PUBLIC INFORMATION BIAS AND PREDICTION MARKET ACCURACY

Abstract

How do prediction markets achieve high levels of accuracy? We propose that, in some situations, traders in prediction markets improve upon publicly available information. Specifically, when there is a known bias in publicly available information, markets provide an incentive for traders to “de-bias” this information. In such a situation, a prediction market will provide a more accurate forecast than the public information available to traders. We test our conjecture using real-money prediction markets for seven local elections in the United States. We find that the prediction market forecasts are significantly more accurate than those generated using the pre-election polls.Previous versions of this paper were presented at the Workshop on The Growth of Gambling and Prediction Markets, the ISBM Conference on Prediction Markets in Marketing: Issues, Challenges and Research Opportunities, the DIMACS Workshop on Markets as Predictive Devices, and the 17th Annual AMA Advanced Research Techniques Forum. The authors thank the participants for their insights which have helped improve this research