Three men, PredictIt traders, sit at a bar table in New York City awaiting presidential primary results.
FixGov

Political prediction markets: What are they good for?

Philip A. Wallach

My name is Philip, and I am an IEM-aholic. I was on the wagon for nearly four years, but in the interest of being able to write this blog post, this past summer I allowed myself to fall off. Allow me to tell of the joys and frustrations of my sordid habit, and about how you ought to think about the political prediction markets that exist today.

Let us start with a brief tour of the landscape. “IEM” is the Iowa Electronic Markets, a real-money prediction market run by the University of Iowa’s business school. Ever since the Unlawful Internet Gambling Enforcement Act of 2006 was passed by Congress (attached to the once-infamous port security bill without deliberation), it has been mostly illegal for U.S. citizens to engage in internet betting on politics or anything else. Whereas once there was a large and growing volume of participation in political betting markets (especially Tradesports), with the prohibition in place it now requires some real determination for U.S. citizens to wager serious money on political outcomes—either using bookies in Nevada or opening an overseas bank account to facilitate use of various international sites, such as BetFair.

I am not so determined, and so I turn to IEM. Because IEM limits investments to $500, is attached to a good university, professes “research” purposes, and in general does not give off the feel of a seedy gambling joint, it is allowed to accept checks from U.S. residents without any legal troubles. IEM hardly makes the most of its special status; to put it very gently, it has a great many technical and administrative shortcomings. Its contract definitions are atrocious, making it extremely difficult to speculate on the future control of Congress; markets are not always closed when they are supposed to be; and the trading machinery is, well, let’s say clunky. (Register here, in spite of all that!)

Still, IEM is a real market with real money, and all it takes is a one-time $5 fee to open an account, after which there are no more transaction costs. For many people, that makes it the best option to get in on the action. (Full disclosure: I am feverishly pushing around the proceeds of an initial investment of $100 on IEM, and very much hope that more money comes into the markets.) There is now also PredictIt, based out of Victoria University in New Zealand and limiting investments to $850, and it claims to make funding accounts easy for U.S. citizens while also having put its legal ducks in a row. It takes a 10 percent of cut of all trading profits. Another active prediction market is Hypermind, which uses fake money and selectively admits its traders.

How should you think about these markets, and their current prices?

If you take their own rhetoric at face value, you might think that you can simply read off the probability of various political outcomes straight from their current prices. Right now, on IEM’s “Winner Take All” Presidential Market, the Democratic shares trade for around 68 cents, and if the Democrat wins more votes than the Republican on November 8, those shares will pay off at $1 each. IEM invites you to interpret this price as meaning that Democrats have a 68 percent chance of winning the 2016 election.

While such a straightforward reading is undoubtedly convenient, a few big caveats are in order:

  1. Many of these markets are small and shallow. IEM’s winner-take-all market—by far its largest and deepest—now has just shy of 90,000 shares outstanding, meaning there is just $90,000 at risk. The market thus reflects the beliefs of some (more or less obsessive) hobbyists, and not all that many of them, surely less than a thousand. Daily trading volume is almost always less than $1,000. PredictIt is somewhat larger and more active, with something like $700,000 at risk in its party market and something like $1 million in the individual market, and with daily volumes sometimes in the tens of thousands in recent days, but that is still not much money in the grand scheme of things. So, unlike the prices in commodities or equity markets, there is no reason to think of prices in these markets as reflecting all of the best information available; for now, in large part because of the legal impediments to participating, political prediction markets reflect only the judgments of niche communities. Sometimes that may limit their predictive ability, as some have argued regarding their misfire in the Brexit referendum. Other times they may be so small as to be nearly worthless—a fair characterization of IEM’s “Vote Share” market, with just $3,220 at risk and paltry activity.
  2. “Dumb money” can dictate prices, at least for a while. Because there is not all that much money involved and because there are relatively few bids and asks at any time, these markets can experience wild short-term price swings just because a few people (or even, in IEM, a single person) decides to impatiently buy or sell like crazy. Eventually, other traders will be glad to take this dumb money, but inefficiencies abound and corrections aren’t always as fast as you might think. Spikes like the one in IEM’s winner-take-all market that saw Democrats’ chances jump past 90 percent (shown below) should be treated as noise.

gs_09142016-market

  1. It isn’t clear how much “extra” information these markets end up incorporating. We want these prediction markets to be crystal balls, but more often they are mirrors of information already prominently featured in the news media—mostly, polling data. To illustrate, let’s look at smoothed prices from IEM’s big market (a moving 3-day average) and compare them to FiveThirtyEight’s famous polls-only model, which translates all available polling data into a single probabilistic assessment of the candidate’s chances.

gs_20160914-probability-of-democratic-victory-revised

Putting aside the coincidence that these values are now almost exactly equal to each other, these series are meaningfully different: the smoothed out IEM market prices are relatively more stable than the polls-only model. But they move together pretty closely (correlation=0.82). The same is true of the FiveThirtyEight’s Polls-Plus model (again, correlation=0.82).

gs_20160914-probability-of-democratic-victory-2-revised

Probably we shouldn’t hold this quality against the markets. After all, it makes sense to follow polling results, which represent our most useful stream of new data for prediction purposes. But if people expect prediction markets to be some kind of boldly countervailing indicator, even without big money players involved, they are likely to be disappointed.

Where should those qualifications leave you in evaluating prediction market prices?

Well, the first implication is that, if you do want to go from trading prices to probabilities, you should follow the money by looking at the deepest and most heavily traded markets. Sadly, that probably means looking away from IEM and PredictIt and over to the now-Ireland-based BetFair, which says that it has almost $49 million of matched bets on the presidential race. If your work computer system blocks you from looking at such a naughty site, Eli Dourado has kindly set up a tracker that converts current odds to percentage probabilities.

Even then, you should be skeptical about some of the implied probabilities. It is fun to take a long shot on Joe Biden becoming president, while it is much less fun to take the other side of that bet which can return at most 2 or 3 percent over a few months. So take the idea that Joe Biden has a 1 in 35 chance of sitting in the Oval Office in January with a grain of salt. (Similarly, I’m somewhat skeptical of Justin Wolfers’ interesting attempt to derive odds of Trump’s leaving the race, which seems to depend on a higher degree of precision than these markets are likely to have.)

If prediction markets are far from perfect, it nevertheless seems reasonable to throw them into the mix of one’s predictive apparatus. One sensible approach is to mix together everything out there, including both prediction markets and high quality polls-based models—which is what David M. Rothschild’s PredictWise offers. (Some people mistake the site for an aggregator of prediction markets only, but in fact it uses some polling models.)

If, on the contrary, you think the prediction markets are just out to lunch, by all means, jump in and take your cut! The water is, if not warm, at least lukewarm and free of contamination. Having more independent-thinking participants will only make these markets better.

And I am happy to conclude by commending the intellectual discipline that trading on a prediction market imposes. Having to quantify one’s own instincts, and back up professions of confidence with actual money—even just a little bit—is extremely clarifying and humbling. It makes you feel, in your guts, how 70 percent and 60 percent (or even 68 percent and 62 percent) are different from each other in a way that is very hard to otherwise conceptualize. Unlike a bet that is fixed in time forever, trading on an active market invites you to reconsider your views as maintaining them becomes expensive. As a result, it is likely to make you realize that you are less willing to defy conventional wisdom than you like to think. For providing such epistemic services, we should be grateful to prediction markets (and call for their widespread legality!), whatever their record as predictors.