The Problem with Prediction Markets

Prediction markets — websites where you can bet on politics — continue to rise in prominence. This election season, their “odds” are often quoted in news stories alongside poll averages.

Of course, these odds aren’t worth much if there aren’t many people betting, and there aren’t. This article by Philip Wallach is an excellent survey of the landscape, and as he points out, the US election has 50 times more betting volume at Betfair (a UK sports book) than at PredictIt, the largest current prediction market.

Social scientists have been infatuated with the idea of prediction markets for decades, because they think they’d provide useful information. But gambling is first and foremost a habit, and the problem with these markets is that there’s not enough to bet on to get people in the habit of coming back. The political world simply doesn’t offer frequent enough events that enough people care about.

These two variables — frequency and interest — can explain the popularity of sports betting and day trading, which are the two most widespread forms of gambling that prediction markets could emulate. You probably care a lot about how your favorite team or fantasy players perform, and they play often; meanwhile you may not care as much where the price of gold or Apple stock is in an hour, but the feedback is even more frequent (and traders are highly vulnerable to conspiracy theories about the Fed or market manipulation, because they want to care).

Meanwhile, you may care who wins the election, but if it’s weeks or months away, what’s going to keep you coming back to the site? Suppose you came to PredictIt in June convinced that Clinton would win, saw that she was “trading” at 66% (1/2 or -200) odds, and bet $100 — so you’d win about $50 if she won the election. Now you come back during her post-convention poll bounce and she’s at 75%; what do you do? You could cash out with a $15 profit, but then what? You could have made $50. Or you could add to your bet at worse odds. Neither is particularly tempting. My guess is you’ll do nothing, and pretty soon you’ll stop visiting the site as frequently.

One way to describe that problem is that the site is trying to get you to think about sports-like betting in a financial market-like framework. But even compared to sports, the idea of waiting months for your bet to settle is frustrating. Most of the betting on sports happens much closer to the event. (According to this chart, 50% of the money bet on an average soccer match is wagered within the preceding two days.)

PredictIt (which is run at least in part by John Aristotle Philips, an interesting character in his own right) has gotten a lot of things right in terms of basic market structure and fees that previous prediction markets like Intrade got wrong, but I’m not sure how they can solve this basic problem.

One solution would be to mix in political contracts with sports, like they do at Betfair, or with financial markets, as Intrade tried to do at one point. (“Will the Dow close above x?”) But regulators in the US are not about to allow either of those things. Even these political markets require an exemption from the CFTC to operate.

What they’re doing instead at PredictIt is creating lots of arbitrary questions to bet on, so there’s always something new and always something settling soon — like the exact poll average on some date:


There’s not much trading volume on these questions, because, not to put too fine a point on it, no one cares about them. All you’ve got are the would-be bookies trading with each other.

Let’s go back to our Hillary example for a moment. If you were a perfect rational actor in an economics textbook, you’d go in both times with your exact estimate of her odds of winning, having mentally updated it for events like the convention, and judge the expected value or rate of return of your bet vs. alternative uses of your capital given what the market’s offering, so you’d wind up transacting more frequently before the event itself. You’d also think about the arbitrage opportunities created by closely related markets (will Clinton win, will the Democratic candidate win, will a female candidate win). But very few people really think that way, and for many it’s stressful and unpleasant to be forced to do so.

I’m reminded of another site called Stickk, also started by social scientists, where you can make “commitment contracts” with yourself to quit a bad habit, and attach monetary penalties: for example, you might commit to quit smoking by year end or pay the site $500. When it launched, Tyler Cowen was skeptical, and he didn’t mince words:

I’ve long predicted this won’t work; one group of potential customers doesn’t really want to change, the other group is unwilling to give up control.  It’s not exaggerating to say that human nature is on the line here, and that if I am wrong this is probably the most important idea you will ever encounter.

Almost ten years later, it’s fair to say that Cowen was right. The site is still around, but its actual usage is, I’m sure, a fraction of what the founders had hoped for.

Prediction markets are running into an analagous problem, and it’s a very tough problem. Even if the novelty of the idea gets people to try it, how do you get them to keep coming back?

Remember those two key factors: interest and frequency. If you wanted to play to interest, you’d let people bet on lots of non-political subjects. For some of these there are too many people with advance information (Oscar winners, reality show outcomes, who will die on Game of Thrones) and the market could end up spoiling the surprise. For others, like the outcome of high profile trials, there may be considerations of taste or privacy. But from a moral perspective it’s probably a lot better than focusing on frequency, like lotteries or casinos, and trying to get people hooked on betting on the weather or traffic or some other quasi-random pattern.

I fear that those of us who want to see prediction markets succeed (and I do) have not been focused enough on the core problem: we’re asking people to bet on a subject that 90% of us only care about (in the US) every four years, if even then. Without changing that underlying proposition, I suspect that the current interest in prediction markets, under the klieg lights of a particularly insane election season, is as high as it will ever be.