The Predictive Power of Herds
by Noam Lupu

The past few years have seen a proliferation of so-called "prediction markets." With a bit of cash and Internet access, you can bet your money on whether the Fed will raise interest rates at its next meeting, how much The Incredibles will make in its opening weekend or who will run for president in 2008. The reason is not only that traders are bored with merely predicting the price of Microsoft stock or crude oil. There is something truly fascinating about these kinds of markets: they make surprisingly good predictions.

Take a now-classic example. In 1987, finance professor Jack Treynor conducted a simple experiment in one of his classes. He passed around a jar of 850 jellybeans and asked each student to guess the number. The aggregate of the students' guesses was remarkably good — 871. In fact, only one of the 56 students in the class had made a better guess.

James Surowiecki, who writes the "Financial Page" column for the New Yorker, picked up on this impressive phenomenon in his recent book, "The Wisdom of Crowds: Why the Many Are Smarter than the Few and How Collective Wisdom Shapes Business, Economies, Societies and Nations." Surowiecki essentially argues that under the right circumstances, groups are better at predicting an outcome than any single individual in the group. A more scientific study published by the National Bureau of Economic Research in May basically concurs.

That's why last year the Defense Advanced Research Projects Agency (DARPA) attempted (unsuccessfully) to establish a prediction market in economic and political events, including when and where the next terrorist attack would take place. If these markets could efficiently assess political risks, the argument went, they could inform defense strategy.

But there's an important catch to prediction markets: markets can be wrong. That was the case last Tuesday when prediction markets forecast that John Kerry would win the presidency. Understanding why the markets got it wrong is an important caveat to Surowiecki's thesis and to the hype about prediction markets.

The "political futures markets" — those prediction markets that trade in political outcomes — basically work like the stock market. You buy "winner-takes-all" contracts that are worth $1 if your candidate wins. So suppose on Aug. 1 I expected Kerry to win. I could have bought Kerry contracts that day for 47 cents each. If Kerry won, I would earn 53 cents on each contract.

As more people enter the political futures market and make decisions about buying or selling their contracts, the market becomes an aggregation of expectations. If, as a group, people expect Bush to win, the price of a Bush contract should be higher than the price of a Kerry contract. And if you believe, like Surowiecki, that the group makes good predictions, you would expect the price of each contract to translate, more or less, to the popular vote: a Kerry contract selling at 52 cents would mean Kerry gets 52 percent of the votes.

In fact, the political futures markets came eerily close. On Nov. 1, the two main exchanges for political futures — the Iowa Electronics Markets (IEM) and TradeSports — both showed Bush winning. Bush contracts traded at 51.2 cents on IEM and 53 cents on TradeSports. In the end, Bush in fact won nearly 52 percent of the popular vote.

It might have been a success story for prediction markets. But on November 2, the political futures markets reversed completely. The afternoon of Election Day, exit polls began to leak suggesting a surprise victory for Kerry. Anyone looking at the exit polls had to be skeptical. They gave almost every swing state to Kerry — Florida, Iowa, Ohio, Pennsylvania, Michigan, Minnesota, New Hampshire, New Mexico and Wisconsin. And exit polls are notoriously inaccurate. In 2000, they had Al Gore up by 3 points in Florida and gave Iowa, Pennsylvania and Wisconsin to Bush (all three went to Gore).

By 4:30 p.m. on Election Day, Kerry contracts were trading near 70 cents. TradeSports predicted a 39 percent chance Bush would win Ohio and a 43 percent chance he would carry Florida. The afternoon became what Slate's Mickey Kaus called John Kerry's "seven-hour presidency." But by 10:30 that night, Bush contracts were trading at 68.9 cents and his re-election seemed more than likely.

So even though they had followed 10 months of campaigns, surveys and predictions, political futures traders panicked over a few unreliable bits of information. Traders are just people after all, and people can be irrational. The political futures traders had money staked on the outcome. And when they saw other traders selling off Bush and buying up Kerry, they followed the herd.

It is this apparent herd behavior that economic historians have blamed for countless stock market crashes and bubbles, including the IT bubble of the '90s. Yale economist Robert Shiller, for example, has shown that stock prices are much more volatile than the true values of companies. And other behavioral economists have shown that investors tend to give undue emphasis to recent or high-profile news over long-term trends.

Still, Surowiecki argues that prediction markets are different. In the stock market, there is no finite point at which you can evaluate whether you were right or wrong. Investors can convince themselves that a company is undervalued and continue indefinitely to send the stock price soaring beyond all reason. Not so with prediction markets. They have a finite date by which you know the outcome. On Nov. 2 (or thereabouts), for example, you know who won the election. And that, Surowicki argues, "keeps the crowd tethered to reality." They are less likely to be swayed by passing fads or the latest news reports.

It may in fact turn out that prediction markets are, as Surowiecki suggests, more accurate than financial markets, but there are as yet no comparative studies to back him up. Even so, being more accurate still leaves enormous room for error, certainly more error than government agencies like DARPA can afford. More importantly, if any lesson should be learned from the behavior of political futures markets last week, it is that even prediction markets can blow bubbles.

E-mail Noam Lupu at noam at flakmag dot com.