Introduction to

Game Theory
PROF. SWAGATA BHATTACHARJEE
Today
 Rationalizability and Iterated Dominance
 Best Response
Dominated Strategies
 A pure strategy of player i is (strictly) dominated if there is a strategy (pure or mixed) such that

for all strategy profiles of the other players.

 A strategy (pure or mixed) of player i (strictly) dominates a pure strategy if for all strategy profiles
of the other players

 Players never play dominated strategies.

Example
 Suppose you are player 1. Do you have a dominated
strategy?
Example
 Suppose you are player 1. Do you have a dominated
strategy? No.
 Does player 2 have a dominated strategy?
Example
 Suppose you are player 1. Do you have a dominated
strategy? No.

 Player 2, however, has a dominated strategy: X

 So, there is no belief that player 2 could have about
your strategy that would cause her to play X.

 So, if you are player 1, rational, and you know that

player 2 is also rational, you would know that
player 2 will never play X.

 That is, if you are rational, you should assign zero

probability to her strategy X (your belief
should have pX = 0)
Example
 You are player 1.

 If you are rational, and you know that player 2 is

also rational, you would know that player 2 will
never play X.

 Given that, strategy A is strictly dominated for you

in this reduced game!

 So, playing A can not be rationalizable for player 1.

Example
 You are player 1.

 If you are rational, and you know that player 2 is

also rational, and player 2 also knows that you
know that player 2 is rational, then player 2 knows
that you will never play A.

 Given that, strategy Y is strictly dominated for

player 2 in this further reduced game!

 So, playing Y can not be rationalizable for player 2.

 Which means, the outcome of the game is (B,Z),

and the payoff is (1,2).

 (B,Z) is the only Rationalizable Strategy!

Iterated Deletion of Strictly Dominated
Strategies
Iterated Deletion/ Removal of
Strictly Dominated Strategies
(Iterated Dominance)
 We can apply the procedure to any normal-form game as follows.

 Delete all of the dominated strategies for each player, because a rational player would never use one of these.

 Then define R1 to be the strategy profiles that remain.

 Common knowledge of rationality therefore implies that it is common knowledge that a strategy profile in R1 will
be played.

 In other words, essentially the players are interacting in a “reduced” game in which R1 is the set of strategy
profiles.


Iterated Deletion/ Removal of
Strictly Dominated Strategies
(Iterated Dominance)
 We can apply the procedure to any normal-form game as follows.

 Now, essentially the players are interacting in a “reduced” game in which R1 is the set of strategy profiles.

 The next step is to remove any strategies that are dominated in this reduced game and define R2 as the strategy
profiles that remain.

 Continue this process to identify smaller and smaller sets of strategy profiles, R3 ,R4, …, until no more strategies
can be deleted.

 Let R denote the resulting set of strategy profiles—those that survive this iterated dominance procedure.

 The set of strategies R that survive iterated dominance is therefore called the rationalizable strategies.
Example
 Any pure strategy that is dominated another
pure strategy?
Example
 Any pure strategy that is dominated another
pure strategy?
 No!
 Any dominated strategy at all?

Example
 Any pure strategy that is dominated another
pure strategy?
 No!
 Any dominated strategy at all?
 Yes!
 Player 2: X is dominated by the mixed
strategy (0, ½, ½ )
 Iterated deletion step 1:
R1 = {{U, M, D} ˟ {Y, Z}}.
Example
 Within R1, any dominated strategy?
 Yes! U is dominated by D for Player 1.
 Iterated deletion step 2:
R2 = {{M, D} ˟ {Y, Z}}.

Example
 R2 = {{M, D} ˟ {Y, Z}}.
 Within R2, for 2, Y is dominated.
 Step 3: R3 = {{M, D} ˟ {Z}}.

 Step 4: for Player 1 D is dominated. So, R3

=R = {(M,Z)}.

 Notice, it is a single strategy profile, so we

can predict the outcome of the game in
rationalizable strategies.
Rationalizability and
Common Knowledge of Rationality
 The players have to hold the common knowledge of rationality: “ you and I are both rational, I
know that we are rational, I know that you know that we are rational, I know that you know that I
know that we are rational, I know that you know that I know that you know that we are rational,
…..”
 It is an essential assumption to use Rationalizability as a solution concept.
Example: Guessing the Average
Guessing the Average
Rationalizable Strategies in GA
Common Knowledge of Rationality in
GA
Common Knowledge of Rationality in
GA
Example: Tragedy of Roommates
Example: Roommates
Example: Roommates
Example: Roommates
Example: Roommates
Beliefs and Best Response
 The concept of dominance and Rationalizability is free from belief!

 “Whatever the other players do, a strategy for player i is dominated by another”

 Often, no strategy is dominated. Then, players have to form belief about what the others are going to
do and respond to that: Best response.

 To maximize the payoff that you expect to obtain—which we assume is the mark of rational behavior
—you should select the strategy that yields the greatest expected payoff against your belief.

 Such a strategy is called a best response (or best reply).

Beliefs and Best Response
 To maximize the payoff that you expect to obtain, you should select the strategy that yields the greatest
expected payoff against your belief.

 Such a strategy is called a best response (or best reply).

 Definition:
Suppose player i has a belief about the strategies played by the other players. Player i’s strategy is a best
response if

for every
Best Response
 Definition:
Suppose player i has a belief about the strategies played by the other players. Player i’s strategy is a best
response if

for every

 Best response is always chosen based on a certain belief.

 If a strategy is best response to all possible beliefs, then it is the dominant strategy!

 There can be more than one best responses. Set of best responses for a given belief is denoted by
Best Response: Example
 Suppose Player 1 believes that 2 will use the
strategy (1/3, ½, 1/6)=

 Then, expected payoffs for player 1

if chooses U:

If chooses M:

If chooses D:

 So,
Best Response: Example
 Suppose Player 2 believes that 1 will use the
strategy (½, ¼, ¼ )=

 Then, expected payoffs for player 2

if chooses L:

If chooses C:

If chooses R:

 So,
Never Best Response and Dominance
 Definition:
Suppose player i has a belief about the strategies played by the other players. Player i’s strategy is a best
response if

for every

 A strategy is a Never Best Response if it is not a best response to any belief by player i.

 Result: A strategy in a finite game is a Never Best Response if and only if it is a Strictly Dominated
strategy.
 Example
 Suppose that two people decide to form a partnership firm.
 The revenue of the firm depends on the amount of effort expended on the job by each person and is given by:

 = effort level of i th player;s are known positive constants

 The contract that was signed by the partners stipulates that person 1 receives a fraction t (between 0 and 1) of the
firm’s revenue, and person 2 receives a 1 - t fraction.

 Each person dislikes effort, measured by a personal cost of

 Person i ’s utility in this endeavor is the amount of revenue that this person receives, minus the effort cost.
Example
 Effort levels (assumed nonnegative) are chosen by the people simultaneously and independently.
(a) Define the normal form of this game (by describing the strategy spaces and payoff functions).

Payoff functions:
Example
 (b) Using dominance, compute the strategies that the players rationally select (as a function of t, a1 and a2).
 In this game, each player has a strategy that dominates all others. To see
this, observe how player 1’s payoff changes as is varied.

 As increases, changes as:

 is maximum when FOC is satisfied:

 So,
 Similarly,
Example
 Suppose that you could set t before the players interact. How would you set to maximize the
revenue of the firm?
 From (b), we know, for every t, ,
So, revenue =
 This is linear in t, so corner solution!
 either t=1 or t=0
 If so t=1 otherwise t=0
 What if the firm wants to maximize revenue minus the effort costs by both the partners?
Next Class
 Nash Equilibrium