Computer Scientists Track Prediction Markets In Run-up To US Elections

The computer scientists have devised Web sites that display continuously updated color maps predicting the outcome of the 2006 gubernatorial and senate races. The predictions are linked to the prices of securities at Tradesports.com, an Irish trading site that runs a market for each state with a gubernatorial or senate race.

"The prices of the securities have in the past shown to be a surprisingly accurate prediction of future events," said Lance Fortnow, Professor in Computer Science at the University of Chicago. "In 2004, these markets correctly predicted all but one of the senate races and every state correctly in the electoral college. We put the map together to highlight the importance of these markets and let people get a quick view of what the markets say."

The maps, a joint project of Fortnow and David Pennock and Yiling Chen of Yahoo! Research, may be viewed at http://www.fortnow.com/governor/ and http://www.fortnow.com/senate/. Solid red states are those that are heavily favored to go Republican, according to Tradesports data, while solid blue states are likely to go Democratic. States of lighter shades indicate more tightly contested races.

Electoral-vote.com also posts colored maps showing the political leanings of the states. Those maps are based on conventional opinion polling and are unrelated to prediction markets, Pennock said.

Information markets such as Tradesports, also called prediction markets, react extremely quickly to news. This enables visitors to the Web sites to obtain an immediate view of how a breaking event, like the one that led to the resignation of Florida Congressman Mark Foley, might affect an election, Fortnow said.

But as a theoretical computer scientist, Fortnow also is interested in analyzing the computational power of the markets to see if he can understand how it works and predict what it will do. "When is it actually going to do a very good job of aggregating information? When is it not going to do a very good job of aggregating information based on the market setup?" Fortnow asked.

Information markets have proven so efficient in combining the collective wisdom of their participants that some companies operate them internally to help make decisions, including HP, Google and Microsoft. Company executives can, for example, learn from the markets whether or not their employees think that a new product will be released on time. But unlike Tradesports, which may be illegal in the United States, the markets are structured so that employees can win prizes without taking financial risk.

In 2003, a terrorism futures market at the Pentagon created a furor that led to the resignation of retired Admiral John Poindexter, who headed the project. "It put a damper on things," Fortnow said. But prediction markets are once again generating interest in some circles. The interest in prediction markets is fueled in part by James Surowiecki's book The Wisdom of Crowds, Pennock said.

"This area is definitely gaining traction, both in the academic world and maybe even more so in the corporate world," Pennock said.

Yahoo! Research, the research and development division of Yahoo!, runs a prediction market called the Tech Buzz game (see http://buzz.research.yahoo.com/bk/index.html).

"It's a play-money market where people try to predict what technologies people will be searching for," Pennock said, whether it be an internet browser or an MP3 player. The game evaluates both the power of prediction markets to forecast high-tech trends and a Yahoo! Research system for conducting electronic commerce.

Prediction markets also have spawned some startups, including Inklingmarkets.com and NewsFutures.com. "Inklingmarkets offers a hosted solution where it's easy for any company to create a private prediction market for internal corporate forecasts and decisions," Pennock said. "Newsfutures provides software and services for companies to run internal markets, and also will host special challenge markets on their website."

Fortnow, meanwhile, sees interesting computational issues in the ability of a group of people to access and process information that goes into the buying and selling of securities and other financial decisions.

"My main research is studying different models of computation, and to me information markets are another model of computation," he said. "We still know so little about why they work well. There are big questions about how much traffic they need to be interesting, and how easy or how hard are they to manipulate."

Prediction markets will be among the topics addressed at the Conference on Electronic Commerce, sponsored by the Association of Computing Machinery, June 11 to 15, 2007, in San Diego, Calif. The Center for Discrete Mathematics and Theoretical Computer Science in New Jersey has also organized a workshop series that is focused on issues pertaining to the social and economic sciences.

"There's been a growing interaction between theoretical computer science and economic issues," Fortnow said.