Bargaining. Divide the melting cake Often bargaining is similar to a game where players make...

26
Bargaining Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting Sometimes this is referred to a divide the dollar game, where time is money so if you wait you have less to divide Consider just one period and two players (A) and (B) The game works as follows: (A) proposes a division; then (B) either agrees or not - if (B) agrees then they divide the cake according to (A)s proposal – otherwise it melts What does game theory predict? – think about subgame perfection or using backwards induction. (B) knows that if he turns down (A) s offer, he gets nothing Therefore, if (B) is playing optimally in every subgame he

Transcript of Bargaining. Divide the melting cake Often bargaining is similar to a game where players make...

Page 1: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Bargaining Divide the melting cake

Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is meltingSometimes this is referred to a divide the dollar game, where time is money so if you wait you have less to divide

Consider just one period and two players (A) and (B) The game works as follows: (A) proposes a division; then (B) either agrees or not - if (B) agrees then they divide the cake according to (A)s proposal – otherwise it melts

What does game theory predict?– think about subgame perfection or using backwards induction.

(B) knows that if he turns down (A) s offer, he gets nothing

Therefore, if (B) is playing optimally in every subgame he will accept any positive offer and would even be indifferent between getting nothing and having the cake melt

Page 2: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

The SPNE (B) will accept any offer; (A) will best respond by proposing the

following division: (A) gets everything; (B) gets nothing

SPE: (A) proposes: (A) everything and (B) nothing (B) accepts anything with a non-negative payoff, rejects

otherwise

What about NE? There are an infinite number of NE Let xB be some fraction between 0 and 1, then an NE takes the form (A) proposes: (A) 1- xB and (B) xB

(B) accepts anything ≥ xB , rejects otherwise

What happens in experiments? Typically, offers range from 50-50 to 70-30

(B) rejects offers below 30% of the pie

The SPNE is not descriptive and is not prescriptive for (A) either if you are playing against someone who rejects offers below 30%

Page 3: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Bargaining with Complete information

Two players bargain to determine the split of V dollars

The game begins in period 1Player (1) makes an offer of a split (a share of V for each player) If Player (2) accepts (1)s offer, bargaining ends and payoffs are realized.

If Player (2) rejects (1)s offer, then (2) makes an offer Then (1) can accept or reject …The game ends in period T and if there is no agreement, both players get 0

Each player has a discount factor (0 < δ < 1) - so a dollar in period t is worth δt-1 terms of period 1

dollars

There is a unique SPNE in this game and the path has no bargaining process– player (1)s first offer is accepted and the game ends

Finite horizon bilateral bargaining

Page 4: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

1st case: T is an odd number When is odd, (1) makes the final offer. (2) is indifferent between accepting zero or not, - but must accept if (1) is to have a BR, otherwise (2) will get arbitrarily close to zero

The uniq

T

1

ue equilibrium in the final subgame: (2) accepts any non-negative number Back-up (1) offers a (V,0) split and the resulting payoffs would be , 0

Now consider the bargaining in stage T-1:

t V

1 1

1 2

(2) knows (1) will accept any offer , because in stage , (1) can get (2) offers , 1 , (1) accepts, game ends w/ payoffs , 1

Continue backing up until per

t T

T T

V T VV V V V

iod 1 - as gets smaller the pie grows, so the player making the offer can always increase his payoff, while holding the other player's payoff at his reservation value

(1) gets a p

t

1 1ayoff of ; (2) gets payoff of

1 1

T T

V V V

Page 5: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

2nd case: T is an even number

1 1

When is an even number, (2) makes the final offer.

A similar analysis demonstrates that

1 1(1) gets a payoff of ; (2) gets payoff of

1 1

Fo

T T

T

V V V

1

r the game from T-1 on it is just like the T odd case except with role reversal.

