Chapter 3 Games, strategies and equilibrium.

Master’s in economics, 2nd year

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

1 Games and strategies.

Games are mathematical objects used to analyze strategic

interactions.
As soon as we depart from the perfectly competitive environment,
an agent’s satisfaction depends on some other agents’ behavior.
Oftentimes an agent’s optimal choice depends on what she expects
other agents’ behavior to be.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

1.1 Trees, nodes and extensive forms.

Describing a strategic situation can be done in an exhaustive

manner by describing the sequence of moves and the set of
actions, the information available to a player when she gets to
make a move, as well as each player’s Payoff as a function of the
outcome of the game.
In a finite game where there are a finite number of players who
have access to a finite set of actions over the entire game
sequence, this exhaustive description can be materialized through a
game tree.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Example 1

Consider a situation with two player, each having access to two

actions, L and R, where player 1 chooses 1st and then player 2
chooses after having observed the choices of player 1.
Resulting payoffs are (2, 0), (2, −1), (1, 0), (3.1) associated with
(L, L), (L, R), (R, L) and (R, R) respectively.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Nodes and branches

Player 1 is associated with one node reflecting that she moves only
once and no other player moves before her.
Player 2 is associated with 2 nodes, because, although she moves
once, the impact of her choice is conditioned on the two possible
choices of action made earlier in the sequence of moves.
From each node start 2 branches corresponding to the two
available choices.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Terminal nodes and payoffs.

There are 4 terminal nodes associated with the 4 possible

outcomes of the game.
At each terminal node, payoffs are specified for each player.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Example 2: information sets.

Example 2 has the same specification as Example 1 in terms of

nodes, branches and payoffs: it only differs in that the 2 nodes of
player 2 are in the same information set.
Difference between the two games is that, when 2 chooses, she
knows the choice of 1 in Example 1 but not in Example 2.
As a result, although the play of 2 is represented by 2 nodes in
both examples, 2 has 2 information sets iin Example 1 and only 1
information set in Example 2.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Lotteries.

We want to allow for lotteries on the set of terminal nodes to

acount for two sources of uncertainties.
1 Uncertainty resulting from randomness in players’ choices: we
will define mixed strategies where a player randomizes
between the available choices at an information set (e.g. in
Example 1, player 2 might choose L with probability 1/3 and
R with probability 2/3 after observing that 1 chose L.
2 Uncertainty resulting from moves by nature which can be
incorporated in the tree as nodes where each branch
represents a different state of nature: important strategic
issue here is whether some or all agents observe the state of
nature at some point before reaching a terminal node.
Here we assume probabilities are objective: this will be part of the
equilibrium concepts we use.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Payoffs.

Players have preferences over the set of lotteries on terminal nodes,

which can be represented by a v.N-M expected utility.
Payoffs specified at terminal nodes are the expected utility
associated with the degenerate lottery where this node happens
with certainty.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Extensive form.

An extensive form representation of a game specifies:

1 the set of players including nature,
2 a probability over the states of nature (if relevant),
3 the set of information sets for each player,
4 the set of choices available at each information set,
5 set of outcomes (terminal nodes) with associated payoffs for
all players other than nature.

Note that the sequence of moves is not explicitly described: it is

reflected in the information sets whenever it is relevant.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Strategies.

A pure strategy for some player i specifies her choice at all her
information sets.

In Examples 1 and 2, player 1 has 2 pure strategies (s11 = L and

s12 = R) whereas player 2 has 4 strategies in Example 1
(s21 = (L, L), s22 = (L, R), s23 = (R, L), s24 = (R, R)) but only 2
in Example 2 (s21 = L and s22 = R).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

A mixed strategy for player i is a probability defined over the set of
her pure strategies: pure strategies can then be viewed as
degenerate mixed strategies.

Pure strategies are sometimes called actions though it might be

less confusing to call “actions” the choices at each information set
(L and R in Examples 1 and 2).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Timing and information structure.

In practice, it is usually tedious to provide a full description of the

extensive form.

It is however very useful to clarify the sequence of moves and the

information available to each player when she has to make a
choice.

Applied articles often provide such a description, although it is

often somewhat loose, under the term timing of the game or
information structure.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

What matters is not so much the timing but what information is
available to a player when she makes a choice.

For instance, the timing might say,

in a first stage, this happens,
in a second stage that happens.

What is important is what players who make choices in the 2nd

stage know about what happened in the 1st stage: the information
can be only partial.

If no information is revealed, the distinction between the two

stages is irrelevant strategically: one might as well say that
everything happens simultaneously.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

1.2 Normal form representation of a game.

From discussion of the extensive form what matters for the

outcome of a game is the profile of strategies selected by all the
players involved.

It determines a probability distribution over the set of outcomes

and hence, the expected payoff of players.

This suggests a more compact description of the game that links

strategy profiles to payoffs, the normal form.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Payoff matrix.

For games with 2 players and a finite set of strategies for each
player, the normal form can be fully describes through a payoff
matrix.

We can illustrate the concept with examples 1 and 2: for Example

1 we have a 2 × 4 matrix and for Example 2 we have a 2 × 2
matrix, where 1 chooses the rows and 2 chooses the columns.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

The normal form representation of a game with N players indexed
by i = 1, ..., N specifies
the pure strategy space of each player Si ,
the payoff function Π, ×N N
i=1 Si ↣ R , that gives, for each
N
strategy profile s = (s1 , ..., sN ) ∈ ×i=1 Si , the players’ payoffs
Π(s) = (Π1 (s), ..., ΠN (s)), where Πi is player i’s payoff
function.

