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 agent’s indiÆerence curves and ask how
she makes tradeoÆs between diÆerent goods. Finally, in Section 4 we look at some examples of
preferences, applying the insights of the earlier theory.

1 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 diÆerent 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 agent’s preferences by adopting two axioms.1

Completeness Axiom: For every pair x, y 2 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.

1
Eco11, Fall 2009 Simon Board

Transitivity Axiom: For every triple x, y, z 2 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. Let’s 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 agent’s 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 diÆerent
numbers to diÆerent 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 2 X, and is denoted by u(x) 2 <. We say a utility function u(x) represents an
agent’s 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 agent’s preferences, <, are com-
plete and transitive, and that X is finite. Then there exists a utility function u(x) : X ! <
which represents <.

2
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 2 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 agent’s 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 2 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 2 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) µ 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) µ N BT (x) or N BT (x) µ 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
More advanced.
3
z 2 N BT (y) means that z is an element of N BT (y).

3
Eco11, Fall 2009 Simon Board

N BT (y) µ N BT (x). Completeness means that a good is weakly preferred to itself, so that
y 2 N BT (y). Since N BT (y) µ N BT (x), we conclude y 2 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 agent’s preferences, <, and f : < ! < is a strictly
increasing function. Then the new utility function v(x) = f (u(x)) also represents the agent’s
preferences <.

The proof of Theorem 2 is simply a rewriting of definitions. Suppose u(x) represents the agent’s
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

15
u(x) = ° 1/2 1/2
(x1 + x2 + 10)3

Solving the agent’s problem with this utility function may be be algebraically messy. Using

4
Eco11, Fall 2009 Simon Board

Theorem 2, we can rewrite the agent’s utility as

1/2 1/2
v(x) = x1 + 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 agent’s 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 µ <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 2 X

for each i, and suppose the sequence converges to x 2 X. If xi < y for each i, then x < y.

Theorem 3 (Representation Theorem for Budget Sets). Suppose the agent’s preferences, <,
are complete, transitive and continuous, and that X µ <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 ! 1, 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.

2 Properties of Preferences

In this Section we introduce two key properties of preferences: monotonicity and convexity.
Throughout, we suppose X µ <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 indiÆerent between x and y and we write x ª y. In
terms of utilities, an agent is indiÆerent 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 indiÆerent 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
implies x ¬ y.
xi > yi for some i

In words: preferences are monotone if more of any good makes the agent strictly better oÆ.

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

6
Eco11, Fall 2009 Simon Board

As we will see, the assumption of monotonicity is very useful. It implies that indiÆerence
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 agent’s utility
function which is useful when we analyse the agent’s utility maximisation problem.

2.2 Convexity

Preferences are convex if whenever x < y then

tx + (1 ° t)y < y for all t 2 [0, 1]

Convexity says that the agent prefers averages to extremes: if the agent is indiÆerent 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 2 [0, 1] (2.1)

Slightly confusingly, a utility function that satisfies (2.1) is called quasi–concave.

The assumption of convexity if important when we analyse the consumers utility maximisa-
tion problem. Along with monotonicity, it means that any solution to the agent’s first–order
conditions solve the agent’s problem.

3 IndiÆerence Curves

An agent’s indiÆerence curve is the set of bundles which yield a constant level of utility.
That is,
IndiÆerent Curve = {x 2 X|u(x) = const.}

An agent has a collection of indiÆerence curves, each one corresponding to a diÆerent level of
utility. By varying this level, we can trace out the agent’s entire preferences.

To illustrate, suppose an agent has utility

u(x1 , x2 ) = x1 x2

7
Eco11, Fall 2009 Simon Board

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

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

k
x2 = (3.1)
x1

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

In Section 3.1 we derive five important properties of indiÆerence 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 IndiÆerence Curves

We now describe five important properties of indiÆerence curves. Throughout, we assume that
preferences satisfy completeness, transitivity and continuity, so a utility function exists. We
also assume monotonicity.

1. IndiÆerence curves are thin. We say an indiÆerence 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

8
Eco11, Fall 2009 Simon Board

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

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

2. IndiÆerence curves never cross. Suppose, by contradiction, that two indiÆerence curves
cross, as shown in figure 3. Since they lie on the same indiÆerence curve, the agent is indiÆerent
between points A and D, and indiÆerent 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,

A¬BªC¬DªA

We conclude that A is strictly preferred to itself, which is false. Intuitively, two indiÆerence
curves describe the bundles that yield two diÆerent utility levels. By monotonicity, one indif-
ference curve must always lie to the northeast of the other.

3. IndiÆerence curves are strictly downward sloping. If an indiÆerence curve is not strictly
downward sloping, then we can find points x and y on the same indiÆerence 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. IndiÆerence curves are continuous, with no gaps. We cannot have gaps in the indiÆerence
curve, as shown in figure 5. This follows from preferences being continuous which, by Theorem
3, implies that the utility function is continuous.

9
Eco11, Fall 2009 Simon Board

Figure 3: IndiÆerence Curves that Cross.

Figure 4: An Upward Sloping IndiÆerence Curve.

10
Eco11, Fall 2009 Simon Board

Figure 5: An IndiÆerence Curve with a Gap.

5. If preferences are convex then indiÆerence curves are convex to the origin. Suppose x and y
lie on an indiÆerence curve. By convexity, tx + (1 ° t)y lies on a higher indiÆerence curve, for
t 2 [0, 1]. By monotonicity, this higher indiÆerence curve lies to the northeast of the original
indiÆerence curve. Hence the indiÆerence curve is convex, as shown in figure 6.

3.2 Marginal Rate of Substitution

The slope of the indiÆerence 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,

dx2 ØØ
M RS = ° Ø (3.2)
dx1 u(x1 ,x2 )=const.

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 agent’s utility function. First, we need to introduce the idea of
marginal utility
@u(x1 , x2 )
M Ui (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 IndiÆerence Curves.

Figure 7: Marginal Rate of Substitution.

12
Eco11, Fall 2009 Simon Board

Let us consider the eÆect of a small change in the agent’s bundle. Totally diÆerentiating the
utility u(x1 , x2 ) we obtain

@u(x1 , x2 ) @u(x1 , x2 )
du = dx1 + dx2 (3.3)
@x1 @x2

Equation (3.3) says that the agent’s 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 indiÆerence 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 U1
M RS = (3.4)
M U2

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 indiÆerence curve is downward sloping. Using equation (3.2), this means that MRS is
positive.

Under the assumption of monotonicity, convexity says that the indiÆerence curve is convex. This
means that the MRS decreasing in x1 along the indiÆerence curve. Formally, an indiÆerence
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
@v/@x1 f 0 (u)@u/@x1 @u/@x1
M RS v = = 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
tradeoÆs under the two utility functions, has identical indiÆerence 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
indiÆerence curve is

k
x2 = (3.5)
x1

DiÆerentiating,

dx2 k
M RS = ° = 2 (3.6)
dx1 x1

We can now verify preferences are convex. DiÆerentiating (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. DiÆerentiating the utility function

M U1 x2
M RS = = (3.7)
M U2 x1

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 Simon Board

4 Examples of Preferences

4.1 Cobb Douglas

The Cobb–Douglas utility function is given by

u(x1 , x2 ) = xÆ1 xØ2 for Æ > 0, Ø > 0

A special case if the symmetric Cobb–Douglas, when Æ = Ø. Using Theorem 2, we can then
normalise the symmetric Cobb–Douglas to Æ = Ø = 1.

The Cobb–Douglas indiÆerence curve has equation xÆ1 xØ2 = k. Rearranging,

°Æ/Ø
x2 = k 1/Ø x1

These indiÆerence curves look like those in figure 1.

The marginal utilities are

M U1 = ÆxÆ°1
1 xØ2
M U1 = ØxÆ1 xØ°1
2

As a result the MRS is,

M U1 Æx2
M RS = =
M U2 Øx1

4.2 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 agent’s 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 indiÆerence curves are L–shaped with the kink where Æx1 =
Øx2 .

In general, preferences are perfect complements when they can be represented by a utility
function of the form
u(x1 , x2 ) = min{Æx1 , Øx2 }

The resulting indiÆerence curves are L–shaped, as shown in figure 8, with the kink along the
line Æx1 = Øx2 . Note that the indiÆerence 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 Æx1 > Øx2 ,

M U1 0
M RS = = =0
M U2 Ø

When Æx1 < Øx2 ,

M U1 Æ
M RS = = =1
M U2 0
At the kink, when Æx1 = Øx2 , then MRS is not defined because the indiÆerence curve is not
diÆerentiable.

16
Eco11, Fall 2009 Simon Board

Figure 9: Perfect Substitutes. These indiÆerence curves are linear with slope °Æ/Ø.

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 ) = Æx1 + Øx2

The resulting indiÆerence curves are straight lines, as shown in figure 10. As a result, preferences
are only weakly convex. The marginal rate of substitution is

M U1 Æ
M RS = =
M U2 Ø

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 indiÆerence curve intersects with the
axes.

4.4 Constant Elasticity of Substitution (CES) Preferences

CES preferences have the form

x±1 x±2
u(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 approxi-
mate perfect substitutes. As ± ! °1, the preferences approximate perfect complements.

The MRS is
M U1 ±x±°1 x1°±
M RS = = 1±°1 = 21°± .
M U2 ±x2 x1
The last expression is convenient since 1 ° ± > 0. Substituting for x2 in this equation and
diÆerentiating, one can show that MRS is decreasing in x1 , so the preferences are convex.

18
Eco11, Fall 2009 Simon Board

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

For example, suppose we have

u(x1 , x2 ) = x21 + x21

DiÆerentiating, 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 indiÆerence curve. DiÆerentiating,

M U1 2x1 x1
M RS = = = (4.1)
M U2 2x2 x2

The equation of an indiÆerence curve is x21 +x22 = k. Rearranging, x2 = (k°x21 )1/2 . Substituting
into (4.1),
x1
M RS =
(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 indiÆerence curves which are concentric circles around the bliss
point of (x1 , x2 ) = (10, 10). These preferences violate monotonicity; as a result the indiÆerence

19
Eco11, Fall 2009 Simon Board

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

curves are sometimes upward sloping.

The marginal rate of substitution is

M U1 10 ° x1
M RS = =
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 agent’s 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). IndiÆerence curves are circles around this
bliss point.

The marginal rate of substitution equals

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

Observe that MRS only depends on x1 , and not x2 . This means that the indiÆerence 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 agent’s income (ignoring boundary constraints). This makes the
consumer’s problem simple to analyse and provides an easy way to calculate consumer surplus.

21
Eco11, Fall 2009 Simon Board

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

22
 Utility Maximisation Problem

Simon Board§

This Version: September 20, 2009

First Version: October, 2008.

The utility maximisation problem (UMP) considers an agent with income m who wishes to
maximise her utility. Among others, we are interested in the following questions:

• How do we determine an agent’s optimal bundle of goods?

• How do we derive an agent’s demand curve for a particular good?

• What is the eÆect of an increase in income on an agent’s consumption?

1 Model

We make several assumptions:

1. There are N goods. For much of the analysis we assume N = 2, but nothing depends on
this.

2. The agent takes prices as exogenous. We normally assume prices are linear and denote
them by {p1 , . . . , pN }.

3. Preferences satisfy completeness, transitivity and continuity. As a result, a utility func-

tion exists. We normally assume preferences also satisfy monotonicity (so indiÆerence
curves are well behaved) and convexity (so the optima can be characterised by tangency
conditions).
§
Department of Economics, UCLA. http://www.econ.ucla.edu/sboard/. Please email suggestions and typos
to [email protected].

1
Eco11, Fall 2009 Simon Board

4. The consumer is endowed with income m.

The utility maximisation problem is:

N
X
max u(x1 , . . . , xN ) subject to pi xi ∑ m (1.1)
x1 ,...,xN
i=1
xi ∏ 0 for all i

The idea is that the agent is trying to spend her income in order to maximise her utility. The
solution to this problem is called the Marshallian demand or uncompensated demand. It is
denoted by
x§i (p1 , . . . , pN , m)

The most utility the agent can attain is given by her indirect utility function. It is defined
by

N
X
v(p1 , . . . , pN , m) = max u(x1 , . . . , xN ) subject to pi xi ∑ m (1.2)
x1 ,...,xN
i=1
xi ∏ 0 for all i

Equivalently, the indirect utility function equals the utility the agent gains from her optimal
bundle,
v(p1 , . . . , pN , m) = u(x§1 , . . . , x§N ).

1.1 Example: One Good

To illustrate the problem, suppose N = 1. For example, the agent has income m and is choosing
how many cookies to consume. The agent’s utilities are given by table 1.

In general, we solve the problem in two steps. First, we determine which bundles of goods are
aÆordable. The collection of these bundles is called the budget set. Second, we find which
bundle in the budget set the agent most prefers. That is, which bundle gives the agent most
utility.

Suppose the price of the good is p1 = 1 and the agent has income m = 4. Then the agent can
aÆord up to 4 units of x1 . Given this budget set, the agent’s utility is maximised by choosing
x§1 = 4, yielding utility v = 28.

2
Eco11, Fall 2009 Simon Board

Units of x1 Utility
1 10
2 18
3 24
4 28
5 30
6 29
7 26
8 21

Table 1: Utilities from diÆerent bundles. Observe that this agent is satiated at 5 units.

Next, suppose the price of the good is p1 = 1 and the agent has income m = 8. Then the
agent can aÆord up to 8 units of x1 . Given this budget set, the agent’s utility is maximised by
choosing x§1 = 5, yielding utility v = 30. In this example, the consumer can aÆord 8 units but
chooses to consume 5. If the agent’s preferences are monotone, then she will always spend her
entire budget.

Finally, suppose the price of the good is p1 = 2 and the agent has income m = 8. Then the
agent can aÆord up to 4 units of x1 , as in the original case. This illustrates that the budget
set is determined jointly by the prices and income: doubling both does not change the agent’s
budget set. When maximising her utility, the agent once again chooses x§1 = 4.

2 Budget Sets

As in Section 1.1, we will solve the agent’s problem in two steps. First, we determine which
bundles of goods are aÆordable. Second, we find which of these bundles yields the agent the
highest utility. In this section we look at the first step.

2.1 Standard Budget Sets

In the standard model, we assume there are unit prices {p1 , p2 } for the 2 goods. The budget
set is the collection of bundles (x1 , x2 ) such that (a) the quantities are positive; and (b) the
bundle is aÆordable. Mathematically, the budget set is

{(x1 , x2 ) 2 <2+ : p1 x1 + p2 x2 ∑ m}

3
Eco11, Fall 2009 Simon Board

Figure 1: Budget Set.

where <+ is the positive part of the real line, and <2+ is the positive orthont in <2 .

Figure 1 illustrates such a budget set. The equation where the budget binds is given by

p1 x1 + p2 x2 = m (2.1)

We can rearrange this to be in the form of a standard linear equation

m p1
x2 = ° x1 (2.2)
p2 p2

Hence the budget line is linear with intercept m/p2 and slope °p1 /p2 . Crucially, the slope only
depends on the relative prices.

The two endpoints are easy to calculate. If the agent spends all her money on x1 she can aÆord

m
x1 = and x2 = 0
p1

If the agent spends all her money on x2 she can aÆord

m
x1 = 0 and x2 =
p2

Figure 2 shows that an increase in the agent’s income leads the budget line to make a parallel

4
Eco11, Fall 2009 Simon Board

Figure 2: An Increase in Income.

shift outwards. Mathematically, this can be seen from equation (2.2). Intuitively, if the agent’s
budget doubles then she can double her consumption of both goods. Since relative prices do
not change, the new budget line is parallel to the old one.

Figure 3 shows that an increase in p1 leads the budget curve to pivot around it’s left endpoint.
Mathematically, this can be seen from equation (2.2). Intuitively, if the agent only buys x2 ,
then her purchasing power is unaÆected by the increase in p1 . As a result, the left endpoint does
not move. If the agent only buys x1 , then the increase in p1 reduces the amount she can buy,
forcing the right endpoint to shift in. As a result, the budget line become steeper, reflecting
the change in the relative prices.

2.2 Nonlinear Budget Sets

While we focus on linear budget constraints, agents often face nonlinear prices. Here we present
some examples.

Figure 4 shows an example of quantity discounts. In this example, the agent has income m = 30.
Good 1 has per–unit price p1 = 2 for x1 < 10, and per–unit price p1 = 1 for x1 ∏ 10. Good 2
has a constant price, p2 = 2. Lets consider 2 cases. First, when the agent buys x1 < 10, the
price of good 1 is p1 = 2 and the equation of the budget line is therefore 2x1 + 2x2 = 30 or

5
Eco11, Fall 2009 Simon Board

Figure 3: An Increase in the Price of Good 1.

x2 = 15 ° x1 . For example, when the agent spends all her money on good 2, she can aÆord
x2 = 15. Second, when x1 ∏ 10 the agent spends $20 on the first 10 units of x1 and $1 per unit
thereafter. Hence her budget constraint is

20 + (x1 ° 10) + 2x2 = 30

Figure 5 shows an example of rationing. In this example, the agent has income m = 30. Good
1 has per–unit price p1 = 2 for x1 ∑ 10, but she is only allowed to purchase 10 units. Good 2
has a constant price, p2 = 2. When the agent buys x1 ∑ 10, the price of good 1 is p1 = 2 and
the budget line is 2x1 + 2x2 = 30. For example, when the agent spends all her money on good
2, she can aÆord x2 = 15. The agent is unable to buy more than 10 units of x1 , so the budget
set is cut oÆ at x1 = 10.

Exercise: Excess tax on x1 . Suppose m = 30, p2 = 2, and p1 = 2 for the first 10 units and
p1 = 3 for each additional unit. Draw the agent’s budget set.

Exercise: Food stamps. Suppose m = 30, p2 = 2, and p1 = 0 for the first 10 units and p1 = 2
for each additional unit. Draw the agent’s budget set.

6
Eco11, Fall 2009 Simon Board

Figure 4: Quantity Discounts. The dark line shows the agent’s budget set with p1 = 1 for x1 ∏ 10.
The dotted line shows her budget set if p1 = 2 for all units.

Figure 5: Rationing. The dark line shows the agent’s actual budget set given she can only buy 10
units of x1 . The dotted line shows her budget set without rationing.

7
Eco11, Fall 2009 Simon Board

3 Solving the Utility Maximisation Problem

In this section we solve the agent’s utility maximisation problem. We make a number of
simplifying assumptions which we explore in Section 4. In particular, we assume:

• The agents utility function is diÆerentiable. As a result there are no kinks in the indiÆer-
ence curve.

• The agent’s preferences are monotone. As a result, she spends her entire budget.1

• The agent’s preferences are convex. As a result, any solution to the tangency conditions
constitute a maximum.

• There is an interior solution to the agent’s maximisation problem.

3.1 Solution Method 1: Graphical Approach

The agent wishes to choose a point in her budget set to maximise her utility. That is, the agent
wishes to choose a point in her budget set that lies on the highest indiÆerence curve.2

Figure 6 characterises the agent’s optimal choice. Graphically, one can imagine the indiÆerence
curve flying in from the top right corner (where utility is highest) and stopping when it touches
the budget set.

To understand this further, consider figure 7. There are 3 indiÆerence curves. I1 yields the
highest utility, but never intersects with the budget set. I2 is corresponds to the agent’s optimal
choice (point A). I3 yields a lower level of utility which is attainable but not desirable.

