GAME THEORY

Sonya Javadi

İTÜ 2020
 2

Introduction
What is Game Theory?!
Study of rational behavior in situations in which your choices
affect others and their choices affect you, (we call them
‘’games’’).

Games We Play:
• Driving
• Group projects
• Dating
• Price wars
• Standard wars
 3

Introduction

Decisions:
You take the world as given and make the best decision for
yourself.
Examples: selecting a field at university or choosing the city to
live, etc.

Games:
Your best decision depends on what others do, and what they do
may depend on what they think you do …
Examples: Pricing of the batch pie for a pie manufacturer and
hunting either stag or hare of two hunters, etc.
 4

Introduction
Optimization theory versus Game theory:
In Optimization Theory the purpose is to optimize a single
objective over a decision variable while in Game Theory the
purpose is the study of multi-person decision problems.

• Large-scale networks.
• Online advertising on the Internet.
• Information evolution and belief propagation in social networks.
 5

Introduction
Game Theory over Time:

• The first studies of games in the economics literature were the papers
by Cournot (1838), Bertrand (1883), Edgeworth (1925) on oligopoly
and pricing and production. But these were seen as special models
that did little to change the economics world.

• The idea of general theory of games was introduced by John Von

Neumann and Oskar Morgenstern in their famous book 1944’ Theory
of Games and Economic Behavior’ which proposed that most
economic questions.
 6

Introduction

John von Neumann(1903-1957) Oskar Morgenstern(1902-1977) John Nash(1928-2015)

 7

Introduction

Key Elements of a Game

Players: who is interacting?

Strategies: what are their options?
Payoffs: what are their incentives?
Information: what do they do know?
Rationality: how do they think?
 8

Introduction
How to Anticipate others’Behavior?

Evolution: If non-strategic and adaptive, play repeatedly (or

observe past play).

Dominance: If never play a strategy that is always worse than

another.

Rationalizability: If play optimal given some beliefs about what

others play (and what others believe).

Equilibrium: If play optimal given correct beliefs about others.

 9

Introduction
Mechanism Design (MD):
Design of a game (or incentives) to achieve an objective (eg.
system-wide goal or designer’s selfish objective).

In Economics, MD is all about designing the right incentives.

In Engineering, focus is more on the design of efficient
decentralized protocols that take into account incentives.
 10

Game Theory (Example)

• Consider the decision problem of a pie manufacturer -let’s call it
Piemax- that must choose between a high and a low price for
today’s batch of pies.

• She should think about what the prices of other pies.

• Piemax knows that the other firms choose their own prices on
the basis of their own predications of the market.

• The game-theoretic way for Piemax to use this knowledge is to

build a model of the behavior of each individual competitor, and
perhaps to look for behavior that forms an equilibrium of the
model.
 11

Game Theory (Example)

What kind of model Piemax should use?!

1. The simplest case is one where Piemax and all its competitors
operate for only one day, where all the firms know the
demand for pies, and where each firms knows the production
technologies of the others.

This kind of situation can be studied with the tools of strategic-

form games and Nash equilibrium.
 12

Game Theory (Example)

2. If the industry will continue for a longer days, then Piemax will
want to consider some other objectives besides maximizing
today’s net profit. For example, low prices today may induce
consumers to buy more pies of Piemax. Or producing a large
batch of pies may help the production staff be more
experienced.

This kind of situation can be studied as extensive-form model and

be solved by using the subgame perfection concept.
 13

Game Theory (Example)

3. If Piemax is uncertain of the cost functions or the long-run
objectives of its competitors.

This kind of situation is analyzed in incomplete information in a

static context.
 14

Overview of the Course

• John von Neumann and Oskar Morgenstern introduced the

ideas of the extensive-form and normal-form (or strategic-form)
representations of a game, defined the minmax solution, and
showed that this solution exists in all two-player zero-sum
games.

(In a zero-sum game, the interests of the players are directly opposed,
with no common interests at all).
 15

Overview of the Course

• Nash (1950) proposed what come to be known as ‘’Nash

equilibrium’’ as a way of extending game-theoritic analyses to
non-zero-sum games.

• It requires that each player’s strategy be a pay off-maximizing

response to the strategies that he forecasts that his opponents
will use, and further that each player’s forecast be correct.
 16

Overview of the Course

• Selten (1965) and Harsany (1967-68) introduced concepts that
have been widely used in recent years.

• Selten argued that in games where the players choose

contingent plans not all the Nash equilbiria are equally
reasonable, because some of them may rely on the ability of
players to make ‘empty threats’ that is, contingent plans that
would not in fact be optimal to carry out.
 17

Overview of the Course

• For example, Piemax used the contingent plans ‘’If you do not let
me have ¾ of the market today, I will give away free pies for the
next ten years.”

