Subgames and their

Chapter 5 equilibria

aThe concept of subgames

Extensive Form aEquilibrium of a subgame
aCredibility problems: threats you have no
Games with incentives to carry out when the time
comes
Perfect aTwo important examples
Information `Telex vs. IBM
`Centipede
1 2

Subgame Perfection
Game in Extensive Form (Selten, 1965)

aWho plays when? Nash Equilibrium: each player must act

aWhat can they do? optimally given the other players'
aWhat do they know? strategies, i.e., play a best response to
the others' strategies.
aWhat are the payoffs?
Problem: Optimality condition at the
beginning of the game. Hence, some
Nash equilibria of dynamic games involve
incredible threats.
3 4

Game in Extensive Form:

Backward Induction Game in Normal Form
1 2
L R ll lr rl rr

2 2 L 1, 3 1, 3 2, 0 2, 0
1
l r l r R 4, 2 0, 1 4, 2 0, 1

1: 1 2 4 0 a Three Nash equilibria in pure strategies: {R,ll}, {L,lr},

2: 3 0 2 1 and {R,rl}.
a {L,lr}, and {R,rl} involve incredible threats.
Unique equilibrium path
5 6
Subgame Perfection
with Perfect Information Definition

Consider a game Γ of perfect information

consisting of a tree T linking the A Nash equilibrium of Γ is subgame perfect
information sets i ∈ I (each of which if it specifies Nash equilibrium strategies
consists of a single node) and payoffs at in every subgame of Γ. In other words,
each terminal node of T. A subtree Ti is the players act optimally at every point
the tree beginning at information set i, during the game.
and a subgame Γi is the subtree Ti and
the payoffs at each terminal node of Ti.

7 8

Subgame Perfection with Telex vs. IBM, extensive form:

Imperfect Information subgame, perfect information
1 Subgame
L R 0, 0
2 Smash

l r l r Enter
IBM

Telex Accommodate
1: 3 1 2 4 2, 2
2: 1 4 3 2 Stay Out
With imperfect information, each information set consisting
of a single node determines a subgame. Hence, there are 1, 5
no (proper) subgames in this example.
9 10

Telex vs. IBM, extensive form: Telex vs. IBM, normal form:
no subgame The payoff matrix

0, 0 IBM
Smash Telex Smash Accommodate

Enter
Enter 0, 0 2, 2
Accommodate 2, 2
IBM
Telex
Stay Out Smash 1, 5
Stay Out 1, 5 1, 5

Accommodate 1, 5 11 12
Telex vs. IBM, normal form: Telex vs. IBM, normal form:
Strategy for IBM Strategy for Telex
IBM IBM
Telex Smash Accommodate Telex Smash Accommodate

Enter 0, 0 2, 2 Enter 0, 0 2, 2

Stay Out 1, 5 1, 5 Stay Out 1, 5 1, 5

13 14

Telex vs. IBM, normal form: Telex vs. IBM, extensive form:
Two equilibria noncredible equilibrium
IBM
Telex Smash Accommodate 0, 0
Smash

Enter IBM
0, 0 2, 2 Enter

Telex Accommodate
2, 2
Stay Out 1, 5 1, 5 Stay Out
1, 5
15 16

Telex vs. IBM, extensive form:

credible equilibrium Centipede, extensive form

Take the
money 1, 0
0, 0
Smash
IBM 1 Take the
Enter 0, 4
money
Wait
Telex Accommodate
2, 2 2
Stay Out
Split the
1, 5 money 2, 2
17 18
 Centipede, normal form:
Centipede, extensive form The payoff matrix
Take the Player 2
money 1, 0 Take the Split the
Player 1 money money

1 Take the Take the

0, 4 money 1, 0 1, 0
money
Wait

2
Wait 0, 4 2, 2
Split the
money 2, 2
19 20

Centipede, normal form: Centipede, normal form:

Strategy for player 1 Strategy for player 2
Player 2 Player 2
Take the Split the Take the Split the
Player 1 money money Player 1 money money

Take the Take the

money 1, 0 1, 0 money 1, 0 1, 0

Wait 0, 4 2, 2 Wait 0, 4 2, 2

21 22

Centipede, normal form:

The equilibrium Look Ahead and Reason Back
Player 2
Take the Split the aThis is also called Backward Induction
Player 1 money money aBackward induction in a game tree leads to
a subgame perfect equilibrium
Take the
money 1, 0 1, 0 aIn a subgame perfect equilibrium, best
responses are played in every subgames

Wait 0, 4 2, 2

23 24
 Credible Threats and
Promises Telex vs. Mean IBM

aThe variation in credibility when money is

all that matters to payoff Smash
0, 4
aTelex vs. Mean IBM
aCentipede with a nice opponent Enter
IBM
aThe potential value of deceiving an
opponent about your type Telex
Accommodate
2, 2
Stay Out
1, 5

25 26

Centipede with a nice opponent, Centipede with a nice opponent,

extensive form normal form: The payoff matrix
Take the Player 2
money 1, 0 Take the Split the
Player 1 money money

