Politics

Bookies, the Best Bet for Ballot Prediction

ON ELECTION EVE, political campaign managers wanting to know how their candidate or issue will fare could pay big money and call a pollster. Or better yet, suggests new research by Justin Wolfers, call a sportsbook. Especially if the race is tight.

Wolfers, an assistant professor of political economy who as a youth worked for a bookmaker in his native Australia, followed a hunch about the predictive power of betting markets in forecasting the outcome of political elections. With Andrew Leigh of Harvard’s John F. Kennedy School of Government, he coauthored a study examining the effectiveness of three tools for forecasting the outcome of Australia’s 2001 federal elections: economic modeling, opinion polling, and betting odds.

While the study found that all three methods performed reasonably well, it introduced, for the first time in Australia, a new source of data for predicting elections—betting statistics from one of the country’s largest bookmakers. The study’s provocative conclusion is this: Particularly in marginal seats, the press might better serve by reporting betting odds than by conducting polls.

Wolfers sets the scene: “Throughout much of the election cycle the candidate on the left, Kim Beazley of the Australian Labor Party, had been expected to win as prime minister in the November election. John Howard, the incumbent and leader of the Australian Liberal Party, saw the tide turn in his favor in the days following the Sept. 11 terrorist attacks on America when the population rallied around its leader. At about the same time—so it’s hard to untangle the two events—a boatload of Afghan refugees was found off the coast of Australia. Howard took a strong stand against allowing them to immigrate, while Beazley chose something in the middle ground and was perceived to be a weak leader. This was argued by many political commentators in Australia to be the turning point of the election.”

When the election was held on Nov. 10, 2001, the Liberal-National Party gained 50.5 percent of the vote and John Howard was re-elected prime minister.

So, how did the three forecasting tools perform? In the medium term, which Wolfers identifies as one to two years out, economic modeling—based on predictions of how voters will react to various economic conditions—can be effective in picking the election-day winner. This is somewhat surprising, he notes, because election forecasting models have had mixed success, confounding political science researchers studying the impact of such economic indicators as unemployment and inflation on 18 postwar elections in Australia. However, if accurate economic measures are available, the forecasting power of economic modeling is quite substantial.

Using election-eve measures of economic indicators, Wolfers found that three econometric models performed extremely well, nailing precisely the predictions of an incumbent victory in one model and missing by 0.1 percent and 0.4 percent in two others.

Pre-election opinion polls should be more accurate in Australia than in countries like the United States, he points out, because voting in Australia is compulsory, eliminating the key variable of whether respondents will actually show up at the voting booth. Past experience indicates that opinion polls taken close to the election are quite accurate. Yet the lesson from Australia in 2001, like America’s 2000 election imbroglio in Florida, points to the potential pitfalls of polling, particularly in very tight races. “The night before the election, Howard was ahead in two of three major polls,” says Wolfers.

In contrast, data from Centrebet, Australia’s largest bookmaker, demonstrated the impact of current events on the betting odds throughout the nine months leading up to the election by reflecting immediately the electorate’s seesawing response to such events as leaders’ televised debates and the Sept. 11 attacks on the United States. “By election day an enormous amount of money had been pumped into the betting market. More money had been bet on the election than on the football grand final, and Howard was the favorite with odds of $1.55, suggesting a 64 percent probability of his winning the election,” says Wolfers. Howard won and Centrebet lost money on the election.

Digging deeper, Wolfers found the data yielded even more impressive results for oddsmakers. Centrebet also offered odds on the outcomes in 47 regional races. In 43 of 47 cases, the betting favorite won the election, even in the marginal seats. Moreover, in the three regional races where opinion polls had been conducted, the polls correctly predicted the winner in two contests; the betting market got all three right.

“The rationale for this happening comes from finance, which says that markets are efficient aggregators of information and equilibrium prices should reflect all available information,” says Wolfers. “The data suggests the Australian betting market is extraordinarily efficient. And why not? There’s a huge incentive for participants to do their homework, collect reliable information, and make sure the price is right.”

Stateside, Wolfers anticipates political scientists will look closely at the rich data sources in the Las Vegas betting markets during the next major u.s. election. Anyone care to make a wager on it?

—HELEN K. CHANG

Three Tools for Forecasting Federal Elections: Lessons from 2001, Justin Wolfers, GSB Research Paper #1723, November 2001; Australian Journal of Political Science, July 2002