GAME THEORY

SPRING 2022

LECTURE NOTES SET 1: INTRODUCTION

Eleftherios Filippiadis
Office: ΓΔ3, 310
Phone number: 2310-891770
Email: [email protected]
What Is Game Theory* …?

➢ Game Theory is a methodological tool.

➢ Philosophers, economists and other social scientists were getting their hands
around it for centuries
➢ Mathematically formulated for the first time by John von Neumann and Oscar
Morgenstern (1944)
➢ Widespread application of game theory in economics (among other fields) in
the last 60 years
➢ It allows us to formally describe specific type of situations and “solve”
them. It models strategic situations, that is, situations where each agent’s
action has consequences on other agents’ well-being.
➢ Agents might try to resolve the situation acting individually (non-cooperatively)
➢ Agents might try to coordinate their actions to resolve the situation (cooperatively)
@ We will only consider non-cooperative game theory in this class @
➢ In such situations traditional optimization fails.

2
*Game Theory is a misnomer for Multi-person Decision Theory
Some Applications Of Game Theory
➢ Oligopolistic competition, where one firm’s price or quantity choices
affect another firm’s profits through a common market demand.
➢ Auctions, where one bidder’s bid affects another bidder’s profit.
➢ Bilateral and multilateral bargaining.
➢ Externalities where, by definition, one agent’s action affects another
agent’s well-being.
➢ Contract theory where a contract has to be designed in an incentive
compatible fashion.
➢ Moral hazard problem occurs when the actions of one party may
change to the detriment of another after a financial transaction has
taken place (informational asymmetry, principal-agent problem).
➢ Market design (one-sided/two-sided, one-to-one, many-to-one
matchings).
➢ Coalition formation (cartels, international environmental
agreements, clubs, trading blocks, coalitional governments). 3

➢ Cost sharing/surplus division.

Representation Of Games
Strategic interactions (games) can be analyzed once we define
• all the agents (players) whose decisions must be taken into account (be
careful: there might be “others” whose actions must be taken into account but THEY are not
players)
• the actions and the information available to them
• actions are the choices available to a player anytime she has to make a decision
affecting the final outcome; a collection of a players actions, one every time she
has to make a choice, is a strategy.
• Information may be complete or incomplete, perfect or imperfect; we will properly
describer them later.
• the “protocol” according to which players choose their actions in order
to reach an outcome
• their preferences over all possible outcomes (based on payoffs)
There are two ways to represent situations of strategic interactions
1.strategic (or normal) form games, where players choose
“simultaneously,” and 4

2.extensive form games, where players choose their actions sequentially

according to a specific order (i.e., a “protocol”).
PART I: STRATEGIC (OR NORMAL FORM) GAMES OF
COMPLETE INFORMATION
A strategic situation involves a number of agents where each agent
must take at least one action while having specific preferences over the
set of potential outcomes (payoffs). The payoff each agent receives at
the end depends not only on her own actions but on the actions of all
other players as well.

We first consider static games of complete information, that is

• players simultaneously choose actions (static game)
• each player in the game is aware of the sequence, strategies, and
payoffs throughout gameplay (complete information)

Definition: A game has a finite set of “players” N, and each player 𝒊 ∈ 𝑵

has a non-empty actions/strategies set {𝑨𝒊 }𝒊∈𝑵 . For each 𝒊 ∈ 𝑵 there is
a preference relation ≽𝒊 on the set 𝑨 =×𝒋∈𝑵 𝑨𝒋 . Hence, a game can be
5
stated as 〈 𝑵, 𝑨𝒊 𝒊∈𝑵 , {≽𝒊 }𝒊∈𝑵 〉
Notes
• In normal form games the terms action and strategy do not differ. This
is no longer true in extensive form games that we will analyze later.
A strategy 𝒔𝒊 ∈ {𝑺𝒊 }𝒊∈𝑵 of a player 𝒊 ∈ 𝑵 is a function that assigns an
action to each point (node or information set) where the specific player
has to make a choice.
• We define the preference relation of a player 𝒊 not on the set of his
actions alone but rather on the set 𝑨 =×𝒋∈𝑵 𝑨𝒋 , the product of all
players’ action sets. The reason is because the choices of others affect
the payoffs of player 𝒊.
• If 𝑨𝒊 is finite for all the players 𝒊 ∈ 𝑵 then the game is finite.

