Chapter 13

Duopoly

13.1 Introduction
In this chapter, we study market structures that lie between perfect competition
and monopoly. As before we assume, at least in most of this chapter, that there is
one homogeneous good that is the same no matter who makes it. We assume that
everyone has perfect information about the good and its price. In our discussion
of monopoly, we assumed that there were barriers to entry that preserved the
monopolist’s position. In this chapter, we also assume that there are barriers to
entry that prevent other firms from entering the market. However, we now assume
that there are already two (or more) firms in the market.
An oligopoly is a market with just a few firms. For instance, the market for cell
phone service in our part of the United States is currently dominated by Verizon
Wireless, AT&T, and Sprint, a total of three large companies. (There are also some
smaller companies.) In this market, each of these large firms realizes that its own
output and the output of each of its competitors will affect the market price. In
contrast, in a competitive market (such as the markets for wheat, corn, or cattle),
there are hundreds or thousands of firms supplying the good, and each firm can
safely ignore the possible effect of its own output or each competitor’s output on
the price. In this chapter, we assume that each firm takes into account how its own
output, and its competitors’ outputs, affects the price, and through the price, its
own profit.
Because one firm’s output decision will affect the profits of the other firms, firms
in an oligopoly are likely to act strategically. Two firms are acting strategically when
each looks at what the other is doing, and thinks along these lines: “I have to make
my production decisions contingent on what he does. If he sells 1,000 units, then
to maximize my profits, I have to sell 1,100 units. And if I produce 1,100 units,
does it make sense for him to produce 1,000 units?” The firms are reacting to each
other. In a competitive market, in contrast, the firms do not react to each other; they
react only to the market price, which they take as predetermined or fixed.

221
222 Chapter 13. Duopoly

In this chapter, we assume that there are only two firms in the market. A market
with just two firms is called a duopoly. Obviously, a duopoly is the simplest sort of
oligopoly; many of the concepts and results that we will describe can be extended to
the case of an oligopoly with more than two firms. Duopoly analysis by economists
dates back to the nineteenth century. Some of the central concepts of duopoly
analysis have to do with strategic behavior, and the analysis of strategic behavior
is the heart of the twentieth-century discipline called game theory. Game theory
builds on duopoly theory. We will turn to game theory in the next chapter.
There are two fundamentally different approaches to duopoly theory. The first
assumes that duopolists compete with each other through their choices of quantity:
each firm decides on the quantity it should produce and sell in the market, contingent
on the other firm’s quantity. The second assumes that duopolists compete with each
other through their choices of price: each firm decides on the price it should charge,
contingent on the price the other firm is charging. The first approach was taken by
the French mathematician and economist Antoine Augustin Cournot (1801–1877),
who wrote about duopoly in 1838. The second approach was developed by another
French mathematician, Joseph Louis François Bertrand (1822–1900), in 1883.
We start in Section 13.2 by describing the basic Cournot duopoly model, and we
develop that model in Sections 13.3 and 13.4. The crucial behavioral assumption
of the Cournot model is that each firm assumes that the other firm’s output is given
and fixed, and maximizes its own profit based on that assumption. Other behavioral
assumptions might be made about the two firms. One is the assumption made by the
German economist Heinrich von Stackelberg (1905–1946). Stackelberg assumed,
as did Cournot, that the firms make decisions about quantities. But he also assumed,
unlike Cournot, that the two firms act differently; one of the duopolists acts as a
follower (as in Cournot’s model), taking the other firm’s output as given and fixed,
and choosing its own output based on that assumption, but the other duopolist
acts as a leader, by anticipating that its rival will act as a follower, and choosing
its own output based on that knowledge. We describe the Stackelberg model in
Section 13.5. In Section 13.6, we describe the Bertrand model, in which the firms
compete with each other through their choices of price, instead of competing, as
in the Cournot (and Stackelberg) models, through the choices of quantity. We will
see that there are two rather different versions of Bertrand’s model, depending on
whether the good produced by the two firms is exactly the same (the homogeneous
good case), or somewhat different (the differentiated goods case, e.g., Coke and
Pepsi).

13.2 Cournot Competition

Assume there are two firms in the market. Firm 1 produces y1 units of the good;
firm 2 produces y2 units. The total amount produced is y = y1 + y2 . We assume
there is a downward-sloping inverse market demand curve p(y) = p(y1 + y2 ). We
 13.2 Cournot Competition 223

assume firm i has a cost curve Ci (yi ), for i = 1, 2. Firm 1 wants to maximize its
profit π1 , given by
π1 (y1 , y2 ) = p(y1 + y2 )y1 − C1 (y1 ).
Similarly, firm 2 wants to maximize its profit π2 , given by
π2 (y1 , y2 ) = p(y1 + y2 )y2 − C2 (y2 ).
The basic Cournot assumption is this: When firm 1 chooses its output y1 to
maximize its profit, it takes firm 2’s output y2 as given and fixed; similarly, when
firm 2 chooses its output y2 to maximize its profit, it takes firm 1’s output y1 as
given and fixed. Therefore, when firm 1 differentiates its profit function π1 (y1 , y2 ),
it treats y2 as a constant. This leads to the first-order condition
∂π1 dp(y) dC1 (y1 )
= p(y) + y1 − = 0.
∂y1 dy dy1
Firm 1 can solve this equation for y1 as a function of y2 . We write the result as
y1 = r1 (y2 ).
The function r1 is called firm 1’s reaction function. It shows, for any output level y2
of firm 2, the quantity of the good that firm 1 should produce in order to maximize
its profit.
Similarly, firm 2 maximizes its profit subject to the assumption that y1 is a
constant. This leads to
∂π2 dp(y) dC2 (y2 )
= p(y) + y2 − = 0.
∂y2 dy dy2
Firm 2 can solve this equation for y2 as a function of y1 , and we write the result as
y2 = r2 (y1 ).
The function r2 is firm 2’s reaction function. It shows, for any output level y1 of
firm 1, the quantity of the good that firm 2 should produce in order to maximize its
profit.
If the two firms randomly choose their output levels y1 and y2 , it is almost
certain that neither would be maximizing its profits subject to what the other one
is doing. Neither firm would be behaving in a clever way. The result would not
make sense; it would be doubly stupid. And if firm 2 randomly chooses an output
level y2 , and then firm 1 uses its reaction function r1 to choose its output level y1 ,
the result would be half sensible – sensible on the part of firm 1, but stupid on
the part of firm 2. However, suppose the reaction functions intersect at a point y1∗
and y2∗ , and suppose firm 1 chooses y1∗ and firm 2 chooses y2∗ . This outcome does
make very good sense for both firms, because firm 1 is making the best choice it
can, subject to what firm 2 has chosen, and firm 2 is making the best choice it can,
subject to what firm 1 has chosen. A Cournot equilibrium in a duopoly model is
224 Chapter 13. Duopoly

