Correlated Strategies and Correlated Equilibrium

Sanjay Singh

1
Department of Information and Communication Technology
Manipal Institute of Technology, MAHE
Manipal-576104, INDIA
sanjay.singh@manipal.edu
2
Centre for Artificial and Machine Intelligence (CAMI)
MAHE, Manipal-576104, INDIA

April 29, 2021

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Definition (Non-cooperative Games)

In non-cooperative games
Players have no binding agreement on joint actions that are
optimal for the group
Every player maximizes his own payoff by choosing his best
response on the basis of what he knows about other’s actions
Players are not involved in any pre-play communication stage

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 Cooperative and Non-Cooperative Games
Is the game cooperative or non-cooperative?
Even if players negotiate, the question is whether the results of
the negotiations can be enforced
If not, a player can always move differently from what was
promised in the negotiation (non-cooperative game)
A cooperative game is one where the results of the negotiations
can be put into contract and be enforced

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Module Outline

Games with contracts

Games with communication
Correlated strategies
Correlated equilibrium

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 2
1
x2 y2
x1 2,2 0,6
y1 6,0 1,1

Above game has a unique equilibrium, (y1 , y2 ) that yields a

payoff (1,1)
Non-equilibrium outcome (x1 , x2 ) yields higher payoffs (2,2)
In such a situations, players may like to transform the game to
extend the set of equilibria to include better outcomes
There could be several ways of achieving this transformation:
1 Players may communicate among themselves to coordinate their
moves
2 Players may formulate agreements in the form of contracts
3 Players may decide to play the game repeatedly

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Games with Contracts

In a game with contracts, a player who signs a contract is

required to play according to a designated strategy, called
correlated strategy
Contract transform games with less desirable equilibria to games
with more desirable equilibria

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Modified Prisoner’s Dilemma with a Contract
Let us say the two players sign the following contract (call it contract 1)
1 If both players sign this contract, then player 1 (player 2) chooses to
play the strategy x1 (x2 )
2 If the contract is signed by only player 1, player 1 would choose y1
3 If the contract is signed by only player 2, player 2 would choose y2

Let ai (i = 1, 2) denotes action of signing the contract by player i

We can expand the strategy set as S1 = {x1 , y1 , a1 } and
S2 = {x2 , y2 , a2 }

2
1
x2 y2 a2
x1 2,2 0,6 0,6
y1 6,0 1,1 1,1
a1 6,0 1,1 2,2

Sanjay Singh Correlated Strategies and Correlated Equilibrium

2
1
x2 y2 a2
x1 2,2 0,6 0,6
y1 6,0 1,1 1,1
a1 6,0 1,1 2,2

Transformed game has a new equilibrium (a1 , a2 ) which is a

weakly dominant strategy equilibrium yielding payoffs (2, 2)
Profile (y1 , y2 ) continues to be an equilibrium but it is not a
dominant strategy equilibrium
Even better payoff could be achieved if a second contract (call it
contract 2) is introduced in addition to contract 1 above

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Modified Prisoner’s Dilemma with Additional Contract
Contract 2 commits the players to a correlated strategy. It is as follows:
If both players signs this new contract, then a coin will be tossed
On event heads they will implement (x1 , y2 ), and
On event tails they will implement (y1 , x2 )
If player 1 alone signs contract 2, then player 1 chooses y1
If player 2 alone signs contract 2, then player 2 chooses y2
If b1 and b2 denotes the actions of player 1 and 2 corresponding to signing
contract 2, the extended matrix is given by

2
1
x2 y2 a2 b2
x1 2,2 0,6 0,6 0,6
y1 6,0 1,1 1,1 1,1
a1 6,0 1,1 2,2 1,1
b1 6,0 1,1 1,1 3,3

Sanjay Singh Correlated Strategies and Correlated Equilibrium

2
1
x2 y2 a2 b2
x1 2,2 0,6 0,6 0,6
y1 6,0 1,1 1,1 1,1
a1 6,0 1,1 2,2 1,1
b1 6,0 1,1 1,1 3,3

The new game has following equilibria:

