Prediction Markets Are Very Hot, But Here's Why They Can Be So Wrong

The primary season is drawing to a close. The general election looms. And that roaring sound in your ears? It's the hurricane of predictions swirling around the process. Some of the forecasts come from self-proclaimed experts, some from polls. Here's another prophecy: This year, you'll also be hearing a lot more from the latest trendy political handicapping tool — the prediction market.

As you've no doubt heard, prediction markets are online trading sites that let people buy and sell shares tied to, among myriad other things, the fortunes of candidates and parties. The price of these shares, in turn, reveals which candidates are most likely to win in a real election. For the 2008 political season, The Wall Street Journal, CNN, and the National Journal have all started running their own trading sites. These join the ranks of established political markets such as Intrade, founded in 1999, and the venerable Iowa Electronic Markets, launched in 1988.

Prediction markets can be spookily accurate. The Iowa Electronic Markets, for example, has proven more accurate than the polls at least 75 percent of the time since its inception. But like the Democrats, prediction markets will come into the 2008 general election battered from a tough primary season. The big blow came in January, when the markets gave Barack Obama a 91 percent chance of beating Hillary Clinton in New Hampshire. Clinton won, leaving prediction-market boosters looking every bit as chuckleheaded as the pollsters and 24-hour-news blowhards. "Nobody Knows Anything," New York Times columnist Paul Krugman titled a blog post the next day. "But to be more specific, the prediction markets — which you see, again and again, touted as having some mystical power to aggregate information — know no more than the conventional wisdom."

What went wrong? After all, markets are renowned for picking up on last-minute swings like the one that swept Clinton to victory. That's why they are so useful and why they seem to have cropped up everywhere in the past few years. Corporations from Google to Chrysler are running markets to tap the collective knowledge of employees. There are trading sites for sports, Hollywood, even avian flu. Have we all been led astray?

Sort of. Like financial markets, prediction markets are big information processors, distilling the collective wisdom of their traders. But the success of any market depends upon the stakes and the pool of traders. Most prediction markets aren't anywhere near as robust as those they emulate on Wall Street. "They are thin, trading volumes are anemic, and the dollar amounts at risk are pitifully small," market analyst Barry Ritholtz wrote in January. That opens them up to all kinds of problems as information processors. Political markets, for example, have a lot of political junkies but few real insiders or outsiders, so they're not very good at catching something the polls might miss. This is a problem in other markets as well. When Justin Wolfers, a Wharton School economist and leading prediction-market specialist, and two other economists studied Google's employee market, they found that traders tended to make choices similar to those of their pod-mates and neighbors.

So how can prediction markets be rectified? For starters, they need to have real stakes. There is some debate about whether this means money (Wolfers suggests it might not be necessary) or something else, like reputation. But cash is definitely the surest way to grease a market. In June 2007, 25 economists signed a letter urging legislators to grant these markets "safe harbor" from Internet gambling regulations, given the sites' value as forecasting tools.

Beyond this, it's important to improve the pool of traders. According to economists, this requires a certain alchemy of expertise and stupidity. With more experts and insiders, the markets can get out ahead of conventional wisdom. But forecasting also needs more so-called noise traders, who do business with almost no information. Noise traders boost accuracy by increasing volume and the potential profits of informed traders.

Diversity helps, too. If you can get different types of people to play, experts say, not only do you get a bigger pool and more information, but differing random guesses will cancel each other out, leaving real signals to rise above the noise. Plus, if you have a critical mass of investors with a variety of backgrounds, locations, and interests, they are less likely to move as a herd.

As the presidential election draws near and the pundits start talking about the magic of prediction markets, pay attention to whether a given market is likely to have a good variety of traders and that they're playing for something meaningful. If not, stay away. Unless, of course, you happen to have some inside information yourself. In that case, by all means jump in and clean up.

Walter