The set of mixed strategies of player i is the set of probabilities on

Si denoted ∆(Si ): e.g. if pure strategies are real numbers, then
mixed strategies are real valued random variables.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

1.3 Non cooperative, versus cooperative games.

We have described a non cooperative game where the outcome is

determined by individual choices made independently.

This is to be contrasted with cooperative games where the

outcome is determined by a collective choice of allocation, which is
often interpreted as the result of a negotiation process which is not
explicitly modeled.

We illustrate with the case of bilateral bargaining.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Bilateral bargaining.

Two individuals, 1 and 2, bargain over a pie of size 1.

Feasible outcomes are (x1 , x2 ) with xi ≥ 0 and x1 + x2 ≤ 1.

Player i’s payoff is xi .

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Nash bargaining.

A popular solution to the problem is to assume that, if the two

players reach an agreement, it should have certain characteristics:
e.g. the outcome should be Pareto efficient so that x1 + x2 = 1.

Then the solution is the set of ordered pairs (x1 , x2 ) that satisfy a
list of axioms.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Divide the dollar game.

An alternative approach is to use a non cooperative game setup

and characterize equilibria for that game.

Consider the game where each individual selects a share si ∈ [0, 1]

and receives xi = si if (s1 , s2 ) is feasible and receives 0 otherwise.

Formally, Πi (s1 , s2 ) = si if s1 + s2 ≤ 1 and Πi (s1 , s2 ) = 0 if

s1 + s2 > 1.

Later we study a sequential version of this game.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

2 Nash equilibrium.

In Example 1, 1 must anticipate what 2 will do in order to identify

his optimal choice.

Looking at the game tree it seems natural that he expects 2 to

choose S22 = (L, R).

Looking at the payoff matrix we see that, no matter what 1 does,

2’s payoff is maximized by s22 : we say that s22 is a dominant
strategy: then 1 should expect 2 to play s22 and should choose s12 .

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Prisoners’ dilemma

Each of two prisoners is offered either to testify against her/his

accomplice or not. Testifying (the fink strategy) yields a reward of
1. Each goes to prison, with a corresponding utility of -3, iff the
other finks, whereas being let out without charge (which happens
if the other prisoner cooperates) without a reward has utility 2.

The payoff matrix shows that no matter what the other prisoner’s
strategy is, each prisoner is better off choosing fink: it yields 3 (as
opposed to 2) if the other cooperates and -2 (as opposed to -3) if
the other finks.

“Fink” is therefore a dominant strategy for both and (F , F ) seems

like the only reasonable solution.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Iterated elimination of dominated strategies.

Games such as the prisoners’ dilemma, where all players have a

dominant strategy are not so widespread.
There are however much more relevant games where the outcome
of play can be restricted through iterated elimination of dominated
strategies.
Rather than looking for dominant strategies, we look for
strategies for which there is some other strategy that
dominates them no matter the other players’ strategies: they
are dominated strategies.
After eliminating all dominated strategies for all players, we
look again for dominated strategies in the restricted game.
We iterate until there is no dominated strategy left.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Common knowledge of rationality.

The argument for the solution in Example 1 assumes much more

than the solution to the PD game;
in the latter, the proposed solution will prevail as long as each
player is rational;
in the former, we also need that player 1 knows that player 2
is rational.

Player 1 might feel unable to formulate beliefs about 2’s behavior,

in which case she might choose L to guarantee herself a payoff of 2
(remember ambiguity aversion).

More generally, iterated elimination or the more powerful concept

of rationalizability require common knowledge of all players’
rationality.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Best response.

To avoid speculations about players’ beliefs, we now assume that

each player i knows the strategies plaid by all the other players,
which we denote s−i if they are pure strategies.

Strategy si ∈ Si is a best response to strategies si if

Πi (si , s−i ) ≥ Πi (si′ , s−i ) for all si′ ∈ Si .

Similarly we say that a mixed strategy σi ∈ ∆(Si ) is a best

response to mixed strategies by other players denoted σ−i if

Es (Πi (s) | σi , σ−i ) ≥ Es (Πi (s) | σi′ , σ−i ) for all σi′ ∈ ∆(Si ).

where Es (Πi (s) | σi′ , σ−i ) means the expected payoff for i over all
pure strategy profiles drawn according to a probability induced by
mixed strategy (σi′ , σ−i ).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Matching pennies.

Two player hold one coin each and have to choose to show one
side of their coin: heads (H) or tails (T ).

Player 1 receives 1 from player 2 if choices match, (H, H) or

(T , T ) and must give 1 to player 2 if choices are mismatched.

Then, for pure strategies, a best response for player 1 is to make

the same choice as player 2 whereas a best response for player 2 is
to make a choice opposite to that of player 1.

For mixed strategies, a best response for 1 involves choosing the

most likely choice of 2 while a best response for 2 is to choose the
least likely choice of 1.

If one of the players selects H or T with equal probabilities of .5,

then any mixed strategy for the opponent is a best response.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Game can be described by denoting πi the probability of choosing
H: set of mixed strategies is then completely describe by having
πi ∈ [0, 1].

If our criterion is that both players should behave rationally while

knowing what the other player is doing, then π1 should be a best
response to π2 and π2 should be a best response to π1 .

From our discussion above, if π2 > 1/2 1 chooses π1 = 1 but if

π1 > 1/2 2 chooses π2 = 0.

So it is not possible that both players are rational with π2 > 1/2
and an analogous argument shows that it is not possible that both
players are rational if π1 ̸= 1/2 or π2 ̸= 1/2.

