28/10/2013

1
Lecture 5
Subgame Perfect Nash Equilibrium
and Backward Induction
Example: Mini Ultimatum Game
Proposer (Player 1) can suggest one of two splits of 10:
(5,5) and (9,1).
Responder (Player 2) can decide whether to accept or
reject (9,1), but has to accept (5,5). Reject leads to 0 for
both
Player 1
(9,1)
a r
9
1
0
0
(5,5)
5
5
Player 2
2
28/10/2013
2
Mini Ultimatum Game in Strategic Form
Player 2
accept (9,1) reject (9,1)
Player 1 propose (5,5) 5,5 5,5
propose (9,1) 9,1 0,0
There are two equilibria:
1. (propose (9,1), accept (9,1))
2. (propose (5,5), reject (9,1)).
Equilibrium 2 is in weakly dominated
strategies (reject (9,1) is weakly dominated)
3
Note that equilibrium 2 (propose (5,5), reject (9,1)) is not
convincing because it relies on a non-credible threat: if
the 1 proposes (9,1) player 2 has an incentive to deviate
(i.e. to accept).
Formally, the decision taken in the subgame starting at
player 2s decision node is not optimal
4
Player 1
(9,1)
a r
9
1
0
0
(5,5)
5
5
Player 2
28/10/2013
3
equilibrium 1 (propose (9,1), accept) is more convincing
because the decision taken in the subgame starting at
player 2s decision node is optimal
We say that equilibrium 1 is subgame perfect because in
every subgame the decisions are optimal
5
Player 1
(9,1)
a r
9
1
0
0
(5,5)
5
5
Player 2
Definition:
Subgame perfect Nash equilibrium
A Nash equilibrium is subgame perfect (Nash
equilibrium) if the players strategies constitute a
Nash equilibrium in every subgame.
(Selten 1965)
Note that every finite sequential game of complete
information has at least one subgame perfect
Nash equilibrium
We can find all subgame perfect NE using backward
induction
6
28/10/2013
4
Backward Induction in dynamic games of
perfect information
7
Procedure:
We start at the end of the trees
first find the optimal actions of the last player to
move
then taking these actions as given, find the optimal
actions of the second last player to move
continue working backwards
If in each decision node there is only one optimal
action, this procedure leads to a Unique Subgame
Perfect Nash equilibrium
Player 1 choose action a
1
from the set A
1
={Left, Right}
Player 2 observes a
1
and choose an action a
2
from the
set A
2
={l, r}
Payoffs are u
1
(a
1
, a
2
) and u
2
(a
1
, a
2
)
8
Player 1
Left Right
Player 2 Player 2
l l r r
U
1
(L,l)
U
2
(L,l)
U
1
(L,r)
U
2
(L,r)
U
1
(R,l)
U
2
(R,l)
U
1
(R,r)
U
2
(R,r)
28/10/2013
5
When Player 2 gets the move, she observes players 1
action a
1
and faces the following problem
Solving this problem, for each possible a
1
A
1
, we get
the best response of Player 2 to Player 1s action.
We denote it by R
2
(a
1
) , the reaction function of Player 2.
Player 1 can anticipate Player 2s reaction, then Player
1s problem is:
9
) a , (a u Max
2 1 2
A a
2
} {
2

)) (a R , (a u Max
1 2 1 1
A a
1
} {
1

The backwards induction outcome is denoted by

It is different from the description of the equilibrium
To describe the Nash equilibrium we need to describe
the equilibrium strategies:
Action of Player 1 (Left or Right)
Action of Player 2 after Left, action of player 2 after Right
10
)) (a R , (a
*
1 2
*
1
28/10/2013
6
Consider Player 2, the optimal action is accept
Taking accept as given, we see that (9,1) is the optimal
action for player 1
Player 1
(9,1)
a r
9
1
0
0
(5,5)
5
5
Player 2
11
Example: Mini Ultimatum Game
The Ultimatum Game
Proposer (Player 1) suggest (integer) split of a fixed pie, say 10.
Responder (Player 2) accepts the proposal or rejects (both receive 0)
There is no unique best response following (10,0), so we have two SPNE
Player 1
Player 2
(10,0)
a r
10
0
0
0
(9,1)
a r
9
1
0
0
a r
(5,5)

5
5
0
0
(0,10)

r a
0
10
0
0
12
28/10/2013
7
First SPNE:
Player 1 proposes (10, 0)
Player 2 accepts in all of his decision nodes (a, a, a, a, a, a, a, a, a, a, a)
Player 1
Player 2
(10,0)
a r
10
0
0
0
(9,1)
a r
9
1
0
0
a r
(5,5)

5
5
0
0
(0,10)

r a
0
10
0
0
13
The Ultimatum Game
Player 1
Player 2
(10,0)
a r
10
0
0
0
(9,1)
a r
9
1
0
0
a r
(5,5)

