Reading
Osborne: Chapter 4
1 Mixed Strategies
3 Best Responses
4 Dominance
Matching pennies:
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
A Different Interpretation
Mixed Strategies
played.
Probability Distributions
Example
T.
A mixed strategy of player i , i , can specify that i chooses H
with probability
Thus, i (H) =
1
2
1
2
Example
i0 (H) =
1
3
and i0 (T ) = 32 .
and i0 = 31 , 23 ,
where the order of the probabilities are understood.
1 1
2, 2
Example
1 1
2, 2
1
2
1 2
3, 3
Expected Payoff
strategy profiles?
We can define players preferences over lotteries.
A (mixed) strategy profile induces a lottery over the action
profiles.
Expected Payoff
1 1
2, 2
, 13 , 32 , and if they
randomize over their actions independently (which is always
assumed to be the case in our course), then:
with
with
with
with
probability
probability
probability
probability
1
2
1
2
1
2
1
2
1
3
2
3
1
3
2
3
they
they
they
they
will
will
will
will
end
end
end
end
up
up
up
up
playing
playing
playing
playing
(H, H);
(H, T );
(T , H); and,
(T , T ).
Expected Payoff
1 1
2, 2
, 13 , 23
induces a lottery, which offers (the outcome induced by)
(H, H) with probability 16 , (H, T ) with probability 31 , (T , H)
with probability 61 , and (T , T ) with probability 13 .
events (H, H), (H, T ), etc., and calculate the expected value
of these payoffs based on the probabilities that these events
will occur.
Example
table:
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
1 1
2, 2
(1)
1
6
+1
1
3
+1
1
6
+ (1)
1
3
= 0.
1 2
3, 3
Example
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
1 1
1 2
,
2, 2 , 3, 3
1
1
2
6 + 1 3 = 3.
1
6
+2
1
3
= 65 .
player 1s
Expected Payoff
a1 A1 a2 A2
1 Mixed Strategies
3 Best Responses
4 Dominance
Ui i , i
Ui i , i
for all i Ai
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 2
3, 3
equilibrium?
No. Why?
Because by deviating to (0, 1), player 1 can strictly increase
U1 (0, 1) ,
1 2
3, 3
1
2 ,
1 2
3, 3
= 0, and
= (1)
1
3
+1
2
3
= 13 .
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 2
3, 3
equilibrium?
No. Why?
Because by deviating to (0, 1), player 1 can strictly increase
U1 (0, 1) ,
1 2
3, 3
1
2 ,
1 2
3, 3
= 0, and
= (1)
1
3
+1
2
3
= 13 .
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 2
3, 3
equilibrium?
No. Why?
Because by deviating to (0, 1), player 1 can strictly increase
U1 (0, 1) ,
1 2
3, 3
1
2 ,
1 2
3, 3
= 0, and
= (1)
1
3
+1
2
3
= 13 .
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 2
3, 3
equilibrium?
No. Why?
Because by deviating to (0, 1), player 1 can strictly increase
U1 (0, 1) ,
1 2
3, 3
1
2 ,
1 2
3, 3
= 0, and
= (1)
1
3
+1
2
3
= 13 .
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 1
2, 2
equilibrium?
Yes. Why?
Because, given that her opponent plays
1 1
2, 2
, a players
payoff is always 0 no matter what mixed strategy she
chooses.
Later on, we will show that this is in fact the unique mixed
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 1
2, 2
equilibrium?
Yes. Why?
Because, given that her opponent plays
1 1
2, 2
, a players
payoff is always 0 no matter what mixed strategy she
chooses.
Later on, we will show that this is in fact the unique mixed
Example
Lets return to Matching Pennies:
Head
Tail
Is (1 , 2 ) =
Head Tail
1,-1 -1,1
-1,1 1,-1
1 1
2, 2
1 1
2, 2
equilibrium?
Yes. Why?
Because, given that her opponent plays
1 1
2, 2
, a players
payoff is always 0 no matter what mixed strategy she
chooses.
Later on, we will show that this is in fact the unique mixed
Existence of NE
More Examples
More Examples
Up
2,2
0,0
Down 0,0
1,1
A mixed
Nash equilibrium is where both players play
strategy
i =
1 2
3, 3
to 23 .
