Mash-ups pick the next president

I’ve barely finished watching the Daily Show’s analysis of Hillary’s comeback Tuesday and already I’m getting twitchy for fresh punditry and predictions. I want to know today what’s going to happen on April 22 — when Pennsylvania, a delegate-loaded state, holds its primary.

That’s why I’m on Intrade.com, one of a slew of prediction markets that have become all the rage this election season. By simulating a stock market, where Obama and Clinton are both future contracts that can be traded by anyone who opens an account, the site tells me that Clinton has a 72 percent chance of winning Pennsylvania.

Not what I want to hear. So I flip the market to the North Carolina primary on May 6. It turns out Obama has a 75 percent chance of beating Clinton here.

Now, no sane pollster (and with this primary season dragging into summer, there might not be a lot of them left) would dare to predict a primary almost two months away. That’s the genius of these prediction markets, a clever system that avoids the problems with “expert bias” by turning to the “wisdom of the crowd.”

It’s pretty much agreed that prediction markets are more accurate than individual polls. Take for instance the Iowa Electronic Markets, which, on average of three times out of four, has out-matched polls in predicting the presidential winner for the past five elections. Don’t just take my word for it; columnists over at Wall Street Journal, Financial Times and The New York Times are clambering over each other for original superlatives to lavish on this little innovation.

But here’s where it actually gets interesting. What if you compare prediction markets to composite polls?

Known as mash-ups, composite polls combine the myriad of individual polls using little more than basic arithmetic. Yet these rather lazy statisticians — the most popular of which are RealClearPolitics.com and Pollster.com — are becoming the industry standard. In the Missouri primary, Zogby, the noted polling firm, predicted Clinton to win by 11 percent while other polls showed Obama winning by large margins. The mash-ups showed a truer picture — Obama winning — but barely.

And in this week’s Ohio primary, one poll had Clinton winning by 14 percent while others had the two in a dead heat. The composites won out again. Most convincingly, a recent matchup among all polls showed SurveyUSA, a composite poll, at the top of the heap, with the lowest margins of error at two percent.

So which one should you go by, prediction markets or mash-ups? The answer could be neither. Exhibit A is the New Hampshire primary, for which both went off the deep end. Intrade, the prediction market, had Obama winning by 70 percent before the polling stations closed. Then, as ominous election results for Obama trickled in, his chance slid to 56 percent. Clinton ended up winning.

The composite polls didn’t predict Clinton’s win either. So what happened? The roller-coaster prices for the futures of Obama and Clinton on that night revealed a fatal flaw in the system: They’re no better than the best of the polls, for that’s where the traders get their information.

The futures markets weren’t able to predict the future — they just told you what happened in the past. When television pundits shifted sides to Clinton, so did the traders (thus the decline for Obama from 70 to 56 percent). And that wasn’t the only time. Prediction markets mistakenly called the 2006 congressional elections for the Republicans, Iowa for Howard Dean in 2004 and even a guilty verdict for the Michael Jackson case.

As for the mash-ups, they threw in the good polls with the bad. They also tend to average out the extremes so that the composite is rather tame. Every pollster thought Obama had New Hampshire in the bag; the composite polls merely reflected this mob mentality.

What’s a guy to do? The natural evolution is of course toward composite prediction markets! But ack, Slate.com has already beaten me to the punch.

Jerry Guo is a junior in Timothy Dwight College. His column runs on alternate Fridays.

Comments