Collaborative Research: Robust Predictions In Games With Private Information

Princeton University, Princeton NJ

Investigators

Abstract

In many economic environments the participants do not have the same information about the relevant aspects of their situation. For example, in a financial market each investor may have different pieces of information about the underlying value of an asset or a company. In product markets, each competiting firm may have private information about its cost structure. In auctions, such as the wireless spectrum auctions organized by the Federal Communications Commission or the offshore oil tracts auctioned off by the U.S. Department of the Interior, the bidders have private information about their valuations and also some, possibly noisy, information about the valuations of their competitors. The private information of an agent, whether it is about his/her own valuations or the valuations of the competing agents will determine the behavior and the strategy of the agent. But in practice, the theoretical and/or empirical analysts do not have a clear understanding of what the economic agents know and, in particular, what they do or do not know about each other. The missing and incomplete information si also highly relevant from a policy perspective as the supervising or regulatory authority in many markets face similar informational constraints. The PIs develop methods that allow predictions about the behavior of economic agents and the reulting market allocations in such settings. These predictions are robust to a large class of private information structures by the agents. They identify conditions on the nature of the interaction and the market under which it is possible to make unique predictions about economic behavior. When these conditions are not met, the PIs establish robust bounds on the distribution of possible economic outcomes, especially the mean and the variance of the outcomes. They use these bounds to predict the welfare impact of information sharing in markets. They also use the bounds to design robust mechanisms, such as auction and voting methods, that perform well in many information environments. This project provides an economic theory that can lead to practical solutions to market and mechanism design problems. It also provides a theoretical foundation for future work in econometrics aimed at testing models in situations where the researcher does not know the market's information structure. This research will have broader impacts for the many different areas of social science that use game theory as a fundamental tool. In addition, the project has direct implications for antitrust policy, since whether or not to allow sellers to share information is a question for price-fixing policy.