Recall that (U, U) and (D, D) are also Nash equilibria. In
More Examples
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
A mixed
strategy Nash equilibrium
is where player 1 plays
2 1
3, 3
1 2
3, 3
. In equilibrium, both
1 Mixed Strategies
3 Best Responses
4 Dominance
Expected Payoff
A (prob. p)
a,b
c,d
B (prob. 1-p)
e,f
g,h
Expected Payoff
A (prob. p)
a,b (pq)
B (prob. 1-p) e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Example
Example
A (prob.
B (prob.
1
4)
3
4)
2,2
3,0
D (prob.
1
1
3 = 12
3
1
1
4 3 = 4
4
0,3
1,1
4
3
4
2
3
2
3
2
3)
= 16
= 12
Example
C (prob.
A (prob.
B (prob.
1
4)
3
4)
2,2
3,0
1
3)
D (prob.
1
1
3 = 12
3
1
1
4 3 = 4
4
0,3
1,1
1
U1 = 2 12
+ 0 16 + 3 14 + 1 12
1
U2 = 2 12
+ 3 16 + 0 14 + 1 12
4
3
4
2
3
2
3
2
3)
= 16
= 12
Head
Head
Tail
1
3
Tail
1,-1
-1,1
Is (1 , 2 ) =
2
3
-1,1
1,-1
1 1
2, 2
1 2
3, 3
equilibrium?
No. Because by deviating to (0, 1), player 1 can
strictly
1 2
3, 3
1 1
2, 2
is not a
1
3
Tail
2
3
Head (p)
1,-1
-1,1
Tail (1 p)
-1,1
1,-1
In fact, let p be the probability that player 1 chooses H. Her
expected payoff given her opponents mixed strategy is:
1
2
1
2
p + (1)
p + (1)
(1 p) + 1
(1 p)
3
3
3
3
1 2
p
3 3
1
=
Head
Tail
Head
1,-1
-1,1
Is (1 , 2 ) =
Tail
-1,1
1,-1
1 1
2, 2
1 1
2, 2
equilibrium?
Yes. Because, given that her opponent plays
1 1
2, 2
,a
players payoff is always 0 no matter what mixed strategy she
chooses.
1 1
2, 2
, any
Best Responses
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
Expected Payoff
Expected Payoff
Best Responses
Best Responses
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
blackboard.
(The graph is on P.112 of Osborne)
fashion.
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
Head
Tail
Head
1,-1
-1,1
Tail
-1,1
1,-1
U2 ((p, 1 p) , H) = (1) p + 1 (1 p) = 1 2p
U2 ((p, 1 p) , T ) = 1 p + (1) (1 p) = 2p 1
Thus, player 2s best response is H if p < 12 , T if p > 21 , and
function.
By graphing the best response functions of the two players,
we
canfind
theunique point where they intersect,
1 1
1 1
.
2, 2 , 2, 2
This is our Nash equilibrium.
BoS
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
BR1 = S
When q > 13 , U1 (B, (q, 1 q)) > U1 (S, (q, 1 q)),
BR1 = B
When q = 13 , any mixed strategy is a best response.
BoS
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
BR1 = S
When q > 13 , U1 (B, (q, 1 q)) > U1 (S, (q, 1 q)),
BR1 = B
When q = 13 , any mixed strategy is a best response.
BoS
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
2s best response?
U2 ((p, 1 p) , B) = p, U2 ((p, 1 p) , S) = 2 (1 p)
When p < 23 , U2 ((p, 1 p) , B) < U2 ((p, 1 p) , S),
BR2 = S
When p > 23 , U2 ((p, 1 p) , B) > U2 ((p, 1 p) , S),
BR2 = B
When p = 23 , any mixed strategy is a best response.
BoS
BoS:
Bach
Stravinsky
Bach
2,1
0,0
Stravinsky
0,0
1,2
2s best response?
