ICES: Small: Information Elicitation and Aggregation in Market Mechanisms

Stanford University, Stanford CA

Investigators

Abstract

Situations in which a decision maker has less information relevant for her decision than does some other agent occur in most areas of economic activity. In such situations, the less informed decision maker would like to obtain information from the more informed agent, or a group of agents. This award funds research on the properties of various mechanisms that can be used to elicit such information, both in the case of a single informed agent and in the case of multiple partially informed agents. The first part of the project studies the single-agent case, the origins of which go back to the literature on scoring rules. While most of the past work has focused on eliciting either the full information about the distribution of the random variable or only the information about its mean, the project's goal is to generalize this theory to the elicitation of any set of characteristics of the distribution, such as, e.g., the mean and the variance, the quantiles, confidence intervals, etc., and to provide a full characterization of sets that can be elicited on their own using a limited number of questions and those that cannot. The second part of the project studies multi-agent mechanisms in a one-period setting. Each informed agent makes a prediction only once, though unlike the single-agent case, the payment that the agent receives now depends not only on his prediction and the outcome, but also on the predictions of other agents. The goal is to provide a full axiomatic characterization of static mechanisms that satisfy several desirable properties, and analyze the performance of such mechanisms. The third part of the project considers dynamic, multi-period market mechanisms, in which strategic agents observe each other's actions, learn from them, and adjust their forecasts accordingly. The research into the properties of such mechanisms is important both for the design of new market-based information elicitation mechanisms (such as, e.g., prediction markets operated by companies like Hewlett-Packard, Electronic Arts, and Google to forecast demand and other variables) and for understanding the informational properties of financial markets that already exist. In particular, this subproject bridges a gap between the non-strategic literature on the informativeness of prices in rational expectations equilibria and the fully strategic literature on the behavior of traders in dynamic market models. The broader impact of this award has several components. The results of this research will be included in the courses taught to graduate students in economics, finance, computer science, and operations research. The research itself will involve students from these disciplines exploring the role of information in markets and single-agent settings, its aggregation and elicitation, and other related issues. Finally, this research has potential for a direct impact on the practical design of information elicitation mechanisms, such as, e.g., corporate prediction markets or information exchanges that allow traders to buy and sell information.