Hence, the only possible prediction for this game where both player
would be rational while knowing what the other player is doing is
π1 = π2 = 1/2: then players are rational since they are indifferent
over the whole set of mixed strategies.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Nash equilibrium.
Informally, we want to say that the game is at an equilibrium if all
players behave rationally while knowing what other players are
doing (though this knowledge may involve uncertainty if other
players play mixed strategies).

Definition
(Nash equilibrium) A mixed strategy profile σ ∗ ∈ ×N
i=1 ∆(Si ) is a
Nash equilibrium of the game if

Es (Πi (s) | σi∗ , σ−i

∗ ∗
) ≥ Es (Πi (s) | σi , σ−i ) for all σi ∈ ∆(Si ),

for all i = 1, ..., N.

All player should be playing a best response to other player’s

strategies: e.g. π1 = π2 = 1/2 is the unique Nash equilibrium in
matching pennies.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Pure strategy equilibrium.
Although mixed strategies provide an intuitive and convincing
prediction for matching pennies and for many other games, this is
not always the case, in particular when we consider economic
applications.
So we are often interested in deriving Nash equilibria in pure
strategies.

Definition
(Pure strategy Nash equilibrium) A pure strategy profile
s ∗ ∈ ×N
i=1 Si is a pure strategy Nash equilibrium of the game if

Πi (si∗ , s−i
∗ ∗
) ≥ Πi (si , s−i ) for all si ∈ Si ,

for all i = 1, ..., N.

A pure strategy Nash equilibrium is a Nash equilibrium (no

deviation to a mixed strategy is profitable).
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Divide the dollar with only two share sizes.

This game could also be construed as Cournot with only 2

quantities and a rectangular demand: also strategically analogous
to famous games such as “The battle of the sexes.” or “Chicken.”.

Assume that each party must pick si ∈ {.6, .4}.

From the payoff matrix we can see that there are two pure strategy
equilibria: (s1 , s2 ) = (.6, .4) and (s1 , s2 ) = (.4, .6): just check for
profitable deviations in each of the 4 boxes.

Illustrates that there can be multiple Nash equilibria.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Checking for mixed strategy equilibria.
In the two by two divide the dollar game, there is also a mixed
strategy equilibrium.
Denoting πi the probability that i chooses .6, we can write
expected payoff from choosing .6 or .4, for j ̸= i. They are
(1 − πi ).6 if sj = .6
.4 if sj = .4.

In order for j to play a mixed strategy, he must play .6 and .4 with

positive probabilities so the two choices must yield the same
expected payoff: so we must have πi = 1/3.
By symmetry, the unique fully mixed strategy equilibrium is
(π1 , π2 ) = (1/3, 1/3).
As for matching pennies, we can draw best responses in the
(π1 , π2 ) space, which shows the 3 equilibria.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Chicken.
Consider the following “chicken game”. Each of two players sits in
a car on each side of a bridge and must decide whether to start the
car or not: if both start, the 2 cars crash, but if only one starts,
she/he wins and the other loses.

Winning earns a payoff of a > 2, while a tie earns 2, losing earns 1

and dying yields -1.

2 pure strategy equilibria: (start, notstart) and (notstart, start).

In the fully mixed strategy equilibrium, denoting πi the probability

for i of starting the car, we should have

−π2 + a(1 − π2 ) = π2 + 2(1 − π2 )

−π1 + a(1 − π1 ) = π1 + 2(1 − π1 )

so the equilibrium is (π1 , π2 ) = ( a−2 a−2

a , a ).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Problem with mixed strategies: commitment.

Mixed strategies allow for obtaining an equilibrium when there is

no pure strategy equilibrium as in matching pennies.

They also allow for finding a symmetric equilibrium in symmetric

games where all pure strategy equilibria are asymmetric (e.g. the
chicken game).

However, once choices have been made, players can end up in

situations where they would like to modify their choices (e.g. in
matching pennies one of them would want to switch).

Hence mixed strategies are credible only in situation where players

are fully committed to their choices.

This seems hard to justify in some economic applications like price

competition.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Problems with mixed strategies: comparative statics.
In the symmetric chicken game an increase in the payoff from
winning, a, increases the probability that either player starts the car
in the symmetric mixed strategy equilibrium, which seems intuitive.

This is however misleading: consider now the asymmetric game

where the payoff from winning is ai > 2 for player i.

In the mixed strategy equilibrium, denoting πi the probability for i

of starting the car, we should have

−π2 + a1 (1 − π2 ) = π2 + 2(1 − π2 )
−π1 + a2 (1 − π1 ) = π1 + 2(1 − π1 )

so the equilibrium is (π1 , π2 ) = ( a2a−2

2
, a1a−2
1
).

If payoff from winning for player 1 increases, it has no impact on

the probability that 1 starts the car but it increases the probability
that 2 does start the car, which is not so intuitive.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Infinite game: continuous strategy sets.

Consider the following duopoly model with price competition.

Two firms with identical and constant marginal costs c > 0 sell
differentiated products, 1 and 2, for which demands are
D1 (p1 , p2 ) = D̄(p1 − p2 ) and D2 (p1 , p2 ) = 1 − D̄(p1 − p2 ), where
Di and pi denote demand for and price of product i respectively
and D̄ is strictly decreasing and differentiable: covered market
where a measure 1 of consumers buys either product.

Firms maximize profits and are risk neutral so payoff for Firm i is
πi (p1 , p2 ) = (pi − c)Di (p1 , p2 ).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

First order conditions and best responses.

