Lectures 5 and 6

University of St. Gallen, HS23 Alia Gizatulina 117

 Mixed strategies Nash Equilibrium

University of St. Gallen, HS23 Alia Gizatulina 118

Mixed Strategies
Matching Pennies

I We have seen how to find a Nash equilibrium of a game. The equilibria

we were searching for were ”pure strategy Nash equilibria”.
I Why is there a qualifier – ”pure”? Well, because there also exist Nash
equilibria that are ”mixed strategy Nash equilibria”.
I Why ”mixed” and what does this mean?
Consider the following game, our known already ”Matching Pennies”:

Player 2
Heads Tails
Player 1 Heads 1, 1 1, 1
Tails 1, 1 1, 1

By looking at the best reply functions, we can see that the game does not have
any (pure) Nash equilibrium
A mixed strategy is a probability distribution over 2 or more pure strategies.

The players will choose randomly among their options in equilibrium.

University of St. Gallen, HS23 Alia Gizatulina 119

Mixed Strategies
Matching Pennies Player i is uncertain about the behavior of his opponents for many reasons.

I The key strategic feature: each player would like to outguess the other
I Similar situations arise in poker, battles, competitive games (football,
tennis...), etc:
I in poker, an analogous question is how often to blu↵: suppose that
player i is known never to blu↵, then i’s opponent will fold whenever
i bids aggressively, but this will make it worthwhile for i to blu↵...
I on the battlefield, an analogous question is where to attack:
suppose that the attackers can choose between two locations (by
land or by sea) and that the defense can counter either attack if
(and only if) it is anticipated correctly...
I Recall our exercise on a kicker and the goalie – the same spirit
The common features in such strategic situations: there is no Nash equilibrium
(in the sense we used so far – ”pure strategy Nash Equilibrium”).

But it does not mean that we cannot say anything about agents’ eventual
behaviour.

University of St. Gallen, HS23 Alia Gizatulina 120

Mixed Strategies
Matching Pennies

I How do we proceed in such setting to make a prediction about agents’

behaviour?
I We allow players to chose ”random” strategies: e.g., choosing some time
play si0 but some time si00
I The concept of mixed-strategy Nash equilibrium was introduced by
Harsanyi (1973)
I Distinction between pure strategies and mixed strategies
I pure strategies = di↵erent actions a player can take, e.g. ‘Heads’ or
‘Tails’
I mixed strategies = the probability distribution (q, 1 q), where
q is the probability of choosing Heads,
1 q is the probability of choosing Tails, and 0  q  1
I notice: pure strategy ‘Heads’ corresponds to mixed strategy (1,0)
and pure strategy ‘Tails’ corresponds to mixed strategy (0,1)

University of St. Gallen, HS23 Alia Gizatulina 121

Mixed Strategies
Matching Pennies
The probability of any event happening must be nonnegative,
and the sum of the probabilities of all the possible events must add up to one.

Definition
I Consider the n-player normal-form game G = {S1 , ..., Sn ; u1 , ..., un }.
I Suppose that player i has K pure strategies: Si = {si1 , ..., siK }.
I A mixed strategy for player i is a probability distribution i = ( i1 , ..., iK )
where ik is the probability that player i will play strategy sik ,
for k = 1, ..., K .
I Since ik is a probability, it must be that 0  ik  1 for k = 1, ..., K and
K
X
ik =1
k=1

i.e. i = { i (si1 ), ..., i (siK )} is an element of a simplex (Si )

University of St. Gallen, HS23 Alia Gizatulina 122

Mixed-Strategy Nash Equilibrium

Lets extend the definition of Nash equilibrium to allow for mixed strategies
I the earlier definition that we have used so far requires that each player’s
pure strategy is a best response to the other players’ pure strategies
I the extended definition requires that each player’s mixed strategy is a
best response to the other players’ mixed strategies
I since any pure strategy can be represented as a mixed strategy that puts
zero probability on all of the player’s other pure strategies, the extended
definition is more general – it covers the earlier one
A mixed strategy will be given by a cumulative distribution function

A belief for player i is a probability distribution over the strategies of his opponents.

University of St. Gallen, HS23 Alia Gizatulina 123

 By allowing a
Mixed-Strategy Nash Equilibrium player to
choose
Applying the extended definition to Matching Pennies probability
distributions,
Player 2 we do not
Heads (”q”) Tails (”1 q”) prohibit her
Heads 1, 1 1, 1 from choosing
Player 1 deterministic
Tails 1, 1 1, 1 actions.