a pair of output levels y1∗ and y2∗ that are consistent in this sense – each firm i is
maximizing its profit at yi∗ , subject to what the other firm j has chosen, yj∗ . The
Cournot equilibrium is Augustin Cournot’s brilliant solution to the duopoly puzzle.
In short, a Cournot equilibrium is a consistent, self-sustaining, and self-
reinforcing outcome in the duopoly model. We now turn to an example to show
how the Cournot equilibrium can be found.
Example 1. Assume that the inverse demand curve is p(y1 + y2 ) = 100 − y =
100 − y1 − y2 . Assume that the cost curves are C1 (y1 ) = 25y1 and C2 (y2 ) = 25y2 .
Marginal cost for either firm is a constant 25. To find firm 1’s reaction function,
we find the y1 that maximizes π1 (y1 , y2 ), under the assumption that y2 is constant.
Firm 1’s profit is
π1 (y1 , y2 ) = (100 − y1 − y2 )y1 − 25y1 .
Differentiating with respect to y1 while holding y2 constant, and setting the result
equal to zero, gives
∂π1
= 100 − 2y1 − y2 − 25 = 0.
∂y1
Solving for y1 as a function of y2 gives firm 1’s reaction function:
y1 = r1 (y2 ) = 37.5 − y2 /2.
Firm 2’s profit π2 (y1 , y2 ) is
π2 (y1 , y2 ) = (100 − y1 − y2 )y2 − 25y2 .
Differentiating with respect to y2 while holding y1 constant, and setting the result
equal to zero, gives
∂π2
= 100 − 2y2 − y1 − 25 = 0.
∂y2
Therefore, firm 2’s reaction function is
y2 = r2 (y1 ) = 37.5 − y1 /2.
In Figure 13.1, we show the reaction functions and the Cournot equilibrium in
Example 1. The Cournot equilibrium is the point at which the two reaction functions
intersect. Solving the two reaction function equations simultaneously (y1 = 37.5 −
y2 /2 and y2 = 37.5 − y1 /2) easily gives (y1∗ , y2∗ ) = (25, 25). At (25, 25), each firm
is maximizing its profit, given what the other firm is doing. The market price is
100 − 25 − 25 = 50. The reader can easily check that profit levels for the firms are
(π1 , π2 ) = (625, 625). The output levels are mutually consistent; neither firm has an
incentive to change, given what the other firm is doing. The Cournot equilibrium
(25, 25) makes sense for firm 1, and simultaneously makes sense for firm 2.
 13.2 Cournot Competition 225

y2 (firm 2)

75

r1

37.5

Cournot equilibrium
25

r2

y1 (firm 1)
25 37.5 75

Fig. 13.1. The reaction functions and Cournot equilibrium in Example 1.

Comparison with Monopoly and Competition

We can use Example 1 to show how a Cournot equilibrium in a duopoly compares
with a monopoly outcome and a competitive outcome. The general result is that in
a duopoly (and, more generally, an oligopoly), total output and price lie somewhere
between what they would be under competition or under monopoly.
In Example 1, remember that the Cournot equilibrium price is $50, and the total
quantity is 50. How would we describe the competitive outcome? We would have
the same demand curve, but price would be equal to marginal cost. That is, the
competitive supply curve would be a horizontal line at p = 25. Combining this
with the inverse demand curve p = 100 − y gives a competitive equilibrium at
pC = $25 and yC = 75, where the C subscript means “competitive.”
How would we describe the monopoly outcome? The monopolist would max-
imize profit by setting marginal revenue equal to marginal cost. In the example,
MR(y) = 100 − 2y and MC(y) = 25. Therefore, 100 − 2y = 25, or y = 37.5.
Putting this y in the equation for the inverse demand curve gives p = 100 − y =
62.5. In the monopoly solution, then, we have pM = $62.5 and yM = 37.5, where
the M subscript stands for “monopoly.” We conclude that the Cournot equilibrium
in a duopoly lies between the competitive outcome and the monopoly outcome,
both for quantity and price.
What about efficiency? We will now investigate the social surplus created at the
Cournot equilibrium in the duopoly. As you might expect, the duopoly social surplus
226 Chapter 13. Duopoly

100

C.S., monopoly
Producer’s surplus, monopoly
pm = 62.5 B
Additional social surplus,
50 duopoly
Additional social surplus,
competition
C
pc = 25 MC = 25
A Demand:
MR = 100 – 2y p(y) = 100 – y

y
ym = 37.5 50 yc = 75 100

Fig. 13.2. Welfare analysis of the duopoly, based on Example 1. Social surplus is least for
a monopoly, greater for a duopoly, and greatest for a competitive market.