• Selten introduced his concept of subgame perfection to rule out

the equilibria that rely on these kinds of threats.
 18

Example

Cigarette Advertising on TV(1964-1970)

• All US tobacco companies advertised heavily on television.

• Surgeon General issues official warning.

• Cigarette smoking may be hazardous

• Cigarette companies’ reaction.

• Fear of potential liability lawsuits
 19

Example

• Companies strike agreement.

• Carry the warning label and cease TV advertising in
exchange for immunity from federal lawsuits.
 20

Example
• Players: Reynolds, Philip Morris
• Strategies: Advertise, Do Not Advertise
• Payoffs: Companies’ Profits

Some Structure:
Each firm earns $50 million from its customers.
Advertising costs each firm $20 million.
Advertising captures $30 million from competitor.
 21

Example

How Can We Represent This Game?!!

 22

Example

Game Matrix
PLAYERS
Philip Morris

NO AD AD
NO AD 50 , 50 20 , 60

Reynolds AD 60 , 20 30 , 30

STARTEGIES PAYOFFS
 23

Example
Best Response for Reynolds:
Philip Morris

NO AD AD
Reynolds NO AD
AD
 24

Example

Best Response for Reynolds:

• If Philip Morris does not advertise advertise.
• If Philip Morris advertises advertise.

Advertise is a dominant strategy!

This is a Prisoners’ Dilemma!
 25

Example

What happened?

• After the 1970 agreement, cigarette advertising decreased by

$63 million.

• Profits rose by $91 million!

• How were the firms able to escape from the Prisoners’

Dilemma?
 26

Example
Game- Changer: Gov’t-Enforced Collusion?

Philip Morris
NO AD AD
Reynolds NO AD 50 , 50 20 , x
AD x , 20 Y ,Y

• Government made advertising illegal .

• If penalty large enough X < 50 and Y < 20.
• The dominant strategy is now No Ad!
• All payoffs go down profits go up!
 27

Strategic-Form Games
We begin with a simple game:

If a group of hunters set out to take a stag, they are fully aware
that they would all have to remain faithfully at their posts in order
to succeed; but if a hare happens to pass near one of them, there
can be no doubt that he pursued it without qualm, and that once
he had caught his prey, he cared very little whether or not he had
made his companions miss theirs.
 28

Strategic-Form Games
To make this into a game, we assume that there are only two
hunters, and they must decide simultaneously whether to hunt
for stag or for hare.

If both hunt for hare, they each will catch one hare. If one hunts
for hare while the other tries to take a stag, the former will catch
a hare and the latter will catch nothing. Each hunter prefers half a
stag to one hare.
 29

Strategic-Form Games
• The hunters are the players.
• Each player has two choices or strategies: hunt stag and hunt
hare.
• The payoff is the prey.
• A stag is worth 4 and a hare is worth 1.
• When both players hunt stag each has a payoff of 2.
• A player who hunts hare has payoff 1, and a player who hunts
stag by himself has payoff 0.
 30

Strategic-Form Games

What predication should one make about the outcome of the

game?!

• Cooperation-both hunting stag- is an equilibrium, or more precisely a

“Nah equilibrium”, in that neither player has a unilateral incentive to
change his strategy.

• If each player believes the other will hunt hare, each is better off
hunting hare himself.

• The noncooperative outcome-both hunting hare- is also a Nash

equilibrium.
 31

Strategic-Form Games
• They are two ways of describing games: the strategic (or
normal) form and the extensive form.

• We may think the players’ strategies as corresponding to various

‘buttons’ on a computer keyboard. The players are asked in a
separate rooms to choose a button.

• We assume that all players know the structure of the strategic

form, and know that their opponents know it, and know that
their components know that they know, and so on ad infinitum.
 32

Strategic-Form Games
• Games in which all of the participants act simultaneously and without
knowledge of other players’ actions are referred to as strategic form
games—or as normal form games or matrix games.

• More generally, we also have to define the game form, which

captures the order of play (e.g., in chess) and information sets (e.g., in
asymmetric information or incomplete information situations). But in
strategic form games, play is simultaneous, so no need for this
additional information.

• A fact is common knowledge if all players know it, and know that they
all know it, and so on.
 33

Decision-Making under Uncertainty

• Von Neumann and Morgenstern posited a number of “reasonable”
axioms that rational decision-making under uncertainty should satisfy.
From these, they derived the expected utility theory.