A utility function 𝒖𝒊 : 𝑨 → 𝓡, such that 𝐮𝐢 (𝐚) ≥ 𝐮𝐢 (𝐛) iff 𝒂 ≽𝒊 𝒃, may

represent preferences. In such a case the game can be stated as
〈 𝑵, 𝑨𝒊 𝒊∈𝑵 , {𝒖𝒊 }𝒊∈𝑵 〉
• If a strategic form game consists of 𝑵 ≤ 𝟑 players then it can be 6
represented using payoff tables (matrices)
Example 1: Two members of a criminal gang are arrested and
imprisoned. Each prisoner is in solitary confinement with no means
of communicating with the other. The prosecutors lack sufficient
evidence to convict the pair on the principal charge, but they have
enough to convict both on a lesser charge. Simultaneously, the
prosecutors offer each prisoner a bargain. Each prisoner is given the
opportunity either to betray the other by testifying that the other
committed the crime, or to cooperate with the other by remaining
silent. The possible outcomes are:
• If prisoners 1 and 2 each betray the other, each of them serves
six years in prison
• If prisoner 1 betrays prisoner 2 but prisoner2 remains silent,
prisoner 1 will be set free and prisoner 2 will serve nine years in
prison (and vice versa)
• If prisoners 1 and 2 both remain silent, they will both serve only
one year in prison (on the lesser charge).
7
 … Example 1: The described typical “prisoner’s dilemma”
game can be represented in the table below
Player 2
Lie Confess
Lie -1, -1 -9, 0
Player 1
Confess 0, -9 -6, -6

We can define all parts of the game, that is:

• 𝓝 = {𝟏, 𝟐}
• 𝑨𝟏 = 𝑨𝟐 = 𝑪, 𝑳
• 𝒖𝟏 𝑪, 𝑳 = 𝒖𝟐 𝑳, 𝑪 = 𝟎
• 𝒖𝟏 𝑳, 𝑪 = 𝒖𝟐 𝑪, 𝑳 = −𝟗
• 𝒖𝟏 𝑪, 𝑪 = 𝒖𝟐 𝑪, 𝑪 = −𝟔
• 𝒖𝟏 𝑳, 𝑳 = 𝒖𝟐 𝑳, 𝑳 = −𝟏
8
Example 2: Consider a duopoly where firms produce an identical
product (at zero per unit cost) and “simultaneously” choose
quantities (i.e., competition à la Cournot). Let the inverse market
demand be 𝒑 = 𝟏 − 𝒒𝟏 − 𝒒𝟐 .

We can define all parts of the game, that is:

• 𝓝 = {𝟏, 𝟐}
• 𝑨𝟏 = 𝑨𝟐 = ሾ𝟎, ∞)
• 𝚷𝒊 𝒒𝒊 , 𝒒−𝒊 = 𝟏 − 𝒒𝒊 − 𝒒−𝒊 𝒒𝒊

9
 Dominant and Dominated Strategies
It is possible that
• it is always to the best interest of a player to choose the same
action/strategy → strictly dominant action
• it is never to the best interest of a player to choose a specific
action/strategy → strictly dominated action

Definitions:
1) A strategy 𝒔ො 𝒊 for player 𝒊 is strictly dominant if
𝒖𝒊 𝒔ො 𝒊 , 𝒔−𝒊 > 𝒖𝒊 𝒔𝒊 , 𝒔−𝒊 for all 𝒔−𝒊 ∈ 𝑺−𝒊 and for all 𝒔𝒊 ≠ 𝒔ො 𝒊 ∈
𝑺𝒊
2) A strategy 𝒔ത 𝒊 for player 𝒊 is strictly dominated if there exists
some 𝒔∗𝒊 ∈ 𝑺𝒊 such that 𝒖𝒊 𝒔ത 𝒊 , 𝒔−𝒊 < 𝒖𝒊 𝒔∗𝒊 , 𝒔−𝒊 for all 𝒔−𝒊 ∈
𝑺−𝒊 10
•
 Dominant and Dominated Strategies

* It is intuitively clear that the outcome (i.e., solution) of a

situation that involves strategic interactions can NEVER include
choices (i.e., actions/strategies) of the agents involved that are
strictly dominated → all strictly dominated strategies can be
eliminated

** However, a solution may very well include weakly dominated

strategies (what’s this?)