lies between the social surplus in the monopoly market and the social surplus in
the competitive market. In short, duopoly (and, more generally, oligopoly) creates
some deadweight loss, but not as much as monopoly creates. We show this in
Figure 13.2.
Figure 13.2 is based on Example 1. It shows total output y on the horizontal axis.
The outermost line is the inverse demand curve p(y) = 100 − y. A monopolist in
this market would find the corresponding marginal revenue curve MR(y) = 100 −
2y. This is the steeper downward-sloping line shown in the figure. A monopolist
would set marginal revenue equal to marginal cost, point A in the figure, to get the
quantity yM = 37.5. He would then go up to the demand curve, to point B, and get
the price pM = $62.5. Total social surplus under monopoly in this example would
be consumers’ surplus (the upward-sloping crosshatched triangle) plus producer’s
surplus (the downward-sloping crosshatched square).
If this market were a duopoly, the Cournot equilibrium total quantity would be
50 (shown on the horizontal axis), and the price would be $50 (shown on the vertical
axis). In a transition from a monopoly to duopoly, consumers’ surplus would grow
and producers’ surplus would shrink. However, the sum of the two welfare measures
would definitely grow, by the area of the horizontally cross-hatched trapezoid in
the figure.
Finally, if this were a competitive market, the equilibrium would require that
price equal marginal cost (point C in the figure). In a transition from duopoly
 13.3 More on Dynamics 227

to competition, consumers’ surplus would greatly expand and producers’ surplus

would disappear. However, social surplus would definitely grow, by the area of the
non-cross-hatched triangle.
We conclude that the competitive outcome is best for society in the sense that it
maximizes social surplus. The Cournot equilibrium in a duopoly is worse than the
competitive outcome. The monopoly outcome is the worst of all.

13.3 More on Dynamics

We have been a bit vague about how our two firms get to the Cournot equilibrium.
The sophisticated and modern game theory–oriented economist looks at Cournot’s
model and describes it as a simultaneous move game. This means that firms 1 and
2, with full knowledge of market demand and full knowledge of their own cost
function and their rival’s cost function, choose their output levels, one time only,
and simultaneously. They end up with a pair of output levels (y1 , y2 ). If the pair is a
Cournot equilibrium, the outcome makes sense for both firms; it is doubly sensible.
If it is not a Cournot equilibrium, the outcome fails to make sense for at least one
of the firms, and possibly for both firms. If each firm is a rational profit maximizer
and expects the other firm to also be a rational profit maximizer, they should end
up at the Cournot equilibrium.
It may be useful to discuss some other possible dynamics in Cournot’s model.
These descriptions of dynamics necessarily go beyond the simple assumption of
simultaneity. One possible dynamic has the firms taking turns reacting to each
other. First, firm 1 reacts to firm 2’s output; then firm 2 reacts to firm 1’s output,
and the process goes on until it (hopefully) gets to a Cournot equilibrium. To make
this discussion more understandable, let us assume that there is a time dimension,
and production and consumption are repeated time unit after time unit, say, day
after day.
Let us assume, then, that firms 1 and 2 start at some initial output quantities,
on day 0, say, (y10 , y20 ). On the morning of day 1, firm 1 looks at firm 2’s out-
put, and calculates what it should produce, contingent on what firm 2 produced
on day 0. That is, it goes to its reaction function, and calculates y1 = r1 (y20 ).
This gives (y11 , y21 ) as the firm outputs on day 1, where y11 = r1 (y20 ) and where
y21 = y20 .
On the morning of day 2, firm 2 looks at firm 1’s output, and calculates what
it should produce, contingent on what firm 1 produced on day 1. That is, it goes
to its reaction function, and calculates y2 = r2 (y11 ). This gives (y12 , y22 ) as the firm
outputs on day 2, where y12 = y11 and where y22 = r2 (y11 ).
In other words, the two firms take turns reacting to each other. The process
continues, day after day, until it (hopefully) converges to a point at which neither
firm wants to make further modifications to its daily output. That point is a Cournot
equilibrium. Of course, it is slightly odd to think that each firm will use its reaction
228 Chapter 13. Duopoly

y2 (firm 2)

75
r1

P0 P1

37.5
P3
P2 Cournot equilibrium
25
P4

r2

y1 (firm 1)
25 37.5 75

Fig. 13.3. A dynamic story about the Cournot equilibrium, based on Example 1.

function at each of its turns, because the reaction functions are based on the
assumption that the rival’s output is fixed, and that the rival is changing its planned
output every other day. (A more rigorous treatment of this and other dynamic
adjustment processes is beyond the scope of this book.)
Figure 13.3 illustrates this dynamic story. The process starts at some initial
output levels P 0 = (y10 , y20 ), shown on the vertical axis in the figure. On day 1,
the process moves to P 1 ; on day 2, it moves to P 2 , and so on. As in the story of
Genesis, on the seventh day they rest. In the figure, the process converges nicely to
the Cournot equilibrium. However, this dynamic process would not converge if the
reaction functions had the wrong slopes at the equilibrium. The reader is invited to
relabel the reaction functions to see what happens if r1 is less steep than r2 .

13.4 Collusion
Let us return to Example 1, and assume that our duopolists are at the Cournot
equilibrium. Once again, to make this discussion more understandable, we assume
that there is a time dimension, and production and consumption are repeated
day after day. Because the two firms are at the Cournot equilibrium, they are
producing and selling y1∗ = y2∗ = 25, day after day. Given those production levels,
the market price is p = 100 − 25 − 25 = 50, day after day. Each firm has profits
of πi = pyi∗ − 25yi∗ = 625, day after day.
 13.4 Collusion 229

