University of Illinois at Urbana-Champaign
Microeconomic theory, game theory, mechanism design, market design, collective decision making, information elicitation, prediction markets, scoring rules, statistical decision theory, two-sided matching markets.
This paper studies mechanisms for finding the true answer to a binary question using the opinions of biased agents. Taking majority rule as a baseline, I study peer-prediction decision rules, which ask agents to predict the opinions of others in addition to providing their own. Incorporating first-order beliefs into the decision rule has the potential to recognize the correct answer even when the majority is wrong. However, I show the majority rule is essentially the only deterministic, neutral, anonymous, and interim dominance solvable mechanism. I then characterize all randomized peer-prediction mechanisms with these properties, using this result to show majority rule is the optimal mechanism in this class. Finally, I consider a simple, non-incentive-compatible decision rule based on the median prediction that implements majority rule when all agents are strategic and improves on majority rule when an unknown subpopulation is honest. Current version (October 2015)
In this paper, I introduce a class of mechanisms for eliciting private correlated signals from a group of expected score maximizers without external verification or knowledge about the agents’ belief structure. Built on proper scoring rules, these ask agents to report a signal and a prediction of the signals of others. If two agents with the same signal have the same expectations about the signals of others, the Bayesian incentive compatibility of these mechanisms follows with no further assumptions on the agents’ belief structure. With a slight modification, the mechanism is still feasible and incentive compatible when the prediction portion of the report is optional. Current version (June 2014)
We study decentralized, two-sided matching markets as an asychronous, uncoordinated proposal process. The standard focus of the literature is on stable matchings, such as those produced by the Gale-Shapley algorithm. Since in a decentralized setting we arguable should not expect to observe stable matchings, we investigate the typical outcomes of uncoordinated proposals. Though not stable, there is minimal overall welfare loss, with substantially more equitable outcomes between the long and short sides of the market compared to stable matchings. These results suggest there is little to gain from implementing formal matching procedures in markets that are not obviously failing.