When will prediction markets grow up?
Prediction markets are 30 years old this year: The Iowa Electronic Markets (IEM) opened in 1988. Since then prediction markets have been promoted with almost evangelical fervour as the answer to information aggregation and decision making problems. The American economist Robin Hanson is one of the most enthusiastic proponents of prediction markets, even introducing the idea of Futarchy — a system of government in which prediction markets play a central role — in 2000. In 2007 Michael Abramowicz wrote the book Predictocracy which advocated for the use of prediction markets in both the public and private sectors. Some corporations have experimented with prediction markets to predict sales or whether projects will deliver on schedule but the use of prediction markets is still unusual and they certainly haven’t become the ubiquitous decision making tool their supporters envisaged. I am convinced of the usefulness of prediction markets and I believe that barriers to their widespread adoption can be overcome.
One oft-cited problem is regulation. Prediction markets look a lot like conventional gambling, they certainly do to regulators. In Britain online gambling is legal, there is a non-trivial overhead involved in obtaining a license and complying with regulations but the law is not an insurmountable obstacle to running a real-money prediction market. Regulation is more of a problem in the United States where gambling has been severely restricted, but I wonder whether regulation is really as big a hurdle as we think. Prediction markets can be structured so they are not classed as gambling. Hypermind and the Winton Climate Prediction Market both make use of “play-money” markets with real-money prizes awarded to the top performers.
From Gambling to Prediction
In the U.K. there are online betting exchanges, such as Betfair and Smarkets, that are sometimes described as prediction markets. Ireland was home to the now defunct Intrade which was also described as such. Conflating these types of gambling sites with prediction markets has contributed to misunderstandings about prediction markets. Gambling is primarily for entertainment whereas prediction markets are specifically designed to reveal and aggregate information. Betting odds do provide information about the probabilities of different outcomes but only for a relatively narrow set of topics. Commercial betting focuses disproportionately on sports. After sports the next most popular topics are politics and entertainment, such as the outcome of reality TV shows. Real-money academic prediction markets such as IEM and PredictIt also focus on politics and efforts to introduce markets for other topics have been less successful. Commercial betting sites, even those structured like prediction markets, are negative sum games: the site takes a commission on bets meaning that in aggregate participants lose money. Even PredictIt, an educational project of Victoria University in New Zealand, charges a fee of 10% of trading profits. The IEM only charges a modest $5.00 fee for opening an account and no trading fees, but this is still only a zero-sum game in which the winnings of successful bettors match the losses of unsuccessful bettors. The organisers of such markets are forced to select topics for which there is a lot of interest on both sides of the market. If this interest isn’t there, the markets lack liquidity and don’t deliver any information. Betfair don’t want to know who will win the World Cup but they do know the question will generate a lot of diverse bets enabling them to skim a profit from the trading.
No free lunch
If you have spent any time in a British pub you will know that sports and politics are two areas where people are more than willing to volunteer “information” for free but, in general, we shouldn’t expect to get information for nothing. People who bet on sports and politics are at least partially motivated by non-monetary concerns: They do it to make watching events more fun or even to psychologically hedge against disappointment. The existence of these non-financial reasons to bet mean that markets in these topics can essentially generate information for free, but these topics are unusual. This has been recognised by some advocates of prediction markets but by no means all. Robin Hanson and David Pennock at Microsoft have produced elegant designs for subsidized prediction market in which a market sponsor can inject money into the market in return for information. These designs solve the problem of liquidity and allow markets to function with small numbers of traders. Providing these traders have the information being sought the market can reveal it. I would argue that these types of markets are true prediction markets in which participants are being paid for the quality of the information they provide.
We haven’t had enough of experts
Another way of contrasting genuine prediction markets with conventional betting is to think about who you would like your participants to be: thousands of people who know little about the subject or a handful of people who know an awful lot. If you prefer the former you are running a gambling site, if you find the latter more appealing you are running a bona fide prediction market. Many prediction markets attract people who are interested in gambling but attracting people with domain relevant knowledge is more important.
Once we recognise the distinction between traditional gambling and prediction markets we can design markets to obtain the information we want, in the form that we want it and target these markets at an appropriate audience. Hivemind’s prediction market technology incorporates these insights. We can move beyond markets for simple binary outcomes, which are often of limited use to decision makers, to markets that produce probability distributions for continuous variables. The Winton Climate Prediction Market shows that extracting joint distributions of continuous variables from small groups of informed and motivated experts is completely feasible.
Prediction markets pay for results
There may be institutional barriers to prediction markets. You don’t have to be card-carrying Futarchist to believe that prediction markets are not being used to their full potential within existing organisations. Sometimes this is because people are aware of the shortcomings of gambling markets masquerading as prediction markets discussed above. Sometimes the reticence is due to the misconception that prediction markets are a substitute for expertise. They are not. Prediction markets are a mechanism for harnessing and paying for expertise. When an organisation hires internal experts or external consultants it is paying for expertise. It isn’t much of a leap to make payments contingent on the accuracy of the expertise, which is what prediction markets do. Prediction markets should have more in common with performance-based pay than gambling. Comparisons with online betting are not necessarily helpful and have possibly hindered the spread of prediction markets.
Holding out for a Hero
One final explanation may be that these things take time. Paul David described how it took almost 50 years for electrification to be fully exploited in manufacturing. Companies had to redesign their factories in ways that took advantage of electric motors rather than just used them as straightforward replacements for steam engines. Other technologies, such as the telephone, radio and television also only become widespread decades after they were invented. Hero demonstrated a steam engine in the first century A.D. but the technology had little impact until the industrial revolution. We might have to wait a little longer for the dawn of Futarchy.