Suppose that one day, the owners of the two firms meet for a game of golf. They
have the following conversation: Firm 1 owner: “I’m maximizing my profits at
y1∗ = 25. But this is based on your holding your output constant at y2∗ = 25. What
if we both cut output a little bit? Could we make more money that way?” Firm 2
owner: “Well, if we each cut production by one unit, the market price would rise
to $52, as the price is given by p = 100 − y1 − y2 . This means my revenue would
change from $50 × 25 to $52 × 24. That’s almost no change – it’s a drop of $2, to
be exact.” Firm 1 owner: “But your costs would drop by $25. So your profit would
shoot up.” Firm 2 owner: “That’s right. In short, if we both cut back output by one
unit, your profit would rise by $23, and mine would too!”
Then their caddy speaks up: “I’m an undercover federal agent. You are both
under arrest for colluding and conspiring to fix prices in the market for the gizmos
you are producing.”
As the presence of our fictional caddy/federal agent suggests, it may be illegal
for two duopolists, or, more generally, a group of firms in an oligopoly – to get
together and make plans such as this. A cartel is a group of producers or firms that
organize (or conspire) to raise the price of the good they are selling by restricting
supply. Under the antitrust laws of the United States and other developed nations,
cartels are usually, but not always, illegal. One of the most notorious (but outside the
reach of law) cartels of recent history is the Organization of Petroleum Exporting
Countries (OPEC). This is an organization of countries whose main purpose is to
keep petroleum prices high by controlling production in member countries. Legal
cartels in the United States include sports leagues, such as Major League Baseball
and the National Football League.
What exactly would our two duopolists do if they took it upon themselves to
maximize joint or total profit, rather than simply letting each firm maximize its
own profit, conditional on the other firms’s output? As our preceding discussion
suggests, they might gain a lot if they both agree to reduce output.
Let π(y1 + y2 ) = π1 (y1 , y2 ) + π2 (y1 , y2 ) represent total profit for the two firms
combined. Then

π(y1 , y2 ) = p(y1 + y2 )(y1 + y2 ) − C1 (y1 ) − C2 (y2 ).

The first-order conditions for maximizing this function of two variables are:

∂π ∂p(y1 + y2 ) dC1 (y1 )

= p(y1 + y2 ) + (y1 + y2 ) − =0
∂y1 ∂y1 dy1

and

∂π ∂p(y1 + y2 ) dC2 (y2 )

= p(y1 + y2 ) + (y1 + y2 ) − = 0.
∂y2 ∂y2 dy2
230 Chapter 13. Duopoly

Both these conditions must hold for total profit to be maximized, at least for an
interior maximum. (The first-order conditions for a maximum at a boundary are
slightly different.)
In Example 2, we examine joint profit maximization for the simple duopoly
introduced in Example 1.
Example 2. From Example 1, we have
π(y1 , y2 ) = (100 − y1 − y2 )(y1 + y2 ) − 25y1 − 25y2
= 100y1 + 100y2 − y12 − y22 − 2y1 y2 − 25y1 − 25y2 .
Taking partial derivatives with respect to y1 and y2 , we get
∂π
= 100 − 2y1 − 2y2 − 25 = 0
∂y1
or
y1 + y2 = 37.5.
Similarly,
∂π
= 100 − 2y1 − 2y2 − 25 = 0
∂y2
or
y1 + y2 = 37.5.
The first-order conditions for joint profit maximization are identical, because the
two firms have identical cost curves. We conclude that joint profit maximization
requires y1 + y2 = 37.5. For example, each firm could produce yi = 37.5/2 =
18.75. With these levels of output, π1 (y1 , y2 ) = π2 (y1 , y2 ) = (100 − 37.5)18.75 −
(25)18.75 = 703.125. Each firm would be making $703.125. This is considerably
better than the $625 profit for each firm at the Cournot equilibrium.
In Figure 13.4, we show the joint profit maximization points, the collusion out-
comes, for this duopoly example. The bold line is the set of outcomes that maximize
joint profits; that is, the set for which y1 + y2 = 37.5. Note that (18.75, 18.75) is
one of many possibilities, but they all involve total output of 37.5 units. Finally,
the reader should remember our Figure 13.2 comparison of monopoly, duopoly,
and competition. In conjunction with that figure, we determined that a monopoly
firm would produce 37.5 units. In Example 2 and Figure 13.4, the two duopolists
together are producing a total of 37.5 units. We get the same answer because the
duopolists in Example 2 are acting just like a monopolist.
Happily for consumers of their products, cartels and colluding duopolists are
inherently unstable. Because a collusion agreement is not a Cournot equilibrium,
each firm has an incentive to cheat on the agreement. For instance, to continue our
 13.4 Collusion 231

y2 (firm 2)

75

r1

37.5 Collusion outcomes

Cournot equilibrium
25

r2

y1 (firm 1)
25 37.5 75

Fig. 13.4. The reaction functions, the Cournot equilibrium, and the collusion outcomes,
all based on Example 2. The two duopolists have an incentive to collude.

numerical example, suppose the two duopolists have agreed to be at the joint profit
maximizing point (18.75, 18.75). Some time later, the owner of firm 1 wakes
up one morning, and says to himself, “The hell with that lawbreaking SOB.
If he’s going to produce 18.75 units per day, I shall greatly increase my own
profits by using my reaction function to figure out what I should produce.” The
answer is

y1 = r1 (y2 ) = 37.5 − y2 /2 = 37.5 − 18.75/2 = 28.125.

As soon as the owner of firm 1 figures this out, he produces 28.125 units per
day. This raises firm 1’s profits from $703.125 per day to π1 (y1 , y2 ) = (100 −
28.125 − 18.75)28.125 − 25(28.125) = $791 per day, a gain of nearly $88. Shortly
thereafter, the owner of Firm 2 realizes he has been duped, so firm 2 reacts to firm
1’s output of y1 = 28.125. Firm 2 switches to

y2 = r2 (y1 ) = 37.5 − y1 /2 = 37.5 − 28.125/2 = 23.44.

And so it goes. After a few rounds of this reacting and re-reacting, the duopoly
may end up back at, or near, the Cournot equilibrium of (25, 25).
The point of this discussion is that duopolists – and, more generally, members
of cartels – always have incentives to collude, to get together and plot against the
public, to figure out how they might reduce output and increase their joint profits.
However, having come to some kind of collusion agreement, the duopolists, or the
232 Chapter 13. Duopoly

cartel members, will be tempted to cheat. If they do start to cheat, they are likely
to drift back toward a Cournot equilibrium. This, then, is the big dynamic: inde-
pendent profit maximization leads toward the Cournot solution. Then joint profit
maximization leads toward the collusion solution. Unless the firms can enforce
their collusion agreements, cheating and independent profit maximization lead
back toward the Cournot solution. So turns the world of duopoly, or the world of
cartels. In the absence of collusion enforcement mechanisms, the likely prediction
for a duopoly, or for a cartel, is instability.

