Abstract: Prediction markets are a widely-used, accurate, engaging, and intuitive way to crowdsource probabilistic predictions of various outcomes, from basketball tournaments to political elections. Constructed as financial markets for securities whose payoffs depend on the outcomes in question, prediction markets aggregate the beliefs of the crowd by offering well-aligned financial incentives, allowing one to interpret the market price as a consensus prediction.
A particular prediction market framework has gained popularity in recent years, wherein participants trade not directly with each other but with a centralized computational agent called an automated market maker. We will see that while this automated framework enjoys many appealing properties, it lacks a crucial flexibility possessed by more traditional markets: the ability to adapt the magnitude of the market incentives to trading activity and external information shocks. In particular, we may wish the "depth" of the market to increase as more participants arrive, but decrease when information is released such as the outcome of a primary election.
In this talk, I will briefly discuss the history and theory behind prediction markets, and outline the potential-based automated market making framework due to Abernethy, Chen, and Wortman Vaughan. I will then give a range of theoretical findings about the design of adaptive markets using convex geometry, from impossibility results to new mechanisms and techniques. As we will see, although theoretical, these results directly inform the implementation of such automated markets, not only in the pricing mechanism, but in the design of the securities themselves.
Bio: Rafael Frongillo is an Assistant Professor of Computer Science at CU Boulder. His research lies at the interface between theoretical machine learning and economics, primarily focusing on domains such as information elicitation and crowdsourcing which involve the exchange of information for money, and drawing techniques from convex analysis, game theory, optimization, and statistics. Before coming to Boulder, Rafael was a postdoc at the Center for Research on Computation and Society at Harvard University and at Microsoft Research New York, and in 2013 received his Ph.D. in Computer Science at UC Berkeley, advised by Christos Papadimitriou and supported by the NDSEG Fellowship.