I suppose player 1 believes that player 2 will play Heads with probability q and
Tails with probability 1-q We refer to such a mixed strategy as a pure strategy.
I lets look for player 1’s pure-strategy best response to player 2’s mixed strategies
I given his belief, player 1’s expected payo↵s are: With these
equations in hand,
v1 (H,q) = 1 - 2q q( 1) + (1 q)1 = 1 2q if he chooses Heads we can calculate
the best response
of player 1 for any
v1 (T,q) = 2q - 1 q1 + (1 q)( 1) = 2q 1 if he chooses Tails choice q of player
2.
I 1 2q > 2q 1 if and only if q < 1/2, then player 1’s best response is

If a player is randomizing Heads if q < 1/2

between two alternatives,
then he must be indifferent Heads or Tails if q = 1/2
between them
Tails if q > 1/2
In particular player 1 will prefer to play H over playing T if and only if v1(H, q) > v1(T, q)

University of St. Gallen, HS23 Alia Gizatulina 124

Mixed-Strategy Nash Equilibrium
Applying the extended definition to Matching Pennies
Take the sum of Player 2
each player’s
payoff from an Heads (”q”) Tails (”1 q”)
outcome, Heads (”r ”) rq 1, 1 1, 1 r (1-q)
multiplied by Player 1
the probability of Tails (”1 r ”) (1-r)(q) 1, 1 1, 1 (1-r) (1-q)
the outcome The players’ choices are assumed to be independent
I now let’s find player 1’s mixed-strategy best response to player 2’s mixed
strategy
I q still denotes player 1’s belief that player 2 will play Heads
I (r , 1 r ) denotes player 1’s mixed strategy, where r is the probability with
which player 1 plays Heads
I player 1’s expected payo↵ from playing mixed strategy (r , 1 r ) is
Player 1’s expected payoff = sum of expected payoffs from each outcome
r [q( 1) + (1 q)1] + (1 r )[q1 + (1 q)( 1)] = 2q 1 + r (2 4q)

I the expected payo↵ is increasing in r if 2 4q > 0, i.e., q < 1/2 and decreasing
in r if q > 1/2 , then player 1’s best response is Maximise the linear response
Each player chooses r =1 if q < 1/2 w.r.t r
their strategy independently,
the probability of each outcome Player 1 will definitely
r 2 [0, 1] if q = 1/2 play heads, if he knows that
is the product of the probabilities
of the strategies that make up the player 2 will play heads with
outcome. r =0 if q > 1/2 a probability of < 0.5

University of St. Gallen, HS23 Alia Gizatulina 125

Mixed-Strategy Nash Equilibrium
Applying the extended definition to Matching Pennies
I Let’s compare the result we found when considering only player 1’s
pure-strategy best response to player 2’s mixed strategy
Heads if q < 1/2

Heads or Tails if q = 1/2

Tails if q > 1/2

I with the result we found when considering player 1’s mixed-strategy best
response to player 2’s mixed strategy
r =1 if q < 1/2
To find a Nash equilibrium we are
looking for a pair of choices (p, q) for
which the two best-response r 2 [0, 1] if q = 1/2
correspondences cross.
r =0 if q > 1/2
I this last statement is stronger because it means that
if q < 1/2 Heads is the best response (among all strategies, pure or mixed) and
if q > 1/2 Tails is the best response (among all strategies, pure or mixed)

University of St. Gallen, HS23 Alia Gizatulina 126

Mixed-Strategy Nash Equilibrium
A formal statement of the extended definition

Definition
The expected payo↵ of player i when he chooses the pure strategy si 2 Si
and his opponents play the mixed strategy i 2 (S i ) is
X
vi (si , i ) = i (s i ) · vi (si , s i )
s i 2S i

Similarly:

Definition
The expected payo↵ of player i when he chooses the mixed strategy
i 2 (Si ) and his opponents play the mixed strategy i 2 (S i ) is
0 1
X X X
vi ( i , i ) = i (si )·vi (si , i (s i )) = @ i (si ) · i (s i ) · vi (si , s i )A
si 2Si si 2Si s i 2S i

University of St. Gallen, HS23 Alia Gizatulina 127

Mixed-Strategy Nash Equilibrium
A formal statement of the extended definition

Definition
The expected payo↵ of player i when he chooses the mixed strategy
i 2 (Si ) and his opponents play the mixed strategy i 2 (S i ) is
0 1
X X X
vi ( i , i ) = i (si )·vi (si , i (s i )) = @ i (si ) · i (s i ) · vi (si , s i )A
si 2Si si 2Si s i 2S i

That is,
I player 1’s expected payo↵ from playing the mixed strategy ( 11 , ..., 1J ) is
the weighted sum of the expected payo↵ from playing each of the pure
strategies (s11 , ..., s1J ) where the weights are the probabilities ( 11 , ..., 1J )

University of St. Gallen, HS23 Alia Gizatulina 128

Mixed-Strategy Nash Equilibrium
3. Lets give a formal statement of the extended definition

A profile α* of mixed strategies is a mixed strategy Nash equilibrium if and only if every player’s
mixed strategy is a best response to the other players’ mixed strategies.