13.5 Stackelberg Competition

We now turn to the duopoly model of the German economist Heinrich von Stackel-
berg. Stackelberg assumed that one of the duopoly firms acts as a Cournot duopolist.
That is, it takes the other firm’s output as given and fixed, and it chooses its own out-
put based on that assumption. We call this firm the follower. Stackelberg assumed
that the other firm anticipates this behavior, and maximizes its profit based on the
assumption that its rival is a follower. We call this firm the leader.
Recall that in the analysis of the Cournot equilibrium, we formally assumed
a simultaneous move structure. That is, we assumed the interaction between the
two firms was one time only, and simultaneous. Our extensive informal discussion
of dynamics involved stories about day-by-day interactions, reactions, and re-
reactions, but that discussion was not necessary for the formal definition of the
Cournot equilibrium.
To describe the Stackelberg model, we now formally assume that the interaction
between the two firms is in two steps, sequentially. In the first step, the leader
firm determines its planned output. In the second and final step, the follower firm
determines its output. The firms then produce and sell their outputs, at a market
price contingent on y1 + y2 , and make their profits.
Here is how it works. We will let firm 1 be the leader firm; its output is y1 .
Firm 2 is the follower firm; its output is y2 . The follower firm is the second firm
to act. It knows what the leader firm is producing, because that was determined
(and announced) at step one. The follower firm acts just like a firm in the Cournot
analysis; it takes y1 as given, and determines what maximizes its own profits
given y1 . Therefore, it uses its reaction function to determine its output. That is,
y2 = r2 (y1 ).
But firm 1, the leader firm, knows how firm 2, the follower firm, behaves. That is,
firm 1 anticipates that firm 2 will choose y2 by using its reaction function formula.
Therefore, firm 1 can use its knowledge of firm 2’s behavior, and use the fact that
it goes first and firm 2 goes second. It does this in a simple way: it just substitutes
r2 (y1 ) for y2 in its own profit function. This gets rid of the y2 terms; firm 1’s
profit is now simply a function of y1 , and firm 1 simply chooses y1 to maximize
profits.
 13.6 Bertrand Competition 233

Formally, firm 1’s profit is π1 (y1 , y2 ) = p(y1 + y2 )y1 − C1 (y1 ). Substituting

r2 (y1 ) for y2 makes this a function of one variable only:
π1 (y1 ) = p(y1 + r2 (y1 ))y1 − C1 (y1 ).
The first-order condition for profit maximization is to set the derivative of profit
with respect to y1 equal to zero. This gives
dπ1 (y1 ) dp dr2 (y1 )
= p(y1 + r2 (y1 )) + y1 (1 + ) − MC1 (y1 ) = 0.
dy1 dy dy1
Let us apply the result to Example 1. Recall that in that example, p(y) =
100 − y = 100 − y1 − y2 , C1 (y1 ) = 25y1 , and r2 (y1 ) = 37.5 − y1 /2. Therefore,
dp dr2 (y1 )
dy = −1, and dy1 = −1/2. Plugging into the first-order condition then gives

(100 − y1 − (37.5 − y1 /2)) − y1 (1 − 1/2) − 25 = 0.

This gives y1 = 37.5. Putting this into the follower’s reaction function gives
y2 = r2 (y1 ) = 37.5 − 37.5/2 = 18.75.
We know that a Stackelberg leader firm will never end up with a profit level
lower than what it gets at the Cournot equilibrium. One option always available to
the leader is to announce its Cournot output, to which the follower would respond
with its own Cournot output. This would produce the Cournot profits for the two
firms.
Figure 13.5 shows the Cournot equilibrium, the collusion outcomes, and the
Stackelberg equilibrium for the numerical example we have been using throughout
this chapter.

13.6 Bertrand Competition

We now turn to the model developed by the mathematician Joseph Louis François
Bertrand. Whereas the Cournot model of duopoly assumes that each of the two
firms decides on what quantity to produce, the Bertrand model assumes that each
firm decides on what price to charge. As we shall see, this approach can lead to
a very interesting but somewhat unrealistic model, with implications that are very
different from the implications of the Cournot model, or it can lead to a model that
is perhaps more realistic than Cournot’s, but with implications similar to Cournot’s.
This difference arises because we can develop the Bertrand analysis in either of
two ways:
(1) We can assume, as we assumed for the Cournot model, that the two firms are producing
exactly the same good. That is, whether produced by firm 1 or firm 2, a unit of the
good is a unit of the good, as far as the buyers are concerned. This is the property of
homogeneity. Commodities such electricity, oil, metals, or wheat are homogeneous;
whether produced by firm 1 or firm 2, a gallon of fuel oil is a gallon of fuel oil. For a
234 Chapter 13. Duopoly

y2 (firm 2)

75

r1

Collusion outcomes
37.5

Cournot equilibrium
25
Stackelberg equilibrium
18.75

r2

y1 (firm 1)
25 37.5 75

Fig. 13.5. The reaction functions, the Cournot equilibrium, the collusion outcomes, and
the Stackelberg equilibrium, all based on Example 1.

homogeneous good being produced and sold by two firms, there can be only one price;
if firm 1 tries to sell it at a slightly higher price than firm 2, its sales drop to zero.
(2) Alternatively, we can assume that the two firms produce goods that are similar but
slightly different, or differentiated. Think of Coke and Pepsi, McDonald’s and Burger
King, Bud Light and Miller Lite, or Schick and Gillette. If the two firms produce goods
that are differentiated, they can charge different prices, and in fact they commonly do
so. Each is likely to claim that its good is both better and less expensive.

