Bargaining and Coalition Formation Dr James Tremewan - - PowerPoint PPT Presentation

bargaining and coalition formation
SMART_READER_LITE
LIVE PREVIEW

Bargaining and Coalition Formation Dr James Tremewan - - PowerPoint PPT Presentation

Bargaining and Coalition Formation Dr James Tremewan (james.tremewan@univie.ac.at) Cooperative Game Theory 1 1 These slides are based largely on Chapter 18, Appendix A of Microeconomic Theory by Mas-Colell, Whinston, and Green. Cooperative


slide-1
SLIDE 1

Bargaining and Coalition Formation

Dr James Tremewan (james.tremewan@univie.ac.at) Cooperative Game Theory1

1These slides are based largely on Chapter 18, Appendix A of

”Microeconomic Theory” by Mas-Colell, Whinston, and Green.

slide-2
SLIDE 2

Cooperative Game Theory

Cooperative Games

  • A cooperative game is a game in which the players have

complete freedom of preplay communication and can make binding agreements.

  • In contrast, in a non-cooperative game no communication is

admitted outside the formal structure of the game, and commitments must be self-enforcing, i.e. part of a subgame perfect equilibrium.

  • There are many solution concepts in cooperative game theory,

e.g:

  • The Nash bargaining solution was an example of a solution

concept for a class of cooperative games.

  • The core includes outcomes that may result from coalitional

competition (descriptive).

  • The Shapley value is motivated as a ”fair” division of surplus

(normative).

2/23

slide-3
SLIDE 3

Cooperative Game Theory

Games in Characteristic Form

  • Cooperative games are described in characteristic form. The

characteristic form summarizes the payoffs available to different coalitions.

  • The set of players is denoted I = 1, ..., I. Nonempty subsets

S, T ⊂ I are coalitions.

  • An outcome is a list of utilities u = (u1, ..., uI) ∈ RI. The

relevant coordinates for a coalition S are uS = (ui)i∈S.

  • A utility possibility set is a nonempty, closed set US ⊂ RS where

uS ∈ US and u′S ≤ uS implies u′S ∈ US i.e. utility freely disposable.

  • A game in characteristic form (I, V ) is a set of players I and a

rule V (·) that associates to evry coalition S ⊂ I a utility possibility set V (S) ⊂ RS .

3/23

slide-4
SLIDE 4

Cooperative Game Theory

Characteristic Form: Examples

  • Nash bargaining game (using notation from earlier in course):

I = {1, 2}; V ({1}) = d1, V ({2}) = d2, and V ({1, 2}) = S.

  • Three-player example (I = {1, 2, 3}):

4/23

slide-5
SLIDE 5

Cooperative Game Theory

Superadditive Games

  • A game is superadditive if two (non-overlapping) coalitions can

do at least as well together as alone.

  • Formally: A game in characteristic form (I, V ) is superadditive if

for any coalitions S, T ⊂ I such that S ∩ T = ∅ we have: if uS ∈ V (S) and uT ∈ V (T), then (uS, uT) ∈ V (S ∪ T).

  • We will look only at superadditive games: as in the bilateral

section, we are primarily interested in situations where there are gains from cooperation.

5/23

slide-6
SLIDE 6

Cooperative Game Theory

Transferable Utility (TU)

  • Much of the literature focuses on TU games (as will we), i.e.

where utility can be transfered costlessly between coalition

  • members. Sufficient conditions for TU are:
  • Players can make side-payments.
  • Players utility is linear in money.
  • Excludes, for example, situations where bribes are illegal.
  • In TU games, V (S) = {uS ∈ RS :

i∈S

uS

i ≤ v(S)} for some v(s).

  • i.e. coalition S chooses a joint action to maximise their total

utility, denoted v(S) which can be allocated amongst S in any way.

  • v(S) is called the worth of coalition S.

6/23

slide-7
SLIDE 7

Cooperative Game Theory

TU Games: Example

Boundaries in TU games are hyperplanes in RS:

7/23

slide-8
SLIDE 8

Cooperative Game Theory

TU Games: Simplex Representation

Normalising utilities such that V ({i}) = 0∀i allows us to represent an n-player TU game on an (n-1) dimension simplex:

8/23

slide-9
SLIDE 9

Cooperative Game Theory

Two Games in Characteristic Form

  • A three-player game is defined by:
  • v({1, 2, 3}) = 1,
  • v({1, 3}) = v({2, 3}) = 1, v({1, 2}) = 0,
  • v({1}) = v({2}) = v({3}) = 0.
  • What do you think will/should happen in this game?
  • A three-player game is defined by:
  • v({1, 2, 3}) = 10,
  • v({1, 2}) = 10, v({1, 3}) = 3, v({2, 3}) = 2,
  • v({1}) = v({2}) = v({3}) = 0.
  • What do you think will/should happen in this game?