Now we are ready to give the definition of a mixed-strategy Nash-Equilibrium

for a game with any finite number n of players, each having a finite set of pure
actions Si

Definition
The mixed-strategy profile = ( 1⇤ , ..., n⇤ ) is a Nash Equilibrium if for each
player i⇤ is a best response to ⇤ i , that is for all i 2 N
⇤ ⇤ ⇤
vi ( i , i) vi ( i , i)

for any i 2 (Si ).

University of St. Gallen, HS23 Alia Gizatulina 129

Mixed Nash Equilibrium
Graphical argument
Let’s use a graph to show that any two-player game in which each player has
two pure strategies has a Nash equilibrium (possibly in mixed strategies)
I recall the Matching Pennies game

Player 2
Heads Tails
Player 1 Heads 1, 1 1, 1
Tails 1, 1 1, 1
I we’ve already computed the values of r denoted r ⇤ (q) such that player 1’s mixed
strategy (r , 1 r ) is a best response to player 2’s mixed strategy (q, 1 q)

r ⇤ (q) = 1 if q < 1/2 r ⇤ (q) = 0 if q > 1/2 r ⇤ (q) 2 [0, 1] if q = 1/2

I now lets compute the values of q, denoted q ⇤ (r ), such that player 2’s mixed
strategy (q, 1 q) is a best response to player 1’s mixed strategy (r , 1 r )

q ⇤ (r ) = 0 if r < 1/2 q ⇤ (r ) = 1 if r > 1/2 q ⇤ (r ) 2 [0, 1] if r = 1/2

University of St. Gallen, HS23 Alia Gizatulina 130

Mixed Nash Equilibrium
Graphical argument
I lets represent r ⇤ (q) and q ⇤ (r ) on a graph

r ⇤ (q) = 1 if q < 1/2 r ⇤ (q) = 0 if q > 1/2 r ⇤ (q) 2 [0, 1] if q = 1/2

q ⇤ (r ) = 0 if r < 1/2 q ⇤ (r ) = 1 if r > 1/2 q ⇤ (r ) 2 [0, 1] if r = 1/2

Player 1 At q = 1/2, player 1 is indifferent between playing
r heads or tails as it doesn’t change his payoff. Thus,
any value of r is the best response to q = 1/2.
His behaviour would be random.
Pure Strategy: r = 1 Heads 1
-> Player 1 always play r*(q)
Heads.
r (q) = q - 1 + r (2 - 4q)
1/2 Mixed NE = (0.5, 0.5)
q*(r)

Pure Strategy: r = 0 Tails 0 q Player 2

-> Player 1 always
play Tails. 0 1/2 1
Tails Heads

University of St. Gallen, HS23 Alia Gizatulina 131

Mixed Nash Equilibrium
Graphical argument
I the intersection of the best-response correspondences r ⇤ (q) and q ⇤ (r ) yields the
mixed-strategy Nash equilibrium in the Matching Pennies game
I if player i plays (1/2,1/2), then player j’s best response is (1/2,1/2)
r

Heads 1
r*(q)
There is only one mixed
strategy Nash equilibria. 1/2
q*(r)
(0.5, 0.5) is a proper mixed
strategy.
Tails 0 q
0 1/2 1
Tails Heads

Again, we call r ⇤ (q) and q ⇤ (r ) correspondences instead of functions, since there exists a value of q such
that r ⇤ (q)ofhas
University St.more thanHS23
Gallen, one value, and there exists a value of r such
Alia that q ⇤ (r ) has more than one value. 132
Gizatulina
Mixed Nash Equilibrium
Observation: A player who randomizes between two actions (H,T) is indi↵erent
between playing H or T. This is the general feature of mixed NE:
Mixed strategies are not intuitive:
You randomize to make me indifferent.

Proposition
Let Si+ ⇢ Si denote the subset of pure strategies that player i plays with
⇤
positive probability in mixed strategy profile = ( 1 , .., I ). Strategy profile
is a Nash equilibrium in game G , if and only if for all i = 1, ..., n
If row player is
1. mixing on all of
⇤ her strategies in
vi (si , i) = vi (si0 , ⇤
i) = vi ( ⇤
i ,
⇤
i) for all si , si0 2 Si+ a NE, then each
must yield the
2. same expected
+ payoff.
vi (si , i) vi (si0 , i) for all si 2 Si+ and all si0 62 Si

Intuition: If a player is randomizing between two alternatives, then he must be

indi↵erent between them. If it was not the case, e.g. vi (si , 0
i ) > vi (si , i ),
si , si0 2 Si+ , then by reducing the probability of playing i⇤ (si0 ) to zero and increasing
⇤ ⇤ ⇤ 0 ⇤
i (si ) to i (si ) + i (si ), player i can increase his payo↵. Which means that i could
not be the best response to ⇤ i .