• Under uncertainty, every choice induces a lottery, that is, a probability

distribution over different outcomes.

• E.g., one choice would be whether to accept a gamble which pays $10
with probability 1/2 and makes you lose $10 with probability 1/2.

• von Neumann and Morgenstern’s expected utility theory shows that

(under their axioms) there exists a utility function (also referred to as
Bernoulli utility function) 𝑢 (𝑐), which gives the utility of (outcome) 𝑐.
 34

Decision-Making under Uncertainty

• Then the utility of this choice is given by the expected utility according
to the probability distribution 𝐹𝑎 (𝑐):
𝑈 (𝑎) = ‫)𝑐( 𝑎 𝐹𝑑 𝑐 𝑢 ׬‬.

• In other words, this is the expectation of the utility 𝑢 (𝑐), evaluated

according to the probability distribution 𝐹𝑎 (𝑐).

• More simply, if 𝐹𝑎 (𝑐) is a continuous distribution with density 𝑓 𝑎 (𝑐),

then

𝑈 (𝑎) = න 𝑢 𝑐 𝑓 𝑎 𝑐 𝑑 𝑐 ,

or if it is a discrete distribution where outcome 𝑐𝑖 has probability 𝑝𝑖𝑎

then 𝑈 𝑎 = σ𝑖 𝑝𝑖𝑎 𝑢 𝑐𝑖 .
 35

Decision-Making under Uncertainty

• Given expected utility theory and our postulate of “rationality,” single
person decision problems are (at least conceptually) simple.

• If there are two actions, 𝑎 and 𝑏, inducing probability distributions

𝐹𝑎 (𝑐) and 𝐹𝑏 (𝑐), then the individual chooses 𝑎 over 𝑏 only if
𝑈 (𝑎) = ‫)𝑐( 𝑏 𝐹𝑑 𝑐 𝑢 ׬ = )𝑏( 𝑈 ≥ )𝑐( 𝑎 𝐹𝑑 𝑐 𝑢 ׬‬.
 36

Strategic-Form Games

Definition: Strategic Form Game

A strategic forms game is a triplet

such that ( I , ( Si )iI , (ui )iI )

• 𝐼 is a finite set of players, i.e., 𝐼 = 1,2,3, … , 𝐼 ;

• ( Si ) is the set of available actions for player 𝑖;
• 𝑠𝑖 ∈ 𝑆𝑖 is an action for player 𝑖 ;
• 𝑢𝑖 : 𝑆 → 𝑅 is the payoff (utility) function of player 𝑖 where 𝑆 = ς𝑖 𝑆𝑖 is the
set of all action profiles.
 37

Strategic-Form Games
In addition, we use the notations:

• 𝑠−𝑖 = 𝑠𝑗 : vector of actions for all players except 𝑖 .

𝑗≠𝑖
• 𝑆−𝑖 = ς𝑗≠𝑖 𝑆 is the set of all action profiles for all players except 𝑖.
𝑗
• 𝑠𝑖 , 𝑠−𝑖 ∈ 𝑆 is a strategy profile, or outcome.
 38

Strategic-Form Games
• In game theory, a strategy is a complete description of how to
play the game.

• It requires full contingent planning. If instead of playing the

game yourself, you had to delegate the play to a “computer”
with no initiative, then you would have to spell out a full
description of how the game would be played in every
contingency.
 39

Strategic-Form Games
• For example, in chess, this would be an impossible task (though
in some simpler games, it can be done).

• Thinking in terms of strategies is important.

• But in strategic form games, there is no difference between an

action and a pure strategy, and we will use them
interchangeably.
 40

Strategic-Form Games
• When the 𝑆𝑖 is finite for all 𝑖 , we call the game a finite game.
• For 2 players and small number of actions, a game can be represented
in matrix form.
• Recall that the cell indexed by row 𝑥 and column 𝑦 contains a pair
𝑎, 𝑏 , where 𝑎 = 𝑢1 (𝑥, 𝑦 ) and 𝑏 = 𝑢2 (𝑥, 𝑦 ).

• A two-palayer zero-sum game is a game such that σ2𝑖=1 𝑢𝑖 𝑠 = 0 for

all 𝑠.

• The key feature of these games is that the sum of the utilities is a
constant; setting the constant to equal zero is a normlization.
 41

Strategic-Form Games
• A mixed strategy is a probability distribution over pure
strategies.

• Each player’s randomization is statistically independent of those

of his opponents, and the payoffs to a profile of mixed strategies
are the expected values of the corresponding pure-strategy
payoffs.

