On Relational Interfaces

Julian Romero
CalTech

Friday, November 6, 2009 at 1:00pm
Engr IV Room 57-124

Abstract
The folk theorem for repeated games relies heavily on the assumption of perfect information. This leads to a large set of equilibrium strategies and payoffs. Does this large set of equilibria still remain when monitoring becomes imperfect? In this paper, I study two player, non-discounted, infinitely repeated games with imperfect private monitoring. Players choose finite automata to implement their strategies. I give necessary and sufficient conditions on equilibrium structure in the limit as monitoring become almost perfect. As a result many equilibria from the perfect monitoring case fail to be equilibrium when signals are imperfect. Using the remaining equilibria automata I show that the folk theorem still holds. However, the strategies used for the folk theorem exhibit fast decreases in payoffs as the signal errors become larger. To better understand behavior with larger errors, I restrict the set of strategies to two state automata. In repeated prisoner's dilemma game, for a large range of payoffs and signal errors, there are at most two equilibria, one which cooperates and one which defects. These equilibrium remain equilibrium over a large range of payoff and error parameters. I also give evidence that these equilibria are used in experiments with human subjects.

Biography
Julian Romero is a Ph.D. candidate in the department of Humanities and Social Science at the California Institute of Technology advised by John Ledyard. He is a member of the Social And Information Sciences Laboratory (SISL). He receive his BA from Northwestern University in Economics and Mathematics in 2005. His research is focused on game theory, behavioral economics, and experimental economics with a particular interest in how humans coordinate and cooperate in repeated interactions.