This provides us with guidance how to search for a mixed-strategy NE.

University of St. Gallen, HS23 Alia Gizatulina 133
Mixed Nash Equilibrium
Graphical argument

What is the interpretation of the concept of mixed-strategy Nash equilibrium?

I the mixed-strategy Nash equilibrium does not rely on any player choosing a
strategy truly at random
I we should interpret player i’s mixed strategy (1/2,1/2) in the Matching Pennies
as stemming from player i’s uncertainty about player j’s choice of a pure
strategy
I example: a striker and a goalkeeper face each other at a penalty kick
I the striker may decide whether to kick the ball left or right based, for
example, on his performance during pre-game practice
I if the goalkeeper understands how the striker will make his choice, but did
not observe the striker’s pre-game practice, then he may believe that the
striker is equally likely to kick the ball left or right
I the goalkeeper’s belief are represented by the striker’s mixed strategy
(1/2,1/2)
I in fact the striker chooses a pure strategy based on information
unavailable to the goalkeeper
A mixed strategy Nash equilibrium involves at least one player playing a
randomized strategy and no player being able to increase his or her expected
payoff by playing an alternate strategy.

University of St. Gallen, HS23 Alia Gizatulina 134

Mixed Nash Equilibrium
Graphical argument

Lets look at a second example of a mixed-strategy Nash equilibrium.

Consider the Battles of the Sexes.

Bill
Opera Basketball
Ann Opera 2, 1 0, 0
Basketball 0, 0 1, 2

I let (q, 1 q) be the mixed strategy in which Bill plays Opera with probability q
and Basketball with probability 1 q
I let (r , 1 r ) be the mixed strategy in which Ann plays Opera with probability r
and Basketball with probability 1 r

University of St. Gallen, HS23 Alia Gizatulina 135

Mixed Nash Equilibrium
Graphical argument

q 1-q
Bill
Opera Basketball
Opera 2, 1 0, 0 r
Ann
Basketball 0, 0 1, 2 1-r

I if Bill plays (q, 1 q) then Ann’s expected payo↵s are

q2 + (1 q)0 = 2q if she chooses Opera

q0 + (1 q)1 = 1 q if she chooses Basketball

I then Ann’s best response is 2q > 1 - q if and only if q > 1/3

Opera (r = 1) if q > 1/3

Opera or Basketball (r 2 [0, 1]) if q = 1/3

Basketball (r = 0) if q < 1/3

University of St. Gallen, HS23 Alia Gizatulina 136

Mixed Nash Equilibrium
Graphical argument

Bill
Opera Basketball
Ann Opera 2, 1 0, 0
Basketball 0, 0 1, 2

I if Ann plays (r , 1 r ) then Bill’s expected payo↵s are

r 1 + (1 r )0 = r if he chooses Opera

r 0 + (1 r )2 = 2(1 r) if he chooses Basketball

I then Bill’s best response is

Opera (q = 1) if r > 2/3

Opera or Basketball (q 2 [0, 1]) if r = 2/3

Basketball (q = 0) if r < 2/3

University of St. Gallen, HS23 Alia Gizatulina 137

Mixed Nash Equilibrium
Graphical argument
I lets graph Ann’s best response r ⇤ (q) and Bill’s best response q ⇤ (r )
r

Opera 1

2/3
q*(r)
r*(q)

Basketball 0 q
0 1/3 1
Basketball Opera

I the two best-response functions intersect in three points – three Nash Equilibria:
I (q,1-q)=(1/3,2/3) and (r,1-r)=(2/3,1/3) Mixed Strategy
I (q,1-q)=(0,1) and (r,1-r)=(0,1) Pure Strategy
I (q,1-q)=(1,0) and (r,1-r)=(1,0) Pure Strategy
matrix

University of St. Gallen, HS23 Alia Gizatulina 138

Mixed Nash Equilibrium
Graphical argument

I in any game, a Nash equilibrium (involving pure or mixed strategies)

appears as an intersection of the players’ best-response functions
I this is true even when there are more than two players and when some or
all players have more than two pure strategies
I however, a simple graphic representation is possible only for two-player
games in which each player has only two pure strategies (as in the
examples we have just seen)

University of St. Gallen, HS23 Alia Gizatulina 139

 Mixed Nash Equilibrium: Quick Exercises

Find all (pure and mixed strategy) Nash Equilibria in the following
games. What are the players’ expected payo↵s in games I and II?:
Player 2
q 1-q
Left Right
H D
r
Up 1, 3 1, 2
Player 1
H 1, 1 10, 0
1-r
Down 2, 1 0, 0
D 0, 10 1, 1
Game II
Game I H/W Done:
Game 1 + 2 + 3
L M R
L M R
U 0, 1 1, 2 1, 0
U 6, 10 2, 8 0, 7
C 1, 0 0, 1 1, 2
B 8, 2 4, 4 1, 1
D 1, 2 1, 0 0, 1
Game III
Game IV
In each game, find payo↵s from the mixed strategy equilibrium for each
player

