When do prediction markets return average beliefs? Experimental evidence

In prediction markets prices can be interpreted as the average belief of the traders under restrictive theoretical assumptions, namely specific risk preferences (e.g., log utility) and the prior information equilibrium. Prior information equilibrium is more likely to hold in a call auction, but prediction markets are usually implemented in double auctions that are known to better aggregate information. In this paper we present a laboratory experiment meant to shed some light on this tension, assessing the influence of the main elements that should affect the equilibrium price also manipulating the market institution. We do not find that risk preferences and incorrect beliefs play a significant role in our data. We find instead that in the double auction– where at least partial information aggregation through belief revisions should be expected – prices are closer to the average belief than in the call auction – where, instead, belief aggregation should be expected. We also find evidence that beliefs are updated in the direction of observed prices, rather than of the true state.