Political Prediction Markets: The “Pricing” of Electoral Candidates as “Assets”

I wrote not too long ago about the declining accuracy of electoral polls. Yet, one remarkably reliable source for the possible outcomes of electoral contests has been political prediction markets. For example, the outcome of the 2012 US Presidential election was predicted accurately by prediction markets, despite many respected pollsters portraying a much closer race, or even a Romney win.

The fact that candidates’ chances of an electoral win are “traded” in prediction markets brings up interesting questions of “asset pricing” for these “instruments.” After all, how do you price a candidate for elections, where the price is a reflection of her odds of winning her electoral contest? What kind of information must you take in to “buy” or “sell” a candidate? In particular, two main questions arise:

• Does “politics” trade in an efficient market? Does the probability of a candidate’s win adjust quickly as expected to new information on the candidate and her campaign?
• Why are prediction markets so accurate, in particular in relation to traditional polls conducted by respected pollsters conducting surveys with tested statistical methods? Are “predictors” better informed than voters and the pollsters who survey them? What is the transmission mechanism—if any—between polls and prediction markets (since presumably polls are a large part of the bettors’ information set), and why is the results from this information set usually so different? In short, why does a mismatch exist between polls and prediction markets?

The second question to some extent already presumes the answer to the first question, in that in order for prediction markets to be accurate, “prices” for these candidates must accurately capture the chances of that candidate’s win and, in the long-run (since by definition of odds, there’s always uncertainty and the most “favored” outcome in each individual race may not actually win), be able to predict electoral outcomes with the level of probability expressed for the most favored candidate in each electoral contest (a prediction market can be only as accurate as the probabilities implied for each contest and whether they actually occur).

Of course, prediction markets are more akin to options in that they expire at either 0 or the strike—or in this case, “prediction”—price. As such, prediction market “instruments” are quite different from stocks and bonds when it comes to their payout structure (in that for the latter two, some form of payout is almost guaranteed to the investor in the form of dividends or selling the stock, or coupons or principal payment—with of course much more certainty for the bond than the equity holder). Moreover, “prices” in prediction markets invariantly move in relation to one another as they must—in total—represent the 100% certainty of one discrete event happening from a set of options. Thus, topics such as asset pricing and market efficiency can differ non-trivially between prediction markets and more “traditional” financial and capital markets.

However, attempting to apply the intuition from these more traditional markets to prediction markets, the simplest possible answer for the above questions is the factors that typically promote market efficiency. For example, the volume of “trade” may be a reason for the higher accuracy of prediction markets compared to polls: prediction markets are more frequently and continuously open to trade, with larger number of participants than the more periodic and sporadic polls with smaller sample sizes.  Moreover, there may actually be a sample selection effect. After all, people who participate in prediction markets might in fact have better information; people who participate in polls do not typically volunteer to participate in these polls, whereas investors have skin in the game and actively choose to participate in these markets, presumably because they feel confident in their beliefs/ pricing of these electoral candidates. Of course, “investors” in prediction markets are subject to the same behavioral biases as investors in more traditional markets, such as overconfidence and speculative bubbles. Also, the same question arises of what kind of different information prediction market participants face that make them good “candidate pickers,” seeing as how, presumably, polls themselves are part of their information set. Are these investors just politically more savvy and aware, i.e. they follow electoral news more closely than poll participants and, as such, have a larger information set that incorporates other factors that poll participants—in aggregate—may not?

Lastly, in order to determine whether these prediction markets actually “do” trade “efficiently,” a simple event study of sorts could be considered, where we study a cause-and-effect relationship over the period of time of a campaign between unexpected developments in a candidate’s campaign and profile, and an immediate response in that candidate’s odds of winning the elections in political prediction markets. Of course, how to accurately control for other factors in a regression model is less clear. For example, one might control for news on the candidate’s political party or opposing party that could be driving her odds in the market. But considering how factors affect one candidate individually as opposed to all candidates—and considering how these odds must invariably move in relation to each other—is less immediately obvious.