University of St. Gallen, HS23 Alia Gizatulina 140

 Static Games of Complete Information: Various Topics

University of St. Gallen, HS23 Alia Gizatulina 141

Outline

The rest of the lecture: will collect various dribs and drabs in static games of
complete information:
I We extend the notion of rationalizability to allow for dominance of mixed
strategies
I We will study the question of existence of a Nash Equilibrium (informally)
I We will discuss can one select an equilibrium when there are many of
them

University of St. Gallen, HS23 Alia Gizatulina 142

Rationalizability and IESDS

Equipped with the mixed strategies, we now can extend the notion of
”dominated strategy” by allowing for mixed strategies

Definition
Let i 2 (Si ) and si0 2 (Si ) be possible strategies for player i, we say si0 is
strictly dominated by i if
Thus, player i can ignore strategy (σi).

ui ( i , s i ) > ui (si0 , s i )

for any s i 2S i
If player 2 is playing each of his strategies, L and C, with a
and relatedly probabilities p and 1 - p, respectively, then player 1 can calculate his
expected utility for each of his own strategies.

Definition
A strategy i is never a best response if there are no beliefs i 2 (S i ) for
player i for which i 2 BR( i )

A set of strategies is rationalizable if they survive (iterative) elimination of

never best-responses (including playing mixed strategies).
A rationalizable strategy is a strategy that is the best response for a
player, given some beliefs about the other players.
University of St. Gallen, HS23 Alia Gizatulina 143
 (σ2(L), σ2(C), σ2(R)) =
Rationalizability and IESDS Player 2 (0, 0.5, 0.5) strictly dominates
choosing L for sure because
q1 = 0 q2 = 0.5 1- q1 - q2 = 0.5
Mixing between this mixed strategy gives
Center and Right Left Center Right player 2 an expected payoff
gives a higher r1 = 0 Up 5, 1 1, 4 1, 0 of 2 if player 1 chooses U, of
payoff 2.5 if player 1 chooses M,
Player 1 r2 = 0.5 Middle 3, 2 0, 0 3, 5 and of 3.5 if player 1 chooses
1 - r1 - r2 = 0.5 Down 4, 3 4, 4 0, 3 D

I No pure strategy is strictly dominated by another pure strategy for any

player
I If we restrict attention to pure strategies then IESDS suggests that
anything can happen in this game
I Yet we can find mixed strategies that dominate pure strategies
I For example:
To consider a strategy as
1 1 strictly dominated,
These are random numbers ( 2 (L), 2 (C ), 2 (R)) = (0, , ) we no longer require that
2 2 some other pure strategy
strictly dominates choosing L for sure dominate it,
but allow for mixed
I Once only C and R are left, for player 1, strategy strategies to dominate it
This mixed strategy gives as well
player 1 an expected payoff of 2 1 1
if player 2 chooses C and ( 1 (U), 1 (M), 1 (D)) = (0, , )
1.5 if player 2 chooses R. 2 2
strictly dominates choosing U.
I Remaining 2 x 2 game cannot be reduced further: what are the equilibria
in this game? In this game, there is no strictly dominant strategy (i.e. no pure NE)
University of St. Gallen, HS23 Alia Gizatulina 144
 Steps to Derive Rationalisability:
Rationalizability 1. Start with the full action set for each player.
2. Remove actions which are never a best responses to any
belief about the opponents’ actions.
3. Repeat process with the opponents’ remaining actions
until no further actions are eliminated.
4. This leaves a non-empty set of actions for each player
those are the rationalizable actions.

We can connect three solution concept – Rationalizability, IESDS and NE in

SDS – as follows:

Result 1: In a two player-game the set of rationalizable strategy profiles

coincides with the set of strategies that survive iterated deletion of strictly
dominated strategies.

Result 2: In general, the set of rationalizable strategies is contained in the set

that survives iterated strict dominance since, with any number of players, a
strictly dominated strategy is never a best response.
Result 1: In games with more than 2 players, there may be strategies that are not strictly
dominated that are nonetheless never best responses.

By iterating on the knowledge of rationality, we iteratively delete strategies that are never best responses.

The set of strategies for a player that survives this iterated deletion of never best responses is called her set of
rationalizable strategies.

University of St. Gallen, HS23 Alia Gizatulina 145

Multiplicity of Nash Equilibria: Selection of a Specific
Equilibrium

I Many games have several Nash equilibria.

I When this is the case the assumption that a particular Nash equilibrium is
played relies on there being some mechanism or process that leads all
players to expect the same equilibrium.