1 Take the Take the

0, 0 money 1, 0 1, 0
money
Wait

2
Wait 0, 0 2, 2
Split the
money 2, 2
27 28

Centipede with a nice opponent, Centipede with a nice opponent,

normal form: Strategy for player 1 normal form: Strategy for player 2
Player 2 Player 2
Take the Split the Take the Split the
Player 1 money money Player 1 money money

Take the Take the

money 1, 0 1, 0 money 1, 0 1, 0

Wait 0, 0 2, 2 Wait 0, 0 2, 2

29 30
 Centipede with a nice opponent,
normal form: The equilibrium Mutually Assured Destruction
Player 2
Take the Split the aThe credibility issue surrounding weapons
Player 1 money money of mass destruction
aA game with two very different subgame
Take the
money 1, 0 1, 0 perfect equilibria
aSubgame perfection and the problem of
mistakes
Wait 0, 0 2, 2

31 32

MAD, extensive form: path to final

MAD, extensive form: entire game backing down
Player 2
Player 1 Doomsday Back down

Doomsday -L, -L -L, -L -0.5, -0.5

Escalate
Back
down -L, -L -0.5, -0.5 Escalate 2
Escalate
Back down
1
Escalate 1, -1
2 Ignore
Back down
1 1, -1 0, 0
Ignore
33 34
0, 0

MAD, normal form: b = Back down; e = Escalate;

MAD, extensive form: path to D = Doomsday; i = Ignore; = equilibrium;
Doomsday = subgame perfect equilibrium
Country 2
e, D e, b b, D b, b
Country 1
-L, -L
Escalate e, D -L, -L -L, -L 1, -1 1, -1

Escalate 2 e, b -L, -L -0.5, -0.5 1, -1 1, -1

Back down
1
1, -1 i, D 0, 0 0, 0
Ignore 0, 0 0, 0

0, 0
i, b 0, 0 0, 0 0, 0 0, 0
35 36
Credible Quantity Competition: Cournot-Stackelberg Equilibrium:
Cournot-Stackelberg Equilibrium firm 2’s best response
aThe first mover advantage in Cournot- X2 = q(x1)
Stackelberg competition
Monopoly
60
aOne firm sends its quantity to the market Cournot Point
first. The second firm makes its moves
40
subsequently. Stackelberg Point

aThe strategy for the firm moving second is 30

a function 20
aIncredible threats and imperfect equilibria Stay out
X1
0 40 60 120
37 38

Cournot-Stackelberg Equilibrium Cournot-Stackelberg Equilibrium for

for two firms two firms: Firm 2 maximizes its profits
Market Price, P = 130 - Q Firm 2 faces the demand curve,
P = (130 - x1) - x2
Market Quantity, Q = x1 + x2
Firm 2 maximizes its profits,
Constant average variable cost, c = $10
max u2(x) = x2(130 - x1 - x2 - 10)
Firm 1 ships its quantity, x1, to market first
Differentiating u2(x) with respect to x2:
Firm 2 sees how much firm 1 has shipped and 0 = ∂u2/∂x2 = 120 - x1 - 2x2
then ships its quantity, x2, to the market
⇒ x2 = g(x1) = 60 - x1/2

39 40

Games like Chess:

Tic-Tac-Toe Player draft

aEach team takes turn choosing players

aIs it best to always chose most preferred
player?

41 42
Cournot-Stackelberg Equilibrium: Firm 1 The Cournot-Stackelberg
also wants to maximize its profits Equilibrium for two firms
Firm 1’s profit function is given by: The Cournot-Stackelberg equilibrium value of
u1(x) = [130 - x1 - g(x1) - 10] x1 firm 1’s shipments, x1* = 60
Substituting g(x1) into that function: Firm 2’s shipments, x2* = 60 - 60/2 = 30
u1(x) = (120 - x1 - 60 + x1/2) x1 Market Quantity, Q = 60 + 30 = 90
∴ Firm 1’s profits depend only on its Market Price, P = 130 - 90 = $40
shipment
This equilibrium is different from Cournot
Taking the first order condition for u1(x): competition’s equilibrium, where x1* = x2* = 40,
0 = 60 - x1 Q = 80 and P = $50
43 44

Credible Price Competition: Bertrand -Stackelberg Equilibrium

Bertrand-Stackelberg Equilibrium for two firms
Market Price, P = 130 - Q and
aPrice is the strategic behavior in Bertrand- Constant average variable cost, c = $10
Stackelberg competition Firm 1 first announces its price, p1
aThe strategy for the firm moving second is
a function Firm 2’s profit maximizing response to p1:
aFirm 2 has to beat only firm 1’s price which p2 = $70 if p1 is greater than $70
is already posted p2 = p1 - $0.01 if p1 is between $70 and $10.02
aThe second mover advantage in Bertrand-
Stackelberg competition p2 = p1 if p1 = 10.01
p2 = $10 otherwise
45 46
Get competitive outcome; no extra profits!

