Modeling a Presidential Prediction Market

Abstract

Prediction markets now cover many important political events. The 2004 presidential election featured an active online prediction market at Intrade.com, where securities addressing many different election-related outcomes were traded. Using the 2004 data from this market, we examined three alternative models for these security prices, with special focus on the electoral college rules that govern U.S. presidential elections to see which models are more (or less) consistent with the data. The data reveal dependencies in the evolution of the security prices across states over time. We show that a simple diffusion model provides a good description of the overall probability distribution of electoral college votes, and an even simpler ranking model provides excellent predictions of the probability of winning the presidency. Ignoring dependencies in the evolution of security prices across states leads to considerable underestimation of the variance of the number of electoral college votes received by a candidate, which in turn leads to overconfidence in predicting whether that candidate will win the election. Overall, the security prices in the Intrade presidential election prediction market appear jointly consistent with probability models that satisfy the rules of the electoral college.