A prediction market primer

If there is any one aspect of the Web 2.0 era that I find most fascinating, it is the idea that groups or “crowds” can collectively offer more intelligence about current and future events than individuals–even experts.

No one describes the power of a crowd’s decision making ability more eloquently than James Surowiecki in his book The Wisdom of Crowds. As Surowiecki’s examples illustrate, this concept is not new; rather, it is a natural consequence of human behavior.

One way that this concept is being put to use in real-world applications is through prediction markets (sometimes called decision markets). Prediction markets have tremendous potential as a forecasting tool in the enterprise or for any entity that wants to make use of its single greatest source of intelligence—the individuals sitting within its walls.

If you’re familiar with how a futures contract works, that’s a good analogy. A conventional futures contract is commonly used as insurance against unfavorable movements in the price of assets and also for purely speculative investing. Established futures markets exist for a variety of heavily-traded commodities and assets like oil, corn, and stock indices.

Beyond their value to individual participants, futures markets provide interesting insight about the future. Because market participants have a vested economic interest in being correct, they use all (legally obtained) information available when making their bets. For example, the futures price of corn will reflect expectations of future weather patterns, the demand for livestock feed, the value of the US dollar, and other factors that we may not necessarily associate with the tasty, versatile yellow vegetable. As such, futures markets offer some of the highest quality information about the future as seen through the collective eyes of people that have a financial interest in being correct.

In essence, a prediction market is a futures market created for a specific event, not necessarily economic in nature.

For a prediction market to be most effective, however, there should be a group of individuals that each have unique information about the event. As members of the crowd place their individual bets, a consensus view of the future emerges. And as more information becomes available, the forecast evolves–right up to the moment of the event. As with a futures market, participants are rewarded in proportion to the accuracy of their forecasts. The “payoff” can be real money, fake money, or anything that motivates participants to make wise bets.

Dating back to the 1988 US presidential election, the Iowa Electronic Markets (IEM) represent one of the earliest examples of prediction markets. In August of 1995, the Wall Street Journal ran a story on how the IEM was being used to predict the outcome of the 1996 US presidential election. At press time the market was showing increasing confidence that Bill Clinton would ultimately win and was also favoring Bob Dole as the likely Republican nominee:

For the moment, the prices in the IEM presidential market bode well for Mr. Clinton. His stock has enjoyed a 19% rally since May 1, climbing to 45.1 cents last week from the May 1 price of 37.9. If Mr. Clinton wins re-election, holders of his stock will receive $1 a share. If he loses, his backers will get nothing.

Meantime, in the market for speculating on who will get the Republican presidential nomination, Kansas Sen. Dole’s stock is enjoying a sparkling rally. His shares are up nearly 30% since the beginning of May, largely at the expense of Texas Sen. Gramm, whose shares have dropped about the same percentage.

The prediction market described above is elegantly constructed. Since holders of Clinton’s stock would ultimately receive $1 per share, the pre-election value of his shares at any point in time represented the probability of Clinton winning. At press time, the market felt Clinton had a 45.1% chance of beating the future Republican nominee.

If you were trading in that market in late August 1995 and you had reason to believe that Clinton’s chances were better than 45.1%, then 45.1 cents a share would have looked cheap. You would naturally be motivated to buy shares, which in turn would have driven the price of Clinton’s shares higher.

It is this natural pursuit of self interest that gives the IEM an edge over more traditional methods for forecasting election results like polling, where pollsters hope that good morals prevail.

The following is a simulation of what a prediction market might look like over the course of a year. At the end of the year, the outcome is “yes.”

More recently, the IEM has been used to forecast the outcome of non-political events, like movie box office receipts and the future value of economic variables like unemployment rates, the consumer index, and stock prices. A complete list of markets operated by the IEM is available on their website.

Prediction markets seem to be getting even more visibility these days as social technologies make it easier than ever to create prediction markets around everything from sporting events to enterprise projects.

In future posts, I will explore prediction markets in greater detail. If you have used a prediction market for any purpose, I would love to hear from you in the comments or by email.

[Crystal ball photo by Isobel T via Flickr]

Advertisements

, , , ,

  1. #1 by Bill Free on April 14, 2010 - 10:48 am

    Good post, Eddie. I’ve been interested in the concept of crowdsourcing for some time, and this motivates me to invest more energy in the subject. I’ll add Surowieki’s book to my reading list.

    The most robust example of the kind of prediction market you describe is Intrade, which allows participants to take positions in a virtually limitless range of events. Based on my understanding (or lack of it), it seems that at least one of two requirements must be in place for a prediction market to function properly: either participation in the market must be limited, or the outcome of the event cannot be determined by the participants. Perhaps you can clarify.

    A national election is an example. As you point out, participants in the market must be vested in the outcome. Theoretically, as the number of participants increases, doesn’t their ability to determine the outcome also increase?