Market Games with

Differentiated Products Differentiated Products
aPrice and quantity competition when All differentiated products
products are differentiated
have one thing in common:
aCournot and Bertrand equilibrium still
different, but the difference is muted
if the price is slightly above
aMonopolistic competition as the limit of
the average price in the
market game equilibrium market, a firm doesn’t lose all
its sales

47 48
 Two firms in a Bertrand competition Two firms in a Bertrand competition
aThe demand function faced by firm 1: aFirm 1 has profits
x1(p) = 180 - p1 - (p1 - average price) u1(p1,p2) = (p1 - 20) x1
= (p1 - 20) (180 - 2p1 + average price)
aThe demand function faced by firm 2: = (p1 - 20) (180 - 1.5p1 + 0.5p2)
x2(p) = 180 - p2 - (p2 - average price) aFirm 2’s profit function
u2(p1,p2)
= (p2 - 20) (180 - 1.5p2 + 0.5p1)

49 50

Bertrand best responses, two firms,

Maximizing profits differentiated products
Firm 1 maximizes its profits when its p1 = f1(p2) = 70 + p2/6
marginal profit is zero: p2
p2 = f2(p1) = 70 + p1/6
0 = ∂u1/∂p1
= (p1 - 20) (-1.5) + (180 - 1.5p1 + 0.5p2)
⇒ 0 = 210 - 3p1 + 0.5p2 p* = (84, 84)
Firm 1’s best response function:
p1 = f1(p2) = 70 + p2/6
Similarly, Firm 2’s best response function:
p1
p2 = f2(p1) = 70 + p1/6
51 52

Bertrand Equilibrium Bertrand competition with n firms

aThe Bertrand equilibrium of the market game aFirm 1’s market demand
is located at (84, 84) x1= 180 - p1 - (n/2) ( p1 - average price)
aThe market price is $84, significantly higher aFirm 1’s profit function
than the marginal price which is given at $20 u1(p) = (p1 - 20) x1
aEach firm sales (180 - 84) units = 96 units aWhen the first order condition is satisfied
aEach firms profit = (84 - 20) × 96 0 = ∂u1/∂p1 = (p1 - 20)(-1-n/2+1/2)+ x1
= $6144 ⇒ K (p1 - 20) = 180 - p1
aTherefore, each firm could spend over $6000 where K = (n + 1)/2
in differentiating its products and can still ∴ p1* = 180 /(K+1) + 20K/(K+1)
come out ahead
53 54
 Bertrand competition with infinite
number of firms Differentiated Products
an → ∞ ⇒ K → ∞ and 1/K → 0 aProduct differentiation mutes both types of
aTaking limit of p1* as n goes to infinity mover advantage
lim p1* = lim 20/(1 + 1/K) = 20 aA mover disadvantage can be offset by a
aIn this limit, price is equal to marginal large enough cost advantage
cost and profits vanish.
aThis limit is monopolistic competition

55 56

Two firms in a Bertrand-Stackelberg Two firms in a Bertrand-Stackelberg

competition competition: Determining optimum p2
The demand function faced by firm 1: Firm 2 wants to maximize its profits, given p1:
x1(p) = 180 - p1 - (p1 - average p) max (p2 - 20)(180 + 0.5p1 - 1.5p2)
⇒ x1 = 180 - 1.5p1 + 0.5p2 Profit maximizes when the first order condition is
Similarly, the demand function faced by firm 2: satisfied: 0 = 180 + 0.5p1 - 3p2 + 30
x2 = 180 + 0.5p1 - 1.5p2 Solving for optimal price p2, we get
Constant average variable cost, c = $20 p2* = g(p1) = 70 + p1/6

57 58

Two firms in a Bertrand-Stackelberg Two firms in a Bertrand-Stackelberg

competition: Equilibrium prices competition: Profits for the two firms
Knowing that firm 2 will determine p2 by using Firm 1 sells less than firm 2 does:
g(p1), firm1 tries to maximize its profit: x1* = 93.34 and x2* = 96.48
max (p1 - 20)[180 - 1.5p1 + 0.5(70 + p1/6)] Firm 1’s profit, u1* = (93.34)(85.88 - 20)
Profit maximizes when the first order condition is = $ 6149.24
satisfied: 0 = 215 - (17/12)p1 + (p1 - 20) (-17/12) Firm 2’s profit, u2* = (96.48)(84.31 - 20)
∴ p1* = 2920/34 = $85.88 = $ 6204.63
Firm 2, which moves last, charges slightly lower Firm 2, the second mover, makes more
price than p1*:
money
p2* = 70 + p1* /6 = 70 + $14.31 = $84.31
59 60
 This Offer is Good for a An example of “This offer is good
Limited Time Only for a limited time only”

aThe credibility problems behind the aExploding job offers

marketing slogan `An early job offer with a very short
aThe principle of costly commitment time to decide on whether to take the
aIndustries where the slogan is credible job.
`Risk-averse people often end up
accepting inferior job offers

61 62

Ultimatum Games in the

Laboratory

aGames with take-it-or-leave-it structure

aIn experiments, subjects playing such
games rarely play subgame perfect
equilibria
aThe nice opponent explanation vs. the
expected payoff explanation

63