Extensive Form Games

Jonathan Levin
January 2002

1 Introduction
The models we have described so far can capture a wide range of strategic
environments, but they have an important limitation. In each game we have
looked at, each player moves once and strategies are chosen simultaneously.
This misses a common feature of many economic settings (and many classic
games such as chess or poker). Consider just a few examples of economic
environments where the timing of strategic decisions is important:

Compensation and Incentives. Firms rst sign workers to a contract,

then workers decide how to behave, and frequently rms later decide
what bonuses to pay and/or which employees to promote.
Research and Development. Firms make choices about what technologies to invest in given the set of available technologies and their
forecasts about the future market.
Monetary Policy. A central bank makes decisions about the money
supply in response to ination and growth, which are themselves determined by individuals acting in response to monetary policy, and so
on.
Entry into Markets. Prior to competing in a product market, rms
must make decisions about whether to enter markets and what products to introduce. These decisions may strongly inuence later competition.

These notes take up the problem of representing and analyzing these dynamic strategic environments.
1

2 The Extensive Form

The extensive form of a game is a complete description of:
1.
2.
3.
4.

The set of players

Who moves when and what their choices are
What players know when they move
The players payos as a function of the choices that are made.

2.1 Examples

We start with a few examples.

An Entry Model Firm 1 is an incumbent monopolist. A second rm,

Firm 2, has the opportunity to enter. After Firm 2 enters, Firm 1
will have to choose how to compete: either aggressively (Fight), or by
ceding some market share (Accomodate). The strategic situation can
be represented as follows.

Firm 2
Out

In
Firm 1

2, 0
Fight
-1, 1

Accomodate
1, 1

Stackleberg Competition An alternative to the Bertrand or Cournot

models of imperfect competition is to assume that one rm is the
market leader, while the other rm (or rms) are followers. In the
Stackleberg model, we think of Firm 1 as moving rst, and setting
a quantity q1, and Firm 2 as moving second, and setting a quantity q2, after having observed q1. The price is then determined by
2

P (q1 + q2) = 1 (q1 + q2). Lets assume that the two rms have constant marginal costs, c = 0. To keep the picture simple, we think of q
i

as taking only two values, Low and High.

Firm 1
qH

qL

Firm 2
qL

qH

qL

qH

Matching Pennies Or consider two variants of the matching pennies game.

In the rst variant, player one moves rst, and then player two moves
second after having observed player ones action. In the second, player
two does not observe player ones action.
Player 1
H

Player 1

Player 2
H

1,-1

-1,1

-1,1

1,-1

Player 2
H

1,-1

-1,1

-1,1

1,-1

Example We can also have more complicated games, where players move
multiple times, or select which player will move.

1
2

2.2 Formal Denitions

Formally, a nite extensive form game consists of:

A nite set of players i = 1, ..., I .
A nite set X of nodes that form a tree, with Z X being the terminal
nodes.
A set of functions that describe for each x / Z ,

The player i(x) who moves at x.

The set A(x) of possible actions at x.
The successor node n(x, a) resulting from action a.
Payo functions u : Z R assigning payos to players as a function
i

of the terminal node reached.

An information partition: for each x, let h(x) denote the set of nodes
that are possible given what player i(x) knows. Thus, if x h(x),
then i(x ) = i(x), A(x ) = A(x) and h(x ) = h(x).
We will sometimes use the notation i(h) or A(h) to denote the player
who moves at information set h and his set of possible actions.
Matching Pennies, cont. Lets revisit the two versions of matching pennies above. In both, we have seven nodes, three of which are nonterminal. The key dierence is the information partition. In the rst
version, each h(x) = {x} for each x. In the second, for the two middle
nodes we have h(x) = h(x ) = {x, x }.
4

In an extensive form game, write H for the set of information sets at

which player i moves.
i

H = {S X : S = h(x) for some x X with i(x) = i}

Write A for the set of actions available to i at any of his information sets.
i

2.3

Strategies

A pure strategy for player i in an extensive form game is a

function s : H A such that s (h) A(h) for each h H .
Denition 1
i

A strategy is a complete contingent plan explaining what a player will do

in every situation. Let S denote the set of pure strategies available to player
i, and S = S1 ... SI denote the set of pure strategy proles. As before,
we will let s = (s1, ..., sI ) denote a strategy prole, and s i the strategies of
is opponents.
i

In the rst version of matching pennies, S1 =

{H, T } and S2 = {HH, HT, TH, TT }. In the second version, S1 =
S2 = {H, T }.