Assume Firm i knows that Firm j ̸= i is charging p−i : than its best
response Ri (p−i ) is price pi that maximizes πi (pi , p−i ) wrt pi .

It must satisfy FOC

Di (Ri (p−i ), p−i )

Ri (p−i ) − c = − (1)
Dii (Ri (p−i ), p−i )

where Dii denotes the partial derivative of Di wrt pi .

This yields an implicit equation for the best response (can be used
to perform comparative statics on best responses).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Equilibrium.
If (p1∗ , p2∗ ) is a Nash equilibrium, then we must have p1∗ = R1 (p2∗ )
and p2∗ = R2 (p1∗ ).

If a Nash equilibrium exists, it should be a solution to this system

of 2 equations with 2 unknowns.

With our demand specifications we have the following

characterization:
1 − 2D̄(p1∗ − p2∗ )
p1∗ − p2∗ =
D̄ ′ (p1∗ − p2∗ )

It is readily shown that in a symmetric market where D̄(0) = 1/2,

the equilibrium is such that
1
p1∗ = p2∗ = c −
2D̄ ′ (0)

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Back to examples 1 and 2: the power of commitment?

Focusing on pure strategies, Example 2 has 2 Nash equilibria,

(L, L) and (R, R).

In Example 1, we identified (R, L/R) as a reasonable outcome: it

yields the same outcome as (R, R) in Example 2 which is the best
equilibrium for player 1: so, by being able to commit to a choice in
Example 1, player 1 can guarantee herself the most favorable
equilibrium outcome.

Although (R, L/R) is an equilibrium in Example 1, there are 2

others, including (L, L/L) which gives the same outcome as (L, L)
in Example 2.

Hence, Nash equilibrium tells a different story from our basic

intuition or iterated elimination of dominated strategies: power of
commitment seems much weaker.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

3 Perfect equilibrium.

We introduce a strengthening of Nash equilibrium: terminology is

refinement of Nash equilibrium.
Here we explore 3 important ideas:
1 Maybe it is not so rational to be absolutely certain that other
players will play their equilibrium strategy.
2 It seems reasonable to require that a player behaves rationally
even if he/she finds out that there has been some deviation
from the equilibrium (subgame perfection).
3 It seems reasonable to assume that rational players do not
play weakly dominated strategies.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Trembles.
Recall that in Example 1, there is an equilibrium (L, L/L) where
player 2 chooses L no matter what and player 1 chooses L.

In this equilibrium, both are rational because player 1 expects 2 to

choose L if she chooses R and player 2 expects 1 chooses L with
probability 1.

Now if 2 has the slightest doubt that 1 plays R so she puts a small
probability ϵ > 0 on 1 playing R, then her choice of L when 1
chooses R would no longer be rational and the equilibrium would
unravel.

To rule out such “fragile” equilibrium, we will introduce later a

requirement that choices should remain optimal if there is a slight
uncertainty that opponents “tremble”.

First we introduce a related restriction ensuring that player’s

behavior is always “credible”.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Example 3: credible threats.

In Example 1, although 2 could induce 1 to play L by “threatening”

to choose L if 1 plays R, she does not benefit from formulating
such a threat: her payoff is 0 whereas if 1 expects that, after she
has chosen R 2 behaves optimally, 2 would obtain a payoff of 1.

Consider the following example where the only difference with

Example 1 is that 2’s payoff associated with outcome (L, L) is 2
instead of 1.

Then, by threatening to play L rather than R in response to R, 2

could obtain 2 instead of 1.

However, such a threat is not “credible” because once 1 has

chosen R, it is best for 2 to choose R.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Subgames.

In Examples 1 and 3, once 1 has made a choice, the choices of 2 at

each information set define a sub-tree starting with an information
set with one node:
each of these sub-trees defines a game with one player who
has one information set and two choices.

These games are called proper subgames of the initial game.

Each of the strategies of 2 in the initial game can be “truncated”

to the subgame so it is a strategy for that subgame (L or R).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

The concept of subgame can be generalized to any game.

The general idea is that any sub-tree starting with a one node
information set which is internally coherent (in the sense that it
does not interact with the rest of the overall tree) represents a
subgame.
in particular, the game itself is one of its own subgames.

Next we provide a fairly precise definition.

But the point is that you recognize a subgame when you see one.

Later in the chapter, we consider 2 categories of games for which

subgames can be clearly identified.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Consider the extensive form representation of a game. Then a
subgame of the initial game is a subset of the initial tree with the
following properties:
1 it starts with an information set containing one node - this
means that the first mover in that game has perfect
information about all choices that precede that node;
2 a node is in the subgame iff it is a successor of the initial node
in the overall game - this means that the subgame exactly
contains all the choices that are made with the knowledge of
choices that precede the initial node.
3 all information sets in the subgame are information set of the
initial game.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Subgame perfect Nash equilibrium: Selten (1965).

We can now define a refinement of Nash equilibrium that ensures

that no player makes a choice based on the anticipation that if she
made some other choice, she would suffer because of some
irrational behavior in one of the subgames.

Definition
A profile of strategies σ in a game Γ is a subgame perfect Nash
equilibrium if, for any subgame of Γ, the truncation of σ to that
subgame is a Nash equilibrium of the subgame.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Game histories and equilibrium path.

Recall that in a Nash equilibrium, each player knows all the

equilibrium strategies of all the players.

This does not mean that the player knows at any time the
sequence of choices that have been made up to that point, what
we call the game history. This is because
if equilibrium strategies are mixed, a player might not know
the realization of mixed strategies of his opponents.
a player might not know that some opponent has deviated
from his equilibrium strategy: indeed, to check whether we
have an equilibrium, we need to check what happens if each
player deviates.

