Econ 414 Lecture: Mixed Strategies: Jeff Borowitz

Theory and Definitions
Examles
More General Examples
Conclusion
Econ 414 Lecture: Mixed Strategies

Jeff Borowitz
Summer I 2010
Jeff Borowitz
1 / 54

Examles
Conclusion

Motivation
Definitions
John Nash and Mixed Strategies
Examles
Battle of the Sexes
Evolution Example
Matching Pennies
The Prisoners Dilemma

Mixed Strategies in a 3x3 Game
Cournot
Poker
Conclusion
Jeff Borowitz
2 / 54

Examles
Conclusion
Motivation
Definitions
Outline
1

Motivation
Definitions
Examles
Battle of the Sexes
Evolution Example
Matching Pennies
Cournot
Poker
Conclusion
Jeff Borowitz
3 / 54

Examles
Conclusion
Motivation
Definitions
Matching Pennies
What happens if theres no Nash Equilibrium in pure strategies?

Bob
Alice
H
T
H
1,-1
-1,1
T
-1,1
1,-1
Alice and Bob both choose heads or tails, then announce their
choices.
If they choose the same thing, Bob pays Alice $1
If they choose the opposite thing, Alice pays Bob $1
Jeff Borowitz
4 / 54

Examles
Conclusion
Motivation
Definitions
Matching Pennies
Looking at each players best response,

Bob
H
Alice
T
There is no Nash Equilibrium!
Jeff Borowitz
H
1,-1
-1,1
T
-1,1
1,-1
5 / 54

Examles
Conclusion
Motivation
Definitions
Mixed Strategy Nash Equilibrium
By randomly choosing multiple strategies, we can find the Nash

Equilibrium
This is not another equilibrium concept
It is just finding another type of solution using the same concept
Well learn a technique for finding mixed strategy Nash Equilibrium
Jeff Borowitz
6 / 54

Examles
Conclusion
Motivation
Definitions
Mixed Strategies
Definition (Mixed Strategy)

If player i has a strategy set S i = (s1i , s2i , . . . , sni ), A mixed strategy is a set
of probabilities i = (1i , 2i , . . . , ni ) such that the player plays pure
strategy sji with probability ji . Mixed strategies are written as:
MS i =
n
X
ji sji
j=1
.5H + .5T is a mixed strategy in the Matching Pennies game
Jeff Borowitz
7 / 54

Examles
Conclusion
Motivation
Definitions
Uncertainty and Rational Choice

How does a rational agent choose between actions with uncertain
payoffs?
They maximize expected utility
The expected value of a function u(x) when the event x that will
occur is uncertain and will occur with probabilty p(x) is given in the
discrete case by:
X
E [u(x)] =
u(x)p(x)
x
Formally, consider preferences over a lottery, which is different events

with different probabilities:
X
L=
p(xi )xi
i
Jeff Borowitz
8 / 54

Examles
Conclusion
Motivation
Definitions
Axioms of Expected Utility Maximization

The axioms are Completeness, Transitivity, Independence, and
Continuity
Completeness and Continuity are really just technical conditions
Definition (Transitivity)
For every lottery A, B, and C , if u(A) u(B) and u(B) u(C ), then it
must be true that u(A) u(C )
Definition (Independence)
Let A and B be two lotteries with u(A) u(B) and let C be any other
lottery. For p (0, 1],
u (pA + (1 p)C ) u (pB + (1 p)C )
Jeff Borowitz
9 / 54

Examles
Conclusion
Motivation
Definitions
Mixed Strategy Nash Equilibrium

Definition (Mixed Strategy Nash Equilibrium)
A mixed strategy = ( 1 , 2 , . . . N ) forms a Mixed Strategy Nash
Equilibrium of a game if for every player i ,
ui ( i , i ) ui ( i , i )
A pure strategy is just a special case of mixed strategies where one
probability is one and the others are all zero.
An important implication of mixed strategy Nash Equilibria is that the
expected utility received from playing all strategies must be equal
Jeff Borowitz
10 / 54

Examles
Conclusion
Motivation
Definitions
Mixed Strategy Nash Equilibrium in Perspective
How does this fit into the class?