9/23

slide-10
SLIDE 10

The Core

The Core

  • The set of feasible utility outcomes with the property that no

coalition could improve the payoffs of all its members.

  • In TU games the core is the set of utility vectors u = (u1, ..., uI)

such that:

i∈S

ui ≥ v(S)∀S ⊂ I, and

i∈I

ui ≤ v(I)

  • The core may be:
  • Empty (strategic instability, no useful prediction, e.g.

divide-the-dollar majority rules).

  • Large (makes no useful prediction, e.g. divide-the-dollar

unanimity rules).

  • Non-empty and small (makes sharp prediction).
  • Note that the core depends only on ordinal utility.

10/23

slide-11
SLIDE 11

The Core

A TU Game with Non-Empty Core

11/23

slide-12
SLIDE 12

The Core

A TU Game with Empty Core

12/23

slide-13
SLIDE 13

The Shapley Value

The Shapley Value

  • Describes a ”reasonable” or ”fair” division taking as given the

strategic realities captured by the characteristic form.

  • Idea of fairness here is egalitarianism: gains from cooperation

should be divided equally.

  • Two player example (I, v) = ({1, 2}, v):
  • Gains from cooperation = v(I) − v({1}) − v({2}).
  • Shi(I, v) = v({i}) + 1

2(v(I) − v({1}) − v({2}))

  • Can re-write as:
  • Sh1(I, v) − Sh1({1}, v) = Sh2(I, v) − Sh2({2}, v), and
  • Sh1(I, v) + Sh2(I, v) = v(I), where Shi({i}, v) = v({i})
  • The benefit to player one from the presence of player two is the

same as the benefit to player two from the presence of player

  • ne.
  • Now generalize this idea to more players...

13/23

slide-14
SLIDE 14

The Shapley Value

The Shapley Value

  • The Shapley value of a game (I, v) is the outcome consistent

with:

  • Shi(S, v) − Shi(S\{h}, v) = Shh(S, v) − Shh(S\{i}, v), and

i∈S

Shi(S, v) = v(S),

  • for every subgame (S, v) and all players i, h ∈ S.
  • In words: the benefit a member of a coalition (player i) gets

from another (player h) joining is equal to the benefit player h would get if already a member and player i was joining.

  • This outcome is unique.

14/23

slide-15
SLIDE 15

The Shapley Value

The Shapley Value: Axioms

  • The Shapley value can also be derived as the unique value

satisfying three axioms (here loosely defined):

  • Efficiency:

i∈S

Shi(I, v) = v(I), i.e. no utility is wasted.

  • Symmetry: If (I, v) and (I, v ′) are identical except the roles of

players i and h are swapped, then Shi(I, v) = Shj(S, v ′), i.e. labeling doesn’t matter.

  • Additivity: If a game is in a particular sense the sum of two
  • ther games2, then Sh(I, v + w) = Sh(I, v) + Sh(I, w).

2See Shapley (1953) for a precise definition.

15/23

slide-16
SLIDE 16

The Shapley Value

The Shapley Value

  • An ”easy” way to calculate the Shapley value:
  • Shi(I, v) =

i∈S s!(n−s−1)! n!

(v(S ∪ {i}) − v(S)),

  • where s = |S| and n = |I|.
  • Intuition: imagine a coalition of all the players is formed by

including one player at a time in a random order, and each player receives all of the added benefit to the coalition at the time they are included. The Shapley value is the expected value

  • f this process if all orders are equally likely.
  • v(S ∪ {i}) − v(S) is the value the new player adds.
  • s! is number of ways the existing members of S could have

arrived.

  • (n − s − 1)! is number of ways the remaining players can arrive.
  • n! is the number of possible ways of the players arriving overall.

16/23

slide-17
SLIDE 17

Examples

Glove Market: The Shapley Value

  • A three-player game is defined by:
  • v({1, 2, 3}) = 1,
  • v({1, 3}) = v({2, 3}) = 1, v({1, 2}) = 0,
  • v({1}) = v({2}) = v({3}) = 0.
  • Player 1:
  • The value P1 adds to the coalition {3} is 1, and this happens
  • nly if P3 is added first, then P1 second.
  • P1 adds 0 to any other coalition.
  • There are six possible orderings so Shi(I, v) = 1

6.

  • Player 2: as for P1.

17/23

slide-18
SLIDE 18

Examples

Glove Market: The Shapley Value

  • Player 3:
  • adds 1 to the coalition {1} is, and this happens only if P1 is

added first, then P3 second.