(y1 , y2 ) with payoffs (1, 1)
(a1 , a2 ) with payoffs (2, 2)
(b1 , b2 ) with payoffs (3, 3)
2 1 2 1 2 1
((0, 0, , ), (0, 0, , )) where mixed strategy (0, 0, , ) for
3 3 3 3 3 3
2 1
player 1 means a1 with probability and b1 with probability .
3 3
5
This equilibrium leads to a payoff of for both players
3

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 X
Using the relation ui (σ1 , σ2 ) = σ(s1 , s2 )ui (s1 , s2 )
(s1 ,s2 )∈S
S = S1 × S2 = (x1 , y1 , a1 , b1 ) × (x2 , y2 , a2 , b2 )
(a1 , a2 ), (a1 , b2 ), (b1 , a2 ), (b1 , b2 ) ∈ S have non zero value

u1 (σ1 , σ2 ) = σ1 (a1 )σ2 (a2 )u1 (a1 , a2 ) + σ1 (a1 )σ2 (b2 )u1 (a1 , b2 )+
+ σ1 (b1 )σ2 (a2 )u1 (b1 , a2 ) + σ1 (b1 )σ2 (b2 )u1 (b1 , b2 )
2 2 2 1 1 2 1 1
= . .2 + . .1 + . .1 + . .3
3 3 3 3 3 3 3 3
8 2 2 1
= + + +
9 9 9 3
5
=
3
5
Similar calculation yields payoff for player 2, u2 (σ1 , σ2 ) =
3

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Correlated Strategies

Choice of pure strategies by players in a game may be correlated

It is because players are observing the same or related random
events before making their choices
A correlated strategy captures the above feature
A correlated strategy also captures cooperative play by members
of a coalition
Definition (Correlated Strategy)
Let Γ = hN, (Si ), (ui )i be a strategic form game. A correlated strategy
for a non-empty subset C (aka coalition) of the players is any
probability distribution over the set of possible combinations of pure
strategies that these players can choose.

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 A correlated strategy, τC for a given coalition C belongs to ∆(SC ),
where SC = ×i∈C Si
N is called the grand coalition and symbol τN denotes a correlated
strategy of the grand coalition

Example
Let N = {1, 2, 3}; S1 = {x1 , y1 }; S2 = {x2 , y2 }; S3 = {x3 , y3 , z3 }. If
C = {2, 3} then

SC = ×i∈C Si
= ×i∈{2,3}Si
= S2 × S3
= {(x2 , x3 ), (x2 , y3 ), (x2 , z3 ), (y2 , x3 ), (y2 , y3 ), (y2 , z3 )}

A correlated strategy for the coalition C is a probability distribution on SC .

1 1 1 1 1 1
For example, ( , , , , , ) would correspond to (x2 , x3 ) with
4 4 4 12 12 12
1 1
probability , (x2 , y3 ) with probability and so on
4 4
Sanjay Singh Correlated Strategies and Correlated Equilibrium

A correlated strategy τC ∈ ∆(×i∈C Si ) can be implemented as

follows:
A reliable mediator or a referee picks randomly a profile of pure
strategies in SC according to distribution τC
Mediator ask each player to play the strategy chosen in this pure
strategy profile
Let S denote S = SN = ×i∈N Si
Suppose α ∈ ∆(S) is any correlated strategy for all players
Let Ui (α) denote expected
X payoff to player i when α is
implemented, Ui (α) = α(s)ui (s)
s∈S
Let U(α) = (U1 (α), . . . , Un (α)) denote the expected payoff
allocation to players when they implement α
Given any allocation in the set {U(α)|α ∈ ∆(S)}, there exists a
contract such that if all the players signed this contract, then they
would get this expected payoff allocation
When there is no confusion, we will use u(α) instead of U(α)
Sanjay Singh Correlated Strategies and Correlated Equilibrium
Correlated and Mixed Strategies

What is the difference between correlated strategy and mixed

strategies?
A correlated strategy for the grand coalition N is a member of
∆(×i∈N Si )
A mixed strategy profile is a member of ×i∈N (∆(Si ))
Given any mixed strategy profile σ of all players, we can find a
correlated strategy of all players α such that
ui (α) = ui (σ) ∀i ∈ N
However, given a correlated strategy α of all players, it may not
always be possible to find any mixed strategy profile σ such that
ui (σ) = ui (α) ∀i ∈ N

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Contract Signing Game

A vector of correlated strategies of all possible coalitions is called a

contract, formally
Definition (Contract)
Consider the vector τ = (τC )C⊆N . Note that

τ ∈ ×C⊆N (∆(×i∈C Si ))

The vector τ of correlated strategies of all coalitions is a contract.

