Professional Documents
Culture Documents
Spring 2021
PROF. SWAGATA BHATTACHARJEE
Today
Extensive to Normal Form Conversion
Mixed Strategy and Expected Payoffs
Solving a game: Dominance Solvability
Best Response
Example: Sequential Legislators’ Vote
Three legislators vote whether they allow themselves a salary raise of $2000 per year. Since voters are
observing the vote, a legislator would estimate the loss of face by having to vote for a raise as $1000 per
year.
If 2 out of 3 legislators vote for hike, it’s accepted, otherwise there is no hike.
A has to vote first, then B, then C, and all votes are open.
Rewrite the payoffs as: if there is a hike and you voted for hike, payoff is 2, if there is a hike but you did
not vote for it, payoff is 3. If there is no hike and you voted for hike, payoff is 0, if there is no hike and
you did not vote for hike, payoff is 1. (This is called normalization of payoffs)
Example: Sequential Legislators’ Vote
Example: Sequential Legislators’ Vote
I={A,B,C}
𝑆𝐴 = 𝑅, 𝑁𝑅
𝑆𝐵 = 𝑅, 𝑁𝑅 ?
Example: Sequential Legislators’ Vote
I={A, B, C}
𝑆𝐴 = 𝑅, 𝑁𝑅
𝑆𝐵 = 𝑅, 𝑁𝑅 WRONG!
From B’s point of view, 𝑁2 and 𝑁1 are very
different, so taking action R at 𝑁1 is NOT the
same as taking action R at 𝑁2 !
Remember , strategy is a complete
contingent action plan. Strategy is NOT
synonymous to Action!
𝑆𝐵 =
𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑅 𝑎𝑛𝑑 𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑁𝑅,
𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑅 𝑎𝑛𝑑 𝑁𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑁𝑅,
𝑁𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑅 𝑎𝑛𝑑 𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑁𝑅
𝑁𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑅 𝑎𝑛𝑑 𝑁𝑅 𝑖𝑓 𝐴 𝑐ℎ𝑜𝑠𝑒 𝑁𝑅
Example: Sequential Legislators’ Vote
I={A, B, C}
𝑆𝐴 = 𝑅, 𝑁𝑅
Remember , strategy is a complete contingent
action plan. Strategy is NOT synonymous to
Action!
𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑅 𝑎𝑡 𝑁2 ,
𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑁𝑅 𝑎𝑡 𝑁2,
𝑆𝐵 = 𝑁𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑅 𝑎𝑡 𝑁2 ,
𝑁𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑁𝑅 𝑎𝑡 𝑁2
Example: Sequential Legislators’ Vote
I={A,B, C}
𝑆𝐴 = 𝑅, 𝑁𝑅
𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑅 𝑎𝑡 𝑁2 ,
𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑁𝑅 𝑎𝑡 𝑁2,
𝑆𝐵 = 𝑁𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑅 𝑎𝑡 𝑁2 ,
𝑁𝑅 𝑎𝑡 𝑁1 𝑎𝑛𝑑 𝑁𝑅 𝑎𝑡 𝑁2
Recall: 𝑆−𝑖 =
𝑠𝑒𝑡 𝑜𝑓 𝑠𝑡𝑟𝑎𝑡𝑒𝑔𝑖𝑒𝑠 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑙𝑎𝑦𝑒𝑟𝑠 𝑒𝑥𝑐𝑒𝑝𝑡 𝑖
Then, ∆𝑆−𝑖 =
𝑠𝑒𝑡 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛𝑠 𝑜𝑣𝑒𝑟 𝑆−𝑖
=set of beliefs
Beliefs
𝑆−𝑖 = 𝑠𝑒𝑡 𝑜𝑓 𝑠𝑡𝑟𝑎𝑡𝑒𝑔𝑖𝑒𝑠 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑙𝑎𝑦𝑒𝑟𝑠 𝑒𝑥𝑐𝑒𝑝𝑡 𝑖
Then, ∆𝑆−𝑖 = 𝑠𝑒𝑡 𝑜𝑓 𝑎𝑙𝑙 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛𝑠 𝑜𝑣𝑒𝑟 𝑆−𝑖 =set of beliefs
Properties:
1 ≥ 𝜃−𝑖 ≥ 0
𝜃−𝑖 (𝑠−𝑖 ) = 1
𝑠−𝑖 ∈𝑆−𝑖
Beliefs: Example
Suppose player 1 and 2 simultaneously play this following game
If a player has a belief: 𝜃−𝑖 ∈ ∆𝑆−𝑖 , for every strategy 𝑠𝑖 that she chooses, she can not be certain of the
payoff that she will get.