Matching Pennies, cont.

There are two ways to represent mixed strategies in extensive form

games.
A mixed strategy for player i in an extensive form game is a
probability distribution over pure strategies, i.e. some i (Si ).
Denition 2

A behavioral strategy for player i in an extensive form game is

a function i : Hi (Ai ) such that support(i (h)) A(h) for all h Hi.

Denition 3

A famous theorem in game theory, Kuhns Theorem, says that in games

of perfect recall (these are games where (i) a player never forgets a decision
he or she took in the past, and (ii) never forgets information she had when
making a past decision see Kreps, p. 374 for formalities), mixed and
behavioral strategies are equivalent, in the sense that for any mixed strategy
there is an equivalent behavioral strategy and vice versa. Since essentially
all the games we will consider have perfect recall, we will use mixed and
behavioral strategies interchangeably.

3 The Normal Form and Nash Equilibrium

Any extensive form game can also be represented in the normal form. If we
adopt a normal form representation, we can solve for the Nash equilibrium.

Matching Pennies, cont. For our two versions of Matching Pennies, the
normal forms are:

HH HT TH TT
H 1, 1 1, 1 1, 1 1, 1
T 1, 1 1, 1 1, 1 1, 1

H
T
H 1, 1 1, 1
T 1, 1 1, 1

In the rst version, Player two has a winning strategy in the sense that
she can always create a mismatch if she adopts the strategy TH. Any
strategy for player one, coupled with this strategy for player two is a
Nash equilibrium. In the second version, the Nash equilibrium is for
both players to mix 12 H + 12 T .
Entry Game, cont. For the entry game above, the normal form is:

Out In
F 2, 0 1, 1
A 2, 0 1, 1
There are several Nash equilibria: (A, In), (F, Out) and (F + (1
)A, Out) for any 1/2.

Note that in the entry game, some of the Nash equilibria seem distinctly
less intuitive that others. For instance, in the (F, Out) equilibrium, it is the
threat of Fight that keeps Firm 2 from entering. However, if Firm 2 were to
enter, is it reasonable to think that Firm 1 will actually ght? At this point,
it is not in Firm 1s interest to ght, since it does better by accomodating.
Consider another example, where this problem of incredible threats arises
in way that is potentially even more objectionable.

Stackleberg Competition In the Stackleberg model, for any q1 [0, 1],

there is a Nash equilibrium in which Firm 1 chooses quantity q1. To
see this, consider the strategies:

s1 = q1
and

s2 =

1q1
2

1 q1
6

if q1 = q1
if q1 = q1

Lets check that these are all equilibria. First, given Firm 2s strategy,
Firm 1 can either set 1 = 1, or 1 = 1 . If it does the former, the
eventual price will be zero, and Firm 1 will make zero prots. If it
does the latter, then Firm 1 will make prots:

1 

1

1
=
1 1 1
1 1 1 0.
2
2
Now, consider Firm 2. Does it have a strategy that yields strictly
higher payo. Clearly, changing its strategy in response to 1 = 1
will have no eect on its payo given Firm 1s strategy. On the other
hand, in response to 1 = 1, its best response solves:
q

max
q2

q2

1 1
q

q2

The solution to this problem is (1 1 ) 2, so Firm 2 is playing a best

response.
q

As in the previous case, many of the Nash equilibria in the Stackelberg

model seem unreasonable. If Firm 1 sets 1 = 1, then Firm 2 typically
has options that give a positive payo. However, it chooses to ood the
market and drive the price to zero. Thus, o the equilibrium path, unreasonable things are happening. And not only is Firm 2 being allowed to make
incredible threats, we have a huge multiplicity of equilibria.
q

4 Subgame Perfect Equilibrium

In response to the problems of credibility we have seen in the last two examples, we now introduce the idea of a subgame perfect equilibrium. Subgame
perfection tries to rule out incredible threats by assuming that once something has happened, players will always optimize going forward.