1Player (2) can guarantee himself in the T period

1

So (1) has to give him that in period T-1, and so

T

TV

on

Page 6: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

“Bargaining model of War & Peace”Motivation: Build on the literature, which has several gaps

1) War is often modeled as a costly lottery - too simple, does not describe realistic conflict, too short, doesn’t allow for bargaining, negotiation, counter-offers, and defeats usually do not end in the utter defeat of one side

2) The literature treats war onset separately from war outcomes -ignores forward looking behavior that might affect conflict engagement decisions

3) When bargaining has been considered, it has been considered in an unrealistic way - Example: allow bargaining before war starts but once war begins the conflict is back to a costly lottery with no surrender

Basic model structure-parties can bargain before fighting to avoid war; bargain in between battles to end war-the model yields factors that affect onset, duration, outcomes, and negotiated settlements

Page 7: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Approach

Introduce a game-theoretic bargaining model with one-sided incomplete information- this leads to a more realistic bargaining process- there maybe a delay in reaching an agreement

By adding one-sided asymmetric information, (2) is now one of two possible types δ1, δ2 - when (1) determines his proposal, he may aim his proposal at the less demanding type - then (1) bears the risk that a war will begin

Two-sided incomplete information is more realistic but it makes it more difficult to obtain clear predictions

a. If an informed player makes a proposal, the game becomes a signaling game which can have multiple equilibria

b. Alternating bargaining is important in situations where it is not clear who the attacker is. For example, India vs. Pakistan

Page 8: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Model One attacker: A benefits: Ba resources: Ra

One defender: D Bd Rd

Benefits are what the players fight and bargain for and are completely divisible so that it is possible to compromise, while the resources are what the players fight with.

Each player’s objective is to obtain as many benefits as possible while conserving resources. - players may be willing to sacrifice resources to acquire more benefits.

Assume each utility function Ui( Bi , Ri ) is increasing, concave, twice differentiable in both arguments and that the cross-partials are greater than zero (like a Cobb-Douglas fn.).a) each player always wants more resources and more benefits b) marginal utilities of resources and benefits are diminishing c) marginal utility of benefits is increasing in resourcesd) marginal utility of resources is increasing in benefits

-these conditions imply that a disputant’s willingness to risk resources in battle is increasing in his amount of resources and decreasing in size if his benefits

Page 9: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

War & Uncertainty War consists of an alternating sequence of negotiations and battles - If negotiations succeed war ends - If negotiations fail and A does not quit, then players fight another

battle - In each battle D defeats A with probability d and looses with

probability (1-d) - Fighting expends resources; the amount depends on who wins (Ra

w

> Ral); (Rd

w> Rdl)

Rawl represents A’s resources after she wins the first battle but

loses the second - War ends as soon as one side’s resources fall below a minimal

amount of resources necessary to continue fighting.

Uncertainty: A only knows the probability distribution of D’s type (dh

or dl)

1) assume uncertainty is one-sided; D has complete info about A, but A does not have complete info about D.

2) assume that A’s uncertainty about D is confined to uncertainty about D’s military ability (the d parameter). All types of D are assumed to have the same utility function

3) A has beliefs that are updated as D rejects proposals and battle outcomes are observed.

Page 10: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Simplest game At most two battles can occur: Assumption 1: war can extend to a second battle only if A wins the

first battle Assumption 2: D can sustain two battle losses before his resources

are insufficient to continue fighting, which limits the size of the game

tree Assumption 3: there are only two types of D: dh> dl (dh is more likely

to win battles)

To identify conditions necessary for Conflict Termination, -Begin at the conclusion of the game-Let γh

t and γlt denote the highest values of γt acceptable to types dh

and dl in stage t

-Type dl is willing to make a greater concession to A in order to

avoid a fight than is dh

Therefore, γht ≤ γl

t

1) any proposal: γ ≤ γht is acceptable to both types

2) any proposal: γht < γ ≤ γl

t is acceptable to type dl, but

unacceptable to type dh

3) any proposal: γlt ≤ γ is unacceptable to both types

Page 11: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

3 possible choices

both types accept

3h 3l

l- accept, but h-reject

both types reject

At most three relevant options for A in each stage 1) proposes γh

t and the conflict ends

2) proposes γlt and depending on D’s type conflict may continue or end

3) ask for more than γlt and conflict will definitely continue

To analyze the game start at the end and work backwards

Page 12: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Fig. 3 Continuation after A retreats

Page 13: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

When will D accept?

