CUR 412: Game Theory and its Applications,

Lecture 8

Prof. Ronaldo CARPIO

April 26, 2016

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Announcements

▸ HW #3 will be posted later today, due in 2 weeks.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Review of Previous Lecture: Extensive Games

▸ Extensive games can model a situation where players move

sequentially.
▸ We describe extensive games with a game tree.
▸ Each node in the tree corresponds to a player’s turn to move.
▸ Each branch from a node corresponds to an action of the player
who moved.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Review of Previous Lecture: Extensive Games

▸ A history is a sequence of actions (i.e. a path from the root node to

some other node).
▸ A terminal history is a path from the root node to a leaf node (i.e.
an ”ending” to the game).
▸ Players have preferences over terminal histories.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategies in Extensive Form Games

▸ One approach to solving an extensive form game is to

formulate it as a strategic (i.e. simultaneous-move) game,
then use the Nash equilibrium solution concept.
▸ We need to expand the definition of an ”strategy”.
▸ Before, a player’s strategy only had to specify one action.
▸ Now, it must specify the player’s action at all places in the
game where that player moves.
▸ Definition: A strategy of player i in an extensive game with
perfect information is a function that assigns to each history h
after which it is i’s turn to move, an action in A(h) (the
actions available after h).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
▸ In this game, Player 1 only moves at the start (i.e. after the empty
history ∅). The actions available are C , D, so Player 1 has two
strategies: ∅ → C , ∅ → D.
▸ Player 2 moves after the history C and also after D. After C ,
available actions are E , F . After D, available actions are G , H.
▸ Player 2 has four strategies:

▸ In this case, it’s simple enough to write them together. We can

refer to these strategies as EG , EH, FG , FH. The first action
corresponds to the first history C .
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategies in Extensive Form Games

▸ We can think of a strategy as an action plan or contingency

plan: If Player 1 chooses action X, do Y.
▸ However, a strategy must specify an action for all histories,
even if they do not occur due to previous choices in the
strategy.

▸ In this example, a strategy for Player 1 must specify an action

for the history (C , E ), even if it specifies D at the beginning.
▸ Think of this as allowing for the possibility of mistakes in
execution.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategy Profiles & Outcomes

▸ As before, a strategy profile is a list of the strategies of all

players.
▸ Given a strategy profile s, the terminal history that results by
executing the actions specified by s is denoted O(s), the
outcome of s.
▸ For example, in this game, the outcome of the strategy pair
(DG , E ) is the terminal history D.
▸ The outcome of (CH, E ) is the terminal history (C , E , H).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Nash Equilibrium

▸ Definition The strategy profile s ∗ in an extensive game with

perfect information is a Nash equilibrium if, for every player
i and strategy ri of player i, the outcome O(s ∗ ) is at least as
∗
good as the outcome O(ri , s−i ) generated by any other
∗
strategy profile (ri , s−i ) in which player i chooses ri :

ui (O(s ∗ )) ≥ ui (O(ri , s−i

∗
)) for every strategy ri of player i

▸ We can construct the strategic form of an extensive game by

listing all strategies of all players and finding the outcome.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategic Form of Entry Game

▸ The strategic form of the Entry Game is:

▸ There are two Nash equilibria: (In, Acquiesce) and (Out, Fight).
▸ The first NE is the same as the one found with backwards induction.
▸ In the second NE, the incumbent chooses Fight. However, if In is
taken as given, this is not rational. This is called an incredible
threat.
▸ If the incumbent could commit to Fight at the beginning of the
game, it would be credible.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Subgames

▸ The concept of Nash equilibrium ignores the sequential

structure of an extensive game.
▸ It treats strategies as choices made once and for all at the
beginning of the game.
▸ However, the equilibria of this method may contain incredible
threats.
▸ We’ll define a notion of equilibrium that excludes incredible
situations.
▸ Suppose Γ is an extensive form game with perfect information.
▸ The subgame following a non-terminal history h, Γ(h), is the
game beginning at the point just after h.
▸ A proper subgame is a subgame that is not Γ itself.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Subgames

▸ This game has two proper subgames:

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Subgame Perfect Equilibria

▸ A subgame perfect equilibrium is a strategy profile s ∗ in which

each subgame’s strategy profile is also a Nash equilibrium.
▸ Each player’s strategy must be optimal for all subgames that
have him moving at the beginning, not just the entire game.