4.1 Subgame Perfection

Denition 4 Let G be an extensive form game, a subgame G of G consists

of (i) a subset Y of the nodes X consisting of a single non-terminal node x
and all of its successors, which has the property that if y Y , y h(y) then
y Y , and (ii) information sets, feasible moves, and payofs at terminal
nodes as in G.
7

Entry Game, cont. In the entry game, there are two subgames. The entire

game (which is always a subgame) and the subgame after Firm 2 has
entered the market.
Example In game below, there are four subgames: (1) The entire game,
(2) the game after player one chooses R, (3) the game after player
one choose L, and (4) the game after Player 1 chooses L and player 2
chooses l.
1
L

Denition 5 A strategy prole s is a subgame perfect equilibrium of G

if it induces a Nash equilibrium in every subgame of G.
Note that since the entire game is always a subgame, any SPE must also
be a NE.
Entry Game, cont. In the entry game, only (A, In) is subgame perfect.

4.2 Application: Stackleberg Competition

Consider the model of Stackleberg Competition where Firm 1 moves rst and
chooses quantity q1, and then Firm 2 moves second and chooses quantity
q2 . Once both rms have chosen quantities, the price is determined by:
( ) = 1 , where = 1 + 2. So that we can compare this model
to Bertrand and Cournot competition, lets assume that both rms have
constant marginal cost equal to 0 1.
To solve for the subgame perfect equilibrium, we work backward. Suppose that Firm 1 has set some quantity 1. Then Firm 2s best response
solves:
max
2 (1 1 2 )
q2
P Q

c <

The rst-order condition for this problem is:

0= 1 1 2 2
which gives a best response:


1

.
2 ( 1 ) = max 0
2
Now consider the problem facing Firm 1, knowing that if it chooses 1 ,
Firm 2 will respond with a quantity 2 ( 1 ). It solves

max
1 (1 1 2 ( 1 ) )
q1
c

q ,

The rst-order condition for this problem is:

0 = 1 2 2 1
q

Solving this out yields:

q1

= 1 2

and

q2

q1

dq2

dq1

= 1 4

The total quantity is 34 (1 ) and the price is S = (1 + 3 ) 4 In comparison, under Cournot competition, both rms set identical quantity 13 c , so
total quantity is 23 (1 ) and the price is C = (1 + 2 ) 3
c

c / .

c / .

q2
1-c
q1 = BR(q 2)

Cournot
Stackelberg
q2 = BR(q 1)
0
.

1-c

q1

Stackelberg Competition
Relative to Cournot, in Stackelberg Competition, the Leader (Firm 1)
can choose any point on the Followers (Firm 2s) best-response curve. Note
that this game has a rst-mover advantage in the sense that there is an
advantage to being the leader.
9

4.3

Backward Induction

As the previous examples illustrate, a common technique for identifying

subgame perfect equilibria is to start at the end of the game and work back
to the front. This process is called backward induction.

Denition 6 An extensive form game is said to have perfect information if each information set contains a single node.
Proposition 7 (Zermelos Theorem) Any nite game of perfect information has a pure strategy subgame perfect equilibrium. For generic payos,
there is a unique SPE.

Proof. (very informal) In a nite game of perfect information, we can label

each node as belonging to stage 1, stage 2, ..., stage K . To nd a pure

strategy SPE (and for generic payos, the unique SPE), we rst consider all
nodes x in stage K . Each node starts a subgame, so by NE the player who
moves must maximize his or her expected payo. Identify the optimal choice
(generically, it will be unique). Now move back to stage K 1, and consider
the problem facing a player who moves here. This player can assume that at
stage K , play will follow the choices just identied. Since each node starts a
subgame, we look for the payo-maximizing choice facing a player who gets
to move. Once these choices have been identied, we move back to stage
K 2. This process of continues until we reach the beginning of the game,
at which point we will have at least one (and typically no more than one).
Q.E.D.

Example Here is a non-generic game where backward induction reveals

three pure strategy subgame perfect equilibria: (R, A), (R, B ) and
(L, B ).

10

1
L

R
2

0, 2
A

1, 1

0, 1

We can use backward induction even in games without perfect information as the next example demonstrates.