I The idea of requiring more of a ”solution” than just to be a Nash

equilibrium is known in the literature as the program of ”perfecting” or
refining Nash equilibrium.

University of St. Gallen, HS23 Alia Gizatulina 146

How can we justify the play of a particular Nash
equilibrium?

I Weak-dominance: Choose the equilibrium that does not involve the play
of weakly dominated strategies.

I Trembling-hand perfection: Choose the equilibrium that is robust to the

possibility that players make mistakes.

I Pareto-dominance: Choose the equilibrium that pays players more than

any other does.

I Risk-dominance: Choose the equilibrium with the lowest ”strategic risk.”

I Evolution: Focal point established by playing repeatedly over time (we

will not study it here) Not In Syllabus

University of St. Gallen, HS23 Alia Gizatulina 147

Equilibrium Selection: Weak-dominance

I A Nash equilibrium can have a player using a weakly dominated strategy.

I For example, the following game have two Nash Equilibria, one is in
dominated strategies.

1\2 L R
U 1,1 0, 0
D 0, 0 0, 0

I In this game it makes sense to rule out the (D,R) NE.

University of St. Gallen, HS23 Alia Gizatulina 148

Equilibrium Selection: Weak-dominance

However, refining with weak dominance is not always noncontroversial.

Consider the game:

1\2 L R
U 10, 0 5, 2
D 10, 11 2, 0

I Strategy D is weakly dominated for player 1.

I So, using the weak dominance refinement we would rule out the (D,L)
NE.
I The problem with this is that (D,L) is better for both players than is
(U,R).

University of St. Gallen, HS23 Alia Gizatulina 149

Equilibrium Selection: Trembling-hand perfection.

I This caution leads us to define a di↵erent refinement of the concept of

Nash equilibrium, known as

) Trembling-hand perfect Nash equilibrium.

I The key idea is to look for Nash equilibria that remain to be Nash
equilibria even if with some very small probability, players make mistakes.

This concept, when used in a game of cards, can refer to a playing unintentionally playing
the wrong card through error (popularly known as tremble).

If a player acknowledges the possibility of an error occurring,

they can choose a trembling hand perfect equilibrium that will protect them in case
their opponent makes a mistake.

University of St. Gallen, HS23 Alia Gizatulina 150

Equilibrium Selection: Trembling-hand perfection.

Nash equilibrium is trembling hand perfect if, for some small trembles by each
Proposition player, the Nash equilibrium strategies remain best responses.
⇤ ⇤
A Nash equilibrium = ( 1,..., I ) of a normal form game is trembling-hand
perfect if and only if
1. There is some sequence { k }1 k=1 of totally mixed strategies (mixed
strategies in which every pure strategy receives positive probability)
converging to the equilibrium strategies, i.e., limk!1 k = .

2. Each player i’s equilibrium strategy i⇤ is a best response to every

element of the sequence { k i }k>k 0 for some k 0 sufficiently large and all
i = 1, . . . , I .

I Note that Nash equilibrium ⇤ = ( 1⇤ , . . . , I⇤ ) can be a pure strategy

Nash equilibrium, i.e. with degenerate probability measures
I Intuitively, we want for each i play i⇤ remain to be a best reply to players’
choices also when they are made in a ”neighbourhood” of their ⇤ i

University of St. Gallen, HS23 Alia Gizatulina 151

Equilibrium Selection: Trembling-hand perfection.

(L,R) is not a NE as there is another combination of strategies, (U, L) where both players obtain
greater payoffs.

I The strategy profile (U,L) is a trembling-hand perfect Nash equilibrium

but the strategy profile (D,R) is not.

1\2 L R The 2 NE is determined

U 1,1 0, 0 by finding the best response
functions of both row and column
D 0, 0 0, 0 players.

Let us see why:

Nash Equilibria: (U, L)

Strategy D is a weakly dominated strategy for each player.

By playing strategy U or L, each player guarantees a payoff at least as high (and sometimes a higher) than
that obtained by playing D.

Thus, we eliminate D because it is based on weakly dominated strategies.

So we propose the strategy profile (U, L) as the outcome of the game.

University of St. Gallen, HS23 Alia Gizatulina 152

Equilibrium Selection: Trembling-hand perfection.
1 - 1/k 1/k

What happens when 1\2 L R Assume that players can tremble.

players may make
mistakes? 1 - 1/k U 1,1 0, 0 When a player intends to play a
strategy s, with probability k,
1/k D 0, 0 0, 0 Nature switches it to the other
strategy s’.