5
5
0
0
(0,10)

r a
0
10
0
0
14
The Ultimatum Game
Second SPNE:
Player 1 proposes (9, 1)
Player 2 rejects after (10, 0) and accepts in all other decision nodes
(r, a, a, a, a, a, a, a, a, a, a)
28/10/2013
8
Challenger
Ready
Unready
Incumbent
Fight
3
3
1
1
4
3
0
2
Acquiesce Acquiesce Fight
15
Three Nash equilibria
1. Ready, (A, F)
2. Unready, (A, A)
3. Unready, (F, A)
Only {Unready, (A, A)} is subgame perfect
Incumbent
(A, A) (A, F) (F, A) (F, F)
Challenger
Ready 3, 3 3, 3 1, 1 1, 1
Unready 4, 3 0, 2 4.3 0, 2
Backward Induction in dynamic games of
imperfect information
We start at the end of the trees
first find the Nash equilibrium (NE) of the last
subgame
then taking this NE as given, find the NE in the
second last subgame
continue working backwards
If in each subgame there is only one NE, this procedure
leads to a Unique Subgame Perfect Nash equilibrium
28/10/2013
9
Example: two stage game of imperfect information
Stage 1: Players 1 and 2 move simultaneously taking,
respectively, actions a
1
A
1
and a
2
A
2
Stage 2: Players 3 and 4 observe (a
1
, a
2
) , then move
simultaneously taking, respectively, actions a
3
A
3
and
a
4
A
4
Payoffs: u
i
(a
1
, a
2
, a
3
, a
4
) for i = 1, 2, 3, 4
Solution:
We solve the simultaneous - move game between
players 3 and 4 in the second stage:
Players 1 and 2 anticipate the behaviour of players 3
and 4
17
18
Player 1
Player 2
Player 3
Player 4
T B
T B
T B T B T B
L R L R
L R
L R
L R
L R
L R
L R L R
L R
2
2
1
0
1
1
2
1
3
3
0
0
4
4
0
0
3
3
1
0
0
2
1
1
4
4
0
0
2
2
0
1
3
3
1
2
2
2
1
0
0
0
0
1
0
0
0
0
2
2
0
1
2
2
0
0
2
0
1
1
2
2
1
0
example
28/10/2013
10
19
G3 Player 4
L R
Player 3
T 3,3,1,2 2,2,1, 0
B 0,0,0,1 0,0,0, 0
G1 Player 4
L R
Player 3
T 2, 2,1,0 1,1 , 2, 1
B 3,3, 0,0 4,4, 0, 0
G4 Player 4
L R
Player 3
T 2,2,0,1 2,2,0,0
B 2,0,1,1 2,2,1, 0
G2 Player 4
L R
Player 3
T 3,3,1,0 0,2,1, 1
B 4,4,0,0 2,2,0, 1
Player 2
L R
Player 1
T G1 G2
B G3 G4
20
G3 Player 4
L R
Player 3
T 3,3,1,2 2,2,1, 0
B 0,0,0,1 0,0,0, 0
G1 Player 4
L R
Player 3
T 2, 2,1,0 1,1 , 2, 1
B 3,3, 0,0 4,4, 0, 0
G4 Player 4
L R
Player 3
T 2,2,0,1 2,2,0,0
B 2,0,1,1 2,2,1, 0
G2 Player 4
L R
Player 3
T 3,3,1,0 0,2,1, 1
B 4,4,0,0 2,2,0, 1
Player 2
L R
Player 1
T 1,1 , 2, 1 0,2,1, 1
B 3,3,1,2 2,0,1,1
28/10/2013
11
Backward induction outcome:
(B, L, T, L)
Subgame perfect Nash equilibrium
(B, L, (T, T, T, B), (R, R, L, L))
21
Challenger
Ready Unready
Incumbent
Fight
3
3
1
1
4
3
0
2
Out
2
4
Acquiesce Acquiesce Fight
22
Challenger
In
Example 2
Challengers strategies: {out, (In ready), (In Unready)}
Incumbent strategies: Acquiesce, Fight
28/10/2013
12
23
Incumbent
Acquiesce Fight
Out Ready 2, 4 2, 4
Challenger
Out Unready 2, 4 2, 4
In Ready 3, 3 1, 1
In Unready 4, 3 0, 2
Three Nash equilibria:
(Out Ready, Fight);
(Out Unready, Fight)
(In unready, Acquiesce)
24
Consider the subgame starting in the decision node
after Challengers choice In
An unique Nash equilibrium: Unready, Acquiesce
Then, only (In unready, Acquiesce) is subgame
perfect NE
Incumbent
Acquiesce Fight
Challenger
Ready 3, 3 1, 1
Unready 4, 3 0, 2
28/10/2013
13
Example 3
Challenger
Ready Unready
Incumbent
Fight
3
3
1
1
4
3
0
2
Out
2
4
Acquiesce Acquiesce Fight
25
Incumbent
Acquiesce Fight
Challenger Ready 3, 3 1,1
Unready 4, 3 0, 2
Out 2, 4 2, 4
Two Nash equilibria:
(Out, Fight)
(Unready, Acquiesce)
Both are SPNE