Homogeneous Goods Case

Let us now assume that firms 1 and 2 produce exactly the same good. For simplicity,
we assume that the two firms have identical constant marginal cost functions. Let
MC represent the marginal cost of producing a unit of the good, and let MC be
the same for both firms and constant over all output levels. Let y1 and y2 represent
the production levels of the two firms. We assume the firms set prices p1 and p2 ,
and then sell their output to meet demand. Because the good is homogeneous, if
firm i sets a lower price than firm j , then all the customers will buy from firm i.
This implies that in any equilibrium at which both firms are operating, we must
have p1 = p2 = p, where p is the (single) market price. If there is one price p,
the market demand curve is given by a function y = y(p). We assume that if both
firms are charging the same price p, they will split the market demand equally –
each will sell y1 = y2 = y(p)/2 units of the good.
 13.6 Bertrand Competition 235

This model cannot be solved using standard calculus techniques. This is because
although the function y(p) is well behaved, firm i’s demand function yi (pi ) is not.
It has a sharp discontinuity when pi equals pj . If pi < pj , demand for firm i is
y(pi ); if pi = pj , demand for firm i is y(pi )/2; and if pi > pj , demand for firm i
is zero.
Because we cannot use standard calculus techniques, we must reason along more
abstract lines. Recall our definition of a Cournot equilibrium from Section 13.2. In
a model in which the two duopolists are reacting to each other by setting quantities,
a Cournot equilibrium is a pair of output levels y1∗ and y2∗ that are consistent, in
the sense that each firm i is maximizing its profit at yi∗ , subject to what the other
firm j has chosen, yj∗ . Let us now define an equilibrium in a similar way, but for
the current model in which the two duopolists are reacting to each other by setting
prices. A Bertrand equilibrium is a pair of prices p1∗ and p2∗ that are consistent, in
the sense that each firm i is maximizing its profit with the choice of pi∗ , subject to
what the other firm j has chosen, pj∗ .
What can we say about a Bertrand equilibrium in the homogeneous goods case?
Let (p1∗ , p2∗ ) represent the equilibrium prices and let (y1∗ , y2∗ ) the corresponding
equilibrium quantities. We can conclude the following:

(1) The firms must be charging the same price. That is, p1∗ = p2∗ = p ∗ . Suppose, to the
contrary, that they are charging different prices, and without loss of generality, assume
p1∗ < p2∗ . Then firm 1 is selling a positive quantity of the good, and firm 2 is selling
nothing.
(a) If p1∗ < MC, then firm 1 has negative profits and would be better off shutting down.
Therefore, this cannot be an equilibrium.
(b) If p1∗ = MC, firm 1 is making $0 on each unit it produces and sells. It could
increase its price somewhat, while keeping it below p2∗ , and make positive amounts
on all the units it sells. (It would sell fewer units, but it would make money on each
one.) Therefore, this cannot be an equilibrium.
(c) If p1∗ > MC, firm 1 is making positive profits on all the units it produces and sells.
If this were the case, however, firm 2 would gain by entering the market with a price
strictly between MC and p1∗ , taking all of firm 1’s customers away. Therefore, this
cannot be an equilibrium either.
We have established that p1∗ = p2∗ implies that we cannot have a Bertrand equilibrium.
Therefore, at a Bertrand equilibrium, we must have p1∗ = p2∗ = p ∗ . Because we have
assumed that demand is split equally between the two firms when their prices are the
same, y1∗ = y2∗ = y(p ∗ )/2.
(2) Marginal cost cannot be less than price; that is, MC < p ∗ cannot hold. Here is why.
Suppose the inequality held. Assume for concreteness that MC = 25 and that p1∗ =
p2∗ = p ∗ = 26. Then either firm, say, firm 1, could shave its price to p1 = 25.99. By
doing so, it would steal away all of firm 2’s customers (half of the total market) and
make almost a dollar profit on each of those sales, while giving up a penny’s profit on
each of the sales it already had (half of the total market). Its profits would obviously go
236 Chapter 13. Duopoly

way up. This contradicts our assumption that firm 1 is choosing a price that maximizes
its profit subject to what firm 2 has chosen.
(3) Marginal cost cannot be greater than price; that is, MC > p ∗ cannot hold. With constant
marginal costs, for either firm i, MC > p ∗ = pi∗ would imply negative profits, and the
firm would opt to go out of business, rather than sell yi∗ at a price of p ∗ .
(4) For both firms, marginal cost equals price; that is, MC = p ∗ = p1∗ = p2∗ . This obvi-
ously follows from (2) and (3).

We conclude that in a Bertrand equilibrium, in the homogeneous good case,

under the assumptions we have made, firms 1 and 2 will charge the same price, and
the price will be equal to marginal cost. But this means that the duopoly market,
in the Bertrand model with a homogeneous good, looks just like a competitive
market. In particular, there is no inefficiency (no loss of social surplus) in the
duopoly market.

Differentiated Goods Case

Now we assume that firms 1 and 2 produce goods that are differentiated – similar,
but not identical. (Think of Coke and Pepsi.) We continue to let y1 and y2 represent
the outputs of the two firms, and p1 and p2 represent the prices. Because the goods
are different, pi < pj does not imply that firm j ’s sales will drop to zero, and the
firms will not be forced to charge the same price in equilibrium.
Now the demand functions for each of the two firms depend on the two prices:
y1 = y1 (p1 , p2 ) and y2 = y2 (p1 , p2 ). For firm i’s demand function yi , the partial
derivative with respect to pi is assumed to be negative (as i raises its price, demand
for i’s good falls), but the partial derivative with respect to pj is positive (as i’s
competitor j raises its price, demand for i’s good rises).
Firm 1’s profit is

π1 (p1 , p2 ) = p1 y1 (p1 , p2 ) − C1 (y1 (p1 , p2 )),

and firm 2’s profit is

π2 (p1 , p2 ) = p2 y2 (p1 , p2 ) − C2 (y2 (p1 , p2 ).