τC for C ⊆ N gives correlated strategy that would be

implemented by players in C if C were the set of players to sign
the contract
A contract defines an extended game and this extended game is
called the contract signing game

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 Example

2
2 1
1 x2 y2 a2 b2
x2 y2 a2
x1 2,2 0,6 0,6 0,6
x1 2,2 0,6 0,6
y1 6,0 1,1 1,1 1,1
y1 6,0 1,1 1,1
a1 6,0 1,1 2,2 1,1
a1 6,0 1,1 2,2
b1 6,0 1,1 1,1 3,3

Consider the modified prisoner’s dilemma game with contract

Payoff matrix defining the contract signing game induced by contract 1
and 2 is given above
Contract 1 is described by (τ1 , τ2 , τ{1,2} ) where
τ1 = (x1 : 0, y1 : 1); τ2 = (x2 : 0, y2 : 1); τ{1,2} = ((x1 , x2 ) : 1; (x1 , y2 ) :
0; (y1 , x2 ) : 0; (y1 , y2 ) : 0)
Contract 2 is given by (τ1 , τ2 , τ{1,2} ) where
τ1 = (x1 : 0, y1 : 1); τ2 = (x2 : 0, y2 : 1); τ{1,2} = ((x1 , x2 ) : 0; (x1 , y2 ) :
1 1
; (y1 , x2 ) : ; (y1 , y2 ) : 0)
2 2
Sanjay Singh Correlated Strategies and Correlated Equilibrium

Mathematical Digression
Mathematical Analysis

Definition (Metric Space)

A metric space (V, d) consists of a set V and a mapping
d : V × V 7→ R such that ∀x, y, z ∈ V, the following holds:
1 d(x, y) ≥ 0
2 d(x, y) = 0 iff x = y
3 d(x, y) = d(y, x)
4 d(x, z) ≤ d(x, y) + d(y, z).
The mapping d is called a metric or distance function.

Definition (Open Ball)

Given a metric space (V, d), an open ball of radius r > 0 and center
x ∈ V, is the set B(x, r) = {y ∈ V : d(x, y) < r}.

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Definition (Open Set)
An open set X in a metric space (V, d) is a subset of V such that we can find,
at each x ∈ X, an open ball that is contained in X.

Definition (Bounded Set)

A subset X of a metric space (V, d) is said to be bounded if X is completely
contained in some open ball, around 0, with finite radius.

Definition (Closed Set)

A subset X of a metric space (V, d) is said to be a closed set iff every
convergent sequence in X converges to a point which lies in X. That is, for
all sequences {xk } ∈ X such that xk → x for some x ∈ V, it will happen that
x ∈ X. A set X is closed iff the complement set X c = V\X is an open set.

Definition (Compact Set)

Given a subset X of a metric space (V, d), X is said to be compact if every
sequence of points in X has a convergent subsequence. A key result is that if
the metric space V ∈ Rn (under Euclidean metric), then a subset X is
compact iff it is closed and bounded.
Sanjay Singh Correlated Strategies and Correlated Equilibrium

Example (Closed and compact set)

The interval A = (−∞, −2] is not compact because it is not

bounded
The interval C = (2, 4) is not compact because it is not closed
The interval B = [0, 1] is compact because it is both closed and
bounded

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Concept of Individual Rationality