• The space of player 𝑖’s mixed strategies is denoted by σ, where

𝜎𝑖 (𝑠𝑖 ) is the probability that 𝜎𝑖 assigns to 𝑠𝑖 .
 42

Strategic-Form Games
• The support of a mixed strategy 𝜎𝑖 is the set of pure strategies to
which 𝜎𝑖 assigns positive probability.
• Player 𝑖’s payoff to profile is
𝐼

𝑢𝑖 𝜎 = ෍ ෑ 𝜎𝑗 𝑠𝑗 𝑢𝑖 𝑠 .
𝑠∈𝑆 𝑗=1
 43

Strategic-Form Games
Example:
What are the payoffs to profits 𝜎1 and 𝜎2 for players 1, 2?!!

L M R
U 4,3 5,1 6,2
M 2,1 8,4 3,6
D
3,0 9,6 2,8
 44

Dominated Strategies
• In pervious example, no matter how player 1 plays, R gives
player 2 a strictly higher payoff then M does.

• Strategy M is strictly dominated. So, a rational player 2 should

not play M.

• If player 1 knows that play 2 will not play M, than U is a better

choice than M or D.
 45

Dominated Strategies
• Finally, if player 2 knows that player 1 knows that player 2 will
not play M, then player 2 knows that player 1 will play U, and so
player 2 should play L.

• The process of elimination described above is called iterated

dominance, or, iterated strict dominance.
• The order is not the case.
 46

Dominated Strategies
Definition: Pure strategy 𝑠𝑖 is strictly dominated for player 𝑖 if
there exists 𝜎𝑖 ′ ∈ σ𝑖 such that

𝑢𝑖 (𝜎𝑖 ′ , 𝑠−𝑖 ) > 𝑢𝑖 (𝑠𝑖 , 𝑠−𝑖 ) 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑠−𝑖 ∈ 𝑆−𝑖 .

The strategy 𝑠𝑖 is weakly dominated if there exists a 𝜎𝑖 ′ such that

above inequality holds with weak inequality.
 47

Dominated Strategies
• A mixed strategy that assigns positive probability to a
dominated pure strategy is dominated.

• A mixed strategy may be strictly dominated even though it

assigns positive probability only to pure strategies that are not
even weakly dominated.
 48

Dominated Strategies
Example:
L R
U 1,3 -2,0
M -2,0 1,3
D 0,1 0,1

Playing U with probability ½ and M with ½ gives expected payoff -

1/2 regardless of player 2’s play and so is strictly dominated by
playing D, even though neither U or M is dominated.
 49

Dominated Strategies
Example:

L R L R
U U - 1,3 2,1
1,3 4,1
D D 0,2 3,4
0,2 3,4

The dominated strategy is U, and iterated strict dominance

predicts that the solution is (U,L).

Could it help player 1 to change the game and reduce his payoffs
if U occurs by 2 units, which would result in the game shown in
next figure?
 50

Dominated Strategies

Decision theory says that such a change would not help, but in a
game theory may have beneficial effects on the play of
opponents.!!
 51

Application of the Elimination of Dominated

Strategies

A single round of elimination of dominated strategies reduces the

strategy set of each player to a single pure strategy.
 52

Application of the Elimination of Dominated

Strategies
Example 1: Prisoner's Dilemma

The story behind the game:

• Two people are arrested for a crime.
• The police lack sufficient evidence to convict either suspect and
need them to give testimony against each other.
• They are put in a different cell.
• If he testifies against the other, he will be released. (reward)
• If neither suspect testifies, both will be released. (no reward)
 53

Application of the Elimination of Dominated

Strategies
• If one testifies, the other will go to prison.
• If both testify, both will go to prison.
• They simultaneously choose the actions.
• If both cooperate (C) (do not testify), they get 1 each.
• If they both play noncoopetaively (D, for defect), they obtain 0.
• If one cooperates and the other does not, the latter is rewarded
(gets 2) and the former is punished (gets -1).

C D
1,1 -1,2
C
D 2,-1 0,0
 54

Prisoner's Dilemma
What will the outcome of this game be?

• Regardless of what the other player does, playing “cooperate” is

better for each player.
• The action is “cooperate” strictly dominated by the action
“Don’t confess”.
• Prisoner’s dilemma paradigmatic example of a self-interested,
rational behavior not leading to jointly (socially) optimal result.
 55

Nash Equilibrium
A Nash equilibrium is a profile of strategies such that each
player’s strategy is an optimal response to the other player's
strategies.