U2 ((p, 1 p) , B) = p, U2 ((p, 1 p) , S) = 2 (1 p)
When p < 23 , U2 ((p, 1 p) , B) < U2 ((p, 1 p) , S),
BR2 = S
When p > 23 , U2 ((p, 1 p) , B) > U2 ((p, 1 p) , S),
BR2 = B
When p = 23 , any mixed strategy is a best response.
BoS
blackboard.
(The graph can be found on P.113 of Osborne)
functions
intersect:
((1, 0) , (1, 0)), ((0, 1) , (0, 1)), and
1 2
3, 3
2 1
3, 3
1 Mixed Strategies
3 Best Responses
4 Dominance
earlier:
The action ai is strictly dominated, if there exists ai0 , such
that:
ui ai0 , ai > ui (ai , ai ) for all ai j6=i Aj .
such that:
Ui (i , i ) > Ui (ai , i ) for all i j6=i Aj .
Example
Prisoners Dilemma:
Quiet Fink
Quiet
2,2
0,3
Fink
3,0
1,1
u1 (F , Q) = 3 > 2 = u1 (Q, Q);
u1 (F , F ) = 1 > 0 = u1 (Q, F ).
Thus, F strictly dominates Q for player 1 according to the
old definition.
However, can you find any mixed strategy 2 of player 2 such
Example
Prisoners Dilemma:
Quiet Fink
Quiet
2,2
0,3
Fink
3,0
1,1
u1 (F , Q) = 3 > 2 = u1 (Q, Q);
u1 (F , F ) = 1 > 0 = u1 (Q, F ).
Thus, F strictly dominates Q for player 1 according to the
old definition.
However, can you find any mixed strategy 2 of player 2 such
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
A (prob. p)
B (prob. 1-p)
C (prob. q)
a,b (pq)
e,f ((1 p) q)
D (prob. 1-q)
c,d (p (1 q))
g,h ((1 p) (1 q))
Expected Payoff
Strict Dominance
Example
L
3,
1,
0,
there
R
0,
2,
5,
any dominated strategies for player 1?
Example
L R
T 3, 0,
M 1, 2,
B 0, 5,
T is NOT strictly dominated by M or B.
M is NOT strictly dominated by T or B.
B is NOT strictly dominated by T or M.
Example
L
3,
1,
0,
M be
R
0,
2,
5,
played in any mixed strategy Nash equilibrium?
Example
M 1, 2,
B 0, 5,
Can M be played in any mixed strategy Nash equilibrium? No.
M is strictly dominated by the mixed strategy = 21 , 0, 12 .
Example
L R
3, 0,
1, 2,
0, 5,
strictly dominated by the mixed strategy =
1
1
2 , 0, 2
Example
1
1
2 , 0, 2
1 Mixed Strategies
3 Best Responses
4 Dominance
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
An Algorithm
player i :
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
First, find all pure strategy Nash equilibria (i.e., first look at
Example
L
R
T 2,2 2,3
M 0,3 3,2
B 3,1 0,2
Now, are there any MSNE where player 1 randomizes over T
and M and player 2 randomizes over L and R?
Let (p, 1 p) be the probabilities with which 1 chooses T
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
1 1
2, 3
Example
L
R
T 2,2 2,3
M 0,3 3,2
B 3,1 0,2
Now, are there any MSNE where player 1 randomizes over M
and B and player 2 randomizes over L and R?
Let (p, 1 p) be the probabilities with which 1 chooses M
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
1 1
2, 2
When q =
Example
L
R
T 2,2 2,3
M 0,3 3,2
B 3,1 0,2
Now, are there any MSNE where player 1 randomizes over T
and B and player 2 randomizes over L and R?
Let (p, 1 p) be the probabilities with which 1 chooses T
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
But there cannot be any p [0, 1] that can solve the first
Example
L
R
T 2,2 2,3
M 0,3 3,2
B 3,1 0,2
Finally, are there any MSNE where player 1 randomizes over
T , M, and B and player 2 randomizes over L and R?
Let (p, q, 1 p q) be the probabilities with which 1 chooses
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
Example
T
M
B
L
2,2
0,3
3,1
R
2,3
3,2
0,2
is
1 1
2, 2, 0
1 2
3, 3
Example
T
M
B
L
4,2
2,3
1,1
R
2,3
2,2
3,2