l

3 l l

3

Fig. 3, type d will accept if

, 1 + 1-d , d , +

A will increase until this inequality binds

There is a similar inequality for the h

l ll lwd d a d d d d d d a d

l

U R B B U R B U R B B

3

3 3

3

igh type

A's proposal 0,1

only 3 possible choices have any relevance , ,

and something greater than that both types reject

l h

l

Page 14: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Rule out: γ ≥ γlt

3 3

3

R1) A never provokes a fight with both types. This implies that A proposes either or ; we can rule out something greater than

Intuition: 1) fighting is costly

l h

l

3 3

3

4 3 4 h h

2) both sides are risk averse

If A proposes both types accept. Then A's payoff is ,

If A proposes , then A's payoff is

1- , d , 0 1 d

h w ha a a

l

w l wla a a a a a

U R B

U R B Ua R U R

3

4 l l 4 h h

,

Ruling out > : the payoff would be

1- d ,0 1 d , d ,0 1 d ,

wwa

l

wl ww wl wwa a a a a a a a a

B

U R U R B Ua R U R B

Page 15: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Proof from Appendix B

l l l

Jensen's inequality is often used to explain why players are avoiding risk

For concave functions ; Rather have (1-d ) for sure than the average of d (0)+ (1-d ) ( )

W

w ww wld d a a a

a a a a

U E x E U x R R R

B U U B

l 3 3

3

3 l l

e need to show that (1-d ) , note that if 1 then this holds

if 1, then defender is indifferent between accepting and rejecting

, 1- d , (1-d ) ,

Note: if type

l l

l

l l lw lld d a d d d a d d d dU R B B U R B B U R B

l l

l l

d fights she receives benefits w/ prob. d and benefits w/ prob. (1-d ), her expected amount of benefits is d .

a d d

a d

B B BB B

Page 16: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Jensen’s inequality

l l l

l l l

By Jensen's inequality ,d d , (1-d ) ,

Given that , - this implies that ,d d , (1-d ) ,

Jensen inequality is weak when we ho

l l ld d a d d d a d d d d

l lw lld d d

l lw lld d a d d d a d d d d

U R B B U R B B U R B

R R RU R B B U R B B U R B

3

l 3

ld resources constant, but strict when we allow resources

to vary, fighting being costly.

Because the right-hand side of this inequality is equal to , 1- ,

this implies that d 1- or

l ld d a d

la a

U R B B

B B

l 3

3 h h

3 h

that 1-d , which was what was to be proved.

A similar argument establishes that , d , 0 (1-d ) ,

A proposes if type d is the only type left.

l

w h wl wwa a a a a a a a

h

U R B U R U R B

QED

Page 17: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Result 1

R1) In equilibrium A never provokes a fight with both types

At each negotiating opportunity, - if type dl remains then A prefers proposing γl

t to any γt > γlt

- if type dh remains then A prefers proposing γht to any γt > γh

t

- A makes a proposal that some type of defender will accept

War does not start/endure simply because A is aggressive or doggedly committed to fighting.

- in this model, the attacker never just provokes a fight

Battles are risky and both players are risk averse- factors that affect this choice - γ h

3 , γ l3

- trade-off γ h3 < γ l

3 high risk-return payoff γ l3

This result generalizes to n types

Page 18: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Result 2 R2) A is more likely to propose γt = γlt

- if A is optimistic that D is type dl - if A anticipates low resource losses from fighting - if dl is low

Intuitive , if A’s settlement proposals are viewed in terms of a risk-return tradeoff,

it depends on how A assesses the value of the gamble

1) The more confident A is that the defender is the weak type, the less risky it is to make a tough demand because it is more likely that the demand will be accepted.

2) If A anticipates low resource losses from fighting, then it is not so costly if she is wrong in her assessment of D’s type

3) The lower type dl’s probability of victory is, the tougher a demand A can make because type dl is willing to make more concessions the lower the probability of victory

(making a demanding proposal is risky, but the large reward may make it worth the risk)

Page 19: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Result 3

R3) Beliefs evolve through Bayes rule in response to proposal rejections and battle outcomes

- Proposal rejections and battle losses make A more pessimistic

- Battle victories make A more optimistic

This result implies that war is an important source of information to disputants

In addition, the possibility that diplomacy continues throughout the war also provides a means of revealing information

Page 20: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Results 4 and 5 R4) A’s subjective belief about the likelihood of winning battles falls as the war progresses

Implies that proposal rejections reduce A’s subjective belief about her likelihood of winning battles because weaker defenders are screened out as they accept proposals (R3).

