Cooperative Game Theory

(N-Person Games)
N-Person Coalitional Games

… Payoffs can usually be increased by forming coalitions.

… A subset of players coordinate their strategies to maximize the total
payoff of coalition members.

… Additional payoff can be shared appropriately among

coalition partners.
… Important questions:
… What coalitions are likely to be formed?
… Are they stable? What are the conditions of stability?
… What is the worst case payoff for a coalition?
… How can the coalition maximize this worst case?
… How to share the payoffs resulting from cooperation?
•Important questions:
… What coalitions are likely to be formed?
… Are they stable? What are the conditions of stability?
… What is the worst case payoff for a coalition?
… How can the coalition maximize this worst case?
… How to share the payoffs resulting from cooperation?
Coalitions and Coalition Structures

… Coalition:
… a set of players who decide to cooperate and coordinate their
strategies.
… Singleton Coalition:
… A player playing independently without entering any coalition.
… Coalition Structure: A possible combination of coalitions.
(A), (B), (C) (AB), (C) (AC), (B) (BC), (A) (ABC)
… Grand Coalition:
… When all players decide to cooperate and coordinate their
strategies according to some understanding.
… Solution Set:
… Set of coalition structures that yield the maximum possible
total payoff to all players.
… What is the worst case payoff for a coalition?
… How can the coalition maximize this worst case?

Characteristic Function
• The function describes how much collective payoff a set of
players can gain by forming a coalition
• The characteristics function of an n-person game assigns to
each subset S of players the maximum value v(S) that
coalition S can guarantee itself by coordinating the strategies
of its members, no matter what other players do.
Land Development Game

… Player 1 owns a piece of land with value of $10,000.

… Player 2 is interested in buying and can develop the land
increase its value to $20,000.
… Player 3 is interested in buying and can develop the land
increase its value to $30,000.
Characteristic Function:
Any coalition that does not contain player 1 has a worth of
$0.
v(Ø) = 0, v({2}) = 0, v({3}) = 0, v({2,3}) = 0
v({1}) = 10000, v({1,2}) = 20000
v({1,3}) = 30000, v({1,2,3}) = 30000
Ice-Cream Game: Characteristic Function

A: 6$ B: 4$
C: 3$

w = 500 w = 750 w = 1000

Cost = $7 Cost = $9 Cost = $11

Characteristic Function:
v(Ø) = v({A}) = v({B}) = v({C}) = 0
v({A, B}) = 750, v({A, C}) = 750, v({B, C}) = 500
v({A, B, C}) = 1000
A Voting Game: Characteristic Function

… Consider that the Parliament of a certain Nation has four

political parties 1, 2, 3, 4 with 45, 25, 15, 12 members
respectively. To pass any bill, at least 51 votes are
required.
This situation could be modeled as:
N = {1, 2, 3, 4}
Characteristic Function:
v(Ø) = 0,
v({1}) = 0, v({2}) = 0, v({3}) = 0, v({4}) = 0
v({1,2}) = v({1,3}) = v({1,4}) = v({1,2,3}) = v({1,2,4}) =
v({1,3,4}) = v({2,3,4}) = v({1,2,3,4}) = 1,
v({2,3}) = v({2,4}) = v({3,4}) = 0
Garbage Game

… Each of 4 property owners has one bag of garbage and

must dump it on somebody’s property.
… Owners can form at most two coalitions.
… If b bags of garbage are dumped on a coalition of
property owners, then the coalition receives a reward of
4 - b.
Characteristic Function:
… The best strategy for a coalition S is to dump all of their
garbage on the property of owners who are not in S.
… v(Ø) = 0, v({1,2,3,4}) = 0 and
… v(S) = 4 -|S| for 1<= |S| <= 3
Three Person Games

Player 2

P Q

Player 3 Player 3

X Y X Y

Pla A 6, 6, 6 7, 7, 1 7, 1, 7 0, 0, 0
yer
1 B 0, 0, 0 4, 4, 4 4, 4, 4 1, 7, 7
Three Person Games

