Monday, October 13, 2008

Prediction Markets: US elections and Economics Nobel Prize

What do the Nobel Prize in Economics (actually the Riskbank prize in honor of Alfred Nobel), to be announced in a few hours, and US elections have in common? There are active markets for both where anyone can bet on the outcome of each of these events.

Prediction markets, as such markets are known, offer a wide spectrum, from large for-profit exchanges (Betfair and Intrade) to the non-profit (Iowa Electronic Markets). There are even prediction markets where all winnings go to charity (Bet2Give) and where no actual money is used (Hollywood Stock Exchange, NewsFutures). In fact, there is even a Prediction Markets Industry Association now.

Political prediction markets have been with us for quite a while; a recent paper traces the origins of betting on political outcomes to 16th century Italy and notes that betting on US elections was common until WWII. So we are in fact seeing the revival of a quite old trade. As I have been looking at what prediction markets/betting exchanges are predicting for US elections, I've come across an interesting phenomenon; in almost all markets covering elections, I've been seeing the reversal of fortunes in McCain and Obama, with a slight McCain lead in early September now transformed into a huge Obama lead. Of course, political analysts can come up with many reasons on this shift, but let me present an economic view: Yale economist Ray Fair has had an econometric model that predicts the outcome of US elections for more than two decades. You have to read the one of his papers to get all the details (I recommend the 1996 version), but his main insight is that current economic conditions determine the fate of the incumbent party. In my first year studying Economics (with Ray Fair himself), I was really confused by the results of the model after learning that the US economy is affected by business cycles rather than policy decisions in the short-run. Talk about the rational voter... Anyway, Fair allows us to compute our own predictions for the 2008 election using his 2004 equations. It is true that the US outlook for 2008 has really worsened in the past month, with all the forecasters including the IMF adjusting their outlook for the economy sharply downwards. However, to get the numbers currently in prediction markets, we would need a huge contraction in 2008! What is going on? For one thing, prediction markets are merely predicting the odds a candidate will win, not the number of electoral seats or percentage of votes he is expected to win, so direct mapping from prediction markets to the FairModel is not possible although it is possible to extract probabilities from odds. Call it wishful thinking if you want to, but although I am aware that the 30-point jump is difficult to be justified by economic outcomes alone, I feel that at least some of it is due to the the rapidly changing perceptions on the fate of the US economy.

The discussion in the previous paragraph, however, raises a more general question on the reliability of prediction markets. This is the subject of a literature survey piece, into which I do not dare to delve, but my quick skim of the literature did not provide a definite answer: For every paper that shows that prediction markets work, either theoretically or empirically, I was able to find another one that reports they don't. We may decide to heed the old saying that people will put their money where their mouth is and look only at markets where actual money is being exchanged, but there is research that shows that virtual markets fare no worse than actual ones (Disclaimer: Two of the authors are affiliated with NewsFutures, a virtual market and a firm that set us private virtual markets for companies, but anyone questioning the validity of their findings is full of bollocks, sorry Justin, just had to squeeze this in). One can also argue that some prediction markets are suffering from lack of liquidity, but in a recent paper, Paul Tetlock of Columbia University finds that limit orders may cause more liquid markets to be less efficient, i.e. deviate further from outcomes. So where does all this bring us to? My understanding of the literature is "prediction markets definitely reveal some information, but use it at your own risk", at least for now... I'll definitely continue tracking them, but I am unlikely to depend on prediction markets as my sole sources for forecasts.

Coming back to the Nobel Economics prize, what are prediction markers revealing? NYT Economix blog is listing the latest odds from some betting exchanges, whereas Greg Mankiw is reporting the results of the Harvard Nobel Prize Pool. I guess we'll know which market is the most efficient for this particular event soon...

No comments: