Learning in Prediction Markets

This paper builds on the ideas of Manski (2006) and Wolfers and Zitzewitz (2005) to show when learning is allowed, an electronic market (a prediction market) may aggregate information. In particular, adding learning to the model used in Manski (2006) causes the market price to converge to the mean of the distribution of beliefs. The paper also presents a similar result under less restrictive assumptions on the preferences of individual traders. The paper ends with a discussion of various assumptions about how traders update their beliefs. The paper presents results which suggest that prices from prediction markets may be used to make inferences about mean beliefs and the true state of the world. However, the results presented here and in previous work suggest caution