At the optimal point, the budget line is tangential to the indiÆerence curve. As a result the
budget line and the indiÆerence curve have the same slope. This tangency condition means
that

p1
MRS(x§1 , x§2 ) = (3.1)
p2
1
We actually assume @u(x1 , x2 )/@xi > 0 for each i. See Section 4.3.
2
Recall monotonicity and convexity implies that indiÆerence curves are thin, downward sloping and convex.

8
Eco11, Fall 2009 Simon Board

Figure 6: The Agent’s Optimal Demand. This figure shows how the agent’s optimal demand is
characterised by the tangency condition.

Figure 7: Understanding the Tangency Condition. In this figure I2 is the highest attainable
indiÆerence curve. I3 is higher but unaÆordable; I1 is aÆordable but not optimal.

9
Eco11, Fall 2009 Simon Board

where the marginal rate of substitution is evaluated at the optimal choice, (x§1 , x§2 ). Since
M RS = M U1 /M U2 ,3 we can write the tangency condition as

M U1 (x§1 , x§2 ) p1
= (3.2)
M U2 (x1 , x2 )
§ § p2

The intuition for this result is as follows. The MRS equals the number of x2 the agent is willing
to give up to get one more unit of x1 . The price ratio equals the number of x2 the agent
has to give up in order to get one more unit of x1 if she wishes to stay within her budget. If
M RS > p1 /p2 , then the agent is more willing to give up x2 than the market requires, so she
can increase her utility by consuming less x2 and more x1 . If M RS < p1 /p2 , then the agent is
less willing to give up x2 than the market requires, so she can increase her utility by consuming
more x2 and less x1 .

One can rewrite the tangency condition as:

M U1 (x§1 , x§2 ) M U2 (x§1 , x§2 )

= (3.3)
p1 p2

Equation (3.3) says that the agent equalises the marginal utility per dollar, or the bang–per–
buck of the two goods. If the bang–per–buck from good 1 is higher than that from good 2,
then the agent buys more of good 1. If the bang–per–buck from good 1 is lower than that from
good 2, then the agent buys less of good 1. At the optimal choice, the bang–per–buck of the
two goods is equal.

3.2 Example: Symmetric Cobb–Douglas.

Suppose u(x1 , x2 ) = x1 x2 . The MRS is

M U1 x2
=
M U2 x1

The tangency condition (3.1) therefore becomes

x2 p1
=
x1 p2
3
Recall M Ui (x1 , x2 ) = @u(x1 , x2 )/@xi .

10
Eco11, Fall 2009 Simon Board

Rewriting, we see that

p1 x1 = p2 x2 (3.4)

which means that the agent’s spends the same money on each good. This is a special property
of the symmetric Cobb–Douglas model and helps make it so tractable. The budget constraint
states that

p1 x1 + p2 x2 = m

Substituting equation (3.4) into the budget constraint implies that p1 x1 = p2 x2 = m/2, so the
agent spends half her income on each good. As a result, the Marshallian demands are

m m
x§1 (p1 , p2 , m) = and x§2 (p1 , p2 , m) = (3.5)
2p1 2p2

We can also calculate the agent’s indirect utility, her utility from the optimal bundle. Using
the demands in equation (3.5), we have
µ ∂µ ∂
m m m2
v(p1 , p2 , m) = u(x§1 , x§2 ) = x§1 x§2 = =
2p1 2p2 4p1 p2

3.3 Solution Method 2: Lagrangian Approach

A second way to solve the agent’s utility maximisation problem is to use a Lagrangian. This
approach is equivalent to the tangency approach but can be more convenient, especially with
complex problems.

At the optimal solution, equation (3.3) tells us that the agent equalises the bang–per–buck from
each good. Let ∏ equal the marginal utility the agent derives from $1 at her optimal bundle.
We then have

M U1 (x§1 , x§2 )
∏= (3.6)
p1
M U2 (x§1 , x§2 )
∏= (3.7)
p2

We can encode equations (3.6) and (3.7) as the first–order conditions of one single equation.

11
Eco11, Fall 2009 Simon Board

Consider maximising the Lagrangian,

L = u(x1 , x2 ) + ∏[m ° p1 x1 ° p2 x2 ] (3.8)

In this equation ∏ is called a Lagrange multiplier. Intuitively, we are maximising the agent’s
utility plus a penalty term which punishes the agent for exceeding her budget (if m ° p1 x1 °
p2 x2 < 0). If the penalty ∏ is too low the agent will spend more than her income; if the penalty
∏ is too high the agent will spend less than her income. We must therefore choose the penalty
∏ so the agent exactly spends her budget.

Mechanically, we solve the problem as follows. First, we derive the first–order–conditions of the
Lagrangian (3.8). This yields:

@L @u(x1 , x2 )
= ° ∏p1 = 0 (3.9)
@x1 @x1
@L @u(x1 , x2 )
= ° ∏p2 = 0 (3.10)
@x2 @x2

We also know the agent’s budget constraint binds:

p1 x1 + p2 x2 = m (3.11)

We now have three unknowns (x1 , x2 , ∏) and three equations: (3.9), (3.10) and (3.11). We can
therefore solve for the agent’s optimal demands.

We can relate these results to those in Section 3.1. Rearranging, equations (3.9) and (3.10) yield
equations (3.6) and (3.7), which means we can interpret the optimal ∏ as the “bang–per–buck”
of the optimal bundle. In addition, dividing (3.9) by (3.10) yields

@u(x1 , x2 )/@x1 p1
=
@u(x1 , x2 )/@x2 p2

which is the tangency condition (3.2) from Section 3.1.

3.4 Solution Method 3: Substitution

The agent’s problem is to maximise her utility subject to her budget constraint. When there are
two goods one can solve for the agent’s optimal bundle by substituting the budget constraint
directly into the objective function.

12
Eco11, Fall 2009 Simon Board

Recalling equation (2.2), the budget constraint can be written as

m p1
x2 = ° x1
p2 p2

Substituting this into the agent’s utility function, she chooses x1 to maximise
µ ∂
m p1
u x1 , ° x1
p2 p2

The first–order condition is

µ ∂
@u(x1 , x2 ) @u(x1 , x2 ) p1
+ ° =0
@x1 @x1 p2

Rearranging,
@u(x1 , x2 )/@x1 p1
=
@u(x1 , x2 )/@x2 p2
which is the tangency condition (3.2) from Section 3.1. Using the budget constraint, we can
therefore solve the for agent’s optimal bundle (x§1 , x§2 ).

3.5 Many goods

The tangency condition (3.3) extends to many goods. In this case, the agent equates the
bang–per–buck from each of the N goods. That is,

M U1 M U2 M UN
= = ... =
p1 p2 pN

One can then derive the agent’s optimal demand using these (N ° 1) equations and the budget
constraint.

Equivalently, one can set up the Lagrangian

L = u(x1 , x2 , . . . , xN ) + ∏[m ° p1 x1 ° p2 x2 ° . . . ° pN xN ]

One can then solve for the agent’s optimal demand and the Lagrange multiplier using the N
first–order–conditions and the budget constraint.

13
Eco11, Fall 2009 Simon Board

4 Problems with the UMP

In this Section we investigate the tangency conditions in more depth. In Section 4.1 we state a
formal version of the theorem we were implicitly using in Section 3. We then look at the four
assumptions we made at the start of the last section: kinks, the monotonicity of preferences,
the convexity of preferences, and boundary problems.

4.1 Kuhn–Tucker Conditions

Suppose the agent is solving the problem:

max u(x1 , x2 ) subject to p1 x1 + p2 x2 ∑ m (4.1)

x1 ,x2
x1 ∏ 0
x2 ∏ 0

Form the Lagrangian

L = u(x1 , x2 ) + ∏[m ° p1 x1 ° p2 x2 ] + µ1 x1 + µ2 x2 (4.2)

where the last two terms are the penalties associated with constraints x1 ∏ 0 and x2 ∏ 0.

Suppose u(x1 , x2 ) is continuously diÆerentiable and (x§1 , x§2 ) solves (4.1). Then four Kuhn–
Tucker conditions hold:

(a) The first order conditions hold: @L

@x1 = 0 and @L
@x2 = 0.

(b) The Lagrange multipliers are positive: ∏, µ1 , µ2 ∏ 0.

(c) The constraints hold: m ° p1 x1 ° p2 x2 ∑ 0, x1 ∏ 0 and x2 ∏ 0.

(d) Complimentary slackness holds: ∏[m ° p1 x1 ° p2 x2 ] = 0, µ1 x1 = 0 and µ2 x2 = 0.

The idea behind these conditions is exactly the same as in Section 3.3. Part (a) say that the
agent is choosing (x1 , x2 ) to maximise her utility plus the penalty functions. Part (b) says
that the penalties are positive. Part (c) says that the agent’s choice must be feasible. Part (d)
says that we cannot have a constraint slack and have the associated Lagrange multiplier being
positive. For example, consider the budget constraint and recall that the Lagrange multiplier

14
Eco11, Fall 2009 Simon Board

can be interpreted as the bang–per–buck. Complimentary slackness says that if the budget
constraint at the optimal solution is slack, then the bang–per–buck equals zero and hence
∏ = 0.

We now present the formal result.

Theorem 1. Suppose the utility function is continuously diÆerentiable.

(i) Any optimal solution satisfies the conditions (a)–(d).
(ii) If preferences are convex and M Ui (x1 , x2 ) > 0 for each i, then the solutions to (a)–(d) are
optimal.

In Section 4.2 we look at the issue of kinks, where the optimal solution may not satisfy the
Kuhn–Tucker conditions. In Sections 4.3–4.4 we investigate what happens if monotonicity and
convexity do not hold. In this case the there may be multiple solutions to the Kuhn–Tucker
conditions, some of which are not optimal. Finally, in Section 4.5 we look at the issue of
boundary constraints, where the nonnegativity constraints may bind.

4.2 Kinks

When deriving the tangency condition, we assumed that the utility function (and hence the
indiÆerence curve) is diÆerentiable.

Suppose there is a kink in the indiÆerence curve along the as shown in figure 8. At the kink,
the MRS to the left and right are diÆerent. Let the MRS to the left be denoted M RS L and
than to the right be denoted M RS R . Then the solution is at the kink if

p1
M RS L (x§1 , x§2 ) ∏ ∏ M RS R (x§1 , x§2 ) (4.3)
p2

Noting that M RS = M U1 /M U2 , equation (4.3) says that to the left of (x§1 , x§2 ) we have

M U1 M U2
∏
p1 p2

so the agent wishes to increase x1 . While to the right of (x§1 , x§2 ) we have

M U1 M U2
∑
p1 p2

so the agent wishes to decrease x1 . Putting this together, when x1 < x§1 , the agent wishes to

15
Eco11, Fall 2009 Simon Board

Figure 8: IndiÆerence Curve with Kinks. The agent chooses point A. At the optimum, M RS L ∏
p1 /p2 ∏ M RS R .

buy more x1 ; when x1 > x§1 , the agent wishes to buy less x1 . Hence the optimal solution is at
x1 = x§1 .

One way to approach this is to think of the MRS at the kink as being an entire set of numbers
[M RS R , M RS L ]. The tangency condition then says that the solution is at the kink if p1 /p2
falls in the set.

4.2.1 Example: Perfect Complements

Suppose u(x1 , x2 ) = min{Æx1 , Øx2 }. In this case utility is not diÆerentiable along the line
Æx1 = Øx2 , as shown in figure 9. The agent’s optimal bundle clearly has the property that
Æx1 = Øx2 . One can then use the agent’s budget constraint to solve for the optimal bundle.

To illustrate, suppose an agent’s utility is

u(x1 , x2 ) = min{2x1 , x2 }

The agent’s income is m = 30 and prices are p1 = 1 and p2 = 1. At the optimal bundle,
x1 = 2x2 . Using the budget constraint we thus have x§1 = 10 and x§2 = 10.

16
Eco11, Fall 2009 Simon Board

Figure 9: Perfect Complements. With perfect complements there is a kink in the indiÆerence curve
where Æx1 = Øx2 . The agent maximises her utility at the kink, at point A.

4.3 Monotonicity

When deriving the tangency conditions, we assumed that each good has a strictly positive
marginal utility. This is an attractive property for two reasons.

First, if preferences are monotone then indiÆerence curves are thin and downward sloping. From
the perspective of the utility maximisation problem, monotonicity also ensures the agent spends
her entire budget. In Section 4.3.1 we look at an example where monotonicity fails and the
agent does not always wish to spend her budget.

Second, the fact that marginal utilities are strictly positive implies that the Kuhn–Tucker
conditions pick out the optimal solution. In Section 4.3.2 we look at an example where there
are two solutions to the Kuhn–Tucker conditions, only one of which is optimal.

4.3.1 Example: Bliss Point

Suppose u(x1 , x2 ) = ° 12 (x1 ° 10)2 ° 12 (x2 ° 10)2 . Figure 10 plots the corresponding indiÆerence
curves which are concentric circles around the bliss point of (x1 , x2 ) = (10, 10).

Suppose prices are p1 = 1 and p2 = 1 and the agent has income m = 10. At this point, the

17
Eco11, Fall 2009 Simon Board

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

agent cannot aÆord her bliss point (which costs $20), so her budget constraint will bind. The
price ratio is p1 /p2 = 1, so the tangency condition implies

M U1 °(x1 ° 10)
= =1
M U2 °(x2 ° 10)

Rearranging, we see that x1 = x2 . The budget constraint states that x1 + x2 = 10. Hence we
have x§1 = 5 and x§2 = 5.

Next, suppose prices are p1 = 1 and p2 = 1 and the agent has income m = 30. In this case the
agent can aÆord her bliss point and will buy x§1 = 10 and x§2 = 10.

One can derive the same results from the Kuhn–Tucker conditions. First, suppose that the
budget constraint binds. Ignoring boundary constraints (which are not an issue here), the
FOCs of the Lagrangian are

°(x1 ° 10) ° ∏ = 0 (4.4)

°(x2 ° 10) ° ∏ = 0 (4.5)

As above, the FOCs imply x1 = x2 . If m = 10, then the budget constraint implies that
(x§1 , x§2 ) = (5, 5), and we are done. If m = 30, then the budget constraint implies that (x§1 , x§2 ) =
(15, 15), which we know to be wrong. Substituting back into (4.4) or (4.5), we find ∏ = °5,

18
Eco11, Fall 2009 Simon Board

Figure 11: Inflexion. The utility function is strictly increasing but has an inflexion at x1 = 2. As a
result, there are two points that satisfy the Kuhn–Tucker conditions, only one of which is optimal.

which breaks part (c) of the Kuhn–Tucker conditions. Economically, this result means that
the bang–per–buck is negative, so the solution is clearly not optimal. We thus know that the
budget constraint does not bind, so complimentary slackness implies ∏ = 0. The agent thus
maximises
L = °(x1 ° 10)2 ° (x2 ° 10)2

which yields (x§1 , x§2 ) = (10, 10).

4.3.2 Example: Inflexion

Suppose there is one good, x1 . The utility of the agent is given by u(x1 ) = (x1 ° 2)3 + 8. The
agent has income m = 4 and faces prices p1 = 1, so her budget states that x1 ∑ 4.

Figure 11 plots the utility function. Since this is increasing the optimal consumption is clearly
x§1 = 4.

Now consider the Lagrangian:

L = (x1 ° 2)3 + 8 + ∏[m ° p1 x1 ° p2 x2 ]

19
Eco11, Fall 2009 Simon Board

The FOC is

3(x1 ° 2)2 = ∏ (4.6)

There are two solutions to the Kuhn–Tucker conditions. First, suppose ∏ > 0. By complimen-
tary slackness, the agent spends her budget. Hence x§1 = 4. Plugging back into (4.6) we see
that
∏ = 3(x1 ° 2)2 = 12 > 0

as initially assumed. Hence this solution satisfies the Kuhn–Tucker conditions.

Second, suppose ∏ = 0. Equation (4.6) implies that x§1 = 2, and the budget constraint is slack.
Again, one can verify this “solution” satisfies the Kuhn–Tucker conditions, even though is is
clearly wrong.

The problem is that the utility function has a inflexion at x1 = 2. When the derivative fails to
be strictly positive like this, there may be multiple solutions to the Kuhn–Tucker conditions.
By part (a) of Theorem 1, one of these is the real solution, but one has to individually check
which one.

4.4 Convexity

If preferences fail to be convex, then the solution to the tangency condition may characterise a
local maximum or, even worse, a local minimum.

Figure 12 illustrates the problem. Points A, B and C all satisfy the tangency conditions. Point
A is the global maximum; point B is a local maximum; point C is a local minimum.

4.4.1 Example: Addiction

Suppose an agent has utility u(x1 , x2 ) = 12 x21 + 12 x22 . DiÆerentiating, M Ui = xi , 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 13, these preferences are concave. One can see this formally by showing

20
Eco11, Fall 2009 Simon Board

Figure 12: Failure of Convexity. Points A, B and C satisfy the tangency condition.

the MRS is increasing in x1 along an indiÆerence curve. DiÆerentiating,

M U1 x1
M RS = = (4.7)
M U2 x2

The equation of an indiÆerence curve is x21 +x22 = k. Rearranging, x2 = (k°x21 )1/2 . Substituting
into (4.7),
x1
M RS =
(k ° x21 )1/2
which is increasing in x1 .

With this addiction model, the tangency condition yields

x1 p1
=
x2 p2

If m = 10, p1 = 1 and p2 = 1, then one would wrongly conclude that (x§1 , x§2 ) = (5, 5). However,
as we can see from figure 13, this is a local minimum.

Looking at figure 13, one can see that the agent’s optimal bundle is on the boundary. Intuitively,
if the agent becomes addicted, then she wishes to consume only one good. The left and right
endpoints of the budget line are (x1 , x2 ) = (m/p1 , 0) and (x1 , x2 ) = (0, m/p2 ) respectively.
Comparing these two points, we see that (x§1 , x§2 ) = (m/p1 , 0) if p1 ∑ p2 and (x§1 , x§2 ) = (0, m/p2 )
if p1 ∏ p2 .

21
Eco11, Fall 2009 Simon Board

Figure 13: Addictive preferences. Point A satisfies the tangency condition but is not optimal. The
agent’s optimal bundle is point B.

4.5 Boundary Solutions

In Section 3 we assumed that the solution to the agent’s UMP is internal. In practice, there
are many goods that a typical person chooses not to buy.

Figure 14 provides an illustration. In this example,

p1
M RS(x1 , x2 ) > (4.8)
p2

for all internal (x1 , x2 ). Since the indiÆerence curve is always steeper than the budget line, the
agent’s optimal bundle is (x§1 , x§2 ) = (m/p1 , 0). The intuition behind this is straightforward:
rearranging, (4.8) we see that
M U1 M U2
>
p1 p2
which means that the bang–per–buck from x1 is always bigger than that from x2 . As a result
the agent consumes only x1 .

There are two ways to solve problems where boundary constraints may bind. First, one can
insert Lagrange multipliers for the boundary constraints as in equation (4.2). One can then use
the Kuhn–Tucker conditions to derive the agent’s optimal bundle.

22
Eco11, Fall 2009 Simon Board

Figure 14: Boundary Solutions. In this figure the indiÆerence curves are always steeper than the
budget line. As a result, the solution is at point A, on the boundary.