Example To use backward induction in the game below, we rst solve for

the subgame perfect equilibrium after Firm 2 has entered. We see that
the unique equilibrium in this subgame is for both Firm 1 and Firm
2 to play D. (Note that this subgame is a prisoners dilemma. Hence
Firm 2 will choose not to enter at the rst node.
Firm 2
In

Out

Firm 1
2,

D
Firm 2

C
1,1

-1,2 2,-1

D
0,0

Another Entry Game

4.4 Criticisms of Subgame Perfection

We motivated Subgame Perfection as an attempt to eliminate equilibria that

involved incredible threats. As we go on to consider applications, we will
11

use SPE regularly as a solution concept. Before we do this, however, it is

worth pausing momentarily to ask whether SPE might be over -zealous in
eliminating equilibria.

Example: Trusting Someone to be Rational Here the unique SPE is

for Player 1 to choose R and Player 2 to choose B . However, Goeree
and Holt (2001) report an experiment in which more than 50% of
player ones play S .

Player 1
S

R
Player 2

80, 50
A

20, 68

90, 70

The Centipede Game In games with many stages, backward induction

greatly stresses the assumption of rationality (and common knowledge
of rationality). A famous example due to Rosenthal (1981) is the
centipede game. The unique SPE is for Player 1 to start by moving
Out, but in practice people do not seem to play the game this way.

2
In

1, 0

1
In

0, 2

2
In

In

3, 1

2, 4

12

2
In

5, 3

7, 5
In

4, 6

5 Stackelberg (Leader-Follower) Games

Above, we considered Stackelberg competition in quantities. We now consider Stackelberg competition in prices, then t both models into a more
general framework.

5.1 Stackelberg Price Competition

In the Stackelberg version of price competition, Firm 1 moves rst and

commits to a price p1. Firm 2 observes p1 and responds with a price p2 .
Sales for Firm i are then given by Q (p1, p2). Supposing that the two rms
have constant marginal costs equal to c, rm is prots can be written as:
i

i (p1 , p2) = (pi c) Qi (pi , pj ) .

Homogeneous Products. If the rms products are heteregenous, then

rm that sets a lower price gets demand Q(p) (where p = min{p1, p2}), and

the rm that sets a higher price gets no demand. Suppose that if p1 = p2

the consumers split equally between the two rms. In this case, if Firm 1
chooses p1 > c, Firm 2 would like to choose the highest price less than p1 .
Unfortunately, such a price does not exist! So there are subgames in which
Nash equilibria do not exist and no SPE.
To resolve this, assume that if p1 = p2 , then all consumers purchase from
Firm 2. Then after observing p1 c, Firm 2 responds by taking the entire
market either by choosing p2 = p1 if p1 pm (the monopoly price), or by
choosing p2 = pm if p1 > pm . It follows that Firm 2s best response function
is given by:
m if
m

m
p2 (p1) =
1 ifif 11 [ ] .
Thus, for any 1 , Firm 1 gets the entire market but loses money, which
for any 1 , Firm 1 gets no demand. It follows that any pair ( 1 2( 1))
with 1 is an SPE.
Note that compared to the Bertrand (simultaneous move) equilibrium,
the price may be higher. In particular, both Firms 1 and 2 set (weakly)
higher prices. Moreover, Firm 2 (the follower) does better than in the simultaneous game, while Firm 1 does the same. Thus we say the game has
a second mover advantage.
p

> p

c, p

< c

< c

p ,p

13

With heterogeneous products, the situation

is similar to Stackelberg quantity competition, except with Bertrand bestresponses rather than Cournot!
Heterogeneous Products.

p2
p2 = BR(p 1)

Bertrand

Stackelberg
(1+c)/2
p1 = BR(p 2)
0

p1

(1+c)/2

Stackelberg Price Competition

5.2

ollower Games

General Leader-F

The Stackelberg models of imperfect competition are examples of what Gibbons calls Leader-Follower games. These games have the following structure:
1. Player 1 moves rst and chooses an action a1 A1.
2. Player 2 observes a1 and chooses an action a2 A2.
There is a simple algorithm to identify the subgame perfect equilibria
of this sort of game. We just apply backwards induction. We rst dene
player 2s best response to any action by Player 1:
(a , a ).
a2(a1) = arg amax
A 2 1 2
2

We then identity what player one should do, assuming that player two will
best respond. To do this, dene:
a1 = arg amax
(a , a(a )) .
A 1 1 2 1
1

A subgame perfect equilibrium is a pair (a1 , a2(a1)).