▸ (Out, Fight) is a NE, but is not a subgame perfect equilibrium

because in the subgame following In, the strategy Fight is not
optimal for the incumbent.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Subgame Perfect Equilibria

▸ Every subgame perfect equilibrium is also a Nash equilibrium,

but not vice versa.
▸ A subgame perfect equilibrium induces a Nash equilibrium in
every subgame.
▸ In games with finite histories, subgame perfect equilibria are
consistent with backwards induction.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Backwards Induction in Finite-Horizon Games

▸ In a game with a finite horizon (i.e. finite maximum length of all

terminal histories), we can find all SPNE through backwards
induction.
▸ This procedure can be interpreted as reasoning about how players
will behave in future situations.
▸ Procedure:
▸ For all subgames of length 1 (i.e. 1 action away from a
terminal node), find the optimal actions of the players.
▸ Take these actions as given. For all subgames of length 2, find
the optimal actions of the players...
▸ Repeat until we cover the entire tree.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1

C D

2 2

E F G H

2,1 3,0 0,2 1,3

▸ There are 2 subgames with length 1.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1
1

C D

2 2

E F G H

2,1 3,0 0,2 1,3

▸ Consider the left subgame. It is Player 2’s turn to move.

▸ Player 2’s optimal action is E , resulting in payoff (2, 1).
▸ We will assume Player 2 always chooses E, so the payoff of this
subgame is (2, 1).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1

C D

2 2

2,1
G H

0,2 1,3

▸ Therefore, the payoff to Player 1 choosing C is (2, 1).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1
1

C D

2 2

2,1
G H

0,2 1,3

▸ Consider the right subgame. It is Player 2’s turn to move.

▸ Player 2’s optimal action is H, resulting in payoff (1, 3).
▸ We will assume Player 2 always chooses H, so the payoff of this
subgame is (1, 3).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1

C D

2 2

2,1 1,3

▸ Therefore, the payoff to Player 1 choosing D is (1, 3).

▸ Now, Player 1’s optimal action is C .
▸ Backwards induction gives the strategy pair (C , EH).
▸ The outcome of (C , EH) is the terminal history CE with payoff
(2, 1).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 170.1

C D

2 2

E F G H

2,1 3,0 0,2 1,3

▸ We mark the optimal actions at each node with thick lines.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategic Form of 170.1
1

C D

2 2

E F G H

2,1 3,0 0,2 1,3

EG EH FG FH
C 2,1 2,1 3,0 3,0
D 0,2 1,3 0,2 1,3

▸ Let’s compare the backwards induction result (C , EH)to the NE of

the strategic form.
▸ (C , EG ) and (C , EH) are NE of strategic form.
▸ However, (C , EG ) includes a non-optimal action for Player 2 in the
right subgame, so is not a subgame-perfect NE.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1
1

C D

2,0
E F

3,1
G H

1,2 0,0

▸ There is one subgame with length 1, and one subgame with length
2.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1
1

C D

2,0
E F

3,1
G H

1,2 0,0

▸ In this subgame, it is Player 1’s turn to move.

▸ Optimal action is G , resulting in payoff (1, 2).
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1

C D

2,0
E F

1,2 3,1

▸ Assume Player 1 chooses G with certainty.

▸ Then, the payoff to choosing E is (1, 2).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1

C D

2,0
E F

1,2 3,1

▸ In this subgame, it is Player 2’s turn to move.

▸ Optimal action is E , resulting in payoff (1, 2).

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1

C D

1,2 2,0

▸ Optimal action is D, resulting in payoff (2, 0).

▸ Backwards induction gives the strategy pair (DG , E ) resulting in
terminal history D.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 160.1
1

C D

2,0
E F

3,1
G H

1,2 0,0

▸ Backwards induction gives the strategy pair (DG , E ) resulting in

terminal history D.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Strategic Form of 160.1

1
E F
CG 1,2 3,1
C D CH 0,0 3,1
2
DG 2,0 2,0
DH 2,0 2,0
2,0
E F

3,1
G H

1,2 0,0

▸ NE of strategic form are: (CH, F ), (DG , E ), (DH, E ).

▸ Only (DG , E ) is a subgame perfect NE.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 158.1 (Variant of Entry Game)

Challenger

In Out

Incumbent

1,2
Acquiesce Fight

2,1 0,1

▸ What if there are multiple optimal actions in a subgame? Then we

need to keep track of them separately.
▸ This is a variant of the entry game in which the Incumbent is
indifferent between Acquiesce, Fight.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 158.1 (Variant of Entry Game)

Challenger

In Out

Incumbent

1,2
Acquiesce Fight

2,1 0,1

▸ In this subgame, both Acquiesce and Fight are optimal actions.

▸ We cannot eliminate either as an irrational choice. So, we keep
track of both possibilities.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Example 158.1 (Variant of Entry Game)

Challenger Challenger

In Out In Out

2,1 1,2 0,1 1,2

assuming Acquiesce is chosen assuming Fight is chosen

▸ Backwards induction gives (In, Acquiesce) and (Out, Fight).