Second, if preferences are convex, one can simply ignore the boundary constraints. If one finds
that x§i < 0, then set x§i = 0 and resolve.

For a certain class of preferences, boundary problems will never be an issue. An indiÆerence
curve implicitly defines x2 as a function of x1 . Let this function be denoted x2 (x1 ), so that
u(x1 , x2 (x1 )) = k. The Inada conditions state that:

lim M RS(x1 , x2 (x1 )) = 1 and lim M RS(x1 , x2 (x1 )) = 0 (4.9)

x1 !0 x1 !1

Under these assumptions, the agent places a huge value on the first unit of both goods, and so
will consume a positive amount of each. Notably, these Inada conditions are satisfied by Cobb
Douglas preferences, u(x1 , x2 ) = xÆ1 xØ2 , where

M U1 ÆxÆ°1 xØ2 Æx2

M RS = = 1
Ø°1
= (4.10)
M U2 Æ
Øx1 x2 Øx1

°Æ/Ø
Along an indiÆerence curve xÆ1 xØ2 = k, so x2 = k 1/Ø x1 and substituting into (4.10),

Æx2 Æ °(Æ+Ø)/Ø
M RS = = k 1/Ø x1
Øx1 Ø

which satisfies the Inada conditions (4.9).

23
Eco11, Fall 2009 Simon Board

Figure 15: Perfect Substitutes. The slope of the agent’s indiÆerence curve is °Æ/Ø, while the slope
of the agent’s budget line is °p1 /p2 . In this figure Æ/Ø < p1 /p2 , so the bang–per–buck is higher from
good 2 than good 1 and the agent chooses point A on the boundary.

4.5.1 Example: Perfect Substitutes

Suppose an agent has a utility function

U (x1 , x2 ) = Æx1 + Øx2

The indiÆerence curves are straight lines with slope °Æ/Ø. The budget line has slope °p1 /p2 .
As shown in figure 15, the agent’s optimal choice will occur at one of the endpoints. If

Æ Ø
>
p1 p2

then the bang–per–buck from good 1 exceeds that from good 2, so the agent’s optimal bundle
is (x§1 , x§2 ) = (m/p1 , 0). If
Æ Ø
<
p1 p2
then the bang–per–buck from good 2 exceeds that from good 1, so the agent’s optimal bundle
is (x§1 , x§2 ) = (0, m/p2 ). If
Æ Ø
=
p1 p2
then the agent is indiÆerent between all points on the budget line.

24
Eco11, Fall 2009 Simon Board

5 General Properties of Solutions

It is useful to know some general properties of Marshallian demand (1.1) and indirect utility
(1.2).

The Marshallian demand is homogenous of degree zero in prices and income. That is,

x§i (p1 , p2 , m) = x§i (Æp1 , Æp2 , Æm) for Æ > 0

This means that a doubling of prices and income has no aÆect on the agent’s demand. In-
tuitively, the agent buys the same goods whether the currency is denominated in Euros or
Dollars.

The indirect utility function has three attractive properties.

1. The indirect utility function is homogenous of degree zero in prices and income. That is,

v(p1 , p2 , m) = v(Æp1 , Æp2 , Æm) for Æ > 0

The agent’s utility depends on what she buys, which is unaÆected by the form of the currency
(see above).

2. The indirect utility function is increasing in income and decreasing in prices. An increase
in the agent’s income expands the budget set and thereby increases the agent’s utility from her
most preferred choice. Conversely, an increase in a price contracts the agent’s budget set and
thereby decreases the agent’s utility from her most preferred choice.

3. Roy’s Identity: 4

@v(p1 , p2 , m) @v(p1 , p2 , m)
= °x§i (p1 , p2 , m) (5.1)
@pi @m

Suppose p1 increases by 1¢. Then there is a direct and indirect eÆect on the agent’s utility. The
direct eÆect is that, holding demand constant, the agent’s eÆective income falls by x§1 £ 1¢. The
indirect eÆect is that as relative prices change, the agent rebalances her optimal choice. This
indirect eÆect, however, is small since the agent’s initial choice was optimal under the initial
price, so is almost optimal under the new price. Putting this together, we see that the eÆect of
a 1¢ price rise on utility equals x§1 times the eÆect of a 1¢ drop in income. This is exactly what
4
Advanced.

25
Eco11, Fall 2009 Simon Board

equation (5.1) says.

Roy’s identity is useful since it enables us to calculate the agent’s Marshallian demand from
her indirect utility function. To illustrate this result, consider the symmetric Cobb–Douglas
model. The agent has utility u(x1 , x2 ) = x1 x2 , income m and faces prices p1 and p2 . Using the
results in Section 3.2 we know that the agent has indirect utility

m2
v(p1 , p2 , m) =
4p1 p2

Using Roy’s identity (5.1), we therefore have,

∑ ∏∑ ∏
@v(p1 , p2 , m) @v(p1 , p2 , m) °1
x1 § (p1 , p2 , m) = °
@pi @m
∑ ∏∑ ∏°1
m2 2m
=° ° 2
4p1 p2 4p1 p2
m
=
2p1

which is the optimal demand we found in Section 3.2.

The formal proof is in two steps. First, we show that ∏ measures the marginal utility of income,
i.e. the bang–per buck. Observe that an agent’s indirect utility is defined by

v(p1 , p2 , m) = u(x§1 (p1 , p2 , m), x§2 (p1 , p2 , m)) (5.2)

DiÆerentiating with respect to m,

@v(p1 , p2 , m) @u(x§1 , x§2 ) @x§1 (p1 , p2 , m) @u(x§1 , x§2 ) @x§2 (p1 , p2 , m)

= +
@m @x1 @m @x2 @m
@x§1 (p1 , p2 , m) @x§2 (p1 , p2 , m)
= ∏p1 + ∏p2 (5.3)
@m @m

where the second line uses the FOCs (3.9) and (3.10). At the optimum the agent’s budget
holds:

p1 x§1 (p1 , p2 , m) + p1 x§2 (p1 , p2 , m) = m (5.4)

DiÆerentiating (5.4) with respect to m,

@x§1 (p1 , p2 , m) @x§ (p1 , p2 , m)

p1 + p2 2 =1
@m @m

26
Eco11, Fall 2009 Simon Board

Substituting into (5.3) we obtain,

@v(p1 , p2 , m)
=∏ (5.5)
@m

as required.

We can now prove Roy’s identity. DiÆerentiating (5.2) with respect to p1 ,

@v(p1 , p2 , m) @u(x§1 , x§2 ) @x§1 (p1 , p2 , m) @u(x§1 , x§2 ) @x§2 (p1 , p2 , m)

= +
@p1 @x1 @p1 @x2 @p1
@x (p1 , p2 , m)
§ @x (p1 , p2 , m)
§
= ∏p1 1 + ∏p2 2 (5.6)
@p1 @p1

where the second line uses the FOCs (3.9) and (3.10). DiÆerentiating the budget constraint
(5.4) with respect to p1 ,

@x§1 (p1 , p2 , m) @x§ (p1 , p2 , m)

x§1 (p1 , p2 , m) + p1 + p2 2 =0
@p1 @p1

Substituting into (5.6),

@v(p1 , p2 , m)
= °∏x§1 (p1 , p2 , m) (5.7)
@p1

Using (5.7) we obtain,

@v(p1 , p2 , m) @v(p1 , p2 , m)
= °x§1 (p1 , p2 , m)
@p1 @m

as required.

6 Comparative Statics

In this Section we consider how demand for good 1 is aÆected by the agent’s income, the price
of good 1 and the price of good 2. These eÆects are further analysed in the EMP notes. First,
we introduce the notion of elasticities.

27
Eco11, Fall 2009 Simon Board

6.1 Elasticities

Suppose we are interested in how the price of a good, p, aÆects the demand for that good, x(p).
We can measure this change in absolute terms or percentage terms.

First, we might be interested in the impact of a 1¢ increase in the price. This is measured by
the derivative dx(p)/dp. This is simple to calculate but has the disadvantage that the measure
depends on the currency we are using. For example, suppose a consumer is buying a good in
US dollars and has demand:

x(p) = 10 ° 10p (6.1)

DiÆerentiating, dx/dp = °10. Suppose we now change the currency to British pounds, and
suppose $2 = £1. The demand function becomes:

x(p) = 10 ° 20p (6.2)

DiÆerentiating, dx/dp = °20. We see that, while nothing fundamental has changed, the sensi-
tivity of demand to price has doubled, simply because we have relabelled the currency.

In order to overcome this problem, we can measure the eÆect of a 1% increase in price. Define
the price elasticity of demand to equal the percentage increase in demand caused by a 1%
increase in the price. Since the percentage change in x equals the absolute change, ¢x, divided
by the level, the elasticity is given by

¢x/x ¢x p
≤x,p = =
¢p/p ¢p x

In diÆerential terms, we have

dx(p) p
≤x,p = (6.3)
dp x

In example (6.1) we therefore have

p p
≤x,p = (°10) =°
10 ° 10p 1°p

Observe that we obtain exactly the same number if the price is denominated in pounds (6.2)
since percentage changes are independent of the currency. We can also write the elasticity (6.3)

28
Eco11, Fall 2009 Simon Board

as
d ln x(p)
≤x,p =
d ln p
The proof is as follows:

d ln x(p) d dx(eln p ) eln p dx(p) p

= ln x(eln p ) = =
d ln p d ln p dp x(e ) ln p dp x(p)

There are diÆerent kinds of elasticities we can define. In the two good model, the agent’s
demand for good 1, x§1 (p1 , p2 , m), depends on her income and the price of both goods. We can
correspondingly define the income elasticity of demand:

dx§1 (p1 , p2 , m) m
≤x1 ,m =
dm x§1 (p1 , p2 , m)

The own–price elasticity of demand:

dx§1 (p1 , p2 , m) p1
≤x1 ,p1 =
dp1 x1 (p1 , p2 , m)
§

And the cross–price elasticity of demand:

dx§1 (p1 , p2 , m) p2
≤x1 ,p2 =
dp2 x1 (p1 , p2 , m)
§

The fact that demand is homogenous of degree zero (see Section 5) implies that a 1% increase
in income and both prices does not aÆect the agent’s demand. Hence

≤x1 ,m + ≤x1 ,p1 + ≤x1 ,p2 = 0.

6.2 Income EÆects

Suppose an agent’s income increases. Figure 16 shows that her budget constraint shifts out-
wards. The line linking her optimal bundles for diÆerent levels of income is called the income
oÆer curve or the income expansion path.5

Figure 17 show how the consumption of one particular good varies with the agent’s income.
This is called the Engel curve.
5
If the income oÆer curve is linear, preferences are said to be homothetic. Perfect substitutes, perfect
complements and Cobb Douglas preferences are all examples of homothetic preferences.

29
Eco11, Fall 2009 Simon Board

Figure 16: Income EÆects: A Normal Good. This figure shows the eÆect of an increase in the
agent’s income on her demand for both goods. Her choice moves from point A to point B, increasing
her consumption of both goods.

Figure 17: Engel Curve. This figure shows the eÆect of an increase in the agent’s income on her
demand for good 1. Points A and B correspond to those in figure 16.

30
Eco11, Fall 2009 Simon Board

Figure 18: Income EÆects: An Inferior Good. This figure shows the eÆect of an increase in the
agent’s income on her demand for both goods. Her choice moves from point A to point B, increasing
her consumption of x2 but reducing her consumption of x1 . Good 1 is therefore inferior.

When an agent’s income rises her demand may rise or fall. If her demand rises with income, the
good is normal (see figure 16). If her demand falls with income, the good is inferior (see figure
18). Many goods are normal for some ranges of income and inferior for others. For example, for
very poor people in China, rice consumption increases in income; for well oÆ people in China,
rice consumption falls in income as people substitute towards meat.

The normal/inferior distinction concerns the eÆect of income on the absolute consumption of
the good. One may also wonder about the eÆect of income on the budget share of a good.
For example, housing often accounts for around 30% of people’s spending, independent of their
income. A good is a luxury if a 1% increase in income leads to a more than 1% increase in
consumption. That is, ≤x1 ,m > 1. A good is a necessity if a 1% increase in income leads to a
less than 1% increase in consumption. That is, ≤x1 ,m < 1.

6.3 Own Price EÆects

Suppose the price of good 1 increases. Figure 19 shows that the budget line pivots inwards
and, in this case, leads the agent to consume less x1 and more x2 . The line linking her optimal
bundles for diÆerent levels of p1 is called the price oÆer curve.

31
Eco11, Fall 2009 Simon Board

Figure 19: Price EÆects. This figure shows the eÆect of an increase in the price of good 1 on the
agent’s demand for both goods. As p1 rises, the agent’s choice moves from A to B. As a result, the
consumption of x1 falls while the consumption of x2 rises.

Figure 20 shows how the consumption of good 1 varies with p1 . This is, surprisingly enough,
called the demand curve. Good 1 is called ordinary if an the demand curve is downward
sloping, so an increase in p1 causes a reduction in x§1 . Good 1 is a GiÆen good if the demand
is locally upward sloping, so an increase in p1 causes an increase in x§1 .

To understand how price aÆects demand, note that we can decompose the impact of an increase
in p1 into two eÆects. First, holding the agent’s purchasing power constant, there is a change
in change in relative prices causing good 1 to become more expensive relative to good 2. This
is called the substitution eÆect and causes the demand for good 1 to fall. Second, holding
relative prices fixed, the increase in p1 reduces the agent’s purchasing power. This is called
the income eÆect and causes the demand for good 1 to fall if it is normal, and rise if it
is inferior. With a GiÆen good, the increase in p1 causes demand for good 1 to fall a little
via the substitution eÆect and causes demand for good 1 to rise a lot via the income eÆect.
For example, when the price of rice rises in China some poor people experience a cut in their
purchasing power, can no longer aÆord meat and consequently consume more rice.6 For more
on income and substitution eÆects see the EMP notes.

We end with some more jargon. If a 1% increase in price changes demand by less than 1%,
demand is called inelastic. That is, °1 < ≤x1 ,p1 < 1. If a 1% increase in price changes demand
6
See Jensen and Miller (2009), “GiÆen Behaviour and Subsistence Consumption”, American Economic Review.

32
Eco11, Fall 2009 Simon Board

Figure 20: Demand Curve for Good 1. As the price of good 1 falls, the demand for good 1 increases.
Hence the good is ordinary.

by more than 1%, demand is called elastic. That is, ≤x1 ,p1 < °1 or ≤x1 ,p1 > 1.

Exercise: The spending on good 1 is given by p1 x§1 (p1 , p2 , m). Show that an increase in p1
reduces the agent’s spending on the good if and only if the good is ordinary and demand is
elastic.

6.4 Cross Price EÆects

Suppose the price of good 1 increases. Such an increase may cause the demand for good 2 to
rise (as in figure 19) or fall (as in figure 21).

Goods 1 and 2 are gross substitutes if an increase in the price of one increases the demand
for the other. Mathematically,

@x§1 @x§2
>0 and >0
@p2 @p1

For example, when the price of pizza rises, the demand for hamburgers goes up (and vice versa).
Hence pizza and hamburgers are gross substitutes.

Goods 1 and 2 are gross complements if an increase in the price of one decreases the demand

33
Eco11, Fall 2009 Simon Board

Figure 21: Price EÆects. This figure shows the eÆect of an increase in the price of good 1 on the
agent’s demand for both goods. As p1 rises, the agent’s choice moves from A to B. As a result, the
consumption of both goods rises.

for the other. Mathematically,

@x§1 @x§2
<0 and <0
@p2 @p1

For example, when the price of buns rises, the demand for hamburgers goes down (and vice
versa). Hence buns and hamburgers are gross complements.

There is a problem with the idea of gross substitutes/complements: the cross derivatives may
have diÆerent signs. For example, suppose x1 are domestic flights and x2 are international
flights. We may have the following scenario. An increase in p1 causes the agent to become
poorer, since she often flies home to see her parents, leading her to cut back on international
holidays and reducing x§2 . An increase in p2 causes the agent to take fewer international holidays
and more domestic holidays, leading to an increase in x§1 . Hence we have

@x§1 @x§2
>0 and <0
@p2 @p1

It is unclear whether these goods are complements or substitutes. In the EMP notes we intro-
duce the idea of net substitutes, where we can never have problems like this.

34
 Expenditure Minimisation Problem

Simon Board

This Version: September 20, 2009

First Version: October, 2008.

The expenditure minimisation problem (EMP) looks at the reverse side of the utility maximisa-
tion problem (UMP). The UMP considers an agent who wishes to attain the maximum utility
from a limited income. The EMP considers an agent who wishes to find the cheapest way to
attain a target utility. This approach complements the UMP and has several rewards:

• It enables us to analyse the eÆect of a price change, holding the utility of the agent
constant.

• It enables us to decompose the eÆect of a price change on an agent’s Marshallian demand

into a substitution eÆect and an income eÆect. This decomposition is called the Slutsky
equation.

• It enables us to calculate how much we need to compensate a consumer in response to a

price change if we wish to keep her utility constant.

1 Model

We make several assumptions:

1. There are N goods. For much of the analysis we assume N = 2 but nothing depends on
this.

2. The agent takes prices as exogenous. We normally assume prices are linear and denote
them by {p1 , . . . , pN }.

1
Eco11, Fall 2009 Simon Board

3. Preferences satisfy completeness, transitivity and continuity. As a result, a utility func-

tion exists. We normally assume preferences also satisfy monotonicity (so indiÆerence
curves are well behaved) and convexity (so the optima can be characterised by tangency
conditions).

The expenditure minimisation problem is

N
X
min pi xi subject to u(x1 , . . . , xN ) ∏ u (1.1)
x1 ,...,xN
i=1
xi ∏ 0 for all i

The idea is that the agent is trying to find the cheapest way to attain her target utility, u.
The solution to this problem is called the Hicksian demand or compensated demand. It is
denoted by
hi (p1 , . . . , pN , u)

The money the agent must spend in order to attain her target utility is called her expenditure.
The expenditure function is therefore given by

N
X
e(p1 , . . . , pN , u) = min pi xi subject to u(x1 , . . . , xN ) ∏ u
x1 ,...,xN
i=1
xi ∏ 0 for all i

Equivalently, the expenditure function equals the amount the agent spends on her optimal
bundle,
N
X
e(p1 , . . . , pN , u) = pi hi (p1 , . . . , pN , u)
i=1

1.1 Example

Suppose there are two goods, x1 and x2 . Table 1 shows how the agent’s utility (the numbers
in the boxes) varies with the number of x1 and x2 consumed.

To keep things simple, suppose the agent faces prices p1 = 1 and p2 = 1 and wishes to attain
utility u = 12. The agent can attain this utility by consuming (x1 , x2 ) = (6, 2), (x1 , x2 ) = (4, 3),
(x1 , x2 ) = (3, 4) or (x1 , x2 ) = (2, 6). Of these, the cheapest is either (x1 , x2 ) = (4, 3) or
(x1 , x2 ) = (3, 4). In either case, her expenditure is 4 + 3 = 7.

2
Eco11, Fall 2009 Simon Board

x1 \x2 1 2 3 4 5 6

1 1 2 3 4 5 6
2 2 4 6 8 10 12
3 3 6 9 12 15 18
4 4 8 12 16 20 24
5 5 10 15 20 25 30
6 6 12 18 24 30 36

Table 1: Utilities from diÆerent bundles.

