Intrade predicts outcomes

For Intrade.com mavens, Sen. Barack Obama’s victory over Sen. John McCain on Election Day was old news. The futures Web site forecasted the landslide in early October.

Intrade’s successful prediction of the outcome of the 2008 presidential election also came as no surprise to several Yale professors, who created two statistical models based on Intrade contract prices, which measure the likelihood an event will occur. The models — one created by economics professor Ray Fair and the other by School of Management professors Keith Chen, Jonathan Ingersoll, Jr. and Edward Kaplan — assume that prediction markets like Intrade process information to produce the most accurate forecasts of future events, including this week’s presidential election.

“We went out on a limb projecting such a large victory for Obama so early compared to what most pundits predicted,” said Chad Rigetti GRD ’05, a vice president of business development at Intrade, in a telephone interview. “And we got it right.”

Though Missouri and North Carolina remained toss-ups Wednesday night, Intrade — which accurately predicted the electoral vote count in the 2004 presidential election — appears poised to repeat such a feat this year, although it predicted Obama would win Missouri and the state is right now leaning Republican by a thin margin.

On Intrade, people place bets with a potential settlement value of $10 on future events. Before the event occurs, people buy and sell these $10 lots at prices from one cent to $10 based on the perception that the event will occur. For the 2008 presidential election, over $110 million — equal to 11 million bets — changed hands in presidential election-related contracts, increasing the likelihood that the contract price for a candidate’s victory in each individual state accurately reflected voting in the state.

Fair’s model examined the contract prices in each state, ranking them in the order of highest price — for the model, states leaning towards McCain — to lowest. The model then tallied up the electoral votes in McCain-leaning states until the number reached 270, the number of votes needed to win the presidency. For the first state McCain did not need to win the election, Fair called the state’s contract price a “pivot point.”

For the 2008 presidential election, Fair found the pivot point prices over time corresponded closely to the nationwide price on the likelihood of a McCain victory.

“On Election Day, for example, New Mexico’s — the pivot point — price [$1.00] nearly matched the national price for a McCain win [87 cents],” Fair said.

Chen, Ingersoll and Kaplan’s model also used Intrade.com’s state-by-state contract prices to create a model to predict the outcome of the election. But unlike Fair’s model, which assumes that after one state in the ranking model votes for the other candidate, all states with prices lower than that state’s price will vote for that same candidate, their model assumed less correlation between the states’ outcomes.

“The Fair model is a special case of our model,” Chen said. “Our model allows for differential effects driven by geographic, racial distribution, and so forth.”

Chen noted that the Fair model failed slightly in the 2008 election, as Missouri, a state which should have voted for Obama in Fair’s model, is right now tilting toward McCain’s column.

Although statistics for Chen, Ingersoll and Kaplan’s model will not be released for weeks, Chen said preliminary election results match well with their model.

Despite Intrade’s success, some academics warned against placing blind trust in prediction markets. Christopher Wlezien, a professor at Temple University, said prediction markets adjust slowly — and often cautiously — to new information.

“After the Republican convention, Intrade.com predicted a McCain victory for days,” Wlezien said, saying polls during that time showed Obama leading. “Something’s not quite right when those arbitrage opportunities just lay out there on the market.”

Comments