▸ In this case, the NE of the strategic form are the same as the
subgame-perfect NE.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Continuous Action Sets

▸ The action set at a node may be infinite (e.g. if the player chooses
a real number).
▸ In this case, we graphically represent this with an arc between the
lowest and highest possible values.
▸ Effectively, there are an infinite number of branches in the game
tree at this node.
▸ Suppose it is Player i’s turn to move after all of these branches.
Then Player i’s strategy profile must specify an action for all
possible branches.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Continuous Action Sets

a b

▸ If the infinite set of actions is an interval of real numbers [a, b],

then Player i’s strategy profile for this node must be a function over
[a, b].
▸ For a strategy profile to be a subgame perfect NE, it must induce a
NE at each of the infinite subgames.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

▸ A union and a firm are bargaining.

▸ First, the union presents a wage demand w ≥ 0.
▸ The firm chooses an amount L ≥ 0 of labor to hire.
▸ The firm’s output is L(100 − L) when it uses L ≤ 50 units of labor,
and 2500 if L > 50.
▸ The price of output is 1.
▸ The firm’s preferences are represented by its profits.
▸ The union’s preferences are represented by the total wage bill, wL.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)
Union

0 infinity

Firm

0 infinity
L

▸ The firm’s payoff is its profit, given by:

⎧
⎪L(100 − L) − wL
⎪ if L ≤ 50
Π(w , L) = ⎨
⎪
⎪2500 − wL if L > 50
⎩
▸ Union’s payoff: wL

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

⎧
⎪L(100 − L) − wL if L ≤ 50
⎪
Π(w , L) = ⎨
⎪
⎪2500 − wL if L > 50
⎩

▸ For every w ≥ 0, there is a subgame where the firm’s payoff depends

on w .
▸ Profit has a quadratic part (if L ≤ 50) and a linear part (if L > 50),
and is continuous at L = 50.
▸ We want to find the profit-maximizing choice of L.
▸ The linear part is decreasing in L, so we can ignore it (its maximum
is at L = 50).
▸ Quadratic part is maximized at L∗ = 100−w
2
.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

▸ Quadratic part is maximized at L∗ = 100−w

2
.
▸ Firm’s profit is:

100 − w 100 − w 100 − w (w − 100)2

(100 − )−w =
2 2 2 4
▸ Profit is always non-negative. Firm’s best response correspondence
is:
⎧
⎪L ≥ 50 if w = 0
⎪
⎪
⎪
Bf (w ) = ⎨L = 100−w
2
if 0 < w ≤ 100
⎪
⎪
⎪
⎩L = 0
⎪ if w > 100

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

⎧
⎪L ≥ 50 if w = 0
⎪
⎪
⎪
Bf (w ) = ⎨L = 100−w
2
if 0 < w ≤ 100
⎪
⎪
⎪
⎩L = 0
⎪ if w > 100

▸ Now, consider the union’s decision.

▸ If w = 0 or w > 100, union’s payoff is 0.
w (100−w )
▸ wBf (w ) = 2
is maximized at w ∗ = 50.
▸ L∗ = Bf (50) = 25.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

▸ The set of subgame perfect NE is:

▸ Union’s strategy profile: at the empty history, choose w = 50.
▸ Firm’s strategy profile: at the subgame following the history w ,
choose an element of Bf (w ).
▸ Note that the firm has an infinite number of strategy profiles, but
there is only one equilibrium outcome, since only the subgame after
w = 50 will be realized.
▸ Firm’s payoff is 625 and union’s payoff is 1250.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

▸ Is there an outcome that both players prefer to the SPNE outcome

with payoffs (1250, 625)?
▸ Suppose that instead of each player maximizing his own payoff, a
social planner could choose both w and L.
▸ The sum of payoffs is L(100 − L) − wL + wL = L(100 − L) which is
maximized at L = 50. The choice of w then allocates payoffs to the
firm and union.
▸ For example, if w = 30, then the firm’s payoff is 1000 and the
union’s payoff is 1500.
▸ This is an illustration that individual maximization may not achieve
the most efficient outcome.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Exercise 177.1 (Firm-Union Bargaining)

▸ Is there a Nash equilibrium outcome that differs from any subgame

perfect NE outcome?
▸ Suppose the union’s strategy is: offer w = 100 and the firm’s
strategy profile is: for any w , offer L = 0.
▸ The firm has no incentive to deviate, since it will make a negative
payoff for any L > 0.
▸ The union has no incentive to deviate, because it will get a payoff of
0 for any choice of w .
▸ This is not subgame perfect, since the firm’s strategy is not optimal
for w < 100.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Characteristics of Finite Horizon Games

▸ Proposition: In a finite horizon extensive game with perfect

information, the set of strategy profiles isolated by backwards
induction is the set of all subgame-perfect equilibria.
▸ Proposition: Every finite extensive game with perfect
information has a subgame perfect equilibrium.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8
Announcements

▸ Please read Chapter 5.

▸ Also, HW #3 will be posted later today, due in 2 weeks.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 8