Player 2
P Q
Player 3 Player 3
X Y X Y
Pla A 6, 6, 6 7, 7, 1 7, 1, 7 0, 0, 0
yer
1 B 0, 0, 0 4, 4, 4 4, 4, 4 1, 7, 7

v(1) max(0,0) = 0
v(2) max(0,0) = 0
v(3) max(0,0) = 0
v(1,2) AP, AQ, BP, BQ min(12,14), min(8,0), min(0,8), min(8,8) max(12,0,0,8) = 12
v(1,3) AX, AY, BX, BY min(12,14), min(8,0), min(0,8), min(8,8) max(12,0,0,8) = 12
v(2,3) PX, PY, QX, QY min(12,0), min(8,8), min(8,8), min(0,14) max(0,8,8,0) = 8
v(1,2,3) max(18,15,15,0,0,12,12,15) = 18
Network Flow Game

Source Sink

v(Ø) = 0, v({1}) = 0, v({2}) = 0, v({3}) = 0, v({4}) = 0, v({5}) = 0

v({1, 2}) = 2, v({4, 5}) = 3, all other v({a, b}) = 0
v({1, 3, 5}) = 2, v({1, 2, 3}) = 2
v({1, 3, 4, 5}) = 5 = v({1, 2, 4, 5}), v({1, 2, 3, 5}) = 4
v({1, 2, 3, 4, 5}) = 7, and so forth…
A Bankruptcy Game

Find the characteristic function.

Characteristic Function is usually Superadditive

Superadditivity

v(S∪T) ≥ v(S) + v(T) if S∩T = φ

… The value of two coalitions will be no less than the sum of
their individual values.
… By forming a coalition they can never be worse-off than
playing individually.
… Implies that the grand coalition has the highest payoff
Inessential game
… A game is inessential if
v(S∪T) = v(S) + v(T) ∀ S,T
How to share the payoffs resulting from cooperation?

Imputations : Sharing the Payoffs

(meaningful for the games with transferable utility (TU) )

… Two-player cooperative game: each player keeps

his/her payoff.
… In a coalitional game it is important to decide how
the total winnings be shared among its members.
… The ‘reasonable’ share outs of the rewards are called
Imputations.
 ☺ Game: Find the Imputation ☺

… Three Players: A, B, C

1. Characteristic Function:
v(A) = v(B) = v(C) =0
v(AB) =100, v(AC) = 200, v(BC) = 300
v(ABC) =400.
2. Characteristic Function:
v(A) = v(B) = v(C) =0
v(AB) =0, v(AC) = 0, v(BC) = 400
v(ABC) =400.

Find the Imputation in both the cases and submit.

Imputations

An imputation in an n-person TU game with characteristic

function v is a vector x = (x1,x2,…xn) satisfying
1) Σxi = v(N) (x is efficient or collectively rational)

2) xi ≥ v(i) ∀ i = 1,2,…,n (x is individually rational)

Assumptions:
… Payoff is infinitely divisible.
… Side Payments are allowed.
Domination of Imputations

… Let x and y be two imputations, then x dominates y w.r.t.

S (x >S y) if:
1) xi > yi for all i ∈S
2) Σi ∈S xi ≤ v(S)
… We say that x dominates y if x dominates y for some
coalition S.
… Members of S will never accept dominated imputation y.
x: 4 5 4 7 y: 3 4 7 6
… It is possible that both x and y may dominate each other.
x: 3 4 7 6 5 y: 4 5 7 5 4
… y dominates x w.r.t. (1,2), provided v(1,2) ≥ 9
… x dominates y w.r.t. (4,5), provided v(4,5) ≥ 11
The Core

… Under what imputations would the players want to form

the grand coalition?
… which are not dominated for any coalition.
… The core of a game with characteristic function v, is the
set of imputations which are not dominated for any
coalition.
… Therefore, it must satisfy
1) Σxi = v(N) (Collectively rational)

2) xi ≥ v(i) ∀ i = 1,2,…,3 (Individually rational)

3) Σi ∈S xi ≥ v(S) ∀ S (Coalitionally rational)

Stable Coalitions and “The Core”

… A coalition structure is stable if no group can benefit by

withdrawing from their respective coalitions to form a
new coalition (including singleton coalition)
… Does it make sense for any subset of players to drop out
of the grand coalition?
… NO if the imputation is in the core.
… If the core exists, the grand coalition is guaranteed to be
stable.
… There may also exist other coalition structures which are
stable.
Example: Oil Market Game

Country 1 has oil which it can use to run its transport system at a profit
of a per barrel. Country 2 wants to buy the oil to use in its manufacturing
industry, where it gives a profit of b per barrel, while Country 3 wants it
for food manufacturing where the profit is c per barrel. a < b ≤ c.