It is the same solution concept as Nash Equilibrium
Mixed Strategy Nash Equilibria are another type of solution that Nash
Equilibrium specifies
Jeff Borowitz
11 / 54

Examles
Conclusion
Motivation
Definitions
Consider matching pennies where Alice plays H with probability p and

Bob plays H with probability q
Bob
H
T
H
1, 1
1, 1
Alice
T
1, 1
1, 1
We already checked the following for NE:
p
p
p
p
Mixed
= 1, q = 1
= 0, q = 1
= 1, q = 0
= 1, q = 1
Pure
(H, H)
(T , H)
(H, T )
(T , T )
But what about other values of (p, q)

Jeff Borowitz
12 / 54

Examles
Conclusion
Motivation
Definitions
Mixed Strategy vs. Pure Strategy NE

Best Responses Functions
1/2
BR (q)
A
BR (p)
B
0
0
1/2
A mixed strategy with = 0 is a pure strategy

Jeff Borowitz
13 / 54

Examles
Conclusion
Motivation
Definitions
John Nash and Nash Equilibrium
What if there is no mixed strategy NE either?

Nash showed that for games with a finite number of players that can
each play a finite number of different strategies, this is this is
impossible
A Nash Equilbrium always exists in finite games!
More on why this should be true later
Jeff Borowitz
14 / 54

Examles
Conclusion
Motivation
Definitions
John Nash
A video...
Jeff Borowitz
15 / 54

Examles
Conclusion
Motivation
Definitions
The game in the movie is like the Prisoners Dilemma:

How the game is modeled (with no offense to brunettes!):
It is better for an individual to go for the blonde than a brunette
If both go for the blonde, they get no girls
If both go for brunettes, they can both get girls
Assume blonde gives utilty 10, brunette 8, and no girl 0:
Nash
Blonde
Brunette
Friend
Blonde Brunette
0, 0
10, 8
8, 10
8, 8
What was the part about Adam Smith in the video?

The invisible hand would suggest that everyone doing what is best
for them gives the best result for everyone
This isnt true in the game: both choosing the blonde is worse for
everyone
Jeff Borowitz
16 / 54
Nash Equilibrium and Hard Work

Examles
Conclusion
Motivation
Definitions
Nashs Work
Consider the games:
Bob
H
T
H
1, 1
1, 1
Alice
T
1, 1
1, 1
There are two cases:
Bob
C
Alice
RS
C
3, 3
4, 0
RS
0, 4
1, 1
At least 1 player has a dominant strategy: (like in the PD), there is a

Nash Equilibrium at that players dominant strategy
No players have a dominant strategy: (like matching pennies) there is a
critical value where a players best response switches
Let p be the probability that Alice plays H and q be the probability

that Bob plays H
Jeff Borowitz
18 / 54
Best Responses
Best Responses Functions
1/2
BRA(q)
BR (p)
B
0
0
1/2
These functions are continuous (you can choose any probability)

Since they are continuous and cross the domain, they must cross, and
the equilibrium must exist

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Outline
1

Motivation
Definitions
Examles
Battle of the Sexes
Evolution Example
Matching Pennies
Cournot
Poker
Conclusion
Jeff Borowitz
20 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Battle of the Sexes - Mixed Strategy NE
Remember the Battle of the Sexes game:

Bob
F
Alice
O
F
1, 3
0, 0
O
0, 0
3, 1
Alice and Bob randomize over strategies, playing F with probability p and
q respectively.
Jeff Borowitz
21 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Alice maximizes:
E [uA ] =p E [uA (F )] + (1 p) E [uA (O)]
where E [uA (F )] =q uA (F , F ) + (1 q) uA (F , O)
=q (1) + (1 q) (0) = q
E [uA (O)] =q uA (O, F ) + (1 q) uA (O, O)
=q (0) + (1 q) (3) = 3 (1 q)
Jeff Borowitz
22 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies

Note that E [uA ()] doesnt depend on p so Alice cant affect it. Therefore,
Alice chooses p to solve:
max E [uA ] = max p E [uA (F )] + (1 p) E [uA (O)]
p
E [uA ]
=0 = E [uA (F )] E [uA (O)]
p
E [uA (F )] =E [uA (O)]
So Alice must get the same expected utility from playing F and O. The
same general condition applies to Bob.
Jeff Borowitz
23 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies

Solving the maximizing conditions for each player, Alice has:
E [uA (F )] =E [uA (O)]
q =3(1 q)
q =3/4
and Bob has:
E [uB (F )] =E [uB (O)]
3 p + 0 (1 p) =0 p + 1 (1 p)
3p =(1 p)
p =1/4
The mixed strategy Nash Equilibrium is therefore:

F + 3O 3F + O
,
4
4
Jeff Borowitz
24 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
It was a lot of work to solve the Battle of the Sexes with mixed
strategies.
Luckily, theres is a shortcut that we stumbled on that always works
If there is a mixed strategy NE, agents must get the same expected
utility from playing each strategy when they play it
Otherwise, it wouldnt be an equilibrium
The player could play whichever strategy was giving higher expected
utility a little more, which would increase total expected utility
The shortcut:
Calculate the expected utility that each player gets from playing each
strategy and compare them.
If one is always larger than the other, then the player will only play that
strategy in equilibrium (like playing C in the Prisoners Dilemma)
If the two utilities can be made equal, then they will be equal in the
mixed strategy NE
Jeff Borowitz
25 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Fighting over Territory

Consider a population of mountain goats that are territorial
Two goats will meet and each will choose whether to fight or back
down.
They are fighting over a territory worth 8
If they both back down or both fight, they split the territory
If one backs down, the fighter gets the territory
Fighting costs each goat 5.
The matrix is given by:
Goat 1 F
B
Jeff Borowitz
Goat
F
-1,-1
0,8
2
B
8,0
5,5
26 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Fight over Territory, Continued

Underlining, the Nash Equilibria are (F , B) and (B, F )
Goat 2
F
B
F
1,1
8, 0
Goat 1
B
0, 8
5,5
But in our context, this makes no sense.
The context is you pick a strategy and then meet a random goat from
the population and use your strategy
If there was an equilibrium at F , F , we would evolve to a population of
all fighters
Here, in a population of all fighters, a single backer-down would have
an advantage
Same thing in a population of all backers-down
Here, the mixed strategy Nash Equilibrium is like what fraction of the
population would be fighters so that neither fighters nor backers down
have an advantage
Jeff Borowitz
27 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Solution
Let Goat 1 play F with probability p and Goat 2 play F with
probability q.
Ignore that p and q should have to be equal if the goats who fight
come from the same population (a slightly different story could have
two roles of goats with different probabilities)
Goat 1 gets the same utility from playing F and B, otherwise the
Goat 1 population would evolve to get higher utility.
E [u1 (F )] =E [u1 (B)]
q u1 (F , F ) + (1 q) u1 (F , B) =q u1 (B, F ) + (1 q) u1 (B, B)
q (1) + (1 q) (8) =q (0) + (1 q) (5)
3 3q =q
q = 3/4
Jeff Borowitz
28 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Solution, Continued
Goat 1
F
B
Goat
F
-1,-1
0,8
2
B
8,0
5,5
E [u2 (F )] =E [u2 (B)]

p u2 (F , F ) + (1 p) u2 (B, F ) =p u2 (F , B) + (1 p) u2 (B, B)
p (1) + (1 p) (8) =p (0) + (1 p) (5)
3 3p =p
p = 3/4
Jeff Borowitz
29 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Solution, Continued
So the Nash Equilibrium here is:

3F + B 3F + B
,
4
4
This means that 3/4 of the population will be fighters, and 1/4 will
back down.
Neither fighters nor back-downers have an eveolutionary advantage,
so this is stable
Jeff Borowitz
30 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Solution: Implications
This shows that Nash Equilibrium is a versatile concept

I showed you a definition of the equilibrium, and how to solve for
equilibria
Whether or when particular equilibria are of interest depends on the
context
Jeff Borowitz
31 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Matching Pennies Revisited

Remember the matching pennies game?
Bob
H
T
H
1,-1
-1,1
Alice
T
-1,1
1,-1
Let Alice and Bob play H with probabilities p and q
Alices condition is:
E [uA (H)] =E [uA (T )]
(1) q + (1) (1 q) =(1) (1 q) + (1) (q)
2q 1 =1 2q
q =1/2
So in equilibrium, Bob plays each strategy with probability q = 1/2
Jeff Borowitz
32 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Mixed Strategies in Matching Pennies

Now lets find Bobs condition:
E [uB (H)] =E [uB (T )]
(1) p + (1) (1 p) =(1) (1 p) + (1) (p)
p =1/2
The equilibrium is therefore:

H +T H +T
,
2
2
Note: Bobs condition gives a requirement on Alices probability, and
vice versa
Why does this make sense?
Jeff Borowitz
33 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Best Responses in Matching Pennies