I For (U,L) consider the following sequence of totally mixed strategies

✓ ◆ ✓ ◆
1 1 1 1
( 1k , 2k ) = [( Uk , Dk ), ( Lk , Rk )] = 1 , , 1 ,
k k k k

I We have that:
Suppose
that the k k
column
1. limk!1 1 = (1, 0) = [U] limk!1 2 = (1, 0) = [L]
player plays
(1 − 1/k, 1/k),
then the row
2. And U and L are best responses since: For Player 1, playing U
is better than D
player’s
payoff is: ✓ ◆ ✓ ◆
k 1 1 1 k 1 1
u1 (U, 2) =1 1 +0 =1 u1 (D, 2)
=0 1 +0 =0
k k k k k
✓ ◆ ✓ ◆
1 1 1 1 1
u2 ( 1k , L) = 1 1 +0 =1 u2 ( 1k , R) = 0 1 +0 =0
k k k k k
Suppose that the row player plays (1 − 1/k, 1/k), then the column player’s payoff is:
University of St. Gallen, HS23 Alia Gizatulina 153
Equilibrium Selection: Trembling-hand perfection.
1/k 1 - 1/k

1\2 L R
1/k U 1,1 0, 0
1 - 1/k D 0, 0 0, 0

I For (D,R) consider the following sequence of totally mixed strategies

✓ ◆ ✓ ◆
k k k 1 1 1 1
= ( 1, 2) = ,1 , ,1
k k k k

I We have that:
k k
1. limk!1 1 = (0, 1) = [D] limk!1 2 = (0, 1) = [R]

2. But D and R are not best responses since:

Suppose that the column player plays (1/k, 1 - 1/k), then the row player’s payoff is:
✓ ◆
The row player k k 1 1 1
is better off u1 (D, 2) = 0 < u1 (U, 2) =1
+0 1 =
playing U than D k k k
✓ ◆
The column 1 1 1
player
u2 ( 1k , R) = 0 < u2 ( 1k , L) = 1 + 0 1 =
k k k
is better off
playing L than R Suppose that the row player plays (1/k, 1 - 1/k), then the column player’s payoff is:
University of St. Gallen, HS23 Alia Gizatulina 154
Equilibrium Selection: Trembling-hand perfection.

Proposition
If = ( 1 , . . . , I ) is a (normal form) trembling-hand perfect Nash equilibrium,
then i is not a weakly dominated strategy for any i = 1, . . . , I .

I Hence, in any (normal form) trembling-hand perfect Nash equilibrium, no

weakly dominated strategies can be played with positive probability.
I Any finite game has a trembling-hand perfect equilibrium (potentially in
mixed strategies).
I Any strict Nash equilibrium is trembling-hand perfect.
I So trembling-hand perfection rules out Nash equilibria involving play of
weakly dominated strategies.
I The converse, that any NE not involving play of a weakly dominated
strategy is necessarily trembling-hand perfect, turns out to be true for
two-player games but not for games with more than two players.
I Hence, trembling-hand perfection can rule out more than just Nash
equilibria involving weakly dominated strategies.

University of St. Gallen, HS23 Alia Gizatulina 155

How can we justify the play of a particular Nash
equilibrium? Harsanyi and Selten General Theory

I The above suggest that strict Nash Equilibria cannot be ruled out by the
requirement of trembling-hand perfection.
I How can we select among strict Nash equilibria?
I Harsanyi and Selten suggested two refinements:
I Payo↵ (or Pareto) dominance
I Risk dominance
Payo↵ dominance is rather obvious: players will chose as a focal point payo↵
dominant equilibrium.

Risk dominance?

Let us see how both work

University of St. Gallen, HS23 Alia Gizatulina 156

How can we justify the play of a particular Nash
equilibrium? Harsanyi and Selten Theories

The Stag Hunt Game

I Recall it from our first homework or watch a real life version of it
https://www.youtube.com/watch?v=3__L0oAa2T8
(”stag” is depicted, as for ”hare” you can think about e.g. herrings)
The story of the Stag Hunt Game:
I In the stag hunt, two hunters must each decide whether to hunt the stag
together or hunt hares alone.
I Half a stag is better than a brace of hares, but the stag will only be
brought down with a combined e↵ort.
I Hares, on the other hand, can be hunted by an individual without any
trouble.
In game theory, the stag hunt is a game which describes a conflict between safety and
social cooperation.
2 individuals go out on a hunt, each can individually choose to hunt a stag or hunt a hare.
Each player must choose an action without knowing the choice of the other.

University of St. Gallen, HS23 Alia Gizatulina 157

Equilibrium Selection: Pareto-dominance.
This is a symmetric game. Exam will test this - 5 marks will be allocated!