v(1) = a, v(2) = v(3) = 0

v(2,3) = 0, v(1,2) = b, v(1,3) = c
v(1,2,3) = c
Core:
x1+x2+x3 = c …(1)
x1 ≥ a, x2 ≥ 0, x3 ≥ 0
x1+x2 ≥ b (S = {1,2}) …(2)
x1+x3 ≥ c (S = {1,3}) …(3)
(1) And (3) imply x2 = 0. Substituting in (2) gives x1 ≥ b. Hence x3 ≤ c-b.
The core is: {(x, 0, c-x); b ≤ x ≤ c}
Find the Core

Consider the game we played in the class:

… v(A) = v(B) = v(C) = 0,
… v(A,B) = 100, v(A,C) = 200, v(B,C) = 300, v(A,B,C) = 400.
Core:
x1,x2,x3 ≥ 0, x1+x2+x3 = 400
x1+x2 ≥ 100, x1+x3 ≥ 200, x2+x3 ≥ 300
Simplify:
400 – x3 ≥ 100 ⇔ x3 ≤ 300
400 – x2 ≥ 200 ⇔ x2 ≤ 200
400 – x1 ≥ 300 ⇔ x1 ≤ 100
x1+x3 ≥ 200 and x1 ≤ 100 imply x3 ≥ 100

C(v) = {(x1,x2,x3) | 100 ≥ x1 ≥ 0, 200 ≥ x2 ≥ 0, 300 ≥ x3 ≥ 100, x1+x2+x3 = 400}

Game 1 Inputs

In Core Submissions
Core: {100 ≥ x1 ≥ 0, (1, 100, 299) (50, 170, 180)

200 ≥ x2 ≥ 0, (1, 199.5, 199.5) (50, 175, 175)

(2, 199, 199) (60, 170, 170)

300 ≥ x3 ≥ 100,
(10, 195, 195) (66.6, 133.3, 200)
x1+x2+x3 = 400} (25, 100, 275) (70, 140, 190)

(25, 183, 192) (75, 150, 175)

(33.3, 133.3, 233.3) (80, 160, 160)

Non Core (33.3, 183.3, 183.3) (83.3, 133.3, 183.3)
Submissions (34, 133, 233) (90, 155, 155)
(0, 150, 150) (34, 143, 223) (98, 101, 201)

(40, 180, 180) (98, 151, 151)

(48, 53, 299) (100, 100, 200)

(50, 125, 225) (100, 133.3, 166.7)

(100, 150, 150)

Game 2 Inputs

Characteristic Function:
v(A) = v(B) = v(C) =0
v(AB) = 0, v(AC) = 0, v(BC) = 400
v(ABC) = 400.

In Core Submissions Non Core Submissions

(0, 200, 200) (0, 100, 100)

(60, 170, 170)

(70, 140, 190)

Another Example

Consider the game with the following characteristic function:

… v(1) = v(3) = 0, v(2) = 3
… v(1,2) = 3, v(1,3) = 1, v(2,3) = 4, v(1,2,3) = 5.
Core:
x1, x3 ≥ 0, x2 ≥ 3, x1+x2+x3 = 5
x1+x2 ≥ 3, x1+x3 ≥ 1, x2+x3 ≥ 4
Simplify:
5 – x3 ≥ 3 ⇔ x3 ≤ 2
5 – x2 ≥ 1 ⇔ x2 ≤ 4
5 – x1 ≥ 4 ⇔ x1 ≤ 1

C(v) = {(x1,x2,x3) | 1 ≥ x1 ≥ 0, 4 ≥ x2 ≥ 3, 2 ≥ x3 ≥ 0, x1+x2+x3 = 5}

Example: Lilliput U.N. Security Council
Liliput has a small version of the UN Security Council (the numbers get too
hard in the real thing) which has two permanent members, 1 and 2, who
have the veto, plus three ordinary members 3, 4, and 5. For a resolution to
be passed, it requires three votes in favour and no vetos. The payoff is 0 if
the resolution is not passed and 1 to the coalition who wants it passed.