In order for Bob to agree to play a mixed strategy, Alice must be
playing exactly p = 1/2 - otherwise, he would play T if p > 1/2 an H
if p < 1/2
A similar argument holds for Alice
Best response functions are given by:
q > 1/2
1,
p [0, 1], q = 1/2
pBR = BRA (q) =
0,
q < 1/2
qBR
p > 1/2
0,
= BRB (p) =
q [0, 1], p = 1/2
1,
p < 1/2
Jeff Borowitz
34 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
Matching Pennies Best Responses

1
BR (q)
A
BR (p)
1/2
0
0
1/2
p
Jeff Borowitz
35 / 54
Question
Does every game have a mixed strategy Nash Equilibrium?

1
Yes
No
Question

1
Yes
No
Yes: Every game has at least one Nash Equilibrium. That Nash
Equilibrium might or might not be in pure strategies, but since a pure
strategy is a generalization of a mixed strategy, this is true

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
The Prisoners Dilemma with Mixed Strategies

Bob
C
RS
C
3, 3
0, 4
Alice
RS
4, 0
1, 1
Consider the situation where Alice plays C with probability p and Bob
plays C with probability q
Jeff Borowitz
37 / 54

Examles
Conclusion
Battle of the Sexes

Evolution Example
Matching Pennies
The Prisoners Dilemma with Mixed Strategies

Alices utility is given by:
E [uA (p)] =p ((3)q + (0)(1 q)) + (1 p) ((4)q + (1)(1 q))
=p 3q 1
Note that E [uA (p)] is increasing in p, so Alice will chose the highest
p possible, which is 1.
Alice chooses p = 1, and by a similar argument Bob chooses q = 1.
The mixed strategy Nash Equilibrium is
(1 C + 0 R, 1 C + 0 R) = (C , C )
Jeff Borowitz
38 / 54

Examles
Conclusion

Cournot
Poker
Outline
1

Motivation
Definitions
Examles
Battle of the Sexes
Evolution Example
Matching Pennies
Cournot
Poker
Conclusion
Jeff Borowitz
39 / 54

Examles
Conclusion

Cournot
Poker
A 3x3 Game
Consider this game:
T
Alice M
B
Note that there are no Nash
Bob
L
M
R
7, 0 6, 3 7, 0
9, 2 7, 6
0, 8
6, 9 7, 1
2, 8
Equilibria in pure strategies:
T
Alice M
B
Jeff Borowitz
L
7, 0
9, 2
6, 9
Bob
M
6, 3
7, 6
7, 1
R
7, 0
0, 8
2, 8
40 / 54

Examles
Conclusion

Cournot
Poker
What is a mixed strategy here?

Each player can play each strategy with some probability
Since there are 3 strategies for each player, we now need 2
probabilities to describe a strategy
Let Alice play T with probability pT , M with probability pM , and B
with probability 1 pT pM .
Bob similarly plays L with probability qL , M with probability qM , and R
with probability 1 qL qM .
Jeff Borowitz
41 / 54

Examles
Conclusion

Cournot
Poker
Mixed Strategies
Using the conditions for a mixed strategy:
E [uA (T )] = E [uA (M)] = E [uA (B)]
E [uB (L)] = E [uB (M)] = E [uB (R)]
Note that this gives 4 conditions in our 4 unknowns (pT , pM , qL , qM )
Calculating E [uA (T )], E [uA (M)], E [uA (B)] and we find:
E [uA (T )] =7qL + 6qM + 7(1 qL qM ) = 7 qM
E [uA (M)] =9qL + 7qM + 0(1 qL qM ) = 9qL + 7qM
E [uA (B)] =6qL + 7qM + 2(1 qL qM ) = 2 + 4qL + 5qM
Jeff Borowitz
42 / 54

Examles
Conclusion

Cournot
Poker
3x3 Calculations Continued

Calculating E [uB (L)], E [uB (M)], E [uB (R)]:
E [uB (L)] =0pT + 2pM + 9(1 pT pM ) = 9 9pT 7pM
E [uB (M)] =3pT + 6pM + 1(1 pT pM ) = 1 + 2pT + 5pM
E [uB (B)] =0pT + 8pM + 8(1 pT pM ) = 8 8pT
Therefore, to solve for Bobs equilibrium mixed strategies (using
Alices conditions), solve:

7 qM = 9qL + 7qM
7 qM = 2 + 4qL + 5qM
2 equations in 2 unknowns are solvable: qL = 1/11, qM = 17/22
Jeff Borowitz
43 / 54