From our examples, we can make several observations about leaderfollower games.
14

1. The Leader always does (weakly) better than in a simultaneous move

pure strategy equilibrium setting (note that this is not true for mixed
strategy equilibria think about matching pennies).
2. The Leader tends to distort his behavior relative to the simultaneous
move game (how he does so depends on Firm 2s best response function
and on what sort of action he prefers Firm 2 to choose).
3. Whether the Follower does better than in the simultaneous game depends on both Firm 2s best response function and the interdependence
in payos (how is action aects j s payo and vice-versa).

6 Strategic Pre-Commitment
We now turn to a class of problems that arise frequently in industrial organization. These problems have the following structure. First, one player
(Firm 1) has the opportunity to take some sort of action or investment
for instance, installing capacity, investing in a cost-saving technology or in
product development, signing an exclusive contract with a key upstream
suppliers, building a relationship with customers, or so on. Then Firm 2 decides whether or not to enter the market. Finally, rms compete with Firm
1 either operating as a monopoly or the two rms competing as a duopoly.
By taking an action in advance of competition, Firm 1 has the opportunity to strategically pre-commit (just as the Stackelberg leader pre-commits
to a price or quantity). It turns out that it is possible to give a very intuitive
analysis of the economics of pre-commitment by using the idea of strategic
complements and substitutes.1 This analysis can then be used to shed light
on a range of problems in industrial organization and other elds.

6.1 Strategic Complements and Substitutes

Let G be a simultaneous move game in which the players 1 and 2 takes

actions a1, a2 R, and have payos 1 (a1, a2 ), 2 (a1, a2). Let BR1(a2)
and BR2 (a1 ) denote the best-response functions (assume best responses are
unique).
Denition 8
0.

The players actions are strategic complements if BR ()

strategic substitutes if BR () 0.
i

The actions are

1
This section is drawn from Tirole (1988, Chapter 8). See Fudenberg and Tirole (1984,
AER) or Bulow, Geanokoplos and Klemperer (1985, JPE) for the original analyses.

15

a2

a2

Strategic Complements

Strategic Substitutes
a 1 = BR2(a 2)

a2 = BR1(a 1)
a2 = BR1(a 1)
a1 = BR2(a 2)

a1

a1

We have already seen that in Cournot Competition, quantities are strategic substitutes, while in dierentiated products Bertrand Competition, prices
are strategic complements.
Proposition 9 Suppose for i = 1, 2,. is twice continuously dierentiable.
0 and strategic substitutes if
Then G has strategic complements if a2a
2
a a 0.
i

For those of you who took Econ 202, these conditions should look very
familiar from our study of Monotone Comparative Statics. A good exercise
is to try to prove this result using Topkis Theorem.

6.2 Strategic Pre-Commitment

We consider the following model:

Firm 1 moves rst and chooses and investment k.

Firm 2 observes k and decides whether or not to compete (enter).
Firms 1 and 2 choose actions a1 A1, a2 A2.
Payos are given by 1 (k, a1 , a2 ) and 2 (k, a1 , a2).2

Note that Leader-Follower games are a special case of this setting. To get the LeaderFollower case, let A1 be a singleton, and think of k as Firm 1s action and a2 as Firm 2s
action.
2

16

We assume that for any choice k, the competition subgame has a unique
Nash Equilibrium, which we can denote ac1 (k), ac2(k). Payos in this game
are given by:
i (k, ac1 (k) , ac2 (k)) for i = 1, 2.
Thus given a choice of k, Firm 2 will choose to enter if
2 (k, ac1(k), ac2 (k)) > 0.
If Firm 2 does not enter, then Firm 1 sets chooses the monopoly strategy
am
1 . Payos are
for Firm 1 and zero for
investment.
We say that:

m
m
1 (k, a1 (k ))
Firm 2. Let k denote the

subgame perfect level of

Entry is deterred if 2 (k, ac1(k ), ac2(k )) 0.