We write these as functions of the two prices (p1 , p2 ) because each firm chooses
its own price, rather than its own quantity as in the Cournot model. Firm 1 chooses
p1 to maximize π1 (p1 , p2 ), taking p2 as given and fixed; firm 2 chooses p2 to
maximize π2 (p1 , p2 ), taking p1 as given and fixed.
Firm 1’s first-order condition is

∂π1 ∂y1 (p1 , p2 ) dC1 (y1 ) ∂y1 (p1 , p2 )

= y1 (p1 , p2 ) + p1 − = 0.
∂p1 ∂p1 dy1 ∂p1
 13.6 Bertrand Competition 237

We can write this more compactly as

∂π1 ∂y1 ∂y1
= y1 + p1 − MC1 = 0.
∂p1 ∂p1 ∂p1
Similarly, firm 2’s first-order condition is
∂π2 ∂y2 ∂y2
= y2 + p2 − MC2 = 0.
∂p2 ∂p2 ∂p2
Recall that when we analyzed the Cournot model, we used the two firms’ first-
order conditions to derive reaction functions. We can do the same here, using
firm i’s first-order condition to find a reaction function that shows firm i’s profit-
maximizing price as a function of firm j ’s price: pi = ri (pj ).
To find a Bertrand equilibrium, we look for a pair of prices p1∗ and p2∗ that are
consistent in the sense that each firm i is maximizing its profit with the choice of pi∗ ,
subject to what the other firm j has chosen, pj∗ . We will again let y1∗ and y2∗ represent
the corresponding equilibrium quantities. In the differentiated goods case, we use
the first-order conditions for profit maximization (or the reaction functions) to find
the equilibrium. We will illustrate with an example soon. Before we do, however,
let us use firm 1’s reaction function to show that in the differentiated goods Bertrand
model (unlike the homogeneous good Bertrand model), at equilibrium, the price
will be greater than marginal cost.
Here is why. At the equilibrium, firm 1’s first-order condition must be satisfied,
which gives
∂y1 ∗ ∂y1
y1∗ + p1 − MC1 (y1∗ ) = 0.
∂p1 ∂p1
Rearranging gives
∂y1 ∗
y1∗ = − (p1 − MC1 (y1∗ )).
∂p1
∂y1
But y1∗ is positive, ∂p1
is negative, and therefore p1∗ − MC1 (y1∗ ) is positive, or

p1∗ > MC1 (y1∗ ).

That is, in the differentiated goods case, at the Bertrand equilibrium, price is greater
than marginal cost for firm 1, and similarly for firm 2. In short, in the differentiated
goods case, a Bertrand equilibrium has social welfare properties similar to the
Cournot equilibrium discussed in Section 13.2.

Example 3. We again assume the cost curves are C1 (y1 ) = 25y1 and C2 (y2 ) = 25y2 ,
and so MC1 = MC2 = 25. We now assume the demand functions for firms 1 and
2 are

y1 (p1 , p2 ) = 50 − p1 + p2 /2
238 Chapter 13. Duopoly

and

y2 (p1 , p2 ) = 50 − p2 + p1 /2.

Firm 1 wants to choose p1 to maximize

π1 (p1 , p2 ) = p1 (50 − p1 + p2 /2) − 25(50 − p1 + p2 /2),

taking p2 as given. The first-order condition gives firm 1’s reaction function,

p1 = r1 (p2 ) = 75/2 + p2 /4.

Similarly, firm 2 wants to choose p2 to maximize

π2 (p1 , p2 ) = p2 (50 − p2 + p1 /2) − 25(50 − p2 + p1 /2).

The first-order condition leads to firm 2’s reaction function,

p2 = r2 (p1 ) = 75/2 + p1 /4.

Solving the two reaction functions (or the two first-order conditions) simulta-
neously gives the Bertrand equilibrium prices of p1∗ = 50 and p2∗ = 50. The equi-
librium quantities are y1∗ = 25 and y1∗ = 25. Equilibrium profit levels are easily
calculated. For firm 1,

π(p1∗ , p2∗ ) = p1∗ y1∗ − C1 (y1∗ ) = 50 × 25 − 25 × 25 = 625,

and similarly for firm 2. Figure 13.6 shows the reaction functions and the equilib-
rium prices for Example 3.

13.7 A Solved Problem

The Problem
Recall the inverse demand function assumed in Example 1:

p(y1 + y2 ) = 100 − y1 − y2 .

The cost functions in that example were C1 (y1 ) = 25y1 and C2 (y2 ) = 25y2 . To find
the Cournot equilibrium, we used the reaction functions of firms 1 and 2:

y1 = r1 (y2 ) = 37.5 − y2 /2, and y2 = r2 (y1 ) = 37.5 − y1 /2.

Recall from Example 1 that the Cournot equilibrium was (y1∗ , y2∗ ) = (25, 25), and
the profit levels at the Cournot equilibrium were (π1 , π2 ) = (625, 625).

(a) Assume firm 1 is a Stackelberg leader, and firm 2 is a Stackelberg follower. Calculate
the Stackelberg equilibrium price and quantities. In addition, find the Stackelberg
equilibrium profit levels.
(b) What would happen if both firms believed that they were Stackelberg leaders?
 13.7 A Solved Problem 239

p2

$50

r2 Bertrand
$37.5 equilibrium

r1

p1
$37.5 $50

Fig. 13.6. The reaction functions and the Bertrand equilibrium in Example 3.