Now suppose the agent faces prices p1 = 1 and p2 = 3 and still wishes to attain utility u = 12.
The combinations of (x1 , x2 ) that attain this utility remain unchanged, however the price of
these bundles is diÆerent. Now the cheapest is (x1 , x2 ) = (6, 2), and the agent’s expenditure is
6 + 2 £ 3 = 12.

While this “table approach” can be used to illustrate the basic idea, one can see that it quickly
becomes hard to solve even simple problems. Fortunately, calculus comes to our rescue.

2 Solving the Expenditure Minimisation Problem

2.1 Graphical Solution

We can solve the problem graphically, as with the UMP. The components are also similar to
that problem.

First, we need to understand the constraint set. The agent can choose any bundle where (a) the
agent attains her target utility, u(x1 , x2 ) ∏ u; and (b) the quantities are positive, x1 ∏ 0 and
x2 ∏ 0. If preferences are monotone, then the bundles that meet these conditions are exactly
the ones that lie above the indiÆerence curve with utility u. See figure 1.

Second, we need to understand the objective. The agent wishes to pick the bundle in the
constraint set that minimises her expenditure. Just like with the UMP, we can draw the level
curves of this objective function. Define an iso–expenditure curve by the bundles of x1 and x2
that deliver constant expenditure:

{(x1 , x2 ) : p1 x1 + p2 x2 = const}

3
Eco11, Fall 2009 Simon Board

Figure 1: Constraint Set. The shaded area shows the bundles that yield utility u or more.

These iso–expenditure curves are just like budget curves and so have slope °p1 /p2 . See figure
2

The aim of the agent is to choose the bundle (x1 , x2 ) in the constraint set that is on the lowest
iso–expenditure curve and hence minimises her expenditure. Ignoring boundary problems and
kinks, the solution has the feature that the iso–expenditure curve is tangent to the target
indiÆerence curve. As a result, their slopes are identical. The tangency condition can thus be
written as

p1
M RS = (2.1)
p1

This is illustrated in figure 3.

The intuition behind (2.1) is as follows. Using the fact that M RS = M U1 /M U2 ,1 equation
(2.1) implies that

M U1 p1
= (2.2)
M U2 p1
1
Recall: M Ui = @U/@xi is the marginal utility from good i.

4
Eco11, Fall 2009 Simon Board

Figure 2: Iso-Expenditure Curve. This figure shows the bundles that induce constant expenditure.

Figure 3: Optimal Bundle. This figure shows how the cheapest bundle that attains the target utility
satisfies the tangency condition.

5
Eco11, Fall 2009 Simon Board

Rewriting (2.2) we find

p1 p2
=
M U1 M U1

The ratio pi /M Ui measures the cost of increasing utility by one util, or the “cost–per–bang”.
At the optimum the agent equates the cost–per–bang of the two goods. Intuitively, if good 1
has a higher cost–per–bang than good 2, then the agent should spend less on good 1 and more
on good 2. In doing so, she could attain the same utility at a lower cost.

If preferences are monotone, then the constraint will bind,

u(x1 , x2 ) = u, (2.3)

The tangency equation (2.2) and constraint equation (2.3) can then be used to solve for the
two Hicksian demands.

If there are N goods, the agent will equalise the cost–per–bang from each good, giving us N ° 1
equations. Using the constraint equation (2.3), we can solve for the agent’s Hicksian demands.

The tangency condition (2.2) is the same as that under the UMP. This is no coincidence. We
discuss the formal equivalence in Section 4.2.

2.2 Example: Symmetric Cobb Douglas

Suppose u(x1 , x2 ) = x1 x2 . The tangency condition yields:

x2 p1
= (2.4)
x1 p2

Rearranging, p1 x1 = p2 x2 .

The constraint states that u = x1 x2 . Substituting (2.4) into this yields,

p1 2
u= x
p2 1

Solving for x1 , the Hicksian demand is given by

µ ∂
p2 1/2
h1 (p1 , p2 , u) = u (2.5)
p1

6
Eco11, Fall 2009 Simon Board

Similarly, we can solve for the Hicksian demand for good 2,

µ ∂
p1 1/2
h2 (p1 , p2 , u) = u
p2

We can now calculate the agent’s expenditure

e(p1 , p2 , u) = p1 h1 (p1 , p2 , u) + p2 h2 (p1 , p2 , u)

= 2(up1 p2 )1/2 (2.6)

2.3 Lagrangian Solution

Using a Lagrangian, we can encode the tangency conditions into one formula. As before, let us
ignore boundary problems. The EMP can be expressed as minimising the Lagrangian

L = p1 x1 + p2 x2 + ∏[u ° u(x1 , x2 )]

As with the UMP, the term in brackets can be thought as the penalty for violating the constraint.
That is, the agent is punished for falling short of the target utility.

The FOCs with respect to x1 and x2 are

@L @u
= p1 ° ∏ =0 (2.7)
@x1 @x1
@L @u
= p2 ° ∏ =0 (2.8)
@x2 @x2

If preferences are monotone then the constraint will bind,

u(x1 , x2 ) = u (2.9)

These three equations can then be used to solve for the three unknowns: x1 , x2 and ∏.

Several remarks are in order. First, this approach is identical to the graphical approach. Di-
viding (2.7) by (2.8) yields
@u/@x1 p1
=
@u/@x2 p2
which is the same as (2.2). Moreover, the Lagrange multiplier is

p1 p2
∏= =
M U1 M U2

7
Eco11, Fall 2009 Simon Board

is exactly the cost–per–bang.

Second, if preferences are not monotone, the constraint (2.9) may not bind. If it does not bind,
the Lagrange multiplier in the FOCs will be zero.

Third, the approach is easy to extend to N goods. In this case, one obtains N first order
conditions and the constraint equation (2.9).

3 General Results

3.1 Properties of Expenditure Function

The expenditure function exhibits four important properties.

1. The expenditure function is homogenous of degree one in prices. That is,

e(p1 , p2 , u) = e(Æp1 , Æp2 , u)

for Æ > 0. Intuitively, if the prices of x1 and x2 double, then the cheapest way to attain the
target utility does not change. However, the cost of attaining this utility doubles.

2. The expenditure function is increasing in (p1 , p2 , u). If we increase the target utility u, then
the constraint becomes harder to satisfy and the cost of attaining the target increases. If we
increase p1 then it costs more to buy any bundle of goods and it costs more to attain the target
utility.

3. The expenditure function is concave in prices (p1 , p2 ). Fix the target utility u and prices
(p1 , p2 ) = (p01 , p02 ). Solving the EMP we obtain Hicksian demands h01 = h1 (p01 , p02 , u) and
h02 = h2 (p01 , p02 , u). Now suppose we fix demands and change p1 , the price of good 1. This gives
us a pseudo–expenditure function

¥h01 ,h02 (p1 ) = p1 h01 + p02 h02

This pseudo–expenditure function is linear in p1 which means that, if we keep demands con-
stant, then expenditure rises linearly with p1 . Of course, as p1 rises the agent can reduce her
expenditure by rebalancing her demand towards the good that is cheaper. This means that

8
Eco11, Fall 2009 Simon Board

Figure 4: Expenditure Function. This figure shows how the expenditure function lies under the
pseudo–expenditure function.

real expenditure function lies below the pseudo–expenditure function and is therefore concave.
See figure 4.

More formally, the expenditure function is given by the lower envelope of the pseudo-expenditure
functions. That is, for any bundle (x1 , x2 ), the cost of this bundle at prices (p1 , p2 ) is given by

¥x1 ,x2 (p1 , p2 ) = p1 x1 + p2 x2

The expenditure function is then the minimum of these pseudo–expenditure functions given the
bundle (x1 , x2 ) attains the target utility. Mathematically,

e(p1 , p2 , u) = min{p1 x1 + p2 x2 : u(x1 , x2 ) = u} (3.1)

Thus the expenditure function is the lower minimum of a collection of linear functions, and is
therefore concave.2 See figure 5.

4. Sheppard’s Lemma: The derivative of the expenditure function equals the Hicksian demand.
That is,

@
e(p1 , p2 , u) = h1 (p1 , p2 , u) (3.2)
@p1
2
Exercise: Show that the minimum of two concave functions is concave.

9
Eco11, Fall 2009 Simon Board

Figure 5: Envelope Property of Expenditure Function. This figure shows the expenditure func-
tion equals the lower envelope of the pseudo expenditure functions.

The idea behind this result can be seen from figure 4. At p1 = p01 the expenditure function
is tangential to the pseudo–expenditure function. The pseudo–expenditure is linear in p1 with
slope h1 (p01 , p02 , u). Hence the expenditure function also has slope h1 (p01 , p02 , u).

The intuition behind Sheppard’s Lemma is as follows. Suppose an agent wishes to attain target
utility u = 25 and faces prices p1 = $1 and p2 = $1. Furthermore, suppose that the cheapest
way to attain the target utility is by consuming h1 = 5 and h2 = 5. Next, consider an increase in
p1 of 1¢. This change has a direct and indirect eÆect. The direct eÆect is that, holding demand
constant, the agent’s spending rises by h1 £ 1¢ = 5¢; the indirect eÆect is that the agent will
change her demands. However, the tangency condition illustrated in figure 3 shows that the
agent is close to indiÆerent between choosing the optimal quantity and nearby quantities, so the
rebalancing demand will will have a very small impact on her expenditure. We thus conclude
that ¢e = h1 ¢p1 , Rewriting,
¢e
= h1
¢p1
This is the discrete version of equation (3.2).

Here is a formal proof of Sheppard’s Lemma. By definition of the expenditure function,

e(p1 , p2 , u) = p1 h1 (p1 , p2 , u) + p2 h2 (p1 , p2 , u)

10
Eco11, Fall 2009 Simon Board

DiÆerentiating with respect to p1 yields

@ @h1 (p1 , p2 , u) @h2 (p1 , p2 , u)

e(p1 , p2 , u) = h1 (p1 , p2 , u) + p1 + p2 (3.3)
@p1 @p1 @p1

As discussed above, we have decomposed the eÆect of the price change into a direct eÆect (the
first term) and an indirect eÆect (the second and third terms). We now wish to show the
indirect eÆect is zero. From the agent’s minimisation problem in Section 2.3, the FOCs are

@u(h1 , h2 )
pi = ∏
@xi

We also know that the agent’s constraint binds:

u(h1 (p1 , p2 , u), h2 (p1 , p2 , u)) = u (3.4)

Substituting the FOCs into (3.3)

∑ ∏
@ @u(h1 , h2 ) @h1 (p1 , p2 , u) @u(h1 , h2 ) @h2 (p1 , p2 , u)
e(p1 , p2 , u) = h1 (p1 , p2 , u) + ∏ + (3.5)
@p1 @x1 @p1 @x2 @p1

DiÆerentiating (3.4) with respect to p1 yields

@u(h1 , h2 ) @h1 (p1 , p2 , u) @u(h1 , h2 ) @h2 (p1 , p2 , u)

+ =0 (3.6)
@x1 @p1 @x2 @p1

Substituting (3.6) into (3.5) yields Sheppard’s Lemma.

3.2 Properties of Hicksian Demand

Hicksian demand has three important properties. These follow from the properties of the
expenditure function derived above.

1. Hicksian demand is homogenous of degree zero in prices. That is,

h1 (p1 , p2 , u) = h1 (Æp1 , Æp2 , u)

for Æ > 0. Intuitively, doubling both prices does not alter the cheapest way to obtain the target
utility u.

11
Eco11, Fall 2009 Simon Board

Figure 6: Hicksian Demand and Own Price EÆects. This figure shows the eÆect of an increase in
p1 , from p1 to p01 . The optimal bundle moves from A to B.

2. The Law of Hicksian Demand: The Hicksian demand for good i is decreasing in pi . That is,

@
hi (p1 , p2 , u) ∑ 0
@pi

Intuitively, when p1 rises the relative prices become tilted in favour of good 2. The cheapest
way to attain the target utility then consists of less of good 1 and more of good 2. Graphically
this can be seen from figure 6. As p1 rises to p01 , the iso–expenditure function becomes steeper
and the optimal bundle involves less of good 1 and more of good 2.3

A formal proof of this result uses the properties of the expenditure function:

@ @2
h1 (p1 , p2 , u) = 2 e(p1 , p2 , u) ∑ 0
@p1 @p1

where the equality comes from Sheppard’s Lemma and the inequality follows from the concavity
of the expenditure function.

This result highlights a big diÆerence between Hicksian demand and Marshallian demand.
An increase in p1 always reduces the Hicksian demand for good 1 but may, in the case of a
GiÆen good, increase the Marshallian demand. This is because the eÆect of a price change on
Marshallian demand has two eÆects: a substitution eÆect (a change in relative prices) and an
3
The fact that the demand for good 2 always rises is an artifact of there only being 2 goods.

12
Eco11, Fall 2009 Simon Board

income eÆect (a change in the consumer’s purchasing power). In comparison, the change in
Hicksian demand isolates the substitution eÆect.

3. Hicksian demand has symmetric cross derivatives. That is,

@ @
h1 (p1 , p2 , u) = h2 (p1 , p2 , u)
@p2 @p1

The proof of this result also uses the properties of the expenditure function.
∑ ∏ ∑ ∏
@ @ @ @ @ @
h1 (p1 , p2 , u) = e(p1 , p2 , u) = e(p1 , p2 , u) = h2 (p1 , p2 , u)
@p2 @p2 @p1 @p1 @p2 @p1

The first and third equalities come from Sheppard’s Lemma and the second from Young’s
theorem.

We say goods x1 and x2 are net substitutes if

@ @
h1 (p1 , p2 , u) > 0 and h2 (p1 , p2 , u) > 0
@p2 @p1

We say goods x1 and x2 are net complements if

@ @
h1 (p1 , p2 , u) < 0 and h2 (p1 , p2 , u) < 0
@p2 @p1

The symmetry of the cross derivatives means that we cannot have one cross–derivative positive
negative and the opposite cross–derivative negative, as with gross substitutes and complements.4

4 Income and Substitution EÆects

We are often interested in how price changes aÆect Marshallian demand. This matters to firms
when choosing prices, to government when choosing tax rates and to economists when making
forecasts. For example: how much will demand for ethanol increase if we lower the price by
$10?

We saw with the UMP that an increase in p1 may lead to a large decrease in demand (if demand
is elastic), may lead to a small decrease in demand (if demand is inelastic) or may lead to an
increase in demand (in the case of a GiÆen good). One major issue is that an increase in the
4
Exercise: Suppose there are two goods. Show they must be net substitutes.

13
Eco11, Fall 2009 Simon Board

Figure 7: Substitution and Income EÆects with Normal Good. With a normal good, both
substitution eÆect (SE) and income eÆect (IE) are negative.

price of good 1 has two eÆects: it both makes good 1 relatively more expensive (the substitution
eÆect) and reduces the agent’s purchasing power (the income eÆect). This section will separate
these eÆects. In Section 4.1 we do this graphically. In Section 4.3 we do this mathematically.

4.1 Pictures

Suppose we start at point A in figures 7 and 8. When p1 increases, the budget line pivots
around it’s left end and demand falls from A to C. We can decompose this change into two
eÆects.

1. A change in relative prices, keeping utility constant. This is the shift from A to B,
and is called the substitution eÆect. This equals the change in Hicksian demand and,
appealing to the Law of Hicksian Demand, is negative.

2. A change in income, keeping relative prices constant. This is the shift from B to C, and
is called the income eÆect. This eÆect is positive if the good is normal, and negative if
the good is inferior.

Exercise: draw the equivalent picture for a GiÆen good.

14
Eco11, Fall 2009 Simon Board

Figure 8: Substitution and Income EÆects with Normal Good. With an inferior good, substi-
tution eÆect (SE) is negative while the income eÆect (IE) is positive.

4.2 Relation between the UMP and EMP

The EMP and UMP are closely related. To illustrate, suppose the agent has $10 to spend on
two goods. Suppose her utility is maximised when (x1 , x2 ) = (5, 5) and she can attain 25 utils.5
What is the cheapest way for the agent to attain 25 utils? Given this information, the answer
must be (x1 , x2 ) = (5, 5). Moreover, her expenditure is $10. The reason is as follows. First, we
know that the agent can obtain 25 utils from $10, so the cheapest way to obtain 25 utils is at
most $10. That is, e ∑ $10. Now suppose, by contradiction, that the agent can obtain 25 utils
for, say, $8. Then, if preferences are monotone, she will be able to obtain strictly more than 25
utils with $10, contradicting our initial assumptions.

We can state this result formally. Fix prices (p1 , p2 ) and income m. Marshallian demand is
given by x§i (p1 , p2 , m) and indirect utility is v(p1 , p2 , m). Consider the EMP:

min p1 x1 + p2 x2 subject to u(x1 , x2 ) ∏ v(p1 , p2 , m)

x1 ,x2

The induced Hicksian demand is given by hi (p1 , p2 , v(p1 , p2 , m)) while the expenditure function
is e(p1 , p2 , v(p1 , p2 , m)). Then using the reasoning above, one can show that

e(p1 , p2 , v(p1 , p2 , m)) = m (4.1)

hi (p1 , p2 , v(p1 , p2 , m)) = x§i (p1 , p2 , m) (4.2)
5
These numbers come from assuming p1 = 1, p2 = 1 and u(x1 , x2 ) = x1 x2 .

15
Eco11, Fall 2009 Simon Board

Suppose we start with income m. Equation (4.1) says that the minimum expenditure required
to reach v(p1 , p2 , m), the most utility from m, is just m. Equation (4.2) says that an agent who
wishes to maximise her utility from m and one who wishes to find the cheapest way to attain
v(p1 , p2 , m) will buy the same goods. Intuitively, in both cases, they will spend m and will do
so by equating the bang–per–buck from each good.

Equation (4.1) is practically useful. Fixing prices and omitting them from the arguments, it
says that e(v(m)) = m. Since the expenditure function is increasing in u, we can invert it and
obtain:

v(m) = e°1 (m) (4.3)

Hence the indirect utility function equals the inverse of the expenditure function. To illustrate
this result, suppose u(x1 , x2 ) = x1 x2 . From equation (2.6), we know that
p
e(u) = 2 up1 p2

We invert this equation by letting m = e(u) and v(m) = u, and solving for v(m). This yields

m2
v(m) =
4p1 p2

One can verify that this indeed the indirect utility function.

We can also state a second, closely related, result. Fix prices (p1 , p2 ) and target utility u.
Hicksian demand is given by hi (p1 , p2 , u) and the expenditure function is e(p1 , p2 , u). Consider
the UMP:

max u(x1 , x2 ) subject to p1 x1 + p2 x2 ∑ e(p1 , p2 , u)

x1 ,x2

The induced Marshallian demand is given by x§i (p1 , p2 , e(p1 , p2 , u)) while the indirect utility is
v(p1 , p2 , e(p1 , p2 , u)). One can show that

v(p1 , p2 , e(p1 , p2 , u)) = u (4.4)

x§i (p1 , p2 , e(p1 , p2 , u)) = hi (p1 , p2 , u) (4.5)

Suppose we start with target utility u. Equation (4.4) says that the most utility the agent can
get from e(p1 , p2 , u), the money required to reach u, is just u. Equation (4.5) says that an agent
who wishes to find the cheapest way to attain u and one who wishes to maximise her utility
from e(p1 , p2 , u) will buy the same goods. Intuitively, in both cases, they will attain utility u

16
Eco11, Fall 2009 Simon Board

and will do so by equating the bang–per–buck from each good.