Definition: A mixed-strategy profile 𝜎 ∗ is a Nash equilibrium if, for all

players 𝑖,
𝑢𝑖 (𝜎𝑖 ∗ , 𝜎−𝑖 ∗ ) ≥ 𝑢𝑖 (𝑠𝑖 , 𝜎−𝑖 ∗ ) 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑠𝑖 ∈ 𝑆𝑖 .

A pure-strategy Nash equilibrium is a pure-strategy that satisfies

the same condition.
 56

Nash Equilibrium

• A Nash equilibrium is strict if each player has an unique best

response to his rivals’ strategies.

• That is, 𝑠 ∗ is a strict equilibrium if and only if it is a Nash

equilibrium and, for all 𝑖 and 𝑠𝑖 ≠ 𝑠𝑖 ∗ ,

𝑢𝑖 (𝑠𝑖 ∗ , 𝑠−𝑖 ∗ ) ≥ 𝑢𝑖 (𝑠𝑖 , 𝑠−𝑖 ∗ ).

 57

Nash Equilibrium

Why is this a “reasonable” notion?

No player can profitably deviate given the strategies of the other

players. Thus in Nash equilibrium, “best response
correspondences intersect”.
 58

Nash Equilibrium

• This has a “steady state” type flavor. In fact, two ways of

justifying Nash equilibrium rely on this flavor:

• Introspection: what I do must be consistent with what you will

do given your beliefs about me, which should be consistent with
my beliefs about you,...Steady state of a learning or evolutionary
process.

• An alternative justification: Nash equilibrium is self-reinforcing.

• If player 1 is told about player 2’s strategy, in a Nash equilibrium

she would have no incentive to change her strategy.
 59

Example
Majority Voting
There are three players, 1,2,3.
There are three alternatives, A, B, and C.
Players vote simultaneously for an alternatives.
The strategy spaces are 𝑆𝑖 = {𝐴, 𝐵. 𝐶}
The alternative with the most votes wins, if no, then alternative A is
selected.
The payoff functions are
𝑢1 𝐴 = 𝑢2 𝐵 = 𝑢3 𝐶 = 2,
𝑢1 𝐵 = 𝑢2 𝐶 = 𝑢3 𝐴 = 1,
𝑢1 𝐶 = 𝑢2 𝐴 = 𝑢3 𝑏 = 0.
This game has three pure-strategy equilibrium outcomes; A, B, and C.
 60

Nonexistence of a Pure-Strategy Equilibrium

Not all games have pure-strategy Nash equilbria.

Example: Matching Pennies.

• Players 1 and 2 simultaneously announce heads (H) or tails (T).
• If the announcements match, then player 1 gains a unit and player 2
loses a unit.
• If the announcements differ, it is player 2 who wins the unit and
player 1 who loses.
• If the predicted outcome is that the announcements will match, then
plaeyr 2 has an incentive to deviate, while player 1 would prefer to
deviate from any predication in which accouchements do not match.
 61

Nonexistence of a Pure-Strategy Equilibrium

Example: Matching Pennies.

The only stable situation is one in which each player randomizes

between his two pure strategies, assigning equal probability to each.

H T
1, -1 -1,1
H
T -1, 1 1,-1
 62

Focal Points
• What do we do when there are multiple Nash equilibria?
• Our models would not be making a unique prediction.

• Two different lines of attack:

• Think of set valued predictions—i.e., certain outcomes are possible,
and Nash equilibrium rules out a lot of other outcomes.
• Think of equilibrium selection.
 63

Focal Points
• Equilibrium selection is hard.
• Most important idea, Schelling’s focal point.
• Some equilibria are more natural and will be expected.
• Schelling’s example: ask the people to meet in New York,
without specifying the place. Most people will go to Grand
Central. Meeting at Grand Central, as opposed to meeting at
any one of thousands of similar places, is a “focal point”.
 64

Focal Points
Example: Battle of the Sexes (players wish to coordinate but have
conflicting interests).
B F
B 1,4 0,0

F 0,0 4,1

Two Nash equilibria, (Ballet, Ballet) and (Soccer, Soccer).

 65

Multiple Equilibria
• Another example of multiple equilbria is the stag-hunt game,
where each player has to choose whether to hunt hare by
himself or to join a group that hunt stag.
• Suppose that there are 𝐼 players, that choosing hare gives payoff
1 regardless of the other players’ actions, and that choosing stag
gives payoff 2 if all players choose stage and gives payoff 0
otherwise.

• This game has two pure-strategy equilbria: ‘all stag’ and ‘all
hare’.
 66

References
• Game Theory, by D. Fudenberg and J. Tirole, MIT Press, 1991.
• Game Theory with Engineering Applications, by Asu. Ozdaglar.
MIT Open Courses.