2
1
x2 y2
x1 2,2 0,6
y1 6,0 1,1

Players may not be willing to sign any contract that is proposed

For example, in MPD game, player 1 would not agree to sign a
contract that would commit the players to implement (x1 , y2 )
since it gives him 0 payoff
Player 1 can always guarantee himself a payoff of 1 by not
signing any contract and simply choosing y1
We’ll now investigate which contracts a player would be
interested in signing

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Maxmin value (security level) is the minimum guaranteed payoff

the player can assure himself when the rest of the players are free
to play any strategies
Security level is defined as

vi = max min ui (τi , sN−i )

τi ∈∆(Si ) sN−i ∈SN−i

where SN−i = SN\{i} ; sN−i ∈ SN−i

The expected utility ui (τi , sN−i ) is given by
X
ui (τi , sN−i ) = τi (si )ui (si , sN−i )
si ∈Si

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 2
1
x2 y2
x1 2,2 0,6
y1 6,0 1,1

v1 = max min u1 (τ1 , s2 ) where

τ1 ∈∆(S1 ) s2 ∈S2
X
u1 (τ1 , s2 ) = τ1 (s1 )u1 (s1 , s2 )
s1 ∈S1
X
= τ1 (s1 )u1 (s1 , s2 )
s1 ∈{x1 ,y1 }

= τ1 (x1 )u1 (x1 , s2 ) + τ1 (y1 )u1 (y1 , s2 )

u1 (τ1 , x2 ) = τ1 (x1 )u1 (x1 , x2 ) + τ1 (y1 )u1 (y1 , x2 )
u1 (τ1 , y2 ) = τ1 (x1 )u1 (x1 , y2 ) + τ1 (y1 )u1 (y1 , y2 )

u1 (τ1 , x2 ) = 0.2 + 1.6 = 6; u1 (τ1 , y2 ) = 0.0 + 1.1 = 1

v1 = max min{6, 1} = 1, similarly v2 = 1
Sanjay Singh Correlated Strategies and Correlated Equilibrium

Individually Rational Correlated Strategy

It is reasonable for player i to sign a contract to play a correlated

strategy α only if ui (α) ≥ vi
It is called individual rationality or participation constraints for
player i

Definition (Individually Rational Correlated Strategy)

A correlated strategy α ∈ ∆(S1 × . . . × Sn ) for all the players in N is
said to be individually rational if

ui (α) ≥ vi ∀i ∈ N.

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 Using a two player ZSG interpretation of individual rational
correlated strategy, the security level vi also satisfies

vi = min max ui (si , τN−i )

τN−i ∈∆(SN−i ) si ∈Si

where τN−i = τN\{i} ∈ ∆(SN−i )

The expected utility ui (si , τN−i ) is given by
X
ui (si , τN−i ) = τN−i (sN−i )ui (si , sN−i ).
sN−i ∈SN−i

The vi is the best expected payoff that player i is guaranteed to

get against any correlated strategy that other players could use
against player i
A minmax correlated strategy against player i is any correlated
strategy τN−i ∈ ∆(SN−i ) of the rest of the players which forces
the payoff of player i to be equal to his minmax value vi

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Equilibria of the Contract Signing Game

Suppose players make their decisions about which contract to sign

independently, then following proposition holds
Proposition
Given any individually rational correlated strategy α, there exists a
contract τ with τN = α such that all players signing this contract is a
Nash equilibrium of the contract signing game.

Proposition
Consider any Nash equilibrium (s∗i , s∗−i ) of a contract signing game
induced by a correlated strategy α. Then, ui (s∗i , s∗−i ) ≥ vi ∀i ∈ N .

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Example: Payoff Allocations in MPD Game

2
1
x2 y2
x1 2,2 0,6
y1 6,0 1,1

Consider the MPD game

The expected payoff in a correlated strategy is a convex
combination of payoffs in different strategy profiles, hence set of
all payoff allocations for MPD game is the convex set with
extreme points (0,6),(6,0), and (1,1)
Sanjay Singh Correlated Strategies and Correlated Equilibrium

See figure on last slide, and note that set of points (0,6),(6,0), and
(1,1) form a closed set and the point (2,2) is in the interior of this
convex set
The set of possible expected payoff allocations satisfying
individual rationality is the triangle with corners at (1,1), (5,1)
and (1,5) as shown
It is because v1 = 1 and v2 = 1 for this example
Note that this set is also closed and convex