Fixing prices and omitting them from the arguments, equation (4.4) says that v(e(u)) = u.
Since the indirect function is increasing in m, we can invert it and obtain:

e(u) = v °1 (u) (4.6)

Hence the expenditure function equals the inverse of the indirect utility function. Together,
equations (4.3) and (4.6) mean we can move back and forwards between the expenditure function
and indirect utility function.

4.3 Slutsky Equation: Own Price EÆects

Suppose p1 increases by ¢p1 . There are two eÆects:

1. Fixing the agent’s utility, relative prices change causing demand to rise by @p1 ¢p1 .
@h1
Since
@h1
@p1 < 0, this eÆect causes demand to fall. This is the substitution eÆect.

2. Fixing relative prices, the agent’s income falls by x§1 ¢p1 . As a result, her demand falls
@x§
by x§1 @m1 ¢p1 . This is the income eÆect.

Putting these eÆects together, we have

@h1 @x§
¢x§1 = ¢p1 ° x§1 1 ¢p1
@p1 @m

Dividing by ¢p1 yields the Slutsky equation.

Theorem 1 (Own–Price Slutsky Equation). Fix prices (p1 , p2 ) and income m, and let u =
v(p1 , p2 , m) be the indirect utility. Then

@ § @ @ §
x1 (p1 , p2 , m) = h1 (p1 , p2 , u) ° x§1 (p1 , p2 , m) x (p1 , p2 , m) (4.7)
@p1 @p1 @m 1

A formal proof is reasonably straightforward. Using equation (4.5),

hi (p1 , p2 , u) = x§i (p1 , p2 , e(p1 , p2 , u))

17
Eco11, Fall 2009 Simon Board

DiÆerentiating with respect to p1 yields

@ @x§1 (p1 , p2 , e(p1 , p2 , u)) @x§1 (p1 , p2 , e(p1 , p2 , u)) @e(p1 , p2 , u)

h1 (p1 , p2 , u) = +
@p1 @p1 @m @p1
@x§1 (p1 , p2 , e(p1 , p2 , u)) @x§1 (p1 , p2 , e(p1 , p2 , u)) §
= + x1 (p1 , p2 , e(p1 , p2 , u)) (4.8)
@p1 @m

where the second line comes from Sheppard’s Lemma. Using the definition of u and equation
(4.1),

e(p1 , p2 , u) = e(p1 , p2 , v(p1 , p2 , m)) = m (4.9)

Substituting (4.9) into (4.8) and rearranging yields (4.7), as required.

4.4 Slutsky Equation: Cross Price EÆects

Equation (4.7) analyses the eÆect of a change in p1 on the demand for good 1. We can use the
same approach to analyse the eÆect of a change in p2 on the demand for good 1.

Suppose p2 increases by ¢p2 . As before, there are two eÆects:

1. Fixing the agent’s utility, relative prices change causing demand to rise by @p2 ¢p2 .
@h1
Recall
that @h1
@p2 > 0 if the goods are net substitutes and @h1
@p2 < 0 are net complements.

2. Fixing relative prices, the agent’s income falls by x§2 ¢p2 . As a result, her demand falls
@x§
by x§2 @m1 ¢p2 .

Putting these eÆects together, we have

@h1 @x§
¢x§1 = ¢p2 ° x§2 1 ¢p2
@p2 @m

Dividing by ¢p2 yields the Slutsky equation for cross–price eÆects.

Theorem 2 (Cross–Price Slutsky Equation). Fix prices (p1 , p2 ) and income m, and let u =
v(p1 , p2 , m) be the indirect utility. Then

@ § @ @ §
x1 (p1 , p2 , m) = h1 (p1 , p2 , u) ° x§2 (p1 , p2 , m) x (p1 , p2 , m) (4.10)
@p2 @p2 @m 1

18
 Eco11, Fall 2009 Simon Board

The proof is almost identical to that of (4.7). Using equation (4.5),

hi (p1 , p2 , u) = x§i (p1 , p2 , e(p1 , p2 , u))

DiÆerentiating with respect to p2 yields

@ @x§1 (p1 , p2 , e(p1 , p2 , u)) @x§1 (p1 , p2 , e(p1 , p2 , u)) @e(p1 , p2 , u)

h1 (p1 , p2 , u) = +
@p2 @p2 @m @p2
@x§1 (p1 , p2 , e(p1 , p2 , u)) @x§1 (p1 , p2 , e(p1 , p2 , u)) §
= + x2 (p1 , p2 , e(p1 , p2 , u)) (4.11)
@p2 @m

where the second line comes from Sheppard’s Lemma. Using the definition of u and equation
(4.1),

e(p1 , p2 , u) = e(p1 , p2 , v(p1 , p2 , m)) = m (4.12)

Substituting (4.12) into (4.11) and rearranging yields (4.10).

4.5 Slutsky Equation: Example

We illustrate the Slutsky equation with our running example. Let u(x1 , x2 ) = x1 x2 . From the
UMP we know that

m
x§1 (p1 , p2 , m) =
2p1
m2
v(p1 , p2 , m) =
4p1 p2

From the EMP (see Section 2.2) we know that

µ ∂1/2
p2
h1 (p1 , p2 , u) = u
p1
e(p1 , p2 , u) = 2(up1 p2 )1/2

The left hand side of the Slutsky equation states

@ § 1
x1 (p1 , p2 , m) = ° mp°2 (4.13)
@p1 2 1

19
Eco11, Fall 2009 Simon Board

The right hand side is

@h1 @x§ 1 °1/2 1/2 1

° x§1 1 = ° u1/2 p1 p2 ° mp°2
@p1 @m 2 4 1
1 1 °2
= ° mp°2
1 ° mp1 (4.14)
4 4

where the second line uses u = v(p1 , p2 , m).

Observe that (4.13) equals (4.14) as we would hope. Moreover, the two terms in equation
(4.14) are identical. This means that the substitution and income eÆects are of equal size: both
account for 50% of the fall in demand.

5 Consumer Surplus

It is often important to put a monetary value on the eÆect of a price change on an agent’s
utility. For example, the government may wish to evaluate the impact of a tax change; or a
court may wish to evaluate the negative eÆect of collusion on consumers.

To gain some intuition, suppose the consumer has monetary valuations for each unit of the
good. In particular suppose their valuations are given by table 2.

Unit Valuation $
1 10
2 8
3 6
4 4
5 2

Table 2: Agent’s Valuations

Suppose the price of the good is initially p1 = 3. Since the agent buys a unit if and only if her
valuation exceeds the price, she will buy 4 units. Her consumer surplus, the diÆerence between
her willingness to pay and the price she pays, equals

CS = (10 ° 3) + (8 ° 3) + (6 ° 3) + (4 ° 3) = $16

Suppose the price rises to p1 = 7. The agent then consumes 2 units and her consumer surplus
is
CS = (10 ° 7) + (8 ° 7) = $4.

20
Eco11, Fall 2009 Simon Board

Figure 9: Consumer Surplus with Quasilinear Demand. The figure shows the agent’s demand
curve. The shaded area is the loss in CS due to the price increase.

Hence the agent would need to be compensated $12 for this price increase. This is shown in
figure 9.

This exercise is familiar from introductory economics courses: consumer surplus is the area
under the agent’s Marshallian demand curve. However this approach assumes the agent has
quasilinear utility, allowing us to associate a monetary value to each unit demanded by the
agent. In Section 5.1 we show that the welfare eÆect of a price change is determined by the
area under the Hicksian demand curve rather than the Marshallian demand. In Section 5.2 we
see that, when utility is quasi–linear then Hicksian demand and Marshallian demand coincide,
justifying the approach taken above.

5.1 Compensating Variation

Suppose prices and income are initially (p1 , p2 , m), and that p1 increases to p01 . The compen-
sating variation is defined by

CV = e(p01 , p2 , u) ° e(p1 , p2 , u)

21
Eco11, Fall 2009 Simon Board

Figure 10: Compensating Variation and IndiÆerence Curves. This figure shows the eÆect on
an increase in p1 . The Marshallian demand falls from A to C. The Hicksian demand moves from A to
B. The compensating variation equals the diÆerence between the consumer’s original income and the
income she would need to attain u.

The CV is thus the extra spending needed to keep the agent at their original utility level. That
is, an increase in income of CV completely compensates the agent for the price increase.6 This
is shown in figure 10.

The compensating variation can be related to the Hicksian demand curve. Applying the fun-
damental theorem of calculus,7
Z p01
@
CV = e(p̃1 , p2 , u) dp̃1
p1 @p1
Zp01
= h1 (p̃1 , p2 , u) dp̃1 (5.1)
p1

where the second equation follows from Sheppard’s Lemma. Equation (5.1) says that the lost
welfare from the price change equals the area under the Hicksian demand curve. See figure 11.
6
There is a closely related measure of welfare called the equivalent variation. We will not discuss this here.
Rb
7
The fundamental theorem of calculus says that f (b) ° f (a) = a f 0 (x)dx.

22
Eco11, Fall 2009 Simon Board

Figure 11: Compensating Variation and Hicksian Demand. This figure shows that CV equals
the area under the demand curve.

5.2 Quasilinear Utilities

While we may wish to calculate the area under the Hicksian demand, it is often easier to
calculate the area under the Marshallian demand curve. For example, in empirical applications,
it is easy to estimate the Marshallian demand by looking at how much people buy at diÆerent
prices.

Suppose utility is quasilinear in that it can be represented by a utility function of the form

u(x1 , x2 ) = v(x1 ) + x2

where we assume v(·) is increasing and concave. Under this specification, the marginal utility
of the second good is constant. For example, x2 could be a general aggregate good or cash.

When utility is quasilinear we can think of an agent’s utility in terms of dollar valuations, as
at the start of this section. The argument is as follows. The agent’s problem is to maximise
her utility subject to her budget constraint, p1 x1 + p2 x2 ∑ m. Since utility is monotone, the
budget constraint will bind. Using the substitution method, the budget constraint becomes

m p1
x2 = ° x1
p2 p2

23
Eco11, Fall 2009 Simon Board

Substituting this into the utility function, the agent maximises

p1 m
v(x1 ) ° x1 + (5.2)
p2 p2

Notice the last term is a constant and can be ignored. If x2 is interpreted as cash, we can
normalise p2 = 1. The agent then chooses x1 to maximise

v(x1 ) ° p1 x1

The agent’s choice is independent of m, so she acts as if she values x1 units of good 1 at v(x1 ),
independent of the units of x2 being consumed. We can then think of v 0 (x1 ) as her valuation
of the marginal unit of x1 .

Under quasilinear utility, the Hicksian and Marshallian demands coincide. Ignoring boundary
problems, the Marshallian demand is derived by maximising (5.2). The first–order condition
implies that Marshallian demand is implicitly given by

p1
v 0 (x§1 (p1 , p2 , m)) = (5.3)
p2

Turning to the EMP, the agent minimises

L = p1 x1 + p2 x2 + ∏[u ° v(x1 ) ° x2 ]

The first first–order conditions are

p1 = ∏v 0 (x1 )
p2 = ∏

Looking at the ratio of these two equations, Hicksian demand is implicitly given by

p1
v 0 (h1 (p1 , p2 , u)) = (5.4)
p2

From equations (5.3) and (5.4) we see that Marshallian demand and Hicksian demand coincide.
Hence the compensating variation is given by
Z p01
CV = h1 (p̃1 , p2 , u) dp̃1
p1
Z p0
1
= x§1 (p̃1 , p2 , m) dp̃1
p1

24
Eco11, Fall 2009 Simon Board

This result provides a foundation for the classical measure of consumer surplus.

6 Endowments of Goods

In the UMP we assume that agents are endowed with income m and use it to maximise their
utility. While this is a useful model to address demand for retail products, it is sometimes more
accurate to assume agents are endowed with goods which they can sell on the open market.
There are two reasons for analysing this model:

• The model is important for understanding practical problems such as a worker’s choice
of labour supply (Section 6.1), and an agent’s decision to smooth consumption over time
(Section 6.2).

• When we analyse the entire economy, we will want to close the model. Hence we wish the
agents who demand goods to also work for firms that make goods.

Suppose there are N goods and the agent starts with endowments {!1 , . . . , !N }, where !i ∏ 0
for all i. The consumer can sell these goods at market prices {p1 , . . . , pN }. For example, an
agent may own a farm which produces vegetables and may sell the produce to buy meat. The
agent has income

N
X
m= pi ! i (6.1)
i=1

Given equation (6.1) the agent’s problem is the same as that studied so far. We can derive
her Marshallian demand and indirect utility (see figure 12). We can also derive her Hicksian
demand and expenditure function (since these are independent of income)

The one major diÆerence from the model with exogenous income is that a price change now
aÆects the agent’s income as well as the goods she buys. We study this in Section 6.3. We first
consider two applications.

6.1 Labour Supply

Suppose an agent has utility u(x1 , x2 ) = x1 x2 over leisure x1 and a general consumption good
x2 . The agent has exogenous income m and can also work at wage w. She has T hours which

25
Eco11, Fall 2009 Simon Board

Figure 12: Marshallian Demand with Endowments This figure shows the optimal choice when
the agent has endowments of the two goods.

she can allocate to either work or leisure. We normalise the price of x2 to p2 = 1.

The agent’s budget constraint is

x2 = w(T ° x1 ) + m

The left hand side equals the agent’s spending on consumption; the right hand side equals her
income. As a thought experiment, one can imagine the agent selling all T units of her labour
and then buying x1 units of it back at price w to be consumed as leisure. We can thus rewrite
the budget constraint as
wx1 + x2 = wT + m

The left hand side is the goods consumed (including leisure, consumed at price w). The right
hand side is the agent’s endowment income, as in equation (6.1).

Ignoring boundary constraints,8 her problem is

max L = x1 x2 + ∏[m + w(T ° x1 ) ° x2 ]

x1 ,x2

8
With this problem the boundary constraints are slightly diÆerent to normal since the agent cannot consume
more that T units of leisure. We thus have T ∏ x1 ∏ 0 and x2 ∏ 0.

26
Eco11, Fall 2009 Simon Board

The FOCs are

x2 = ∏w
x1 = ∏

Taking the ratio of these FOCs, we see that

x2
=w
x1

As before, the left hand side is the MRS, while the right hand side is the price ratio. Using the
budget constraint the agent’s demands are given by

1
x§1 = [wT + m] (6.2)
2w
1
x§2 = [wT + m] (6.3)
2

Equations (6.2) and (6.3) show that the consumer splits her endowment income of wT + m
equally between leisure and consumption. This is just like the solution to the Cobb Douglas
problem without endowments (see UMP notes), where we found that

1 1
x§1 = m and x§2 = m (6.4)
2p1 2p2

We can now evaluate an eÆect of a change in wages. DiÆerentiating (6.2) and (6.3),

@x§1 1 1 1
= wT ° [wT + m] = ° 2 m (6.5)
@w 2w 2 2w 2 2w
@x§2 1
= T (6.6)
@w 2

From (6.5), we see an increase in the wage reduces the amount of leisure the agent consumes.
There are two eÆects here: an increase in the wage raises the relative price of leisure and reduces
demand (the substitution eÆect); it also makes the agent richer and increases the demand for
leisure (the income eÆect). In this case the substitution eÆect dominates the income eÆect: we
analyse this formally in Section 6.4.

From (6.6), we see an increase in the wage increase the amount of x2 the agent consumes. This
is because an increase in wages increase the value of the agent’s endowment; in comparison,
without endowments, equation (6.4) shows that x§2 is independent of p1 .

27
Eco11, Fall 2009 Simon Board

6.2 Intertemporal Optimisation

Suppose an agent allocates consumption (e.g. money) across two periods. Let the consumption
in period 1 and 2 be x1 and x2 respectively. The agent’s utility is

u(x1 , x2 ) = ln(x1 ) + (1 + Ø)°1 ln(x2 ) (6.7)

where Ø ∏ 0 is the agent’s discount rate.

In periods 1 and 2 the agent is endowed with income m1 and m2 , respectively. The agent can
save at interest rate r ∏ 0, so that $1 in period 1 is worth $(1+r) in period 2. As a result, the
agent’s budget constraint is

m1 + (1 + r)°1 m2 = x1 + (1 + r)°1 x2 (6.8)

The left hand side of (6.8) is the agent’s lifetime income in terms of period 1 dollars. The right
and side is the agent’s lifetime spending. We say they are borrowing when x1 > m1 and saving
when x1 < m1 .

We can solve this problem just as we would solve a regular utility maximisation problem, where
p1 = 1 and p2 = (1 + r)°1 . See figure 13. Using (6.7) and (6.8) the tangency condition,
M RS = p1 /p2 , becomes
1/x1
(1 + Ø) = (1 + r)
1/x2
Rearranging,

1+Ø §
x§1 = x (6.9)
1+r 2

Equation (6.2) immediately implies that if r = Ø then the agent consumes the same in each
period, x§1 = x§2 . Intuitively, since the agent’s per–period utility is concave, she wishes to smooth
her consumption across time. If r = Ø, then the agent is just as impatient as the market, so
she will perfectly smooth her consumption across the two periods. If Ø > r then the agent is
more impatient than the market and she consumes more in the first period, x§1 > x§2 .

Using the budget constraint, demand is given by

1+Ø 1+r
x§1 = [m1 + (1 + r)°1 m2 ] and x§2 = [m1 + (1 + r)°1 m2 ]
2+Ø 2+Ø

28
Eco11, Fall 2009 Simon Board

Figure 13: Intertemporal Optimisation This figure shows an agent who has a high income in period
1 and a low income in period 2. At the optimum, she saves in period 1.

6.3 Own Price EÆects

Suppose there is an increase in p1 . As with a fixed income m, the budget line becomes steeper.
However, since the value of the endowment changes, it is no longer true that the budget set
shrinks. Rather, the budget line pivots around the endowment: see figure 14.

As in Section 4, we can decompose the price change into a substitution and income eÆect.
However, the income eÆect has to be adjusted for the change in the value of the endowment.
Suppose p1 increases by ¢p1 . Then there are two eÆects:

1. Fixing the agent’s utility, relative prices change causing demand to rise by @p1 ¢p1 .
@h1
Since
@h1
@p1 < 0, this eÆect causes demand to fall. This is the substitution eÆect.

2. Fixing relative prices, the agent’s income rises by (!1 ° x§1 )¢p1 . This means that the
agent’s income rises if she is a net seller of the good (as in the labour example), and falls
@x§
if she is a net buyer of the good. As a result, her demand rises by (!1 ° x§1 ) @m1 ¢p1 . This
is the income eÆect.

29
Eco11, Fall 2009 Simon Board

Figure 14: Own–Price EÆects with Endowments. This figure shows the eÆect on a decrease in p1
when the agent is endowed with {!1 , !2 }. Note the income looks like it goes down, even though prices
fall. This is because, at point A, the agent owns more of good 1 than she buys, !1 > x§1 . Hence a
decrease in p1 reduces her purchasing power.

Putting these eÆects together, we have

@h1 @x§
¢x§1 = ¢p1 + (!1 ° x§1 ) 1 ¢p1
@p1 @m

Dividing by ¢p1 yields the Slutsky equation.

Theorem 3 (Own–Price Slutsky Equation with Endowments). Fix prices (p1 , p2 ), income m
and endowments (!1 , !2 ), and let u = v(p1 , p2 , m) be the indirect utility. Then

@ § @ @ §
x1 (p1 , p2 , m) = h1 (p1 , p2 , u) + (!1 ° x§1 (p1 , p2 , m)) x (p1 , p2 , m) (6.10)
@p1 @p1 @m 1

This result follows from the regular Slutsky equation (4.7). All we need to do is define net
demand for good 1 by z1§ (p1 , p2 , m) = x§1 (p1 , p2 , m) ° !1 . We can then apply the regular Slutsky
equation to the agent’s net demand:

@ § @ @ §
z (p1 , p2 , m) = h1 (p1 , p2 , u) ° z1§ (p1 , p2 , m) z (p1 , p2 , m) (6.11)
@p1 1 @p1 @m 1

Since z1§ and x§1 diÆer by a constant term, we can put equation (6.11) back in terms of
x§i (p1 , p2 , m), yielding equation (6.10).

30
Eco11, Fall 2009 Simon Board

6.4 Labour Supply and the Slutsky Equation

We now apply the Slutsky equation (6.10) to the labour supply problem in Section 6.1. From
equation (6.2), the Marshallian demand is

1
x§1 (p1 , p2 , m) = [wT + m] (6.12)
2w

Using the (6.2) and (6.3) the indirect utility is

1
v(p1 , p2 , m) = x§1 x§2 = [wT + m]2 (6.13)
4w

From (2.5) and using p1 = w and p2 = 1, the Hicksian demand is

h1 (p1 , p2 , u) = u1/2 w°1/2 (6.14)

We now have all the elements we need.

The left hand side of the Slutsky equation is

@ § 1
x1 = ° 2 m
@p1 2w

The right hand side of the Slutsky equation is

∑ ∏
@h1 § @x1
§ 1 °3/2 1/2 1 1
+ (!1 ° x1 ) =° w u + T° [wT + m]
@p1 @m 2 2w 2w
1 1 1
= ° w°3/2 w°1/2 [wT + m] + [wT ° m]
2 2 4w2
1 1
= ° 2 [wT + m] + [wT ° m]
4w 4w2
1
= ° 2m
2w

where the first line uses (6.12) and (6.14), and the second uses u = v(p1 , p2 , m) and equation
(6.13). We can therefore see that the substitution eÆect outweighs the income eÆect, and as
m becomes smaller these two eÆects grow closer in magnitude. In the limit, as m ! 0, leisure
demand (and hence labour supply) are independent of the wage.

31
 Firm’s Problem

Simon Board§

This Version: September 20, 2009

First Version: December, 2009.

In these notes we address the firm’s problem. We can break the firm’s problem into three
questions.

1. Which combinations of inputs produce a given level of output?

2. Given input prices, what is the cheapest way to attain a certain output?

3. Given output prices, how much output should the firm produce?

We study the firm’s technology in Sections 1–2, the cost minimisation problem in Section 3 and
the profit maximisation problem in Section 4.

1 Technology

1.1 Model

We model a firm as a production function that turns inputs into outputs. We assume:

1. The firm produces a single output q 2 <+ . One can generalise the model to allow for
firms which make multiple products, but this is beyond this course.
§
Department of Economics, UCLA. http://www.econ.ucla.edu/sboard/. Please email suggestions and typos
to [email protected].

1
Eco11, Fall 2009 Simon Board

2. The firm has N possible inputs, {z1 , . . . , zN }, where zi 2 <+ for each i. We normally
assume N = 2, but nothing depends on this. We can think of inputs as labour, capital or
raw materials.

3. Inputs are mapped into output by a production function q = f (z1 , z2 ). This is normally
assumed to be concave and monotone. We discuss these properties later.

To illustrate the model, we can consider a farmer’s technology. In this case, the output is the
farmer’s produce (e.g. corn) while the inputs are labour and capital (i.e. machinery). There is
clearly a tradeoÆ between these two inputs: in the developing world, farmers use little capital,
doing many tasks by hand; in the developed world, farmers use large machines to plant seeds
and even pick fruit.

In some examples inputs may be close substitutes. To illustrate, suppose two students are
working on a homework. In this case the output equals the number of problems solved, while
the inputs are the hours of the two students. The inputs are close substitutes if all that matters
is the total number of hours worked (see Section 2.3).

In other cases inputs may be complements. To illustrate, suppose an MBA and a computer
engineer are setting up a company. Each worker has specialised skills and neither can do the
other’s job. In this case, output depends on which worker is doing the least work, and we say
the inputs are perfect compliments (see Section 2.2).

The marginal product of input zi is the output from one extra unit of good i.

@f (z1 , z2 )
M Pi (z1 , z2 ) = ,
@zi

The average product of input i is

f (z1 , z2 )
APi (z1 , z2 ) = .
zi

1.2 Isoquants

An isoquant describes the combinations of inputs that produce a constant level of output.
That is,
Isoquant = {(z1 , z2 ) 2 <2+ |f (z1 , z2 ) = const.}

2
Eco11, Fall 2009 Simon Board

Figure 1: Isoquant. This figure shows two isoquants. Each curve depicts the bundles that yield
constant output.

A firm has a collection of isoquants, each one corresponding to a diÆerent level of output. By
varying this level, we can trace out the agent’s entire production possibilities.

To illustrate, suppose a firm has production technology

1/3 1/3
f (z1 , z2 ) = z1 z2

1/3 1/3
Then the isoquant satisfies the equation z1 z2 = k. Rearranging, we can solve for z2 , yielding

k3
z2 = (1.1)
z1

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

1.3 Marginal Rate of Technical Substitution

The slope of the isoquant measures the rate at which the agent is willing to substitute one good
for another. This slope is called the marginal rate of technical substitution or MRTS.
Mathematically,

dz2 ØØ
M RT S = ° Ø (1.2)
dz1 f (z1 ,z2 )=const.

3
Eco11, Fall 2009 Simon Board

We can rephrase this definition in words: the MRTS equals the number of z2 the firm can
exchange for one unit of z1 in order to keep output constant.

The MRTS can be related to the firm’s production function. Let us consider the eÆect of a
small change in the firm’s inputs. Totally diÆerentiating the production function f (z1 , z2 ) we
obtain

@f (z1 , z2 ) @f (z1 , z2 )
dq = dz1 + dz2 (1.3)
@z1 @z2

Equation (1.3) says that the firm’s output increases by the marginal product of input 1 times
the increase in input 1 plus the marginal product of input 2 times the increase in input 2. Along
an isoquant dq = 0, so equation (1.3) becomes

@f (z1 , z2 ) @f (z1 , z2 )
dz1 + dz2 = 0
@z1 @z2

Rearranging,
dz2 @f (z1 , z2 )/@z1
° =
dz1 @u(z1 , z2 )/@z2
Equation (1.2) therefore implies that

M P1
M RT S = (1.4)
M P2

The intuition behind equation (1.4) is as follows. Using the definition of MRTS, one unit of z1
is worth MRTS units of z2 . That is, M P1 = M RT S £ M P2 . Rewriting this equation we obtain
(1.4).

1.4 Properties of Technology

In this section we present three properties of production functions that will prove useful.

1. Monotonicity. The production function is monotone if for any two input bundles z = (z1 , z2 )
and z 0 = (z10 , z20 ),

zi ∏ zi0 for each i o

implies f (z1 , z2 ) > f (z10 , z20 ).
zi > zi0 for some i
In words: the production function is monotone if more of any input strictly increases the
firm’s output. Monotonicity implies that isoquants are thin and downwards sloping (see the
Preferences Notes). As a result, it implies that MRTS is positive.

4
Eco11, Fall 2009 Simon Board

2. Quasi–concavity. Let z = (z1 , z2 ) and z 0 = (z10 , z20 ). The production function is quasi–concave
if whenever f (z) ∏ f (z 0 ) then

f (tz + (1 ° t)z 0 ) ∏ f (z 0 ) for all t 2 [0, 1] (1.5)

Suppose z and z 0 are two input bundles that produce the same output, f (z) = f (z 0 ). Then
(1.5) says a mixture of these bundles produces even more output. That is, mixtures of inputs
are better than extremes.

Under the assumption of monotonicity, quasi–concavity says that isoquants are convex. This
means that the MRTS decreasing in z1 along the isoquant. Formally, an isoquant defines an
implicit relationship between z1 and z2 ,

f (z1 , z2 (z1 )) = k

Convexity then implies that M RT S(z1 , z2 (z1 )) is decreasing in z1 . This is illustrated in Pref-
erences Notes.

3. Returns to Scale. A production function has decreasing returns to scale if

f (tz1 , tz2 ) ∑ tf (z1 , z2 ) for t ∏ 1 (1.6)

so that doubling the inputs less that doubles the output. A production function has constant
returns to scale if
f (tz1 , tz2 ) = tf (z1 , z2 ) for t ∏ 1

so that doubling the inputs also doubles output. Finally, a production function has increasing
returns to scale if
f (tz1 , tz2 ) ∏ tf (z1 , z2 ) for t ∏ 1

so that doubling the inputs more than doubles the output.

We will sometimes use the assumption that the production function f (z1 , z2 ) is concave. That
is, for z = (z1 , z2 ) and z 0 = (z10 , z20 ),

f (tz + (1 ° t)z 0 ) ∏ tf (z) + (1 ° t)f (z 0 ) for t 2 [0, 1] (1.7)

Concavity implies that the production function is quasi–concave (1.5) and hence that isoquants
are convex. This follows immediately from definitions: if f (z) ∏ f (z 0 ) then concavity (1.7)

5
Eco11, Fall 2009 Simon Board

implies
f (tz + (1 ° t)z 0 ) ∏ tf (z) + (1 ° t)f (z 0 ) ∏ f (z 0 )

so the production function is quasi–concave. In addition, concavity implies decreasing returns

to scale. Applying the definition of concavity (1.7) to the points z = sz 00 and z 0 = 0 for s ∏ 1,
and letting t = 1/s, we obtain
µ µ ∂ ∂ µ ∂
1 1 1 1
f (sz) + 1 ° 0 ∏ f (sz) + 1 ° f (0)
s s s s

Using f (0) = 0 and simplifying, we obtain (1.7)

2 Examples of Production Functions

Here we present some examples of production functions. Many details are omitted since this a
repetition of the examples of utility functions.

2.1 Cobb Douglas

A Cobb–Douglas production function is given by

f (z1 , z2 ) = z1Æ z2Ø for Æ ∏ 0 and Ø ∏ 0

Typical isoquants are shown in figure 1. The marginal products are given by

M P1 = Æz1Æ°1 z2Ø
M P2 = Øz1Æ z2Ø°1

The marginal rate of technical substitution is

M P1 Æz2
M RT S = =
M P2 Øz1

The returns to scale are easy to evaluate.

f (tz1 , tz2 ) = (tz1 )Æ (tz2 )Ø t = tÆ+Ø z1Æ z2Ø = tÆ+Ø f (z1 , z2 )

6
Eco11, Fall 2009 Simon Board

Figure 2: Isoquants for Leontief Technology. The isoquants are L–shaped, with the kink along the
line Æz1 = Øz2 .

Hence there are decreasing returns if Æ + Ø ∑ 1, constant returns if Æ + Ø = 1 and increasing

returns if Æ + Ø ∏ 1.

Exercise: Assume Æ + Ø ∑ 1. Show that f (z1 , z2 ) is concave.1

2.2 Perfect Complements (Leontief )

A Leontief production function is given by

f (z1 , z2 ) = min{Æz1 , Øz2 }

The isoquants are shown in figure 2. These are L–shaped with a kink along the line Æz1 = Øz2 .
This production function exhibits constant returns to scale.

2.3 Perfect Substitutes

With perfect substitutes, the production function is given by

f (z1 , z2 ) = Æz1 + Øz2

1
For the definition of concavity with two variables, see the p. 5–6 of the math notes.

7
Eco11, Fall 2009 Simon Board

Figure 3: Isoquants for Perfect Substitutes. The isoquants are straight line with slope °Æ/Ø.

The isoquants are shown in figure 3. These are straight lines with slope °Æ/Ø. This production
function exhibits constant returns to scale.

3 Cost Minimisation Problem (CMP)

We make several assumptions:

1. There are N inputs. For much of the analysis we assume N = 2 but nothing depends on
this.

2. The agent takes input prices as exogenous. We assume these prices are linear and strictly
positive and denote them by {r1 , . . . , rN }.

3. The firm has production technology f (z1 , z2 ). We normally assume that the production
function is diÆerentiable, which ensures that any optimal solution satisfies the Kuhn–
Tucker conditions. If the production function is quasi–concave and M Pi (z1 , z2 ) > 0 for
all (z1 , z2 ), then any solutions to the Kuhn–Tucker conditions are optimal. See Section
4.1 of the UMP notes for more details.

8
Eco11, Fall 2009 Simon Board

3.1 Cost Minimisation Problem

The cost minimisation problem is

N
X
min ri zi subject to f (z1 , . . . , zN ) ∏ q (3.1)
z1 ,...,zN
i=1
zi ∏ 0 for all i

The idea is that the firm is trying to find the cheapest way to attain a certain output, q. The
solution to this problem yields the firm’s input demands which are denoted by

zi§ (r1 , . . . , rN , q)

The money the firm must spend in order to attain its target output is its cost. The cost
function is therefore
N
X
c(r1 , . . . , rN , q) = min ri zi subject to f (z1 , . . . , zN ) ∏ q
z1 ,...,zN
i=1
zi ∏ 0 for all i

Equivalently, the cost function equals the amount the firm spends on her optimal inputs,

N
X
c(r1 , . . . , rN , q) = ri zi§ (r1 , . . . , rN , q) (3.2)
i=1

Note this problem is formally identical to the agent’s expenditure minimisation problem. The
cost function is therefore equivalent to the agent’s expenditure function.

Given a cost function, the average cost is,

c(r1 , r2 , q)
AC(r1 , r2 , q) =
q

The marginal cost equals the cost of each additional unit,

dc(r1 , r2 , q)
M C(r1 , r2 , q) =
dq

9
Eco11, Fall 2009 Simon Board

Figure 4: Constraint Set. This figure shows the set of inputs that deliver the target output, q.

3.2 Graphical Solution

The firm wishes to find the cheapest way to attain a certain output.

First, we need to understand the constraint set. The firm can choose any bundle of inputs
where (a) the firm attains her target output, f (z1 , z2 ) ∏ q; and (b) the quantities are positive,
z1 ∏ 0 and z2 ∏ 0. If the firm’s production function is monotone, then the bundles that meet
these conditions are the ones that lie above the isoquant with output q. See figure 4.

Second, we need to understand the objective. The firm wishes to pick the bundle in the
constraint set that minimises her cost. Define an isocost curve by the bundles of z1 and z2 that
deliver constant cost:
{(z1 , z2 ) : r1 z1 + r2 z2 = const.}

These isocost curves are just like budget curves and so have slope °r1 /r2 . See figure 5.

Ignoring boundary problems and kinks, the solution to the CMP has the feature that the isocost
curve is tangent to the target isoquant. As a result, their slopes are identical. The tangency
condition can thus be written as

r1
M RT S = (3.3)
r2

This is illustrated in figure 6.

10
Eco11, Fall 2009 Simon Board

Figure 5: Isocost. The isocost function shows the set of inputs which cost the same amount of money.

Figure 6: Tangency. This figure shows that, at the optimal input combination, the isocost curve is
tangent to the isoquant.

11
Eco11, Fall 2009 Simon Board

The intuition behind (3.3) is as follows. Using the fact that M RT S = M P1 /M P2 , equation
(3.3) implies that

M P1 r1
= (3.4)
M P2 r1

Rewriting (3.4) we find

r1 r2
=
M P1 M P1

The ratio ri /M Pi measures the cost of increasing output by one unit. At the optimum the agent
equates the cost–per–unit of the two goods. Intuitively, if good 1 has a higher cost–per–unit
than good 2, then the agent should spend less on good 1 and more on good 2. In doing so, she
could attain the same output at a lower cost.

If the production function is monotone, then the constraint will bind,

f (z1 , z2 ) = q. (3.5)

The tangency equation (3.4) and constraint equation (3.5) can then be used to solve for the
two input demands. In addition, one can derive the cost function using equation (3.2).

If there are N inputs, the agent will equalise the cost–per–unit from each good, giving us N ° 1
equations. Using the constraint equation (3.5), we can again solve for the firm’s input demands.

3.3 Example: Cobb Douglas

1/3 1/3
Suppose a firm has production function f (z1 , z2 ) = z1 z2 . The MRTS is

1 °2/3 1/3
3 z1 z2 z2
M RT S = 1/3 °2/3
=
1 z1
3 z1 z2

The tangency condition from the CMP is thus

r1 z2
=
r2 z1

Rewriting, this says r1 z1 = r2 z2 , so the firm spends the same on both its inputs.

12
Eco11, Fall 2009 Simon Board

Figure 7: Cost curves. This figure shows the cost, average cost and marginal cost curves for the
Cobb–Douglas example.

1/3 1/3
The constraint equation is q = z1 z2 . This means that

r1 z1
q 3 = z1 z2 = z1
r2

where the second equality uses the tangency condition. Rearranging, we find the optimal input
demands are µ ∂1/2 µ ∂1/2
r1 r2
z1§ = q 3/2
and z2§ = q 3/2
r2 r1
The cost function is
c(r1 , r2 , q) = r1 z1§ + r2 z2§ = 2(r1 r2 )1/2 q 3/2

The average and marginal costs are

AC(r1 , r2 , q) = 2(r1 r2 )1/2 q 1/2 and M C(r1 , r2 , q) = 3(r1 r2 )1/2 q 1/2

These are illustrated in figure 7.

3.4 Lagrangian Solution

Using a Lagrangian, we can encode the tangency conditions into one formula. As before, let us
ignore boundary problems. The CMP can be expressed as minimising the Lagrangian

L = r1 z1 + r2 z2 + ∏[q ° f (z1 , z2 )]

13
Eco11, Fall 2009 Simon Board

As usual, the term in brackets can be thought as the penalty for violating the constraint. That
is, the firm is punished for falling short of the target output.

The FOCs with respect to z1 and z2 are

@L @u
= r1 ° ∏ =0 (3.6)
@z1 @z1
@L @u
= r2 ° ∏ =0 (3.7)
@z2 @z2

If the production function is monotone then the constraint will bind,

f (z1 , z2 ) = q (3.8)

These three equations can then be used to solve for the three unknowns: z1 , z2 and ∏.

Several remarks are in order. First, this approach is identical to the graphical approach. Di-
viding (3.6) by (3.7) yields
@u/@z1 r1
=
@u/@z2 r2
which is the same as (3.4). Moreover, the Lagrange multiplier is exactly the cost–per–unit,

r1 r2
∏= = .
M P1 M P2

Second, if preferences are not monotone, the constraint (3.5) may not bind. If it does not bind,
the Lagrange multiplier in the FOCs will be zero.

Third, the approach is easy to extend to N inputs. In this case, one obtains N first–order
conditions and the constraint equation (3.5).

3.5 Properties of Cost Functions

We now develop six properties of the cost function. The first four are identical to the properties
of the expenditure function: see the EMP Notes for more details.

1. The cost function is homogenous of degree one in prices. That is,

c(r1 , r2 , q) = c(tp1 , tp2 , q) for t > 0

14
Eco11, Fall 2009 Simon Board

Figure 8: Concavity of Cost Function in Input Prices. This figure shows how the cost function
lies under the pseudo–cost function.

