If you’ve been following the literature on the predictive power of election betting markets, you may have seen this graph, fromÂ a paper based on data from the first 12 years of the Iowa Electronic Markets. The horizontal axis shows the actual election result, while the vertical axis shows the IEM prediction. A dot on the line shows an accurate prediction; a dot off the line shows an inaccurate prediction.
However, I just stumbled across something that looks surprisingly similar, yet is based on data on the other side of the Atlantic. In “Predictive Accuracy of Political Stock Markets – Empirical Evidence from a European Perspective“,Â Michael Berlemann and Carsten Schmidt look at a bunch of European election prediction markets. Here’s their key graph.
In both cases, the authors look at published opinion polls, and find that the mean absolute error of the poll predictions was higher than the mean absolute error of the betting market predictions. Neither are perfect, but it’s clear which is the superior forecasting tool.
Update: On the topic of prediction markets, my coauthor Justin Wolfers is profiled (along with his research) in today’s New York Times.