The two-player stag hunt game (coordination game, where S corresponds to

stag hunt, and H to hares hunt) (S,S) is payoff / Pareto dominant,
p 1-p where both players obtain the highest
A stag hunt is a game with two payoff if they choose to cooperate
pure strategy Nash equilibria - one 1\2 H S in hunting for the Stag together.
that is risk dominant another that H x,x x, 0 1. Other hunter plays “Stag” →
is payoff dominant.
“Stag” is best response.
S 0, x 1,1
NE: (H,H) and (S,S)
2. Other hunter plays “Hare” →
“Hare” is best response.
Here, 0 < x < 1
I Everyone choosing S is a Nash equilibrium, but reaching it for sure
depends on everyone thinking everyone else is likely to choose S.
I Everyone choosing H is also a Nash equilibrium, but it pays less than if
players could somehow coordinate on everyone choosing S.
I The Pareto-dominance criterion predicts that people will play S in the
stag hunt game. For all players,
No one strictly prefers H to S—that is, for all i,
Ui(H) ≥ Ui(S).

University of St. Gallen, HS23 Alia Gizatulina 158

Equilibrium Selection: Risk-dominance
If the opponent chooses S, you will
be left with choosing H, which
On the other hand: makes you worse off.

I Choosing S is ”riskier” than choosing H in the sense payo↵ drop in the

case that the opponent chooses H (not then an equilibrium action when i
chooses S) is higher as compared to the drop of the payo↵ when playing
(H,H) and opponent chooses S.
The problem is that if one choose
I That is the losses are: S, one will be able to kill a stag
only if the other has also chosen S.
Indifference principle:
Given the other hunter plays S, Choosing H is playing it safe.
equal expected payoff for Stag and Hare 1 0>x x =0
for any x
I Here ”riskier” should be interpreted as strategic uncertainty and not as
the decision theoretical notion of risk.
I So, the risk-dominance criterion, proposed in Harsanyi and Selten (1988),
predicts that if 0 < x < 1, then people should rather play L in the stag
H
hunt game.

University of St. Gallen, HS23 Alia Gizatulina 159

Existence of Nash Equilibrium
Nash Theorem
The following is quite technical and will not be at the exam. But very good
game-theorists as you are, you should be aware of the topic
I We can now discuss the issue of existence of an equilibrium in a given
game – pure or mixed equilibrium
I Nash Theorem (1950) guarantees that any game with a finite number of
players, each of whom has a finite number of pure strategies, has a Nash
equilibrium.

Theorem: Existence of Nash-Equilibrium

In the n-player normal-form game G = {S1 , ..., Sn ; u1 , ..., un }, if n is finite and
Si is finite for every i then there exists at least one Nash equilibrium, possibly
involving mixed strategies.

Proof
It involves a fixed-point theorem.

University of St. Gallen, HS23 Alia Gizatulina 160

Existence of Nash Equilibrium
Nash Theorem
Lets look at a simple example of a fixed-point theorem: Brouwer’s fixed-point
theorem
I suppose f (x) is a continuous function with domain [0, 1] and range [0, 1]
I then, there exists at least one fixed point, that is, there exists at least one
value x ⇤ in [0, 1] such that f (x ⇤ ) = x ⇤

f(x)

f(x*)

45˚
0 x
0 x* 1

University of St. Gallen, HS23 Alia Gizatulina 161

Existence of Nash Equilibrium
Nash Theorem

Lets look at a simple example of a fixed-point theorem.

I if f (x) is not a continuous function, as in the figure below
I then it cannot be guaranteed that there exists at least one fixed point

University of St. Gallen, HS23 Alia Gizatulina 162

Existence of Nash Equilibrium
Nash Theorem

Applying a fixed-point theorem to prove Nash theorem involved two steps:

1. show that any fixed point of a certain correspondence is a Nash
equilibrium
2. use an appropriate fixed-point theorem to show that this correspondence
must have a fixed point
The relevant correspondence is the n-player best-response correspondence.

The relevant fixed-point theorem is the Kakutani fixed-point theorem (Brower’s

fixed point theorem is the special case of Kakutani).

University of St. Gallen, HS23 Alia Gizatulina 163

Existence of Nash Equilibrium
Nash Theorem

1. Show that any fixed point of a certain correspondence is a Nash equilibrium.

I The collection of best response correspondences
BR ⌘ BR1 ⇥ BR2 ⇥ ... ⇥ BRn maps (S) = (S1 ) ⇥ ... (Sn ) the set of
profile of mixed strategies onto itself. That is, BR : (S) ◆ (S) takes
every element 2 (S) and converts it into a subset BR( 0 ) ⇢ (S)
I A mixed strategy profile ⇤ 2 (S) is a Nash equilibrium if and only if it
is a fixed point of the collection of best response correspondences
⇤
2 BR( ⇤ ).
2. Apply the fixed point theorem
I The players’ expected utility is continuous in , this means that BR is
continuous too
I We can apply the fixed-point theorem.

University of St. Gallen, HS23 Alia Gizatulina 164

Homework

Your homework:

Chapter 6 of Tadelis, Exercises 6.3, 6.5, 6.7, 6.9, 6.11

Next week: dynamic games of complete information...

University of St. Gallen, HS23 Alia Gizatulina 165