Examles
Conclusion

Cournot
Poker
3x3 Calculations Continued

Repeating the same calculation for Alices strategies (using Bobs
conditions):

9 9pT 7pM = 1 + 2pT + 5pM
9 9pT 7pM = 8 8pT
We find: pT = 44/65 and pM = 3/65
Therefore, the mixed strategy Nash Equilibrium here is:

44T + 3M + 18B 2L + 17M + 3R
,
65
22
You wont have to do this algebra you should be able to set up this
problem
Jeff Borowitz
44 / 54

Examles
Conclusion

Cournot
Poker
Mixed Strategies in Cournot
Can you have mixed strategies in a game where you choose a

continuous strategy like Cournot? Yes!
A mixed strategy has a probability of playing every infinitesimal value
of q
Obviously, this is very often impractical to solve (you wont have to
do so)
Jeff Borowitz
45 / 54

Examles
Conclusion

Cournot
Poker
Mixed Strategies in Cournot

Probability of Playing q
0.25
0.2
0.15
0.1
0.05
q
Jeff Borowitz
46 / 54

Examles
Conclusion

Cournot
Poker
Mixed Strategies and Poker

Poker (Texas Hold Em in particular) can be modeled as a game
Its much more complex than what weve looked at so far, but the
intuition of mixed strategies holds
If you have a really good hand, should you bet more or not?
Pro: youre likely to win, so betting makes you more money
Con: if your opponent knows you have a good hand, she will fold
What about when you have a pretty good hand?

Pro: if your opponent knows you bet with pretty good hands, she will
play more when you have really good hands
Con: you might lose more than you put in on average
Bottom line: the optimal poker strategy involves mixed strategies
Jeff Borowitz
47 / 54

Examles
Conclusion
Outline
1

Motivation
Definitions
Examles
Battle of the Sexes
Evolution Example
Matching Pennies
Cournot
Poker
Conclusion
Jeff Borowitz
48 / 54

Examles
Conclusion
Equilibrium Concepts in General
Require assumptions
Will select some outcomes of a game as solutions
Have a tradeoff between assumptions and solutions that it chooses
Jeff Borowitz
49 / 54

Examles
Conclusion
Review of Nash Equilibrium as an Equilibrium Concept

Assumptions about what the players know
Both players know the game
Both players prefer higher payoffs
Both players know the last two points (common knowledge)
Both players have expectations that the other player will play their
Nash Equilibrium strategy
Which solutions does it select? Those where:

Each player is playing a best response to the other players strategy
Those where no player can profitably deviate
Tradeoffs between assumptions and predictions

Big new benefit of mixed strategies: existence in finite games!
Jeff Borowitz
50 / 54

Examles
Conclusion
Mixed Strategy Nash Equilbria
This is not a different equilibrium concept than pure strategy NE

We just allowed a more complete set of strategies
In pure strategies, p = 0 or p = 1
Now, 0 p 1
Allowing more strategies gives more equilibria

Setting expected utilities equal is our main technique
Jeff Borowitz
51 / 54

Examles
Conclusion
Techniques for Finding Nash Equilibrium

Big picture: have each player choose a strategy that is a best
response to the other players strategies in Equilibrium
Specific techniques:
Underlining: used in box games, underline each players best responses,
and when they match up there is a NE
Calculating Best Response Functions: used in games with continuous
strategies (like Cournot), you use calculus to set derivative of profit
function equal to zero for all players at once
Thinking: used in games with special structure (like Bertrand): think
through what patterns a best response has, and see what can be NE
Setting expected utilities of each strategy equal: used to find
mixed strategy Nash Equilibria
Jeff Borowitz
52 / 54

Examles
Conclusion
Where Are We?

Done:
Dominant Strategy Equilibrium
Weak DSE
DSE
Nash Equilibrium
Pure Strategy NE
Mixed Strategy NE
Still to Come:
Subgame Perfect Nash Equilibrium
Bayesian Nash Equilibrium
Perfect Bayesian Nash Equilibrium
Jeff Borowitz
53 / 54
The End

Econ 414 Lecture: Mixed Strategies: Jeff Borowitz

Uploaded by

Copyright:

Available Formats

Econ 414 Lecture: Mixed Strategies: Jeff Borowitz

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Econ 414 Lecture: Mixed Strategies: Jeff Borowitz

Uploaded by

Copyright:

Available Formats

Theory and Definitions