At any point in the game, Information sets describe what players

have learned about the game history.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

An information set is on the equilibrium path if the corresponding
information about the game history is consistent with the
equilibrium strategy profile.

Subgame perfection ensures that, in any subgame, players’

behavior is optimal both on the equilibrium path and off the
equilibrium path.

The concept is relevant in game where, when a player learns

something about the history, all players learn the entire history
perfectly (that is, they enter a new subgame).

Such games are sometimes called games of almost perfect

information.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Backward induction

To solve examples 1 or 3, we first check what would be a rational

behavior for the last mover, player 2, at each of her information
sets: this yields strategy L/R.

Then we can work backwards, computing player 1’s payoff, which

is 2 if she plays L and 3 if she plays R, so in the unique subgame
perfect equilibrium she chooses R.

This is a special case of what Kuhn (1951) calls backward

induction

We will show that when, subgame perfection is relevant, the set of

SPNEs can usually be determined using some form of backward
induction.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Eliminating weakly dominated strategies.

Consider the following Bertrand competition game: Firm 1 with

constant marginal cost 0 competes with Firm 2 with constant
marg. cost 2 on a market for a homogeneous product.

Each chooses between prices {1, 2, 3}: total demand is always 2, is

split equally if prices are equal or else goes to the firm with the
lowest price.

Three pure Nash equilibria (1, 2), (2, 2) and (2, 3).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

In the first 2 equilibria, Firm 2 charges its marginal cost, which
yields 0 payoff, which is weakly dominated by charging 3, which
always yield at leas 0 but can yield 1 if Firm 1 charges 3 as well.

(NOTE: with continuous strategies, there is a continuum of pure

strategy Nash equilibria and all involve the high cost firm charging
a price at or below its marginal cost, which is weakly dominated:
when marginal costs are equal, unique pure strategy equilibrium
has both charge marginal cost which is weakly dominated).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Weakly dominated strategies and trembles.

Note that a player would never play a strongly dominated strategy

in a Nash equilibrium.

In the equilibria (1, 2) and (2, 3) Firm 2 finds it optimal to choose

a price of 2 because it is certain that 1 will not price strictly above
2.

Now, if Firm 2 has the slightest doubt that Firm 1 might charge a
price > 2 (say it assigns a small probability ϵ > 0 to this event)
then it should prefer charging 3 just in case it could make a strictly
positive profit.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Trembling hand perfect Nash equilibrium: Selten (1975).
Our discussions of credible threats and weakly dominated
strategies suggests that we might want to rule out Nash equilibria
that exist only because all players are absolutely certain that all
players will play their equilibrium strategies.

This is the idea behind trembling hand perfect Nash equilibrium.

A fully mixed strategy is a mixed strategy that assigns positive

probabilities to all pure strategies.

Loosely, a strategy profile σ is a trembling hand perfect Nash

equilibrium if there exists a sequence of fully mixed strategy
profiles, {σ n } that converges to σ and such that, for every player i
n .
and for all n, σi is a best response to σ−i

The above formulation is not the actual definition by Selten but,

rather, a characterization which is equivalent to Selten’s definition.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Trembling hand perfection and weak dominance.

It can be shown that if σ is a trembling hand perfect equilibrium,

then it involves no weakly dominated strategy.

The link between trembling hand perfection and subgame

perfection is more intricate.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

4 Games of perfect information.

In Example 1, 2 knows the actual choices of 1 when deciding

between L and R.

Since this is the only relevant information for her, we might want
to say she has perfect information: in particular, if 1’s equilibrium
strategy is mixed, then she knows the realization of that strategy
and if 1 has deviated from her equilibrium strategy she knows it.

We want to describe games where “all players have perfect

information”: as soon as there are at least 2 players, it is not
possible that all are in the situation of player 2 in Example 1 (for
instance in Example 1 player 1 could not observe a deviation by
player 2 even though this would be very relevant to her).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

What we want to rule out is a structure where several players are
choosing “simultaneously” as in Example 2.

Loosely we say that a game is of perfect information if, in its

extensive form representation, all information sets have exactly one
node so all choices are made with perfect knowledge of the game’s
history.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Backward induction and the iterated elimination of
strongly dominated strategies.

Perfect information games can be solved through backward

induction, using Kuhn’s algorithm, Kuhn (1951).

Our discussion of Example 1 at the start of Section 3 illustrates

that, in perfect information games, backward induction usually
turns out to be equivalent to iterated elimination of dominated
strategies.

In particular, if no player is indifferent between two outcomes of

the game, then the two methods are equivalent.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Is it better to be a leader or a follower?

At the start of Section 3, Example 1 illustrated that by committing

to a choice before 2 chooses, 1 can ensure that she obtains the
best equilibrium outcome among those that would prevail in the
simultaneous version in Example 2: this illustrates that being a
leader usually allows to do better than if play is simultaneous (in
any case, a leader can do at least as well as with simultaneous play
by choosing an equilibrium strategy of the simultaneous game).

What about the follower?

does she benefit from being able to observe the leader’s
choice?
would she prefer being a leader?

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Consider again the duopoly price competition setting of Section 2:
assume now that D̄(p1 − p2 ) = 21 − α(p1 − p2 ), α ∈ (0, 1).

In the simultaneous move game prices are

1
p1s∗ = P2s∗ = c + 2α = 1/2+αc
α .

Now suppose 1 chooses its price first and then 2 gets to observe
that price before choosing its own.