Intuitively, if the prices of r1 and r2 double, then the cheapest way to attain the target output
does not change. However, the cost of attaining this output doubles.

2. The cost function is increasing in (r1 , r2 , q). If we increase the target output then the
constraint becomes harder to satisfy and the cost of attaining the target increases. If we
increase r1 then it costs more to buy any bundle of inputs and it costs more to attain the target
output.

3. The cost function is concave in input prices (r1 , r2 ). Fix the target utility q and prices
(r1 , r2 ) = (r10 , r20 ). Solving the CMP we obtain input demands z10 = z1§ (r10 , r20 , q) and z20 =
z2§ (r10 , r20 , q). Now suppose we fix demands and change r1 , the price of input 1. This gives us a
pseudo–cost function
cz10 ,z20 (p1 ) = r1 z10 + r20 z20

which is linear in r1 . Of course, as r1 rises the firm can reduce her costs by rebalancing her
input demand towards the input that is cheaper. This means that real cost function lies below
the pseudo–cost function and is therefore concave. See figure 8.

4. Sheppard’s Lemma: The derivative of the cost function equals the input demand. That is,

@
c(r1 , r2 , q) = z1§ (r1 , r2 , q) (3.9)
@r1

The idea behind this result can be seen from figure 8. At r1 = r10 the cost function is tangential

15
Eco11, Fall 2009 Simon Board

to the pseudo–cost function. The pseudo–cost is linear in r1 with slope z1§ (r10 , r20 , q), so the
expenditure function also has slope z1§ (r10 , r20 , q).

The intuition behind Sheppard’s Lemma is as follows. When r1 increases by ¢r1 there are
two eÆects. First, holding input demand constant, the firm’s cost rises by z1§ (r1 , r2 , q) £ ¢r1 .
Second, the firm rebalances its demands, buying less of input 1 and more of input 2. However,
this has a small eÆect on the firm’s costs since it is close to indiÆerent buying the optimal
quantity and nearby quantities.

5. If f (z1 , z2 ) is concave then c(r1 , r2 , q) is convex in q. Intuitively, concavity of the production

function, implies that the marginal product of an input is decreasing in the amount of the input
used:
d d2
M Pi (z1 , z2 ) = 2 f (z1 , z2 ) ∑ 0
dzi dzi
Therefore, as the firm expands, it needs more inputs to produce each additional unit of output.
As a result, the cost of producing this unit increases, and the total cost is convex. When there
is only one input this is easy to see formally: if f (z) is concave, then c(q) = rf °1 (q) is convex.

6. AC(q) is increasing when M C(q) ∏ AC(q), is flat when M C(q) = AC(q) and is and
decreasing when M C(q) ∑ AC(q). Suppose the firm currently produces n units of output, and
that the marginal cost of the (n + 1)st unit is higher than the average cost of the first n. Then
the average cost of producing n + 1 units is higher that producing n units since the costs is
being dragged up by the final unit. To prove this result formally, we can diÆerentiate the AC
curve,
d d c(q) c0 (q)q ° c(q)
AC(q) = =
dq dq q q2
Hence AC(q) is increasing if and only if c0 (q)q ∏ c(q). Rearranging, this condition is just
M C(q) ∏ AC(q), as required.

3.6 Pictures of Cost Functions

Figure 7 shows the cost curves associated with a concave production function. One can see
that the cost function is convex and, as a result, the marginal cost is increasing and exceeds
the average cost.

Figure 9 shows the cost curves associated with a production function which is concave for

16
Eco11, Fall 2009 Simon Board

Figure 9: Cost Curves for a Nonconcave Production Function I: Fixed Cost. This figure
shows the cost, average cost and marginal cost curves when the firm must pay a fixed cost.

positive quantities but requires a fixed cost needed to initiate production.2 The marginal cost
of the first unit is infinite and is therefore not shown in the picture; the marginal cost of each
subsequent unit is increasing. The average cost is U–shaped: it starts at infinity, is minimised at
q 0 and then rises as the higher marginal cost drags up the average cost. Note that the marginal
cost intersects the average cost at its lowest point: this follows from property 6 from Section
3.5.

Figure 10 shows the cost curves associated with a second nonconcave production function.3
The cost curve is S–shaped. As a result, the marginal cost and average cost functions are U–
shaped. For the first unit, the marginal cost and average cost coincide; for low levels of output,
the marginal cost is decreasing and lies below the average cost; for high levels of output, the
marginal cost is increasing, exceeding the average cost for q ∏ q 0 .

3.7 Long Run vs. Short Run Costs

The cost of a firm depend on which factors of production are flexible. We diÆerentiate between
four cases, and then illustrate them with an example.4
2
For example, try f (z) = (z ° 1)1/2 .
3
For example, try f (z) = 100z ° 16z 2 + z 3 .
4
While the idea of short and long run is standard, diÆerent authors mean diÆerent things by the “short run”
and “long run”.

17
Eco11, Fall 2009 Simon Board

Figure 10: Cost Curves for a Nonconcave Production Function II. This figure shows the cost,
average cost and marginal cost curves.

1. In the very short run all the factors of production are fixed, and output is fixed.

2. In the short run some factors are flexible, while others are fixed. For example, the
firm may be able hire some more workers, but may not be able to order new capital
equipment. Any fixed costs are also sunk, so that they cannot be avoided even if the firm
ceases production.

3. In the medium run all factors are flexible, but fixed costs are sunk.

4. In the long run all factors are flexible and fixed costs are not sunk. Hence the firm can
costlessly exit.

In practice, the meaning of short and long run depend on the application. For example, consider
a farmer who wishes to increase her output. It may take her a few days to hire an extra worker,
a few weeks to lease an extra tractor and a few months for a new farmer to buy land and enter
the business (or for an old one to exit).

To illustrate, suppose a firm has production function5

f (z1 , z2 ) = (z1 ° 1)1/3 (z2 ° 1)1/3

This firm has Cobb–Douglas production, except that the first unit of both inputs is useless,
inducing a fixed cost.
5
Since negative outputs are impossible, we should say that q = 0 if either z1 < 1 or z2 < 1.

18
Eco11, Fall 2009 Simon Board

First, let us solve for the long–run cost function. The firm’s Lagrangian is

L = r1 z1 + r1 z2 + ∏[q ° (z1 ° 1)1/3 (z2 ° 1)1/3 ]

DiÆerentiating, this induces the tangency condition r1 (z1 °1) = r2 (z2 °1). Using the constraint,
q = (z1 ° 1)1/3 (z2 ° 1)1/3 ° 1, we obtain
µ ∂1/2 µ ∂1/2
r1 r2
z1§ = (q)
3/2
+1 and z2§ = (q)3/2 + 1
r2 r1

The cost function is

c(r1 , r2 , q) = r1 z1§ + r2 z2§ = 2(r1 r2 )1/2 (q)3/2 + (r1 + r2 )

In addition, the firm can shutdown and produce zero at cost c(r1 , r2 , 0) = 0. Observe that this
cost function is the same as that in Section 3.3 with a startup cost of r1 + r2 .

In the medium run, the fixed cost r1 + r2 is sunk. The medium run cost curve is therefore

c(r1 , r2 , q) = 2(r1 r2 )1/2 (q)3/2 + (r1 + r2 )

where c(r1 , r2 , 0) = r1 + r2 .

In the short run, z1 is flexible but z2 is fixed at z20 . The fixed cost is also sunk. The constraint
in the CMP becomes
q = (z1 ° 1)1/3 (z20 ° 1)1/3

Rearranging,
q3
z1§ = +1
z20 ° 1
The cost function is therefore given by

q3
c(r1 , r2 , q; z20 ) = r1 z1§ + r2 z20 = r1 + r1 + r2 z20
z20 ° 1

Figure 11 illustrates the short run cost curves for three diÆerent levels of z2 . Observe that the
long run cost curve is given by the lower envelope of the short run cost curves. To see why this
is the case, fix an output level q 0 and calculate the optimal input demands when both factors are
flexible, denoted by z10 and z20 . Now suppose we fix z2 at z20 and consider the cost of attaining
diÆerent output levels. If q = q 0 then the firm is using the optimal amount of input 2 and the
short–run cost will coincide with the long–run cost. If q > q 0 then the firm is using too little of

19
Eco11, Fall 2009 Simon Board

Figure 11: Long Run and Short Run Costs. This figure shows the long run cost curve and the
short run cost curves corresponding to three levels of the second input.

z2 and too much of z1 , raising the short–run cost over the long–run cost. If q < q 0 then the firm
is using too much of z2 and too little of z1 , again raising the short–run cost over the long–run
cost.

In the very short run, inputs are fixed at z1 = z10 and z2 = z20 . Hence the firm can produce
q 0 = (z10 ° 1)1/3 (z20 ° 1)1/3 at cost r1 z10 + r1 z20 , but is unable to produce anything else.

4 Profit Maximisation Problem (PMP)

Assumptions:

1. There is one output good, with linear price p. This means that the firm is a price–taker
in the output market.

2. There are two input goods with linear prices r1 and r2 . The firm is therefore a price–taker
in the input market.

3. The firm has production technology f (z1 , z2 ). We normally assume that the production
function is diÆerentiable, which ensures that any optimal solution satisfies the first–order
conditions.

20
Eco11, Fall 2009 Simon Board

The firm’s profit equals its revenue from selling the output minus it’s cost:

º = pf (z1 , z2 ) ° r1 z1 ° r2 z2

We now explore two ways of solving this problem.

4.1 One–Step Solution

The firm’s profit maximisation problem is

max pf (z1 , z2 ) ° r1 z1 ° r2 z2 subject to zi ∏ 0 for all i (4.1)

z1 ,z2

The first–order conditions are

dº @f (z1 , z2 )
=p ° r1 = 0 (4.2)
dz1 @z1
dº @f (z1 , z2 )
=p ° r2 = 0 (4.3)
dz2 @z2

Together (4.2) and (4.3) define the optimal input demands of the firm, z1§ (p, r1 , r2 ) and z2§ (p, r1 , r2 ).
we can then derive the optimal output:

q § (p, r1 , r2 ) = f (z1§ , z2§ )

which is called the supply function. We can also derive the firm’s optimal profit,

º § (p, r1 , r2 ) = pq § ° r1 z1§ ° r2 z2§

which is called the profit function.

Observe that solving (4.1) is much easier than solving the utility maximisation problem. With
the UMP, the consumer maximises her utility subject to spending no more than her income.
With the PMP, the firm’s expenses directly enter the firm’s objective function, so we only have
to solve an unconstrained optimisation problem.

In order for the FOCs (4.2) and (4.3) to characterise a maximum, the second–order conditions

21
Eco11, Fall 2009 Simon Board

must hold. That is, f (z1 , z2 ) must be locally concave, which implies

@2 @
f (z1 , z2 ) = M P1 (z1 , z2 ) ∑ 0
@z12 @z1
@2 @
f (z1 , z2 ) = M P2 (z1 , z2 ) ∑ 0
@z22 @z2

If f (z1 , z2 ) is globally concave, then any solution to the FOCs is a maximum.

4.2 Example: Cobb Douglas

1/3 1/3
Suppose a firm has production function f = z1 z2 . Profit is given by

1/3 1/3
º = pz1 z2 ° r1 z1 ° r2 z2

The FOCs are

1 °2/3 1/3
pz z2 = r1
3 1
1 1/3 °2/3
pz z = r2
3 1 2

Solving these two equations yields input demands:

1 p3 1 p3
z1§ (p, r1 , r2 ) = and z2§ (p, r1 , r2 ) =
27 r12 r2 27 r1 r22

The optimal supply is

1 p2
q § (p, r1 , r2 ) = (z1§ )1/3 (z2§ )1/3 =
9 r1 r2
The profit function is
1 p3
º § (p, r1 , r2 ) = pq § ° r1 z1§ ° r2 z2§ =
27 r1 r2

4.3 Two–Step Solution

Step 1. Find the cheapest way to attain output q. Recall the cost function is given by

c(q, r1 , r2 ) = min r1 z1 + r2 z2 subject to f (z1 , z2 ) ∏ q

z1 ,z2
zi ∏ 0 for all i

22
Eco11, Fall 2009 Simon Board

Step 2. Find the profit–maximising output. Given a cost function, the firm’s problem is

max º = pq ° c(q, r1 , r2 ) subject to q ∏ 0

q

The first–order condition for this problem is

dº d
= p ° c(q, r1 , r2 ) = 0
dq dq

That is,

p = M C(q, r1 , r2 ) (4.4)

The idea behind this result is shown in the left panel of figure 12, which shows the firm’s revenue
and costs as a function of output, q. The firm wishes to maximise the vertical distance between
the two lines so, at the optimum, they are parallel. The slope of the revenue line is p while the
slope of the cost function is M C, which yields (4.4).

One can also look at this result with the right panel of figure 12. The diÆerence p ° M C
equals the profit the firm makes on the last unit. The FOC (4.4) says that the firm will keep
producing while the profit–per–unit is positive and will stop when it falls to zero. Note that, in
this picture, one can measure profits two ways. First, profit equals the price obtained per unit
minus the average cost of a unit multiplied by the number of units sold:

º(q) = pq ° c(q) = pq ° AC(q)q = [p ° AC(q)]q

In the picture, this equals the areas given by A+B+C.

Second, the profit of a marginal unit is p ° M C(q). Hence the total profit of the firm, ignoring
fixed costs, is the area below the price and above the MR curve. That is,
Z q Z q Z q
º(q) = pq ° c(q) = pdq̃ ° M C(q̃)dq̃ ° F = [p ° M C(q̃)] dq̃ ° F
0 0 0

where F is the fixed cost. Hence the firm’s profit is A+B+D+E minus the fixed cost, F .

In order for the FOC (4.4) to constitute an optimum, the second–order condition should hold:

d2 º d2 d
= ° c(q, r1 , r2 ) = ° M C(q, r1 , r2 ) ∑ 0
dq 2 dq 2 dq

So the marginal cost needs to be locally increasing. Conversely, if the cost function is convex,

23
Eco11, Fall 2009 Simon Board

Figure 12: Profit maximisation. The left panel shows that profit is maximised when the revenue line
is parallel to the cost line. The vertical gap, is then equal to the firm’s profit. The right panel shows
that profit is maximised when the price equals to marginal cost. Profit then equals A+B+C.

which is guaranteed by the concavity of f (z1 , z2 ), then any solution to the FOC (4.4) is an
optimum.

4.4 Example: Cobb Douglas

We now return to the example in Section 4.2, deriving the same results using the two–step
approach.

1/3 1/3
Suppose f (z1 , z2 ) = z1 z2 . Using the results in Section 3.3, the cost function is

c(q, r1 , r2 ) = 2(r1 r2 )1/2 q 3/2

The first–order condition (4.4) yields

p = 3(r1 r2 q)1/2

Rearranging, the supply curve is given by

1 p2
q § (p, r1 , r2 ) =
9 r1 r2

24
Eco11, Fall 2009 Simon Board

The profit function is then

1 p3
º § (p, r1 , r2 ) = pq § ° r1 z1§ ° r2 z2§ =
27 r1 r2

as in Section 4.2.

4.5 Examples of Supply Functions

Figure 13 shows the supply function that results from a convex cost function with no fixed
cost.6 The marginal cost is increasing and is always above the average cost. For any given
price, the firm chooses quantity such that p = M C(p). Hence the supply curve coincides with
the M C curve.

Figure 14 shows the supply function that results from a convex cost function with a fixed cost.7
The marginal cost function is increasing so, if the firm produces, its supply curve coincides with
M C(q). However, when the price lies below the average cost, the firm makes negative profits.
Hence the firm’s supply curve coincides with the M C(q) curve above the AC(q) curve and is
zero elsewhere.

Figure 15 shows the supply function that results from a U–shaped marginal cost function
without a fixed cost.8 For prices below p0 the marginal cost is below the average cost, so the
firm cannot make a profit and it chooses to produce q § (p) = 0. At p = p0 the firm is indiÆerent
between producing 0 and q 0 . For price above p0 the firm produces on the increasing part of the
marginal cost function.

Figure 16 shows the supply function that results from a nonconvex cost curve.9 For low prices
the supply curve coincides with the first part of the M C curve. At a price p0 the supply jumps
to the right. Intuitively, if the firm is going to pay to produce the expensive units in region A
then it should also produce the cheap units in region B. At the optimum, the area of A equals
the area of B, so the profit lost by producing the expensive units is exactly oÆset by the profit
gained by producing the cheap units.

One can also use these figures to understand the diÆerence between the short–run and long–
run supply curves. In the very short run, supply is fixed and the supply curve is vertical. In
6
For example, try c(q) = q + q 2 .
7
For example, try c(q) = 1 + q + q 2 .
8
For example, try c(q) = 15q ° 12q 2 + q 3 .
9
For example, try c(q) = 20q 2 ° 8q 3 + q 4 .

25
Eco11, Fall 2009 Simon Board

Figure 13: Supply Curve with Convex Costs. This figure shows how the supply curve coincides
with the marginal cost curve.

Figure 14: Supply Curve with Nonconvex Costs I: Fixed Costs. This figure shows how the
supply curve coincides with the marginal cost curve when it lies above the average cost.

26
Eco11, Fall 2009 Simon Board

Figure 15: Supply Curve with Nonconvex Costs II: U–Shaped Marginal Cost. This figure
shows how the supply curve coincides with the marginal cost curve when it lies above the average cost.

Figure 16: Supply Curve with Nonconvex Costs III. This figure shows how the supply curve
coincides with the marginal cost curve when it lies above the average cost.

27
Eco11, Fall 2009 Simon Board

the short–run, some of the inputs are fixed and the supply curve coincides with the short–run
marginal cost. In the medium–run, the firm can change all its inputs, but cannot close down.
Hence the supply curve coincides with the marginal cost curve above the average variable cost.
In the long–run the firm can shut down, so the supply curve coincides with the marginal cost
above the average cost.

4.6 Properties of the Profit Function

The profit function º § (p, r1 , r2 ) has four key properties:

1. º § (p, r1 , r2 ) is homogenous of degree one in (p, r1 , r2 ). If all prices double then the opti-
mal production choices remain unchanged and profit also doubles. Intuitively, if currency is
denominated in a diÆerent currency this should not aÆect the firm’s choices.

2. º § (p, r1 , r2 ) is increasing in p and decreasing in (r1 , r2 ). An increase in p increases profits

for any output q, and therefore increases profit for the optimal output choice. An increase in
r1 increases costs and decreases profits for any output q, and therefore decreases profit for the
optimal output choice.

3. º § (p, r1 , r2 ) is convex in (p, r1 , r2 ). Let us first consider changes in p, and ignore the input
prices. Fix p = p0 and solve for the optimal output q 0 = q § (p0 ). Now suppose we fix the output
and change p, yielding a pseudo–profit function pq 0 ° c(q 0 ) which is linear in p. Of course, as
p rises the firm can increase her output, so the real cost function lies above this straight line
and is therefore convex. See figure 16. Second, the profit function is convex in (r1 , r2 ) because
profit is equal º = pq ° c(q, r1 , r2 ) and c(q, r1 , r2 ) is concave in (r1 , r2 ).