v(1,2,3) = v(1,2,4) = v(1,2,5) = v(1,2,3,4) = v(1,2,3,5) = v(1,2,4,5) = v(1,2,3,4,5) = 1

v(S) = 0 for all other S
Core:
x1+x2+x3+x4+x5 = 1
x1+x2+x3 ≥ 1, x1+x2+x3+x4 ≥ 1
x1+x2+x4 ≥ 1, x1+x2+x3+x5 ≥ 1
x1+x2+x5 ≥ 1, x1+x2+x4+x5 ≥ 1
xi ≥ 0 ∀ i
This implies x3 = x4 = x5 = 0
The core is: {(x,1-x,0,0,0); 0 ≤ x ≤ 1}
Land Development

Characteristic Function:
Any coalition that does not contain player 1 has a worth of $0.
v(2) = 0, v(3) = 0, v(2,3) = 0
v(1) = 10000, v(1,2) = 20000
v(1,3) = 30000, v(1,2,3) = 30000
Core:
x1+x2+x3 = 30000
x1 ≥ 10000, x2 ≥ 0, x3 ≥ 0
x1+x2 ≥ 20000
x1+x3 ≥ 30000
This implies x2 = 0.
The core is: {(x, 0, 30000-x); 20000 ≤ x ≤ 30000}
Coalitional Games Solutions

… Given a coalitional game we want to find

the distribution of the gains of the
coalition between the agents
… Different solution concepts have different
objectives
… The Core promotes stability
… The Shapley value promotes fairness
… Shapley value is motivated by the need to
have a theory that would predicts a unique
imputation for every given coalitional game
The Shapley value

• Marginal contribution of a player i: v(S ∪ {i}) − v(S)

• Dependent on the set S.
• Using Marginal Contribution of an individual as her
share-out.
• Unfair because it depends on the ordering of the players
• One way to make it fair: average over all possible
orderings.
• Let φ(i, π) be the marginal contribution of i in ordering π
• Then i’s Shapley value is Σπ φ(i, π)/(n!)
Land Development: Shapely Value
Characteristic Function:
v(1) = 10000, v(2) = 0, v(3) = 0
v(1,2) = 20000, v(1,3) = 30000, v(2,3) = 0
v(1,2,3) = 30000
The core is: {(x, 0, 30000-x); 20000 ≤ x ≤ 30000}

π φ(1, π) φ(2, π) φ(3, π)

(1,2,3) 10,000 10,000 10,000

(1,3,2) 10,000 0 20,000

(2,1,3) 20,000 0 10,000

(2,3,1) 30,000 0 0

(3,1,2) 30,000 0 0

(3,2,1) 30,000 0 0

130,000/6 10,000/6 40,000/6

Shapely Value x1=21,666.67 x2=1,666.67 x3=6,666.67
Eat Out Game: Shapely Value
Characteristic Function:
v(A) = 80, v(B) = 56, v(C) = 70
v(A,B) = 80, v(A,C) = 85, v(B,C) = 72
v(A,B,C) = 90

π φ(A, π) φ(B, π) φ(C, π)

(A,B,C) 80 0 10

(A,C,B) 80 5 5

(B,A,C) 24 56 10

(B,C,A) 18 56 16

(C,A,B) 15 5 70

(C,B,A) 18 2 70

235/6 124/6 181/6

Shapely Value x1=39.2 x2=20.7 x3=30.2
Exercise: Find the Core & Shapely Value

Characteristic Function:
V(S) = 0 if |S| = 1
v(1,2) = v(1,3) = v(2,3) = v(1,2,3) = 1

Core:
x1+x2+x3 = 1
x1 ≥ 0, x2 ≥ 0, x3 ≥ 0
x1+x2 ≥ 1
x1+x3 ≥ 1
x2+x3 ≥ 1

The core is Empty

Empty Core: Shapely Value
Characteristic Function:
V(S) = 0 if |S| = 1
v(1,2) = v(1,3) = v(2,3) = v(1,2,3) = 1
The core is Empty

π φ(1, π) φ(2, π) φ(3, π)

(1,2,3) 0 1 0

(1,3,2) 0 0 1

(2,1,3) 1 0 0

(2,3,1) 0 0 1

(3,1,2) 1 0 0

(3,2,1) 0 1 0

2/6 2/6 2/6

Shapely Value x1=0.33 x2=0.33 x3=0.33
Important take aways

… The core of a game is stable but can be empty.

… The Shapley value always exists and is unique.
… The Shapley value may or may not lie in the core.
… Therefore, Shapley value may not always provide
a stable solution but is fair by definition.