Trading Offers Best Election Predictions

Nov. 7, 2006 — Web-based maps of the United States that dynamically update predictions for the 2006 gubernatorial and senate races could prove more accurate than maps based on opinion polls.

The election maps are the first based on prediction markets. These markets are artificially created and work like the stock exchange — people get rewarded or punished financially for their decisions to buy or sell assets.

The current price of each asset (for example, whether a team will win a certain game or whether a candidate will win a particular race) determines its value and the likelihood of its outcome.

Lance Fortnow, a professor in computer science at the University of Chicago, and team members David Pennock and Yiling Chen of Yahoo! Research in Berkeley, based their maps on the values of election assets being traded for real money on the Dublin-based entertainment website Tradesports.com.

In 2004, values for election assets on Tradesports.com correctly predicted every state's Electoral College outcome and all but one of the races for the Senate, said Fortnow.

Traditionally election predictions have come from opinion polls. But polling doesn't necessarily get to the truth. The questions can be skewed to lead a person to unintentional answers.

People who are not informed about a particular issue may be afraid to admit it and provide dishonest answers. Or, sometimes they will admit it, thereby creating some uncertainty in the final results.

But in a prediction market, only the best and strongest opinions make the most money.

"People tend to be more honest about their opinions when you actually put real money on the table," said Fortnow.

Fortnow and his colleagues wrote a program that automatically downloads probability data from Tradesports.com and, in real time, converts that data into a color.

States favored to go Republican are colored red; those favored to Democratic are colored blue. Lighter shades of both indicate tightly contested races.

After the election results come in, Fortnow's team will determine just how accurate the predictions were.

Using data from a site like Tradesports.com to forecast who will win an  election is just scratching the surface, said Robin Hanson, associate professor of economics at George Mason University in Fairfax, VA, and one of the founders of the field of prediction markets.

Although the economic incentive is high for picking a winner, Hanson would like to see prediction markets forecast the consequences of a candidate getting into office. Will unemployment go up or down? Will we have more or less trouble in Iraq? Will we decrease or increase the deficit?

"The social value of telling people who's likely to win is questionable. The social value of telling people the consequences is arguably far more valuable," said Hanson.