Since on average, ex post equal ex ante beliefs, we can focus on proposal rejections where the weaker types drop out

R5) In expectation, A is more likely to propose terms acceptable to stronger types as the war

continues. On average, war is more likely to end the longer it has endured.

Note: in expectation, battles have no effect on A’s beliefs, because by Bayes’ Rule expected posteriors equal priors. In expectation, the effect of proposal rejections determines how A’s proposals evolve over time. Wars exhibit positive duration dependence; they are more likely to end the longer they have endured.

Page 21: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Analysis The possible equilibrium paths from page 834

• A is unwilling to attack in stage 1, so war does not start. It does not matter what A proposes in this

case; all proposals are rejected.

• A proposes γ h1 and both types accept. War does not start.

• A proposes γ l1 , type dl accepts and type dh rejects.

If D rejects, then A attacks. If A wins the first battle, she proposes γ h

2 and D accepts.

• A proposes γ l1 type dl accepts with probability (0 < w < 1), and

type dh rejects. If D rejects, then A attacks. If A wins the first battle, she proposes γ h

2 and D accepts, and does not fight a second battle.

• A proposes γ l1 , type dl accepts with probability (0 < w1 < 1) and

type dh rejects. If D rejects A attacks. If A wins the first battle, she proposes γ l

2, type dl accepts with (0< w1 ≤ 1) and type dh rejects. If D rejects, then A attacks again

Formal results provide intuition about the conditions under which each path occurs.

Page 22: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Extensions

1) Include third parties, more participants

2) What happens when we put more structure on benefits: physical vs. ideological benefits

3) Separate resources into labor/capital; people care more about labor (casualties) than capital

4) We could endogenize resource choices and allow for the production of military resources, GDP growth, etc.

5) We could allow transfers of benefits through fighting itself (taking territory, for example)

6) There are problems with negotiated settlements. How do you enforce the settlement?

Page 23: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Another extension

7) We may want to abandon equilibrium conditions and perfect foresight

or use equilibrium refinements to allow for two-sided incomplete information

- signaling vs. bargaining

- more dyna

mics

, ; A knows , D does not; D knows , A does not

We could also use , , D's type and compute numerical results

A D A D

A D

d X X X X

d R R

Page 24: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Other extensions

1 2 3

indivisible

1 2 3

8) Provide a richer description of benefits:

...additive

or = multiplicative effect

9) Consider decision making

B B B B

B B B B

within a nation

a. a state is not a unitary actor

b. a coalition of supporters needs to be kept happy

c. some kind of aggregation of preferences occurs

d. regime type may affect aggregate preferences

Page 25: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Conclusion In the model,

War starts1) because the attacker believes her power affords her a greater share

of the benefits than she currently has and thus demands some concessions from the defender

2) because the defender believes that he does not need to make as many concessions as the attacker has demanded and thus refuses to concede

War continues So long as attacker continues to overestimate what defender will concede - or in the event of a retreat, underestimate what she must give up

War ends - when the attacker and defender’s beliefs about the defender’s power converge sufficiently for the attacker to make a proposal acceptable to the defender- at each stage the attacker can obtain a concession only if her threat to attack is credible

Page 26: Bargaining. Divide the melting cake Often bargaining is similar to a game where players make alternating offers about how to divide a cake that is melting.

Testable Hypothesis 1) The attacker never retreats after winning first battle. In general, retreats follow losses not victories

2) The attacker is more likely to make proposals that involve a risk that a war begins or continues a. if she is optimistic that the defender is weak b. if she anticipates low resource losses from fighting c. if her probability of winning battles conditional on the defender

being weak is high

3) Since battle victories make the attacker more optimistic we expect that attacker is more likely to offer an acceptable settlement after a loss than a victory

4) Early losses for the attacker are conducive to a short war

5) The failure of negotiations makes the attacker more pessimistic. We expect that on average the duration of war will positively affect the likelihood that the war will end and negatively affect the attacker’s prospects for victory

6) The final settlement responds to military variables. Diplomacy & force are linked.