In a SPNE, 2 best responds to p1 in accordance with (1) which, for

our demand specification yields.
1/2 + αp1
R2 (p1 ) = c + .
2α

Knowing this, Firm 1 maximizes its profit.

Π1 = (p1 − c)D(p1 − R2 (p1 )),
wrt p1 .
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Firm 1’s equilibrium price p1∗ must satisfy FOCs

D̄(p1∗ − R2 (p1∗ ))
p1∗ − c = − .
D̄ ′ (p1∗ − R2 (p1∗ ))(1 − R2′ (p1∗ ))

1/2+α(2+α)c
For our specification we obtain p1∗ = α(1−α) > p1s∗ .

In response, 2 sets a price p2∗ = R2 (p1∗ ) > R2 (p1s∗ ) = p2s∗ .

Firm 1 as a leader has an incentive to increase its price beyond

what it would do in the simultaneous choice setting, because, by
doing so, it induces Firm 2 to increase its price (recall that
R2′ > 0), which has a positive impact on Firm 1’s demand, which
would not exist if Firm 2’s price was anticipated to be unchanged.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Using the fact that R2 has slope 1/2 < 1, we can also show that
p2∗ < p1∗ and hence, D2 (p1∗ , p2∗ ) = 1 − D̄(p1∗ − p2∗ ) > 1/2 and
D1 (p1∗ , p2∗ ) < 1/2: Firm 2 sells more than Firm 1.

Note that Firm 2 could have chosen a price p2 = P1∗ in which case
firms would have shared the market equally at price P1∗ and
obtained a larger profit than the equilibrium profit of Firm 1 who
serves less than half the market at price p1∗ .

By choosing a lower price than p1∗ Firm 2 obtains more profit than
by choosing p1∗ (revealed preference argument): so its profit is
higher than that of Firm 1.

To sum up if one firm becomes a price leader, then both firms

benefit as compared to simultaneous move but the follower
benefits more.

Predictions are quite different when comparing Stackelberg with

Cournot.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Sequential bargaining: Rubinstein (1982)
Consider the bilateral bargaining setting again.

Assume now that player 1’s strategy is s1 ∈ [0, 1], her bid for a
share in the pie and player 2’s choice is Y or N, which means he
decides whether to accept or not 1’s offer after observing it, where
he gets 1 − s1 if he accepts and 0 otherwise tie breaking rule is, he
prefers accept when indifferent.

Unique SPNE is that 2 accepts all offers and 1 selects s1 = 1.

Suppose now that, if he rejects, 2 can make a counter offer and

the game ends after 1 has decided whether to accept or reject.

Then 1 accepts all counter offers, then 2 accepts s1 only if s1 = 0,

else he makes a counter offer s2 = 1: then 1 is indifferent between
all offers s1 ∈ [0, 1] and we assume she offers s1 = 0 and 2 accepts.

Equilibrium payoffs are 0 for player 1 and 1 for player 2.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Discounting.

Assume there is a discount factor δ ∈ (0, 1) common to both

players.

In the 2 rounds case, 1 accepts all counter offers, then 2 accepts

only offers s1 ≤ 1 − δ and else makes a counter offer s2 = 1: then
1 offers s1 = 1 − δ in the first round and 2 accepts.

Following the same logic, in a 3 rounds setting 1 offers

3
s1 = 1 − δ + δ 2 = 1+δ
1+δ in the 1st round and 2 accepts.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Backward induction and many rounds of negotiation.

Using backward induction it can be shown that if there are T ≥ 1

T
rounds, then in the 1st round 1 offers s1 = 1−δ
1+δ if T is even and
1+δ T
s1 = 1+δ if T is odd, and 2 accepts.

1+δ 1−δ 2
It is true for T = 1 and T = 2 since 1 = 1+δ and 1 − δ = 1+δ .

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Now assume the result holds for some T ≥ 1 and consider the
game with T + 1 rounds.

In round 1, player 2 knows that if he rejects the offer, his payoff in

T 1+δ T
the next round will be 1−δ
1+δ if T is even and. 1+δ if T is odd.

Then player 1 offers the highest possible s1 such that 2 accepts:

T +1 δ+δ T +1
i.e. such that 1 − s1 = δ−δ
1+δ if T is even and 1 − s1 = 1+δ if
T is odd.
T +1 1−δ T +1
Hence, s1 = 1+δ
1+δ if T + 1 is odd and s1 = 1+δ if T + 1 is
even, which proves the result.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Infinite number of rounds of negotiation.

Assume now there is no end to the bargaining process: after t

rounds, the player who receives an offer can always make a counter
offer (note that t here denotes the number of offers already made
as opposed to T in the previous analysis, which denoted the
number of rounds remaining).

Now st denotes the offer made at period t.

For a given SPNE, call the value of the game for a player at round
t the equilibrium expected discounted payoff of that player at
round t.

Let Vit denote the value of the game for player i at round t if she
makes the offer at round t

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

The unique SPNE of the infinite horizon sequential
bargaining game.

Define V̄i and V i as the maximum and the minimum, over all
SPNEs and all rounds t, of the value of the subgame starting at
round t.

At period t, the most player 1 can claim is st = 1 − δV 2 and the

least that can be claimed is st = 1 − δ V̄2 . So we have

V̄1 ≤ 1 − δV 2 and V 1 ≥ 1 − δ V̄2 . (2)

Similarly for player 2 we must have:

V̄2 ≤ 1 − δV 1 and V 2 ≥ 1 − δ V̄1 . (3)

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Combining the 1st inequality in (2) and the last equation in (3) we
1
obtain V̄1 ≤ 1 − δ + δ 2 V̄1 so V̄1 ≤ 1+δ .

analogously, combining the last inequality in (2 with the first

1
inequality in (3) we obtain V 1 ≥ 1+δ .

