A Colloquy on Prediction Markets - Amherst College



A Colloquy on Prediction Markets

Geof Woglom and

Michael Abramowicz, AC 1994, Professor of Law, George Washington University

[pic]



Therefore I started an e-mail exchange with Michael that morning. I thought our exchange would be of to our readers.

MA (in response to my e-mailing him the Slate article):

Great to hear from you – and of course I think the Slate article misses the point. Prediction markets can’t be expected to be perfect predictors; otherwise, they would always report 100% or 0%. What I think prediction markets do is they capture the views of people who are among the most informed on a particular issue. (With subsidies, one can get an even more informed view.) I think that yesterday’s New Hampshire results were a surprise to the most informed people – at least this article doesn’t indicate that they weren’t. That surprise is relevant in and of itself, because momentum appears to matter in a presidential campaign. In addition, it causes a revision of assumptions about key underlying variables, like turnout.

 

If Gross really believes what he is saying, then he has a profitable trading strategy – bet against the momentum of the market. I wouldn’t be sure that this would be a failure; the overall financial incentives are currently sufficiently low that it is possible that there are opportunities for sophisticated traders to make money. And even the most efficient markets present profit opportunities. In the long term, that should tend to fix whatever imperfections exist. But the notion that the markets are way off seems wrong to me.

GW:

I understand that efficient markets do not reflect perfect foresight, and also that it is hard to draw inferences from a sample size of 1.  But I actually have a sample size of 2:  the day and early evening of the 2004 election I first assured my class and then my family that Kerry had won because quoted the Kerry contract at .75.  Clearly that was due to the misinformation from the exit polls.  The issue for me (and I think for Gross) is that these were two examples where the traditional information sources got it very wrong provided an opportunity for prediction markets to reflect a different source of information.  In these 2 cases it appears as that they were unable to exploit this opportunity.

MA:

I think that the 2004 exit poll experience reinforces my point. I have friends who were working in the inner circles on both sides that day, and the Kerry people were extremely confident after the exit poll numbers, and the Bush people thought that they had lost. So, I do think that the prediction markets captured the sentiments of the most informed people. So, it seems to me that if anything, the markets were quite conservative in still giving Bush a 25% chance of victory based on that information.

Ultimately, you can’t assess prediction markets anecdotally. You need to look at a large number of cases, and the best data we have are from prediction markets devoted to gambling. Look at the attached graph. It reflects a sample of over 145,000 Major League Baseball transactions betting on games in 2005. As you can see, prediction market forecasts of 90 cents on the dollar correspond pretty closely to 90% probabilities. That is, among all of the 90 cent predictions that a team will win, the team wins close to 90% of the time. It’s certainly possible that some sophisticated players can beat the market. But the most important point for me is that the predictions are accurate enough for government work – and probably corporate work too. I doubt that if you assigned a government or a corporate official to make predictions on a large variety of issues that you would end up with the nice 45-degree line exhibited here.

[pic]

 

GW:

For me the key issue is not whether prediction markets are as accurate as conventional sources of information, such as polls.  Instead, I think the crucial test is whether the markets have different information than traditional sources (if they don’t why bother).  For example, Richard Roll wrote an article ( “Orange Juice and Weather,” Futures markets. Volume 2, 1997, pp. 337-56.) where, if I remember correctly, he tested whether futures prices for orange juice had information that was nor already reflected in long-range weather forecasts.  If I remember correctly the test was based on a regression of subsequent temperature on the long-range weather forecast for that date and on the futures price on the same date the forecast was made.  Roll found that the coefficient on the futures price was significant, and interpreted this as showing that futures price have different (and useful) information from the long-range weather forecast. Have there been any similar studies done on prediction markets?

MA:

I think there is a fundamental institutional point about the benefits of prediction markets that you are missing. In my book, I distinguish between two kinds of aggregation that prediction markets perform – “information aggregation” and “assessment aggregation.” Sometimes, prediction markets may be useful because each person has access to different pieces of information, and the market may perform information aggregation. But sometimes, a number of people may have access to essentially the same information, but reach different conclusions about it. Individuals and institutions are probably better off relying on prediction markets than on alternative algorithms for picking a best prediction given the existence of disagreement (such as trusting one’s own instincts or assigning an individual to make a prediction).

 

Assessment aggregation seems to me to be the primary function served by the election markets. It’s possible that there is some degree of information aggregation – for example, some traders might have exit poll data before everyone else. But even if this isn’t the case, prediction markets are useful for people who need or want to know what a good prediction would be, when different people announce different numbers. The task of assessment aggregation is particularly important in contexts in which individuals may have incentives to make predictions that are different from what they truly believe. Spin seems unlikely to have much effect on prediction markets.

 

I do talk about Roll’s article in my book. As you note, it’s not that orange juice futures are a better predictor of the weather than the National Weather Service, but that they seem to reflect at least some information besides National Weather Service forecasts, so in theory combining the official forecasts with a small adjustment based on OJ futures prices would allow for a better forecast. A weather prediction market similarly could incorporate any information that people have about how to improve on the National Weather Service forecasts (information aggregation), but also could be useful because it would provide a means of aggregating competing views about weather prediction (assessment aggregation). In deciding whether to bring my umbrella with me, I don’t want to look at a series of competing predictions and try to figure out which is best, and I certainly don’t want to try to decode orange juice futures prices. I want a single number that can guide my decision making, and that’s what prediction markets are good at providing.

GW:

Thanks for taking the time. This has been fun and informative (for me). I hope we can get you to visit Amherst in the not too distant future to carry on the discussion.

MA:

Thank you. That would be fun.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download