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Games with Communication

We know that contracts can transform a game with less desirable

equilibria to a game with more desirable equilibria
However, many a times players may not be able to commit
themselves to binding contracts
There could be many reasons for this
Player’s strategies may not be observable to the enforcers of
contracts
There may not be adequate sanctions to guarantee compliance
with contracts
Player’s strategies might actually involve inalienable rights (i.e.,
players may not be able to transfer their rights)
In the above mentioned situations also, it is still possible for
players to communicate and coordinate among themselves, and
achieve a self-enforcing equilibrium with desirable payoff
Such games corresponds to games with communication

Sanjay Singh Correlated Strategies and Correlated Equilibrium

A game with communication is one, in which, in addition to the

strategy options explicitly specified, players have a range of
implicit options to communicate with each other
A game with communication need not have any contracts

Example (A Game with Communication)

2
1
x2 y2
x1 5,1 0,0
y1 4,4 1,5

This game has three Nash equilibria

(x1 , x2 ) with payoff allocation (5,1)
(y1 , y2 ) with allocation (1,5)
1 1 1 1
Mixed strategy Nash equilibrium (( , ), ( , )) which yields
2 2 2 2
the outcome (2.5,2.5)

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Example (.contd)
Payoff with mixed
X strategy is computed by
ui (σ1 , σ2 ) = σ(s1 , s2 )ui (s1 , s2 )
(s1 ,s2 )∈S
Here S = S1 × S2 = {(x1 , x2 ), (x1 , y2 ), (y1 , x2 ), (y1 , y2 )} and
σ1 = σ2 = (1/2, 1/2)

u1 (σ1 , σ2 ) = σ(x1 , x2 )u1 (x1 , x2 ) + σ(x1 , y2 )u1 (x1 , y2 ) +

+ σ(y1 , x2 )u1 (y1 , x2 ) + σ(y1 , y2 )u1 (y1 , y2 )
= σ1 (x1 )σ2 (x2 )u1 (x1 , x2 ) + σ1 (x1 )σ2 (y2 )u1 (x1 , y2 ) +
+ σ1 (y1 )σ2 (x2 )u1 (y1 , x2 ) + σ1 (y1 )σ2 (y2 )u1 (y1 , y2 )
1 1 1 1 1 1 1 1
= . .5 + . .0 + . .4 + . .1
2 2 2 2 2 2 2 2
5
=
2
5
Similarly, we compute u2 (σ1 , σ2 ) =
2
Sanjay Singh Correlated Strategies and Correlated Equilibrium

Example (.contd)
Note that (y1 , x2 ) is not a NE, though it is a desirable outcome
It can be realized through a binding contract
If contracts can not be used, we are note sure whether we can
achieve the payoff profile (4,4 or even anything better than
(2.5,2.5))
Correlated strategies exists that achieve better allocation than
(2.5,2.5) (see next example)

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Example (Correlated Strategy 1)
Let us say players choose to toss a coin and select outcome
(x1 , x2 ) with probability 1/2 and (y1 , y2 ) with probability 1/2
It refers to following correlated strategy
 
1 1
α = (x1 , x2 ) : ; (x1 , y2 ) : 0; (y1 , x2 ) : 0; (y1 , y2 ) :
2 2

players when α is implemented is

The expected utility ofX
Ui (α) = α(s)ui (s)
s∈S
u1 (α) = α(x1 , x2 )u1 (x1 , x2 ) + α(x1 , y2 )u1 (x1 , y2 )+
computed as + α(y1 , x2 )u1 (y1 , x2 ) + α(y1 , y2 )u1 (y1 , y2 )
1 1
= .5 + 0.0 + 0.4 + .1
2 2
=3
Similarly, u2 (α) = 3
Sanjay Singh Correlated Strategies and Correlated Equilibrium