1
It follows that V1t = 1+δ for all t and the same arguments show
1
that V2t = 1+δ for all t.

Hence in the unique SPNE, the player making an offer in the first
1
round chooses s1 = 1+δ > 1/2: impatience, δ < 1, induces the
first mover to offer less than half of the pie to the other party but
her bargaining power decreases as δ goes to 1.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

5 Games of “almost” perfect information.
The essence of perfect information is that players get to observe
the game’s history before making a choice.

Truly perfect information rules out simultaneous moves, so thai a

player who is making a choice needs not worry about some other
players taking decisions at the same time, which would depend on
what they expect his own choice to be;
perfect information allows a player to be committed to a
certain choice vis-a-vis all other players for the rest of the
game.

We now want to consider games where some choices are

simultaneous but still:
all player who play at the same time have the same perfect
information about the history of the game.
the choices of these players is then perfectly known to all
players who have to make a choice at future stages.
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Stages.

An almost perfect information game can be divided into stages, or

equivalently, periods or dates.
At each stage, a group of players play simultaneously knowing
the choices that were made in previous stages: hence they
have common and perfect information about the history.

Importantly, there is no asymmetric information where some

information about history would be known by some players but not
others (among those who are still in a position to make choices).

Two examples with 3 players: one where player 1 plays in the first
stage and players 2 and 3 play in the second stage and one where
players 1 and 2 play in the first stage and player 3 plays in the
second stage.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Repeated games

An important and seemingly simple application is that of repeated

games.

A repeated game involves a simultaneous game, or “one shot”

game (such as example 2 or chicken), which is played T ≥ 2 times:
at date t > 1, all players know perfectly what was played at all
previous dates t ′ = 1, ..., t − 1.

Payoffs for the repeated game are the discounted sum of the
payoffs for the one shot game at each period.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

In repeated games, past choices have no “direct” impact on
current payoffs.
for instance, if we consider the repetition of an “inspection
game” between a supervisor and an employee where the
employee benefits from learning by doing if he exerts a high
effort, then this is not a repeated game because choices in
period 1 impact payoffs in period 2 (e.g. exerting effort in
period 1 decreases the cost of exerting effort in period 2).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Repeated prisoners’ dilemma.

To illustrate, consider the prisoners’ dilemma with discount factor

δ ∈ (0, 1).

Key question in repeated games is whether repetition can yield

SPNE that differ from the repetition of Nash equilibria of the one
shot game.

For the prisoners’ dilemma,

the unique equilibrium is (Fink, fink) because “fink” is a
strongly dominant strategy for both players.
is it possible that, if the game is repeated, there exists a
SPNE where players engage in some cooperation?

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Finitely repeated prisoners’ dilemma.

Suppose T = 2: there are 5 subgames, the full repeated game and

the games played in period 2 after learning what was played in
period 1 (4 possible outcomes).

All subgames in period 2 have a unique equilibrium which is

(fink, fink); then, in period 1, prisoners’ discounted payoffs are
merely the one shot payoffs −2δ so fink is a dominant strategy.

Unique SPNE for the 2 periods repeated game is (fink, fink) in

both periods.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

backward induction and many periods.

Now consider T ≥ 3 and assume the only SPNE for the repeated
game with T − 1 periods has (fink, fink) every period.

Then, discounted payoff for each player in a subgame starting at

period 2 (which has T − 1 periods) is
P −2 k 1−δ T −1
Π2 ≡ −2 T k=0 δ = −2 1−δ .

Hence, discounted payoffs in period 1 are those of the one shot

game + δΠ2 so “fink” is a strongly dominant strategy, which
implies that (fink, fink) at all periods is the only SPNE of the
T -periods repeated game.

Note that the argument does not depend on “fink” being a

strongly dominant strategy: all we need is that the one-shot game
has a unique Nash equilibrium.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Infinitely repeated prisoners’ dilemma.
Assume now an infinite number of repetitions.

The backward induction argument for T finite can be adapted to

show that (fink, fink) at all periods is an equilibrium: discounted
payoffs at date t are the one shot payoffs plus δΠt+1 = −2δ1−δ and so
(fink, fink) at period t is an equilibrium.

Now consider the following strategy: cooperate at period 1 and, at

any period t > 1, cooperate iff nobody has finked in previous
periods, else fink.

Suppose both players play this strategy.

Note that if somebody has finked before some period t, equilibrium

play is that both fink forever, which is a SPNE of the subgame
starting at t (that subgame is merely an infinitely repeated
prisoners’ dilemma).
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Now consider a subgame starting at period 1 or at some period
t > 1 where everybody has cooperated at all previous periods.

Equilibrium play in that subgame involves cooperation forever,

2
which yields a discounted payoff of 1−δ .

If a player deviates in period t and chooses fink with a positive

probability, her/his payoff would be 3 in period t and then a
−2δ
discounted payoff of 1−δ for the remaining periods: so there is an
immediate benefit of 3 − 2 = 1 to be compared to a future loss of
4δ
1−δ .

Then such a deviation is not profitable if δ ≥ 1/5.

This means that a deviation at some later time t ′ > t would not
be profitable either if δ ≥ 1/3 so that (cooperate, cooperate)
forever is indeed a Nash equilibrium of the subgame starting at t.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

So provided that there is not too much discounting, δ large
enough, cooperation can be sustained as a SPNE in the infinitely
repeated prisoners’ dilemma.