Entry is accomodated if 2 (k , ac1(k), ac2(k)) > 0,
If entry is deterred, then the SPE involves Firm 1 choosing am1 (k ), and
achieves prots m1 (k, am1 (k )). If entry is accomodated, Firms choose
ac1(k ), ac2(k ).
An alternative to this model would be a case without pre-commitment
where Firm 1 chooses k at the same time as a1, a2 (or chooses k in advance
but without it being observed). Lets assume that this game also has a
nc
unique Nash Equilibrium, denoted (knc, anc
1 , a2 ). We will say that:
nc
Entry is blockaded if 2 (knc, anc
1 , a2 ) 0.

In what follows, we assume that entry is not blockaded. In addition, we will

assume that m1 and 1 are both concave in k.
6.3

Entry Deterrence

Lets rst consider SPE in which entry is deterred. In this case, Firm 1
need to choose a level of k that makes Firm 2 less protable. Indeed, it will
choose an investment k such that:3
2 (k, ac1 (k), ac2(k)) = 0.
The fact that Firm 1 will choose k to make Firm 2s competition prots exactly zero
follows from the assumption that entry is not blockaded and that payos functions are
concave.
3

17

Lets consider what sort of strategy by Firm 1 works to make Firm 2

unprotable and hence deter entry. From second period optimization:
2 c c
(k , a1 (k ), a2 (k )) = 0
a2

Hence:

d2
dk

2
k



direct eect

2 a1
a1 k

 
strategic eect

To deter entry, Firm 1 wants to choose an investment that will make

Firm 2 less protable. It has two ways to do this. It may be able to invest
in a way that makes Firm 2 directly less protable. It make also be able
to change the nature of competition for example, if k is investment in
capacity, k has no direct eect, but only a strategic eect.
To classify Firm 1s strategies, we adopt the terminology of Fudenberg
and Tirole (1984).
Denition 10 Investment makes Firm 1

makes Firm 1 soft if

d 2 >
dk

0.

tough

if

d2 <
dk

0. Investment

Fudenberg and Tirole suggest the following typology of strategies for

investment.

: Be big (invest a lot) to look tough.

Puppy Dog: Be small (invest only a little) to look soft.
Lean and Hungry Look: Be small to look tough.
Fat Cat: Be big to look soft.

Top Dog

Example: Reducing Own Costs Suppose that Firm 1 has the opportu-

nity to invest to lower its marginal costs. If Firm 2 enters, competition

will be Cournot. There is no direct on Firm 2. But there is a strategic
eect. Investment will shift out Firm 1s best-response function, and
lead to a competitive outcome where Firm 1 produces higher quantity
and Firm 2 lower quantity. This top dog strategy may deter entry.
Example: Building a Customer Base Suppose that Firm 1 has the opportunity to invest in customer relations, building up a loyal customer
base. The direct eect of this is to limit the potential market for Firm
18

q2

q1 = BR(q2;kH)

q1 = BR(q 2;kL)
0
0

q2 = BR(q 1)

q1

Figure 1:
2, making Firm 1 tough. However, there may be a second eect. If
Firm 1 cannot price discriminate, it may be tempted to set a high price
to take advantage of its locked in customers. This strategic eect can
work to make Firm 1 soft. The overall eect is ambiguous. Thus either
a top dog strategy or a lean and hungry look might work to deter
entry depending on the specics.

6.4 Accomodation: Changing the Nature of Competition

Suppose now that Firm 1 nds it too costly to deter entry. How should it
behave in order to make more prots once Firm 2 enters? In this case, Firm
1 is interested in choosing its investment to maximize:
1 (k, ac1 (k) , ac2 (k)) .