Suppose 1’s Belief is 𝜃−1 ∈ ∆𝑆−1 given by: (0.6, 0.2, 0.1, 0.1)
Then, expected payoff for 1 for any pure strategy is given by:
Suppose 1’s Belief is 𝜃−1 ∈ ∆𝑆−1 given by: (0.6, 0.2, 0.1, 0.1)
Then, expected payoff for 1 for any pure strategy is given by:
Note that all the while 1 actually does not know if 2 will behave according to this belief!
Mixed Strategy
Mixed strategy: when a player selects strategies according to a probability distribution
Adam might think: I will Confess half the times and Deny otherwise (that is, choose C with probability
=1/2 )
1 1
Above example: 𝜎𝐴 = (2 , 2) over the set 𝑆𝐴 = {𝐶, 𝐷}
A regular strategy is called “Pure strategy”, it is a mixed strategy where probability of choosing the
particular action is 1 or 0.
So, if Adam always chooses to Confess, it is a “pure strategy”, or a mixed strategy 𝜎𝐴 = (1,0)
Including mixed strategies expand the set of all possible strategies available to a player.
Mixed Strategy: Example
1/2 Black White Yellow Green
Blue 2,1 1,1 3,2 5,8
Purple 9,1 1,3 1,1 1,0
Now, suppose 1 wants to play Bl with 0.6 probability and P with 0.4 probability. So,
𝜎1 = (0.6, 0.4) ∈ ∆𝑆1
Then, the expected payoff for 1 from this mixed strategy 𝜎1 is:
𝑢 𝜎1 , 𝜃−1 = 2.2 0.6 + 5.8 0.4 = 3.64
Mixed Strategy: PD Example
Adam\Bob Deny Confess
Deny -1,-1 -3,0
Confess 0,-3 -2,-2
Now, suppose B wants to play C with 0.5 probability and D with 0.5 probability. So,
𝜎𝐵 = (0.5, 0.5) ∈ ∆𝑆𝐵
Also assume B believes that A will always Confess. So, 𝜃−𝐵 = (1,0)
Then, the expected payoff for B from this mixed strategy 𝜎𝐵 is:
𝐸𝑢 𝜎𝐵 , 𝜃−𝐵 = −3 0.5 + −2 0.5 = −2.5
Do It Yourself
Wilson Ch 4: Guided Exercise, Q 1, 3.
Solving a Game: PD Example
Adam\Bob Deny Confess
Deny -1,-1 -3,0
Confess 0,-3 -2,-2
You are Adam. Suppose you know Bob is going to Deny. What should you do?
Solving a Game: PD Example
Adam\Bob Deny Confess
Deny -1,-1 -3,0
Confess 0,-3 -2,-2
You are Adam. Suppose you know Bob is going to Deny. What should you do?
Choose C, because 0>-1
What if you knew that Bob is going to Confess. What should you do then?
Solving a Game : PD Example
Adam\Bob Deny Confess
Deny -1,-1 -3,0
Confess 0,-3 -2,-2
You are Adam. Suppose you know Bob is going to Deny. What should you do?
Choose C, because 0>-1
What if you knew that Bob is going to Confess. What should you do then?
Choose C, because -2>-3
So, if you are Adam, no matter what Bob does, you should always choose C!
Check for Bob, same is true!
C is dominant strategy for both A and B! D is dominated by C.
Technically, we say that strategy D is dominated by strategy C, and thus D should never be played by a
rational player A.
The Curious Case of Prisoners’ Dilemma
Adam\Bob Deny Confess
Deny -1,-1 -3,0
Confess 0,-3 -2,-2
A pure strategy 𝑠𝑖 ∈ 𝑆𝑖 of player i is (strictly) undominated if there exists no pure or mixed strategy
𝜎𝑖 such that
𝑢 𝜎𝑖 , 𝑠−𝑖 > 𝑢(𝑠𝑖 , 𝑠−𝑖 )
for all strategy profiles 𝑠−𝑖 ∈ 𝑆−𝑖 of the other players.
𝑈𝐷𝑖 = {𝑠𝑖 |∄𝜎𝑖 ∈ ∆𝑆𝑖 𝑠𝑢𝑐ℎ 𝑡ℎ𝑎𝑡 𝑢 𝜎𝑖 , 𝑠−𝑖 > 𝑢 𝑠𝑖 , 𝑠−𝑖 ∀𝑠−𝑖 ∈ 𝑆−𝑖 }
A strategy (pure or mixed) 𝜎𝑖 ∈ ∆𝑆𝑖 of player i (strictly) dominates a pure strategy 𝑠𝑖 if for all strategy
profiles 𝑠−𝑖 ∈ 𝑆−𝑖 of the other players
𝑢 𝜎𝑖 , 𝑠−𝑖 > 𝑢(𝑠𝑖 , 𝑠−𝑖 )
We can not always rule out “weakly dominated” strategies, unlike the strictly dominated ones.
Next Class
Iterated Dominance and Rationalizability
Best Reponse