Strategy used to punish a deviation is sometimes called “grim

trigger strategy”: it is extreme and very harmful both for the
punished and the punisher.

For δ sufficiently close to 1, cooperation can be sustained with

other strategies, such as “tit for tat” where each player finks at
period t if his/her opponent finked at t − 1 but switches back to
cooperating at t + 1 if the other player had cooperated at period t.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Folk theorems.

In the infinitely repeated prisoners’ dilemma, it is actually possible

to sustain many outcomes in a SPNE other than full cooperation:
it can be shown that any strategies that guarantee each player an
average per period payoff that is “feasible” and strictly exceeds −2
on average can be sustained for δ sufficiently close to 1.

There are more general “folk theorems” applying to general

repeated games, where punishments can be harsher than playing a
Nash equilibrium of the one-shot game: in that case, punishment
periods cannot last for an infinite number of periods and this
allows to sustain payoffs which are below the Nash payoffs (in
Cournot for instance).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

6 Existence of equilibrium.

Existence arguments typically rely on the fixed point property of

Nash equilibrium: an equilibrium strategy profile is a description of
how the game is plaid by each player such that, if each player
knows this description, s(he) rationally chooses to play her/his
equilibrium strategy.

Key ingredients for the more standard existence results are:

continuity of payoff functions,
compactness and convexity of strategy spaces,
quasiconcavity of payoff functions.

These are standard sufficient (but not necessary) conditions.

We consider existence of both mixed and pure equilibria.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Continuity, compactness and convexity: the intermediate
value theorem.
Restricting attention to the real line, a compact and convex set is
a closed interval.

To gain intuition about fixed point arguments, consider the

following classic result of real analysis.

Proposition
(The intermediate value theorem) Let f be a real valued function
defined on an interval of the real line [a, b]. If f is continuous, then
for any y between f (a) and f (b) there exists x ∈ [a, b] such that
y = f (x).

A slightly more general version of the result says that the

continuous function f maps [a, b] into a closed interval containing
all values weakly between f (a) and f (b).
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.
Consider now a continuous function which maps [a, b] into itself
and define the function g on [a, b] by g (x) = f (x) − x.

Then, g (a) = f (a) − a ≥ 0 and g (b) = f (b) − b ≤ 0.

By the intermediate value theorem there exists x̄ ∈ [a, b] such that

g (x̄) = 0 or, f (x̄) = x̄: we call x̄ a fixed point of f .

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Brouwer fixed point theorem.

Proposition
(Brouwer fixed point theorem) Let f be a continuous function
mapping a non empty, compact and convex subset of a finite
dimensional euclidian space, K , into itself. Then f has a fixed
point in K .

Compact and convex subsets of euclidian apace Rn are cartesian

products of intervals of real numbers (e.g. [0, 1]2 ).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Best-response function.

Here we denote Ai the set of (pure or mixed) strategies for player

i, where a strategy is ai .

Consider some player i and some profile of strategies for all other
players a−i ∈ ×j̸=i Aj .

Consider the function γi which, to each a−i ∈ ×j̸=i Aj associates

the subset of Ai of all strategies for i which are best responses to
a−i .

So γi is a set valued function (or correspondence) which maps

×j̸=i Aj into the set of subsets of Ai , 2Ai .

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Now consider a set valued function, Γ, defined on the set of
strategy profiles ×N
i=1 Ai which, to each strategy profile a,
associates Γ(a) = (γ1 (a−1 ), ..., γN (a−N )).

Then a∗ is a Nash equilibrium if a∗ ∈ Γ(a∗ ) and we say that a∗ is a

fixed point of Γ.

We are looking for properties of the strategy spaces Ai , i = 1, ..., N

and of the function Γ that ensure that Γ has a fixed point.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Kakutani fixed point theorem.

Proposition
(Kakutani fixed point theorem) Let f be a set valued function
mapping some non empty, compact and convex subset of a finite
dimensional euclidian space, K into the set of subsets of K , 2K .
Then, if f is convex valued and has a closed graph, then it has a
fixed point in K .

Applying the theorem to Γ, assuming that it is convex valued and

has a closed set implies that, for each player i, the set of his best
responses to the other player’s strategies is convex and closed: if
the euclidian space is Rn , then these sets are cartesian products of
closed intervals in R.

Closed graph also ensures some continuity property for Γ (upper

hemi-continuity).

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Standard existence theorems.

In what follows
we assume there is a finite number of players N,
we use the notation Si for the set of pure strategies for player
i,
we call player i’s payoff function denoted Πi (si , s−i ) the
function that gives player i’s payoff as a function of the profile
of pure strategies,
we say that Πi is quasiconcave in si if, for any s−i and any
real number v , the set of pure strategies for player i, si such
that Πi (si , s−i ) ≥ v is convex.
The Kakutani fixed point theorem can be used to prove several
standard existence theorems.

Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.

Proposition
(Nash 1950) If Si is finite for all i = 1, ..., N, then the game has a
(possibly mixed) Nash equilibrium.

Proposition
(Glicksberg 1951) If Si is compact and convex and Πi is
continuous for all i = 1, ..., N, then the game has a (possibly
mixed) Nash equilibrium.

Proposition
If Si is compact and convex and Πi is continuous and quasiconcave
in si for all i = 1, ..., N, then the game has a pure Nash equilibrium.

There are also non standard existence results for mixed equilibria
with discontinuous payoffs (e.g. Dasgupta and Maskin, 1986).
Master’s in economics: 2nd year Chapter 3 Games, strategies and equilibrium.