Popular Posts

Sunday, May 16, 2010

Prediction Markets

[This is a draft from a few months ago, but I'm posting it in hopes that it'll prompt me to finish it. Maybe it will stimulate someone to want to hear more (or, do YOU want to finish it? If so, get a copy of "Infotopia" by Harvard Law prof Cass Sunstein, if you haven't already. I started writing this up after reading this book.).]

Prediction Markets are a fancy name for the age-old idea of betting on future events or outcomes, or "putting your money where your mouth is", in other words. In a prediction market, participants can place bets on whether, say, health care will pass, or whether Obama will get re-elected (if he runs), or (more controversially) when the next major terrorist incident will occur on American soil, and so on. Researchers have found in recent years that Prediction Markets are remarkably accurate. In many cases, they outperform the predictions of subject matter experts, and other types of group-aggregation predictors like polls, or deliberative groups. For example, the Iowa Electronic Markets, run by the University of Iowa since 1998, is a popular prediction market that has psroven effective in guessing election outcomes. The IEM results had Bush with 50.45% of the vote, and Kerry with 49.55%, compared to the actual 51.56% for Bush and 48.44% for Kerry. The result, based simply on people placing bets on the final outcome of the election, outperformed professional polling results.

The question is, why? Why is betting such a powerful tool for predicting the future? Participants in Prediction Markets--bettors--don't have to have a Ph.D. in the electoral process; they don't have to have access to special information; they don't really need any credentials at all, short of agreeing to partipate by putting up some initial money and hoping to get more by guessing correctly. And yet, collectively, the Prediction Markets often outperform the experts.

The story of why Prediction Markets (hereafter PMs) work so well is best told by telling another story, one of how deliberative groups often don't work well. In a deliberative group, members, well, deliberate: they discuss and analyze some topic or set of possible outcomes and decide on a course of action, or a most likely outcome, or a best policy, depending on what is getting deliberated. Deliberative groups often do produce good outcomes (perhaps the most famous example of the success of deliberation is the ratification of the American Constitution, where sustained discussion about the details of the Constitution resulted in the document we have today).


Rebecca said...

I had no idea that prediction markets have become more accurate than perceived prognosticators. I am curious if it because pollsters, deliberative groups, etc. are essentially, stake-holders, and therefore could have “blind-spots” to possible outcomes. Their bias may unconsciously taint the outcome, whereas the prediction marketers are motivated by bottom line and immediate successful outcome--money. I wonder if these guys can make successful predictions now for midterm November elections?

Erik J. Larson said...

I think this is part of what Sunstein is getting at, that deliberation often involves the injection of bias into a decision process, where in a simple prediction market, you don't make any money if you don't bet based on what you think really will occur (as opposed to what you want to occur). The other, related, idea is that deliberative groups often do a poor job of making sure that what everyone knows is successfully taken into account. Some people will speak a lot, and have influence in the group, and others won't speak as much, and have less influence, and so on. The problem is that too often this type of dynamic does not get the relevant information out "on the table", and so the group a whole comes to a sub-optimal decision. Prediction markets often work better because, just as you suggest, there's no incentive to do anything other than to use the information each bettor has to make the best prediction, with hopes of a correct prediction (hence $$$).

One are that is just terrible for decision making is in management situations where lower level employees have valuable information but won't share it, or similarly when managers have an outcome in mind and hence won't listen to contrary views, or they steer the discussion toward their desired outcome. Assuming the managers are omniscient, this strategy works great. In the more realistic case that managers need feedback from other employees to make good decisions in the first place, this strategy works terrible!

Rebecca said...

This idea of prediction markets leads me to think of intriguing, yet dangerous precedents. On the one hand you have Wall Street brokers who made HUGE BANK while the system collapsed. Then you have the public education system....much bickering, while it looks like it will collapse as well. Hmmmmm.