To implement this correlated strategy, players can toss a coin and

choose outcome (x1 , x2 ) in the event of a head and choose
outcome (y1 , y2 ) in the event of tail
To implement this, communication and coordination is required
The outcome suggested by this correlated strategy will be
self-enforcing since neither player will gain by unilaterally
deviating from this

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 The above correlated strategy can also be implemented with the help of a
trusted mediator who helps the players to communicate and share information:
mediator recommends, randomly with probability 0.5 for each profile (x1 , x2 )
and (y1 , y2 )
Assume that each player comes to know only the strategy recommended to him
by the mediator
Player 1, if recommended x1 by the mediator thinks that player 2
is recommended x2 . Believing that player 2 obeys the
recommendation x2 , player 1 finds it a best response to choose x1
and thus accepts the mediator’s recommendation
Player 1, if recommended y1 by the mediator thinks that player 2
is recommended y2 . Believing that player 2 obeys the
recommendation y2 , player 1 finds it a best response to choose y1
and thus accepts the mediator’s recommendation
Player 2, if recommended x2 by the mediator thinks that player 1
is recommended x1 . Believing that player 1 obeys the
recommendation x1 , player 2 finds it a best response to choose x2
and thus accepts the mediator’s recommendation
Player 2, if recommended y2 by the mediator thinks that player 1
is recommended y1 . Believing that player 1 obeys the
recommendation y1 , player 2 finds it a best response to choose y2
and thus accepts the mediator’s
Sanjay Singh
recommendation
Correlated Strategies and Correlated Equilibrium

Example (Correlated Strategy 2)

Consider a different correlated strategy that can be realized with

the help of mediator
Consider the correlated strategy
 
1 1 1
α = (x1 , x2 ) : ; (x1 , y2 ) : 0; (y1 , x2 ) : ; (y1 , y2 ) :
3 3 3

10
Now, u1 (α) = u2 (α) =
3
To implement this correlated strategy, mediator recommends,
randomly, with probability 1/3, each of the profile
(x1 , x2 ), (y1 , x2 ), (y1 , y2 )

Sanjay Singh Correlated Strategies and Correlated Equilibrium

 Assume that each player comes to know only the strategy recommended to him by
the mediator
Suppose mediator recommends x1 to player 1. Then player 1 knows that player
2 is recommended x2 . When player 2 plays x2 , it is a best response for player 1
to play x1 , so he would be happy to play x1 and thus accepts the
recommendation of the mediator
Suppose mediator recommends y1 to player 1. Then player 1 knows that
mediator would recommend the mixed strategy x2 : 0.5; y2 : 0.5 to player 2.
When player 2 plays above mixed strategy, then player 1 gets a payoff of 2.5 if
he plays x1 and gets a payoff of 2.5 even if he plays y1 . Thus player 1 can be
indifferent between x1 and y1 and will not mind accepting the recommendation
of mediator to play y1
Similarly player 2 would be happy to do as recommended by the mediator
under the belief that player 1 would do as recommended by the mediator
Above arguments shows that two player can reach a self-enforcing
understanding to obey the mediator if the mediator recommends the correlated
strategy
 
1 1 1
α = (x1 , x2 ) : ; (x1 , y2 ) : 0; (y1 , x2 ) : ; (y1 , y2 ) :
3 3 3
Even though mediator’s recommendation is not binding on two players, the
two players find it in their best interest to follow it
There is a Nash equilibrium of the transformed game with mediated
Sanjay Singh
communication without contracts Correlated Strategies and Correlated Equilibrium

Correlated Equilibrium
Consider following set up of a game with communication
Let Γ = hN, (Si ), (ui )i be a finite SFG
Assume there is a mediator who chooses a pure strategy profile
(s1 , . . . , sn ) according to a probability distribution α ∈ ∆(S)
which is common knowledge among players
Mediator recommends si to player i but does not reveal s−i to
player i
Based on recommendation, player either obeys it or choose any
other strategy from his strategy set Si
Let δi : Si → Si describe player i’s choice of a strategy based on
mediator’s recommendation
δi (si ) gives the strategy that player i chooses to play when
mediator recommends si to him
δi (si ) = si means that player i obeys the mediator when mediator
recommends si
Sanjay Singh Correlated Strategies and Correlated Equilibrium
 Definition (Correlated Equilibrium)
Given a finite strategic form game Γ = hN, (Si ), (ui )i, a correlated
strategy α ∈ ∆(S) recommended by a mediator is called a correlated
equilibrium if
X
ui (α) ≥ α(si , s−i )ui (δi (si ), s−i ) ∀δi : Si → Si ∀i ∈ N.
(si ,s−i )∈S