11
 Dominant and Dominated Strategies
Note: If a game has a strictly dominant strategy for every player, the
collection of these strategies is a (logical) solution of the game. For
example, in the prisoner’s dilemma game we get
• For player 1
𝒖𝟏 𝑪, 𝑳 = 𝟎 > 𝒖𝟏 𝑳, 𝑳 = −𝟏
𝒖𝟏 𝑪, 𝑪 = −𝟔 > 𝒖𝟏 𝑳, 𝑪 = −𝟗
→ player 1 has a strictly dominant strategy to confess (no matter if her
partner confesses or lies!).
• For player 2
𝒖𝟐 𝑳, 𝑪 = 𝟎 > 𝒖𝟐 𝑳, 𝑳 = −𝟏
𝒖𝟐 𝑪, 𝑪 = −𝟔 > 𝒖𝟐 𝑳, 𝑪 = −𝟗
→ player 2 has a strictly dominant strategy to confess (no matter if his
partner confesses or lies!).
**Hence, a logical solution/outcome of the prisoner’s dilemma game is
that BOTH players choose to confess.
• Is this concept enough to analyze all games? NO. There can be
infinitely many games without a strictly dominant strategy for all 12
players.
Iterated Elimination of Strictly Dominated Strategies
Rational players DO NOT play strictly dominated strategies.
Therefore, in a game we can eliminate all strictly dominated
strategies.
• be careful: once strictly dominated strategies have been
eliminated one should check if, on the remaining non-eliminated
strategies space, there are “revealed” strictly dominated
strategies.
Example
Use the iterated elimination of strictly dominated strategies on
the following game
Player 2
Left Middle Right
Up 1, 0 1, 2 0, 1
Player 1
Down 0, 3 0, 1 2, 0
Two caveats: The process of iterated elimination of s.d.s.
1. additional assumptions about players’ rationality should be 13
imposed.
2. need not lead to a “solution” of the game
Nash Equilibrium
The idea behind the Nash Equilibrium (NE) concept is simple:
given the choices of all other players (no matter if these
choices are “best,” “rational,” etc.) should I change my choice?
No, I shouldn’t IF I have chosen the action that yields the best
for me (BEST RESPONSE) for the given choices of others!

Definition: A best response function (or correspondence) is

defined as
𝑩𝒊 𝒂−𝒊 = 𝒂𝒊 ∈ 𝑨𝒊 ȁ 𝒂𝒊 , 𝒂−𝒊 ≽𝒊 𝒂𝒊 ′, 𝒂−𝒊 , ∀𝒂𝒊 ′ ∈ 𝑨𝒊

Definition: A Nash Equilibrium of a strategic form game is an

action profile 𝒂∗ ∈ 𝑨 =×𝒊∈𝓝 𝑨𝒊 such that
𝒂∗𝒊 , 𝒂∗−𝒊 ≽𝒊 𝒂𝒊 , 𝒂∗−𝒊 for all 𝒂𝒊 ∈ 𝑨𝒊 and for all 𝒊 ∈ 𝓝.
14
Alternatively, A Nash Equilibrium is a profile 𝒂∗ ∈ 𝑨 such that
𝒂∗𝒊 ∈ 𝑩𝒊 𝒂−𝒊 ∀𝒊 ∈ 𝓝
Nash Equilibrium
How do we find the NE of a game? We must check every
action profile if a player wants to deviate from that state for
fixed choices others (i.e., unilateral deviation). If, for a given
profile, there is at least one player that wants to change her
action the given profile CANNOT be a Nash Equilibrium. If, for a
given profile, no player wants to change his action then the
given profile is a Nash Equilibrium.
Alternatively, we use the Best Response Functions and look for
a profile that satisfies the alternative definition of NE.

15
Example 1’: Consider again the typical “prisoner’s dilemma”
game as in the table below
Player 2
Lie Confess
Lie -1, -1 -9, 0
Player 1
Confess 0, -9 -6, -6

16
Example 2’: Consider again a duopoly where firms produce an
identical product (at zero per unit cost) and “simultaneously”
choose quantities (i.e., competition à la Cournot). Let the inverse
market demand be 𝒑 = 𝟏 − 𝒒𝟏 − 𝒒𝟐 .

17
 Nash Equilibrium: None, one, or many?
In the examples we have previously discussed we identified a unique
Nash Equilibrium in each one. Is it always the case?
Example 3
Battle of the Sexes
Player 2
UFC Opera
UFC 2, 1 0, 0
Player 1
Opera 0, 0 1, 2
(an example with two Nash equilibria)

Example 4
Matching the Pennies
Player 2
Heads Tails
Heads 1, -1 -1, 1
Player 1
Tails -1, 1 1, -1 18
(an example with no (?) Nash Equilibrium)
Nash Equilibrium: None, one, or many?
Implications of example 3:
⁻ When there are multiple Nash equilbria there is a tiny little
problem! We cannot be sure as to which Nash equilibrium
will prevail (or, IF any of the Nash equilibria will be realized at
the end!).
⁻ We will discuss later the problem of coordination failure.

Implications of example 4:
⁻ In many cases it is essential to identify if a game has a
solution (i.e., Nash equilibrium) or not!
⁻ However, we will see that the game of “matching the
pennies” has actually a Nash Equilibrium.
19
Time to practice!
Consider a Cournot triopoly where firms have zero marginal and
zero fixed costs and the inverse demand is described by
𝟑
𝑨 − 𝒒𝟏 − 𝒒𝟐 − 𝒒𝟑 , 𝒊𝒇 ෍ 𝒒𝒊 ≤ 𝑨
𝒑= 𝒊=𝟏

𝟎, 𝒐𝒕𝒉𝒆𝒓𝒘𝒊𝒔𝒆

Find all Nash Equilibria of this game.

20