  • adds 1 to the coalition {2} is, and this happens only if P2 is

added first, then P3 second.

  • adds 1 to the coalition {1, 2} is, and this happens if P1 is added

first, P2 second, then P3 third, or if P2 is added first, P1 second, then P3 third.

  • Sh3(I, v) = 1

6 · 1 + 1 6 · 1 + 2 6 · 1 = 2 3

  • The Shapley value suggests the allocation {1

6, 1 6, 2 3}

  • Check! The sum of Shi(I, v) is v(I)!

18/23

slide-19
SLIDE 19

Examples

Glove Market: the Core

  • What is the core of this game?:
  • Consider an allocation {u1, u2, u3}.
  • Suppose u1 > 0. P2 and P3 are better off forming the coalition

{2, 3} and sharing u′

2 = u2 + 1 2u1 and u′ 3 = u3 + 1 2u1.

  • Therefore in any allocation in the core u1 = 0 (similarly u2 = 0).
  • No coalition can improve on {0, 0, 1} for all members.
  • The core is {0, 0, 1}.
  • The strategic environment means P1 and P2 undercut each
  • ther leaving zero profits.
  • The Shapley value ({1

6, 1 6, 2 3}) is more equitable than the core.

  • However the Shapley value is less equitable than an even split

({1

3, 1 3, 1 3}) because it recognises to some extent the individual

contributions (bargaining power?) of the different players.

19/23

slide-20
SLIDE 20

Examples

Three Player Game: The Shapley Value

  • A three-player game is defined by:
  • v({1, 2, 3}) = 10,
  • v({1, 2}) = 10, v({1, 3}) = 3, v({2, 3}) = 2,
  • v({1}) = v({2}) = v({3}) = 0.
  • Player 1:
  • adds 10 to the coalition {2} if P2 is added first, then P1 second.
  • adds 3 to the coalition {3} if P3 is added first, then P1 second.
  • adds 8 to the coalition {2, 3} if P2 is added first, P3 second and

P1 third; or if P3 is added first, P2 second and P1 third.

  • P1 adds 0 to any other coalition.
  • There are six possible orderings so

Sh1(I, v) = 1

6 · 10 + 1 6 · 3 + 2 6 · 8 = 29 6

20/23

slide-21
SLIDE 21

Examples

Three Player Game: The Shapley Value

  • Player 2:
  • adds 10 to the coalition {1} if P1 is added first, then P2 second.
  • adds 2 to the coalition {3} if P3 is added first, then P2 second.
  • adds 7 to the coalition {1, 3} if P1 is added first, P3 second and

P2 third; or if P3 is added first, P1 second and P2 third.

  • P2 adds 0 to any other coalition.
  • Sh2(I, v) = 1

6 · 10 + 1 6 · 2 + 2 6 · 7 = 26 6

  • Player 3:
  • adds 3 to the coalition {1} if P1 is added first, then P3 second.
  • adds 2 to the coalition {2} if P2 is added first, then P3 second.
  • adds 3 to the coalition {1, 2} if P1 is added first, P2 second and

P3 third; or if P2 is added first, P1 second and P3 third.

  • P3 adds 0 to any other coalition.
  • Sh3(I, v) = 1

6 · 3 + 1 6 · 2 = 5 6

  • The Shapley value suggests the allocation {29

6 , 26 6 , 5 6}

21/23

slide-22
SLIDE 22

Examples

Three Player Game: The Core

  • Consider an allocation {u1, u2, u3}.
  • If u1 + u2 < 10 P1 and P2 are better off forming the coalition

{1, 2} and sharing u′

1 = u1 + 1 2(10 − u1 − u2) and

u′

2 = u2 + 1 2(10 − u1 − u2).

  • Therefore in any allocation in the core u1 + u2 ≥ 10.
  • But v(I) = 10 so u1 + u2 ≤ 10 which means u1 + u2 = 10. Thus

u3 = 0 (from now on we take this as given).

  • If u1 < 3 P1 is better off forming the coalition {1, 3} and

sharing u′

1 = u1 + ǫ, u′ 3 = 3 − u1 − ǫ.

  • If u2 < 2 P2 is better off forming the coalition {2, 3} and

sharing u′

2 = u2 + ǫ, u′ 3 = 2 − u2 − ǫ.

  • The core is the set of allocations {x, 10 − x, 0} where x ∈ [3, 8].

22/23

slide-23
SLIDE 23

Examples

Three Player Game

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

✉ ✉ ✉

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

(0,10,0) (0,0,10) (10,0,0) (8,2,0) (3,7,0) Shapley value Core

23/23