Incentives and Reputation
Darwin on reputation Man‘s] motive to give aid […] no longer consists of a blind instinctive impulse, but is largely influenced by the praise and blame of his fellow men.
Indirect Reciprocity
Direct vs indirect reciprocity ‚to help‘ means: confer benefit b at own cost c
Binary model Each player has a binary reputation G good or B bad Individuals meet randomly, as Donor and Recipient Donor can give benefit b to Recipient at cost c If Donor gives, Donor´s reputation G if not, Donor‘s reputation B Discrimination: give only to G-player (SCORING) Undiscriminate stategies AllC and AllD
SCORING vs. AllC and AllD
The paradox of SCORING Why should one discriminate? (it reduces chances of being helped later) Discrimination is costly AllC can invade
Assessment What is ‚bad‘? (rudimentary moral systems) SCORING: bad is to refuse help SUGDEN: bad is to refuse help to good player KANDORI: bad is (in addition) to help bad player
Assessment rules First order: is help given or not? Second order: is recipient good or bad? Third order: is donor good or bad? 256 assessment rules (value systems) (Ohtsuki, Iwasa; Brandt et al;2004)
Assessment rules First order: is help given or not? Second order: is recipient good or bad? Third order: is donor good or bad? Only eight strategies lead to cooperation and cannot be invaded by other action rules, e.g. by AllC or AllD (Ohtsuki, Iwasa 2004)
Assessment What is ‚bad‘? (rudimentary moral systems) SCORING: bad is to refuse help SUGDEN: bad is to refuse help to good player KANDORI: bad is (in addition) to help bad player
The leading eight L3 (SUGDEN) and L6 (KANDORI) are second order assessment rules, the others third order (L1 considered in Panchanathan-Boyd and Leimar-Hammerstein)
SUGDEN (or KANDORI) vs. AllC and AllD
The competition of SUGDEN and KANDORI Must assume private image (Brandt and Sigmund, Pacheco et al) rather than public image (Ohtsuki and Iwasa, Panchanathan and Boyd)
AllC AllD Sugden Stable fixed points (Mixture of K and S) Kandori
Incentives
Ultimatum game Two players can share 10 euros Toss of coin decides who is proposer, who is responder Proposer offers share to Responder Responder accepts, or declines.
What does homo oeconomicus? If each player maximises payoff: Proposer offers minimal share, Responder accepts
What do we do? In real life: 60 to 80 percent of all offers between 40 et 50 percent Less than 5 percent of all offers below 20 percent
Economic anthropology Henrich et al, Amer. Econ. Review 2001
Variants of Ultimatum Against computer Against five responders Against five proposers
Ultimatum for mathematicians strategy (p,q) p size of offer, if Proposer q aspiration level, if Responder (percentage of total)
Mini-Ultimatum Only two possible offers High offer H (40 %) Low offer L (20 %)
Mini-Ultimatum
Asymmetric Games
Conditional Strategies
Conditional Strategies
Conditional Strategies
Conditional Strategies
Conditional Strategies
Mini-Ultimatum Population of players Types (H,H) (social) (L,L) (asocial) (H,L) (mild) (L,H) (paradoxical) Players copy whoever is successful
Mini-Ultimatum
Mini-Ultimatum
Reputation and temptation Suppose that with a small probability Players have information about type of co-player (reputation) and makes reduced offer L if co-player has low aspiration level (temptation)
Mini-Ultimatum with reputation and temptation
Mini-Ultimatum with reputation-temptation Bistability Attractors HH (social) and LL (asocial)
Mini-Ultimatum with reputation-temptation Bistability Attractors HH (social) and LL (asocial) Social stronger if H<1/2
Bifurcation
Back to full ultimatum Evolution leads to minimal offers (as with rational players) With reputation-temptation to values between 40 and 50 percent
Individual-based simulations
Individual-based simulations
An economic experiment Ultimatum with or without reputation (Fehr and Fischbacher, Nature 2004)
What if someone is watching? Experiments by Haley, Fessler By Bateson et al (honesty box)
Trust Game Investor can send amount c to Trustee, knowing it will be multiplied by factor r>1 on arrival Trustee, on receiving b=rc, can send part of it back to Investor
Mini-Trust
Mini-Trust
Mini-Trust with Reputation
Incentives for cooperation First, play a donation game (or a more complex game, involving cooperation), then punish the defector or reward the cooperator (same structure as ultimatum or trust)
PD with Reward
PD with Reward with reputation
PD with Reward with reputation
Payoff
Results:
low information high information