When a correlated equilibrium is recommended, every player

finds it a best response to obey recommendation of mediator
A correlated equilibrium is any correlated strategy for players
which could be self-enforcingly implemented with the help of a
mediator who can make non-binding recommendation to each
player

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Computing Correlated Equilibria

The inequalities
X
ui (α) ≥ α(si , s−i )ui (δi (si ), s−i ) ∀δi : Si → Si ∀i ∈ N
(si ,s−i )∈S

is equivalent to the inequalities

X
α(s)[ui (si , s−i ) − ui (s0i , s−i )] ≥ 0 ∀si ∈ Si ; ∀s0i ∈ Si ; ∀i ∈ N
s−i ∈S−i

The above equation asserts that no player i could expect to

increase his expected payoff by using some disobedient action s0i
when mediator recommends si
The above constraints are called strategic incentive constraints
Strategic incentive constraints to be satisfied by a mediator’s
correlated strategy for ensuring that all players could rationally
obey the recommendations
Sanjay Singh Correlated Strategies and Correlated Equilibrium
Note that
α(s) ≥ 0, ∀s ∈ S
X
α(s) = 1
s∈S
The set of all payoff allocations under correlated equilibria in a
finite game is a compact and convex set
Consider the following linear program
X
maximize ui (α)
i∈N
subject to
X
α(s)[ui (si , s−i ) − ui (s0i , s−i )] ≥ 0∀si ∈ Si ; ∀s0i ∈ Si ; ∀i ∈ N,
s−i ∈S−i

α(s) ≥ 0∀s ∈ S,
X
α(s) = 1
s∈S
Any feasible solution of this LP will give a correlated
equilibrium
Sanjay Singh Correlated Strategies and Correlated Equilibrium

2
1
x2 y2
x1 5,1 0,0
y1 4,4 1,5
Let us write the objective function
X
ui (α)
i∈N
XX
= α(s)ui (s)
i∈N s∈S
= α(x1 , x2 )u1 (x1 , x2 ) + α(x1 , y2 )u1 (x1 , y2 ) +
+ α(y1 , x2 )u1 (y1 , x2 ) + α(y1 , y2 )u1 (y1 , y2 ) +
+ α(x1 , x2 )u2 (x1 , x2 ) + α(x1 , y2 )u2 (x1 , y2 ) +
+ α(y1 , x2 )u2 (y1 , x2 ) + α(y1 , y2 )u2 (y1 , y2 )
On substituting payoff values, we get
6α(x1 , x2 ) + 0.α(x1 , y2 ) + 8α(y1 , x2 ) + 6α(y1 , y2 )

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Now let us write constraints for i = 1
X
α(s)[ui (si , s−i ) − ui (s0i , s−i )] ≥ 0, ∀si ∈ Si ; ∀s0i ∈ Si ; ∀i ∈ N
s−i ∈S−i
X
α(s)[u1 (s1 , s2 ) − u1 (s01 , s2 )] ≥ 0
s2 ∈S2

α(x1 , x2 )[u1 (x1 , x2 ) − u1 (y1 , x2 )] + α(x1 , y2 )[u1 (x1 , y2 ) − u1 (y1 , y2 )] ≥ 0

(5 − 4)α(x1 , x2 ) + (0 − 1)α(x1 , y2 ) ≥ 0
similarly
α(y1 , x2 )[u1 (y1 , x2 ) − u1 (x1 , x2 )] + α(y1 , y2 )[u1 (y1 , y2 ) − u1 (x1 , y2 )] ≥ 0
(4 − 5)α(y1 , x2 ) + (1 − 0)α(y1 , y2 ) ≥ 0

