Interpreting Prediction Market Prices

Abstract

Prediction market prices are often used as estimates of the probability of outcomes in future elections and referendums. I argue that this practice is often flawed, and I develop a model that empiricists can use to partially identify probabilities from prediction market prices. In the special case of log utility, election outcome probabilities can be fully (point) identified by a simple type of futures contract that is not commonly used in practice. Prediction markets are also used to examine whether stock market valuations would be higher under one election outcome than the other. I show that this question cannot be answered without assuming investors' higher-order beliefs are correct. In the case of the 2016 US presidential election, my model suggests that investors had incorrect higher-order beliefs, and that these incorrect higher-order beliefs affected the aggregate value of the S&P 500 by approximately $400 billion, or 2% of its aggregate value.