Trying to use prediction markets to forecast the presidential election

Published: Tuesday, Sept. 4 2012 7:03 a.m. MDT

You may have noticed there is a presidential election coming up. The Republican Party met this past week in Tampa, Florida and officially nominated Mitt Romney as their candidate. Democrats are meeting this week in Charlotte, North Carolina to renominate President Barack Obama. Of course all that really matters is the counting of electoral college votes, and those will be decided on November 6th.

In the meantime, however, it is at least entertaining to try and predict which candidate will win. There is no shortage of opinion, of course. In the past week I have read that an Obama win is a sure thing, that Romney is sure to win, and that the election is too close to call.

Social scientists have been studying presidential elections for decades and have become quite adept at using statistical analysis to predict their outcome. One of the earliest pieces of evidence is a scholarly paper by economist Ray Fair of Yale University that was published in the Review of Economics and Statistics in 1978. Fair showed that the most important predictor of the outcome was the growth rate of the economy in the year preceding the election. High growth rates substantially boosted the probability of the candidate from the incumbent's party winning the election. The evidence was so strong, in fact, that Fair concluded the contribution of other observable factors was negligible by comparison.

More recent studies – and there is a now a small industry producing them – tend to largely confirm Fair's results. Kenneth Bickers and Michael Berry of the University of Colorado – Denver made the news in August when they released a study showing that economic indicator collected on a state-by-state basis predicted a Republican landslide. Based on data through June their model predicted a 320 – 218 margin for Romney in the electoral college with 52.9 percent of the popular vote going in his favor.

A recent technological innovation in presidential elections is the use of prediction markets. Political markets have been around since at least the early 1990s when a small-scale political derivatives market now called the Iowa Electronics Market (IEM) was set up at the Tippie College of Business at the University of Iowa. The most widely cited source these days is the online trading site InTrade.

Prediction markets like these are used to forecast the results of all sorts of uncertain events, from the Oscar for Best Picture, to the odds that extraterrestrial life will be discovered before the end of 2015. The accuracy of these predictions is built on what finance researchers call the efficient markets hypothesis. Boiled down to its basics, the hypothesis says that all available information about an event is reflected in the price of an asset linked to that event. The intuition being that when large sums of money are at stake investors in these markets will be quite thorough in sifting and analyzing all relevant information.

It is often argued that the predictions from these markets are more accurate than even the best conducted polls. Respondents to polls, after all, have no financial incentive to accurately report their true preferences. Indeed, they may fudge their responses substantially in order to give an answer they think the interviewer wants to hear.

Investors in political prediction markets have access to all publicly available information, so they can certainly incorporate information from polls. But they may also have access to useful private information.

For example, suppose a close friend of a Republican candidate knew an embarrassing secret was about to be uncovered. That friend could make a great deal of money if he were to buy shares in an asset on InTrade or IEM that pays out should the Democrat win.

Get The Deseret News Everywhere