Sanjay Singh Correlated Strategies and Correlated Equilibrium

For i = 2, we write
X
α(s)[ui (si , s−i ) − ui (s0i , s−i )] ≥ 0, ∀si ∈ Si ; ∀s0i ∈ Si ; ∀i ∈ N
s−i ∈S−i
X
α(s)[u2 (s1 , s2 ) − u2 (s1 , s02 )] ≥ 0
s1 ∈S1

α(x1 , x2 )[u2 (x1 , x2 ) − u2 (x1 , y2 )] + α(y1 , x2 )[u2 (y1 , x2 ) − u2 (y1 , y2 )] ≥ 0

(1 − 0)α(x1 , x2 ) + (4 − 5)α(y1 , x2 ) ≥ 0
Similarly
α(x1 , y2 )[u2 (x1 , y2 ) − u2 (x1 , x2 )] + α(y1 , y2 )[u2 (y1 , y2 ) − u2 (y1 , x2 )] ≥ 0
(0 − 1)α(x1 , y2 ) + (5 − 4)α(y1 , y2 ) ≥ 0

Sanjay Singh Correlated Strategies and Correlated Equilibrium

The linear program for this problem is

maximize 6α(x1 , x2 ) + 0.α(x1 , y2 ) + 8α(y1 , x2 ) + 6α(y1 , y2 )

subject to
(5 − 4)α(x1 , x2 ) + (0 − 1)α(x1 , y2 ) ≥ 0,
(4 − 5)α(y1 , x2 ) + (1 − 0)α(y1 , y2 ) ≥ 0,
(1 − 0)α(x1 , x2 ) + (4 − 5)α(y1 , x2 ) ≥ 0,
(0 − 1)α(x1 , y2 ) + (5 − 4)α(y1 , y2 ) ≥ 0,
α(x1 , x2 ) + α(x1 , y2 ) + α(y1 , x2 ) + α(y1 , y2 ) = 1

Sanjay Singh Correlated Strategies and Correlated Equilibrium

The linear program

maximize 6α(x1 , x2 ) + 0.α(x1 , y2 ) + 8α(y1 , x2 ) + 6α(y1 , y2 )

subject to
(5 − 4)α(x1 , x2 ) + (0 − 1)α(x1 , y2 ) ≥ 0,
(4 − 5)α(y1 , x2 ) + (1 − 0)α(y1 , y2 ) ≥ 0,
(1 − 0)α(x1 , x2 ) + (4 − 5)α(y1 , x2 ) ≥ 0,
(0 − 1)α(x1 , y2 ) + (5 − 4)α(y1 , y2 ) ≥ 0,
α(x1 , x2 ) + α(x1 , y2 ) + α(y1 , x2 ) + α(y1 , y2 ) = 1

can be re-written as

maximize 6z1 + 0.z2 + 8z3 + 6z4

subject to
z1 − z2 ≥ 0,
−z3 + z4 ≥ 0,
z1 − z3 ≥ 0,
−z2 + z4 ≥ 0,
z1 + z2 + z3 + z4 = 1

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Solving the LP on last slide yields the solution
1 1 1
z1 = ; z2 = 0; z3 = ; z4 =
3 3 3
i.e.
1 1 1
α(x1 , x2 ) = ; α(x1 , y2 ) = 0; α(y1 , x2 ) = ; α(y1 , y2 ) =
3 3 3
The above correlated strategy yields the maximum total payoff to the
two players

Sanjay Singh Correlated Strategies and Correlated Equilibrium

Figure shows sets of all

allocations for this
example
under correlated
strategies (convex hull
with corner points
(0,0),(1,5),(5,1),(4,4))
under individually
rational correlated
strategies (convex hull
with corner points
(1,1),(1,5),(5,1),(4,4)),
and
under correlated
Can we select a small number of equilibria (convex hull
desirable or best outcomes among with corner points
above sets? 10 10
(1,5),(5,1),(3,3),( , ))
3 3
Payoff allocation

Sanjay Singh Correlated Strategies and Correlated Equilibrium