MRS only depends on x1 , and not x2 . This means that the indifference curves
are vertical parallel shifts of each other, as shown in figure 13. As a consequence, preferences
are convex if and only if v(x1 ) is a concave function, so the marginal utility of x1 decreases in
x1 .
As we will see later, quasilinear preferences have the attractive property that the consumption
of x1 is independent of the agents income (ignoring boundary constraints). This makes the
consumers problem simple to analyse and provides an easy way to calculate consumer surplus.

21

Eco11, Fall 2009

Simon Board

Figure 13: Quasilinear Preferences. These indifference curves are parallel shifts of each other.

22