The Solution
(a) If firm 1 is a Stackelberg leader and firm 2 is a Stackelberg follower, firm 2 acts like
a standard Cournot firm; it takes firm 1’s output y1 as given and fixed and chooses its
own output in response. In other words, it chooses y2 to maximize π2 (y1 , y2 ), under the
assumption that y1 is constant. This means that it derives and uses its reaction function
y2 = r2 (y1 ) = 37.5 − y1 /2. Firm 1, on the other hand, knows that y2 is not fixed; in
fact, firm 1 knows exactly how firm 2 chooses y2 based on y1 . In other words, firm 1
knows firm 2’s reaction function and exploits that knowledge.
We can now write firm 1’s profit function as

π1 (y1 , y2 ) = p(y1 + y2 )y1 − C1 (y1 ) = p(y1 + r2 (y1 ))y1 − C1 (y1 )

= (100 − y1 − r2 (y1 ))y1 − 25y1 .

The next step is to substitute for r2 (y1 ), using r2 (y1 ) = 37.5 − y1 /2. After a little minor
algebra and some rearranging, we get π1 (y1 ) = 37.5y1 − y12 /2. We differentiate this
function, set the result equal to zero, and solve, which produces y1∗ = 37.5. The ∗
denotes the Stackelberg equilibrium quantity for firm 1, the leader firm. Plugging y1∗
into firm 2’s reaction function will allow us to solve for the follower firm’s Stackelberg
equilibrium quantity: y2∗ = 37.5 − y1∗ /2 = 18.75.
To find the Stackelberg equilibrium price, we insert y1∗ and y2∗ into the inverse
demand function, which gives p∗ = 100 − 37.5 − 18.75 = 43.75. Profit levels are
given by π1∗ = p∗ y1∗ − 25y1∗ = (p∗ − 25)y1∗ = (43.75 − 25)37.5 = 703.13. Similarly,
π2∗ = (p ∗ − 25)y2∗ = (43.75 − 25)18.75 = 351.56.
240 Chapter 13. Duopoly

(b) If firm 1 is a Stackelberg leader, it knows firm 2’s reaction function and takes advantage
of it. In part (a), we found that this line of reasoning would lead firm 1 to choose y1∗ =
37.5. Let us now assume that firm 2 also acts as a Stackelberg leader. It figures out firm
1’s reaction function y1 = r1 (y2 ) = 37.5 − y2 /2, plugs this into its own profit function
π2 = (100 − r1 (y2 ) − p2 )y2 − C2 (y2 ), and maximizes, which leads to y2∗ = 37.5. Now
the market price would be p∗ = 100 − 37.5 − 37.5 = 25 and the profit level for each
firm is πi∗ = (25 − 25)37.5 = 0. In their quest for leadership, they end up with nothing.
This situation, however, is quite peculiar in that firm 1 believes firm 2 is choosing
its output according to its reaction function, and firm 2 believes firm 1 is choosing its
output according to its reaction function. But both beliefs are false. (We could call this
situation a Stackelberg “disequilibrium.”)

Exercises
1. The Corleone and Chung families are the only providers of good h in the United States.
The market demand for good h is h = 1,200 − 20p. The costs of production for each
of them are represented by the cost functions C1 (h1 ) = 10h1 and C2 (h2 ) = 20h2 ,
respectively. Suppose both families must choose their output levels simultaneously.
(a) Derive their reaction functions.
(b) Calculate the Cournot equilibrium in this market. Indicate output levels, market
price, and individual profits.
2. Consider the Corleone and Chung families from Question 1. Suppose the two families
sign an agreement to restrict the amount of good h in the market. By doing this, market
price and profits will increase. Suppose the agreement specifies that given the cost
differential, Corleone will receive 3/5 of the total profits and Chung will receive 2/5.
(a) Find the solution to this collusion problem. Indicate individual outputs, market
price, total profits, and individual profits.
(b) Does either family have an incentive to break the agreement? Who does, and why?
Hint: Remember that the first-order conditions obtained from differentiation give an
interior solution. If the first-order conditions do not yield a solution, try cases in which
one of the firms produces zero.
3. MBI and Pear are the only two producers of computers. MBI started producing com-
puters earlier than Pear. MBI’s costs of production are given by C1 (y1 ) = y12 . Pear’s
cost function is C2 (y2 ) = 5y2 . The national demand for computers is y = 106 − 105 p.
(a) Calculate the Stackelberg equilibrium in which MBI is the leader in this market.
Indicate output levels, market price, and the profits of each firm.
(b) Suppose that both firms enter this market at the same time. Calculate the Cournot
equilibrium and compare it to the situation in part (a).
4. Reuben and Simeon are duopolists that produce jeans in a differentiated goods market.
The market demand for Reuben’s jeans is y1 = 80 − p1 + 12 p2 , whereas the mar-
ket demand for Simeon’s jeans is y2 = 160 − p2 + 12 p1 . Reuben’s cost function is
C1 (y1 ) = 80y1 , whereas Simeon’s cost function is C2 (y2 ) = 160y2 .
(a) Calculate the Bertrand equilibrium in this market. Indicate each firm’s price, output
level, and profits.
(b) Find prices and output levels that would maximize joint profits, and calculate the
maximum joint profits.
 Exercises 241

5. Laban and Jacob are sheep farmers in a differentiated goods market. The market
demand for Laban’s sheep wool is y1 = 34 − p1 + 13 p2 , whereas the market demand
for Jacob’s sheep wool is y2 = 40 − p2 + 12 p1 . Laban’s cost function is C1 (y1 ) = 24y1 ,
whereas Jacob’s cost function is C2 (y2 ) = 20y2 . They compete with each other through
their choices of price.
(a) Calculate the equilibrium in which Laban is the price leader in this market, and
Jacob is the price follower. Indicate prices, output levels, and individual profits.
(b) How do prices, output levels, and individual profits change if Jacob is the price
leader in this market, and Laban is the price follower?
6. Compare the social welfare properties of the following models of duopoly behavior:
simultaneous quantity setting (Cournot), quantity leadership (Stackelberg), simultane-
ous price setting (Bertrand, both homogeneous goods and differentiated goods cases),
price leadership, and collusion. Which model results in the highest output? The lowest
output? The highest price? The lowest price?