The total derivative is given by (using the envelope theorem):

d1
dk

1
k



direct eect

1 da2
.
a2 dk

 
strategic eect

That is, the rst term is the direct eect. This would exist even if investment
was not observed by Firm 2. The strategic eect results from the fact that
Firm 1s investment can change the way Firm 2 will compete.
Lets investigate the strategic eect further. To do this, lets assume
that the second-period actions of the two rms have the same nature in the
19

sense that 1/a2 has the same sign as 2/a1. For instance, either both
rms choose quantities or both choose prices. Then:
Thus,

dac2
dk


dac2 da1
da1 dk

dBR2 (a1 ) da1

da1
dk


1 da2
dBR2 (a1 )
1 da2
= sign
sign
a2 dk
a2 dk
da1
Assuming that 2/k = 0, we can identify the sign of the strategic eect

sign

with two things: (1) whether investment makes Firm 1 Tough or Soft (the
rst term) and (2) whether there are strategic substitutes or complements
(the second term). We have four cases:
If investment makes Firm 1 tough and reaction curve slope down,
investment by Firm 1 softens Firm 2s action thus Firm 1 should
overinvest (top dog).
If investment makes Firm 1 tough and reaction curves slope up, Firm
1 should overinvest so as not to trigger an aggressive response by Firm
2 (puppy dog).
If invesmtnent makes Firm 1 soft and reaction curves slope down, Firm
1 should stay lean and hungry.
If invesment makes Firm 1 soft and reaction curves slope up, Firm 1
should overinvest to become a fat cat.
To summarize,
Investment makes Firm 1
Tough
Soft
Strategic Complements Puppy Dog
Fat Cat
Strategic Substitutes
Top Dog Lean and Hungry
Example: Reducing Costs Suppose Firm 1 can invest to reduce its costs
before competing. With Cournot competition, the strategic eect to
make Firm 1 more aggressive. The equilibrium changes so that Firm 2
ends up producing less. Thus Firm 1 wants to be a Top Dog and invest
heavily. On the other hand, if Firm 1 can reduce its costs before price
competition, the strategic eect is to make Firm 1 more aggressive
so that in equilibrium Firm 2 ends up pricing more aggressively as
well. Thus, Firm 2 might want to be a Puppy Dog to soften price
competition.
20

6.5

Applications in Industrial Organization

1. Product Dierentiation. Suppose Firm 1 can choose its products location prior to competing in price with Firm 2. Producing a
product that is close to Firm 2s will tend to make price competition
more intense, lowering prices and prots. To deter entry or to drive
Firm 2 from the market, Firm 1 might want to adopt a Top Dog
strategy. But to change the nature of competition in a favorable way,
Firm 1 might want to adopt a Puppy Dog ploy and dierentiate its
oering from Firm 2s.
2. Most-Favored Customer Clause. With price competition, if Firm
1 wants to accomodate Firm 2, it wants to look inoensive so as to
keep Firm 2 from cutting price. In particular, it would like to commit
itself to charging high prices (a Puppy Dog ploy). One way to do
this is most-favored customer clauses. Firm 1 can write contracts with
its customers promising that if it ever oers a low price to another
customer, the original customer can get the new low price. This makes
it very costly for Firm 1 to drop prices, eectively committing itself
to be unaggressive in competing with Firm 2.
3. Advertising. Suppose Firm 1 can invest in advertising that makes
customer more excited not just about its own product, but about
the whole market. This kind of advertising makes Firm 1 soft. To
deter entry, Firm 1 should not do much of this sort of advertising
rather it should run advertisements that increase the demand for its
own product and decrease the demand for other rms product. But
what if Firm 2 surely plans to enter. If competition is in prices, then
Firm 1 will want to advertise in a way that increases demand (direct
eect) and in a way that softens price competition (for example by
establishing separate niches for dierent products in the market). This
is a fat cat approach to advertising.
4. Leverage and Tying. An old story in IO is that a rm with monopoly
power in one market can leverage this power to monopolize a second.
(Think Microsoft and browsers.) One way to do this is to tie the two
products together. Suppose that there are two markets, that Firm 1
has a monopoly in the rst, and that Firms 1 and 2 may compete in
prices in the second. Firm 1 must decide whether to bundle or tie its
two products. The question is how this action will eect its pricing
behavior. This depends on how we model demand, but in many cases,
21

bundling will make demand more elastic. This leads Firm 1 to price
more aggressively in response to Firm 2s prices. It follows that from
a strategic point of view, bundling is a top dog strategy that works to
deter entry. But if entry is to be accomodated, it may be better to use
the puppy dog ploy of not bundling.

22