4. Hotelling’s Lemma: The derivative of the profit function with respect to the output price
equals the optimal output. That is,

@ §
º (p, r1 , r2 ) = q § (p, r1 , r2 ) (4.5)
@p

The idea behind this result can be seen from figure 16. At p = p0 the profit function is tangential
to the pseudo–profit function. The pseudo–profit is linear in p with slope q § (p0 ). Hence the
expenditure function also has slope q § (p).

The intuition behind Hotelling’s Lemma can be seen in figure 17. We start at p = p0 , with

28
Eco11, Fall 2009 Simon Board

Figure 17: Convexity of Profit Functions This figure shows how the profit function equals the upper
envelope of the pseudo–profit functions, pq ° c(q).

profit equal to area A.10 When the price increases to p00 there are two eÆects. First, holding
output constant, the firm’s profit rises by q § (p) £ (p00 ° p0 ), illustrated by area B. Second, the
firm increases its output, yielding extra profit C. However, for small price changes this second
eÆect is small, which yields Hotelling’s Lemma. One can also see from this picture that profit
is convex in price: output is higher when the price is higher, so the change in profit induced by
a 1¢ increase in the price is higher when the price is higher.

4.7 Properties of Supply Functions

There are two important properties of the supply function.

1. Supply q § (p, r1 , r2 ) is homogenous of degree zero in (p, r1 , r2 ). If prices are denominated in

a diÆerent currency this will not aÆect the firm’s optimal output.

2. Law of Supply: q § (p, r1 , r2 ) is increasing in p. The supply curve is always upward sloping.
Intuitively, an increase in the price increases the benefits to producing and so increases the
optimal output. Formally, Hotelling’s Lemma implies that

d § d2
q (p, r1 , r2 ) = 2 º § (p, r1 , r2 ) ∏ 0
dp dp
10
Note there are no fixed costs in this picture

29
Eco11, Fall 2009 Simon Board

Figure 18: Convexity of Profit Functions This figure shows how the profit function is convex in the
price and that the derivative equals the current supply.

where the inequality come from the convexity of the profit function.

30
 Partial Equilibrium: Positive Analysis

Simon Board§

This Version: November 28, 2009

First Version: December 1, 2008.

In this Chapter we consider consider the interaction between diÆerent agents and firms, and
solve for equilibrium prices and quantities.

Section 1 introduces the idea of partial equilibrium. Section 2 looks at how we aggregate
agent’s demand curves and firm’s supply curves to form market demand and market demand
supply. Section 3 defines an equilibrium, and discusses basic properties thereof. Section 4 looks
at short-run equilibrium, where entry and exit are not possible. Finally, Section 5 considers
long-run equilibrium, where entry and exit are possible.

1 Partial Equilibrium

When studying partial equilibrium, we consider the equilibrium in one market, taking as ex-
ogenous prices in other markets and agents’ incomes, as well as preferences and technology.
The main advantage of this model is simplicity: the equilibrium price is found by equating
supply and demand. The model can also be used for welfare analysis, evaluating the eÆect of
tax changes or the introduction of tariÆs. However, the assumption that we can analyse one
market independently of others can be dubious in some cases.

As an alternative, economists sometimes study general equilibrium. In this model, we fix

preferences and technology and suppose agents are endowed with goods and shares. We then
§
Department of Economics, UCLA. http://www.econ.ucla.edu/sboard/. Please email suggestions and typos
to [email protected].

1
Eco11, Fall 2008 Simon Board

solve for all prices simultaneously, equating supply and demand in each market. While this
approach is far more general (hence the name), it is harder to analyse.

To illustrate the diÆerence between partial and general equilibrium consider the worldwide
market for cars.1 A partial equilibrium analysis would add up the world’s demand for cars
to form a market demand curve. It would also add up the diÆerent firms’ supply curves to
form a market supply curve. The price of cars can then be found where demand equals supply.
Intuitively, if the price were lower the would be excess demand and the price would be bid
up; if the price were higher there would be excess supply and competition among suppliers
would drive the price down. An exogenous increase in demand from China, due to Government
construction of highways, would then shift up the demand curve, raising equilibrium price and
quantity. In a general equilibrium model, there would be many other eÆects. First, the value of
car firms would rise, increasing the income of their shareholders. Second, the increased demand
for cars would push up the price of complements, such as oil. Third, there would be an increase
in demand for inputs, such as steel, so commodity prices would rise. Ultimately, there is no
single correct model: rather, there is a tradeoÆ between complexity and realism which depends
on the markets at hand and the questions one is interested in.

1.1 Partial Equilibrium Model

We make the following assumptions:

1. We are interested in market 1. The price of this good, p1 , is to be determined.2 We

assume that all consumers and firms face the same prices (the law of one price), and that
they are all price takers.

2. On the demand side, there are J agents who desire good 1. Each agent j has income mj
and utility uj (x1 , . . . , xN ). The consumer spends her income on N outputs which have
prices {p1 , p2 . . . , pN }, where {p2 . . . , pN } are exogenous. For simplicity, we often take
N = 2.

3. On the supply side, there are K firms who sell good 1. Each firm k has a production tech-
nology f k (z1 , . . . , zM ), where the M input prices are given exogenously by {r1 . . . , rM }.
For simplicity, we often take M = 2.
1
We assume that there is one type of car. This assumption is highly stylised but, recall, the purpose of a
model is not to describe the real world exactly, but to make useful abstractions.
2
For simplicity, we will sometimes denote the price of this good by p.

2
Eco11, Fall 2008 Simon Board

Figure 1: Summing Demand. Market demand is the horizontal sum of individual demands.

2 Market Demand and Supply

2.1 Market Demand

Solving the consumer’s utility maximisation problem, we can derive agent j’s Marshallian de-
mand for good 1,
xj1 (p1 , . . . , pN ; mj )

We can then form market demand by summing the individual agent’s demands:

I
X
X1 (p1 , . . . , pN ; m1 , . . . , mJ ) = xj1 (p1 , . . . , pN ; mj )
j=1

Market demand depends on the tastes of the agents, the prices of goods, the number of agents
and the distribution of income in the economy. This means that if we redistribute money
from every agent to one special agent (think Bill Gates), then this will change the demand for
high-value items like yachts.

Figure 1 shows that when we sum demand, we are eÆectively adding the demand curves hori-
zontally.

3
Eco11, Fall 2008 Simon Board

2.2 Market Supply

Solving the firm’s profit maximisation problem, we can derive firm k’s supply of good 1,

q k (p1 , r1 . . . , rM )

We can then form market supply by summing the individual firm’s supply functions:

I
X
Q(p1 , r1 . . . , rM ) = q k (p1 , r1 . . . , rM )
k=1

Note that market demand depends on the technologies of the firms, the price of good 1, the
prices of the inputs and the number of firms. This means that if the number of firms increases
then the market supply curve will shift out; similarly, if the number of firms decreases then the
supply curve will shift in. We will see examples of this below.

3 Equilibrium

The equilibrium price of good 1, p§1 , is found by equating supply and demand,

X1 (p§1 , p2 ; m1 , . . . , mI ) = Q(p§1 , r1 , r2 ) (3.1)

where, for simplicity, we assume there are 2 output goods and 2 input goods. Figure 2 shows
the classic demand and supply picture.

The idea behind equation (3.1) is that if there is excess demand then consumers will bid the
market price up, and if there is excess supply firms will compete for customers and reduce the
market price.3

We now consider two general questions concerning the equilibrium price. First, does there exist
a price p§1 that satisfies (3.1)? Second, is there one one price p§1 that satisfies (3.1) or many?
3
This process is called tatonnement. One problem with this intuition is that it contradicts our assumption of
price taking: how can firms undercut each other when they cannot aÆect the price? This suggests that the correct
interpretation of price taking is not that firms cannot oÆer a diÆerent price, but that they have no incentive to
do so.

4
Eco11, Fall 2008 Simon Board

Figure 2: Equilibrium. In equilibrium, supply equals demand.

3.1 Existence

Does a price, p§1 , exist that equates supply and demand? We can ensure existence if three
conditions hold, all of which are satisfied in Figure 2.

1. For high prices, supply exceeds demand. This is trivially satisfied since demand must fall
to zero as the price rises to infinity.

2. For low prices, demand exceeds supply. This is easy to satisfy since one would expect
supply to fall to zero as prices converge to zero.

3. The supply and demand functions are continuous in p1 .

Figure 3 shows an example with one firm and one consumer where an equilibrium does not
exist. The problem in this example is that at price p0 the consumer wants 2 units, while the
firm wishes to supply either 0 or 3 units. This problem is caused by the fixed cost: if the
firm’s production function is concave then its supply curve will be continuous. However, even
if the individual firms’ supply functions contain jumps, this is not a problem if the market is
su±ciently large. In the example, if there are 3 identical agents then we can have two firms
producing q = 3. The total demand then equals 2 + 2 + 2 = 6, while the total supply equals

5
Eco11, Fall 2008 Simon Board

Figure 3: Nonexistence of Equilibrium. Due to a fixed cost, supply jumps at price p0 . As a result,
no equilibrium exists.

3 + 3 = 6, and we have an equilibrium.4

3.2 Uniqueness

In Figure 2 there is a unique equilibrium price. Can there be more? From the law of supply we
know that the supply curve is upward sloping. If the good is ordinary for each agent then the
market demand is downward sloping and there can at most be one equilibrium. In contrast, if
the good is a GiÆen good, then there may be multiple equilibria, as shown in Figure 4.

4 Short-Run Equilibrium

In Section 3 we discussed equilibrium in general. It is useful to diÆerentiate between the

following time periods:5
4
Without fixed costs, equilibrium may also fail to exist if the marginal cost curve is first increasing, then
decreasing, and increasing again. This causes a discontinuity in the supply function, as shown in the Production
Chapter. Again, this is not a problem if the market is su±ciently large.
5
In the Production Chapter, we considered a fourth case when some factors are flexible, but others are not.
We omit this here for simplicity.

6
Eco11, Fall 2008 Simon Board

Figure 4: Multiple Equilibria. In this figure, the demand curve has an upward sloping component.
As a result, there are three equilibria with quantity Q1 , Q2 and Q3 being sold.

1. In the very short run all the factors of production are fixed, and output is fixed.

2. In the short run all factors are flexible, but fixed costs are sunk. Firms cannot enter or
exit.

3. In the long run all factors are flexible and fixed costs are not sunk. Hence firms can
enter and exit freely.

In the very short-run, the analysis is straightforward since output is fixed. We solve a numerical
example in Section 5.3.

In this Section we analyse short-run equilibrium, assuming the number of firms is fixed. In this
case, each firm operates on its individual supply curve, and market supply equals the sum of
individual firm supply.

4.1 Example

Suppose there are 15 agents, ten of whom have income mj = 10 and five of whom have mj = 20.
Each agent has symmetric Cobb-Douglas utility,

uj (x1 , x2 ) = x1 x2

7
Eco11, Fall 2008 Simon Board

Solving the agent’s utility maximisation problem, the demand function of each agent for good
1 is
mj
xj1 =
2p1
Summing over demand curves, market demand is given by

200 100
X1 = =
2p1 p1

Next, suppose there are 9 firms with production function

f k (z1 , z2 ) = (z1 ° 1)1/3 (z2 ° 1)1/3

Solving the firm’s cost maximisation problem, its cost function is

ck (q, r1 , r2 ) = 2(r1 r2 )1/2 q 3/2 + (r1 + r2 )

Solving the profit-maximisation problem, its supply function is given by

1 p21
qk =
9 r1 r2

Summing over firms, market supply is thus

p21
Q=
r1 r2

Equating supply and demand,

p§1 = (100r1 r2 )1/3

4.2 Shifts in Supply and Demand

We are often interested in how shifts in supply or demand aÆects prices and quantities. A
shift in demand may be due to changes in income (e.g. a recession), changes in the price of
substitutes or complements, changes in the quality of the goods, and so forth. Similarly, a shift
in the supply curve may be due to changes in input prices, the number of producers or changes
in technology.

Figure 5 shows the eÆect of an increase in demand. The left-hand side shows the market from
the firm’s perspective, while the right-hand side shows the entire market. The increase in
demand causes the equilibrium price to rise and each firm to move up its supply function. As

8
Eco11, Fall 2008 Simon Board

Figure 5: Increase in Demand. This figure shows the eÆect on an increase in demand from the firm’s
perspective (left), and the market’s perspective (right). The original demand is D, with price p and
quantities q and Q. When demand shifts up to D0 , the price rises to p0 and quantities rise to q 0 and
Q0 . Note that, in this picture, entry and exit are impossible so the firm’s supply curve coincides with its
marginal cost curve.

a result, each firm’s profit increases.

Figure 6 shows the eÆect of a shift in demand for diÆerent supply functions. When supply
is inelastic there is a small change in quantity and a large change in price. Conversely, when
supply is elastic, there is a large change in quantity and a small change in price.

Exercise: How does a shift in the supply curve aÆect price and quantity when the demand curve
is elastic/inelastic?

5 Long-Run Equilibrium

In the long run, we assume there are many identical potential entrants. This means that entry
will occur if there exists a quantity q such that p > AC(q). Conversely, exit will occur if
p < AC(q). As a result price must be driven down to the minimum average cost

p = min AC(q).
q∏0

9
Eco11, Fall 2008 Simon Board

Figure 6: Shifts in Demand. This picture shows the eÆect of an increase in demand when supply is
inelastic (left) or elastic (right). When supply is inelastic, the price rises substantially, but there is little
extra output. When supply is inelastic, the price rises a little, and there is a lot of extra output.

We can thus think of the long-run supply function as a horizontal line equal to the minimum
average cost. Note: this does not contradict the first-order condition of each firm since AC(q) =
M C(q) when AC(q) is minimised. Hence we have p = AC(q) = M C(q). See Figure 7.

In this class, we adopt a simple model of “the long-run”, where there are an infinite number of
identical firms. One could imagine an alternative model, where news firms are not as e±cient as
current firms, implying that the long-run supply function is upward sloping.6 As an extension,
one might suppose there is a diÆerence between demand in the short-run and the long-run. For
example, consider the market for oil. In the short-run, demand is very inelastic as consumers
have fixed commutes. In the long-run consumers can change their cars, organise car-pools
and even change jobs, in reaction to a change in the price of oil, making demand much more
inelastic.

5.1 Example

Here we continue the example from Section 4.1. The algebra can get a little involved, so I urge
you to work with example out, on your own, with r1 = r2 = 1.
6
This would also occur if new firms bid up the price of common inputs.

10
Eco11, Fall 2008 Simon Board

Figure 7: Long-Run Equilibrium. In the long-run equilibrium, each firm produces at the minimum
average-cost, and the firms enter until aggregate supply equals aggregate demand. In this figure, LS is
the long-run supply curve, while SS is the short-run supply corresponding to output Q0 .

The firm’s cost function is

c(q, r1 , r2 ) = 2(r1 r2 )1/2 q 3/2 + (r1 + r2 )

Dividing by q, average cost is

AC(q) = c(q)/q = 2(r1 r2 )1/2 q 1/2 + (r1 + r2 )q °1

As argued above, average cost is minimised in the long-run. DiÆerentiating,

dAC(q)
= (r1 r2 )1/2 q °1/2 ° (r1 + r2 )q °2
dq

Setting this equal to zero and rearranging, the AC-minimising quantity is

q § = (r1 r2 )°1/3 (r1 + r2 )2/3

The long-run supply curve is a horizontal line at the minimum average cost. Substituting q §
into the average cost, the price in the long-run is

p§ = AC(q § ) = 3(r1 r2 )1/3 (r1 + r2 )1/3

11
Eco11, Fall 2008 Simon Board

We now switch to the market level. Since we know the price level, we can calculate the market
demand. Using the demand function from Section 4.1,

100
X(p§ ) = 100/p§ = (r1 r2 )°1/3 (r1 + r2 )°1/3
3

In equilibrium, aggregate supply equals aggregate demand. Since each firm produces q § , we
know the number of firms is

100
K § = X(p§ )/q § = (r1 + r2 )°1
3

Note that K § may not be an integer, but we won’t worry about this.

5.2 Shifts in Demand

Suppose we start in long-run equilibrium, as shown in figure 7. Suppose there in an increase in

demand (e.g. due to government expenditure).

In the very short run, quantity is constant, and the market price rises a lot. This is shown in
figure 8.

In the short run, firms increase their output in response to the increase in demand. Each firm
moves up its individual supply curve (i.e. their marginal cost curve) and the market moves up
its aggregate supply curve. As a result, firms now make positive profits, as shown in figure 9.

In the long-run the positive industry profits attracts entrants. This causes the industry short-
run supply curve to shift out, and the price level to move back to the minimum average cost.
The new aggregate quantity is given by the intersection of the long-run supply function and
the demand function. Each individual firm returns to operating at minimum average cost, and
making zero profits. See figure 10.

5.3 Example

Suppose market demand is linear:

X(p) = 1500 ° 50p.

12
Eco11, Fall 2008 Simon Board

Figure 8: Very-Short Run. In the very-short run, each firm produces the same quantity q0 and no
firms enter, so market quantity is given by Q0 . The price rises to p1 to clear the market.

Figure 9: Short Run. In the short run, the individual firms increase their quantity to q2 . As a result,
the market moves up its short-run supply curve and the equilibrium is given by p2 .

13
Eco11, Fall 2008 Simon Board

Figure 10: Long Run. In the long run, new firms enter, the industry reverts to the long run supply
function and the short run supply function shifts to SS 0 . At this point, each firm again produces at
long-run minimum average cost, q0 .

There are many potential firms, each with cost function

c(q) = 100 + q 2 /4.

Let’s calculate the initial long-run equilibrium. The average cost is

AC(q) = c(q)/q = 100q °1 + q/4

In the long-run equilibrium, firms operate at minimum average cost. DiÆerentiating,

dAC(q)
= °100q °2 + 1/4 = 0
dq

Rearranging, each firm produces

q § = 20

Substituting into the average cost, the market price is then

p§ = AC(q § ) = 10

Market demand is
X(p§ ) = 1500 ° 50p§ = 1000

14
Eco11, Fall 2008 Simon Board

which equals market supply. Hence the number of firms is

K § = X(p§ )/q § = 1000/20 = 50.

Suppose demand falls (e.g. the price of a complement rises). In particular, new demand is given
by
X(p) = 1200 ° 50p.

In the very short run, each firm’s output is fixed. Hence market output is fixed at 1000.
Equating supply and demand,
1200 ° 50p = 1000.

Hence the equilibrium price is p§ = 4.

In the short run, each firm reduces its output due to the reduction in demand. Each firm
maximises profits:
º(q) = pq ° c(q) = pq ° 100 ° q 2 /4

Taking the first-order condition, the optimal output is q § (p) = 2p. This gives this individual
firm’s supply curve. There are 50 firms, so the market supply curve is

Q(p) = 100p

Equating supply and demand,

1200 ° 50p = 100p

Rearranging, the short-run equilibrium price is p§ = 8. Each firm produces q = 2p§ = 16, and
makes profits
º = pq ° c(q) = 8 £ 16 ° 100 ° 162 /4 = °36.

In the long run, the negative profits cause firms to exit the industry. In the long-run, the price
returns to p§ = 10, and each active firm produces q § = 20. Given the price, market demand is

X(p§ ) = 1200 ° 50p§ = 700

which also equals market supply. Thus the number of firms is given by

K § = X(p§ )/q § = 700/20 = 35

which is less than the 50 firms we started with.

15