Models of bounded rationality often lead to sharper predictions about real world outcomes than their full rationality counterparts. Full rationality in repeated interactions allows a plethora of equilibrium outcomes. In this paper, I examine the effect of bounded rationality in infinitely repeated games. In particular, does the introduction of boundedly rational agents lead to a smaller set of outcomes in equilibrium?
I show that the number of equilibrium outcomes is smaller when agents are boundedly rational. Importantly, cooperative outcomes are still possible in equilibrium, even when players can't use sophisticated strategies and are not able to perfectly monitor their opponents. The strategy that leads to cooperation is called "Win-Stay, Lose-Shift". Using this strategy, I show that cooperation is possible in equilibrium for a large class of 2x2 games.
I also give necessary and sufficient conditions on equilibrium structure for two-player N-by-M games. These conditions suggest that in equilibrium, players must be able to cooperate without getting caught in long periods of conflict.
We propose a decision making process meant to mimic human behavior. This process is implemented with computational agents. We use this computational testbed to run simulations of two coordination games, the minimum-effort coordination game and the battle of the sexes game. We find that the computational agents exhibit behavior similar to human subjects from previous experimental work. We then use the computational testbed to develop experimental hypotheses, which are then confirmed in the laboratory using human subjects. In particular, we show that higher cost may actually lead to higher average payoffs in the minimum-effort coordination game.
Institutions designed to increase turnout appeal to democratic sentiments but are highly debated as they entail two potentially countervailing effects. While generating more pieces of information, they may decrease the average voter's informedness level. We examine two commonly discussed institutions inducing participation: abstention penalties (used in 32 countries around the world) and lotteries providing a prize to one random participant (as proposed on the 2006 Arizona ballot). We analyze a benchmark rational choice model in which voters vary in their information quality and participation is costly. We illustrate that both institutions can improve collective outcomes, though lotteries are a more effective instrument asymptotically. In an array of lab experiments we empirically assess institutional performance. We find strong evidence for selective participation: lab voters participate more when better informed or when institutionally induced. Furthermore, when subjects are sufficiently experienced, lottery prizes (and not abstention sanctions) improve overall welfare in the lab.
In auctions where advertisers compete to place ads in online content, some types of bids are proportional to estimates of probabilities of clicks or conversions. The highest bidder wins the auction. As a result, the auction winner is more likely to have a bid based on an over-estimate than on an under-estimate of a probability, even if the probability estimates are unbiased. This paper explores the impact of this effect on auction revenue and fairness, and it outlines some methods to improve revenue and fairness.
Many auctions for online ad space use estimated offer values and charge the winner based on an estimate of the runner-up offer value. By awarding the ad space to the winner, the auctioneer produces more information about the value of the winning offer but not about the runner-up offer. Since price is based on an estimate of the runner-up offer value, failing to improve the estimate can harm revenue for the publisher and auctioneer. Also, failing to improve the estimate of the runner-up value can cause a less valuable offer to be repeatedly selected over a more valuable one, harming fairness for advertisers. This paper explores the dynamics of the second price and shows that learning the second price can improve revenue and fairness.
This paper looks at the dynamics of a class of coordination games. In particular we propose experiments which will give predictions for equilibrium selection dependent on the previous games played. The predictions are formulated by examining bifurcations in the quantal response equilibrium correspondences. We predict that for the exact same game, we should be able to induce two opposite equilibrium outcomes based on the dynamics leading up to that game. Experimental results confirm this hysteresis in large group minimum effort coordination games.
California Institute of Technology
1200 E. California Blvd. MC 228-77
Pasadena, CA 91125