    The stock market is another. It is supposed to function as a predictive market: participants invest in companies based on their expectation of future performance. But in modern equities markets, huge amounts are invested not in bets on future outcomes, but to extract gains from short-term movement in prices. Hasn’t the market’s predictive value has been diluted because its participants have redefined the outcome?

    • #2 by Eddie on April 14, 2010 - 7:53 pm

      Bill, you ask some great questions. First of all, I’m definitely not a an expert in prediction markets. I’ll offer my thoughts based on what I’ve read and also discussions I’ve had with those in the prediction market industry.

      I think for a prediction market to be successful, the outcome in question must be fairly straightforward and short-term in nature. The pool of participants should be as diverse as possible but ideally they each possess useful knowledge that can inform an individual prediction. The “market” mechanism provides a way of encouraging unbiased, truthful “bets.” The goal, I think, is to let a consensus view arise from a crowd of individuals, who by themselves would not offer their individual contributions in a constructive way. In other words, each person has a piece of the puzzle, and the prediction market lets those pieces join.

      I’m not sure what correlation the number of participants has on the outcome. I think it probably comes down to the composition of the crowd. Simply adding a lot of people to the pool may dilute the decision power of the group if the people being added aren’t informed. In other words, if you add a lot of wild guesses, it probably muddies the picture to some extent. But if you add more people who are informed, I suspect the group’s predictive power increases. I’m not sure if there is any empirical evidence for this, but I’m going to look.

      Regarding equities markets, there are a lot of angles. I think even if you’re taking a short-term gain, you’re still making a statement about the future to some extent. Some might argue that by selling (even for short-term gains), you’re implicitly hinting that the upside of an asset is more limited than the downside. Of course, there is much irrationality, tax strategy, and general noise in the short-term movement of equities.

      I think the ultimate goal is to get a group to work together by harnessing individuals’ natural instinct to pursue their own self-interest. Money does this well, but based on talks I’ve had with people closer to prediction market applications, it’s not an absolute requirement. Just the basic human need to feel valued and important is often enough.

  2. #3 by Bill Free on April 14, 2010 - 4:35 pm

    Oh, by the way. Corn is a grain.

    • #4 by Eddie on April 14, 2010 - 7:55 pm

      You are absolutely correct!

  3. #5 by Philip J. Loree Jr. on April 14, 2010 - 4:50 pm

    Eddie,

    Nice, well-written post — I visited just to make sure you were blogging away, and sure enough, you have been!

    Keep up the great work,

    Phil

    • #6 by Eddie on April 14, 2010 - 7:55 pm

      Thanks, Phil. Great to see you.

  4. #7 by Paul Hewitt on April 18, 2010 - 11:04 am

    Bill commented: ” it seems that at least one of two requirements must be in place for a prediction market to function properly: either participation in the market must be limited, or the outcome of the event cannot be determined by the participants.”

    I don’t think it is necessary to limit participation in order for prediction markets to function properly. In a money-based market, the uninformed will quickly lose their funds, which naturally limits their participation (they may have to exit the market). Even in a non-monetary market, the most successful traders will gain paper-wealth, which increases their “say” in the prediction relative to the “chimps”. Also, it is not necessary for the event to be non-determinable by the participants. Some (or many) of the participants could (or should) be able to determine the outcome, just not with certainty.

    I agree with most of Eddie’s comments about the stock market, except with the comment: “ultimate goal is to get a group to work together by harnessing individuals’ natural instinct to pursue their own self-interest.” One of the other conditions for prediction markets is that the participants act independently. Each brings diverse pieces of information with minimal influence from the others, avoiding information cascades and group-think.

    This raises an interesting point, where none of the participants have sufficient information to make a reasonable prediction. Maybe some problems are like jigsaw puzzles, where everyone has a piece, but no one can link them together. Climate change predictions might fall into this category of problems. Robin Hanson’s Futarchy involves forecasting future GDP+. I argued that this would be virtually impossible to do, even with the world’s foremost experts.

    In most cases, adding participants will improve accuracy. Even uninformed (“chimps”) help by making poor predictions which provide incentives for informed participants to trade. Logically, there must be some limit where adding more chimps degrades the prediction. To my knowledge, there are no research papers on this topic. The industry researchers assume that all predictions follow a normal distribution (i.e. all errors cancel out). Some markets are not likely to follow normal distributions, invalidating this assumption.

    On my blog (http://torontopm.wordpress.com), I have written, extensively, on many of the issues related to prediction markets, including: information completeness, accuracy and general principles. I welcome all comments.

    Paul S. Hewitt

    • #8 by Eddie on April 19, 2010 - 10:58 am

      Thanks for your very informative comment, Paul. I look forward to reading more of your blog.

  1. Risk + 2.0: The week in links 8/6/2010 « Risk + 2.0
%d bloggers like this: