Hayek and the Price System

To understand prediction markets, we should begin with Friedrich Hayek, the great twentieth-century critic of socialism and economic planning. Hayek emphasized the importance of free markets, above all as a way of collecting and combining information. His most important contribution to social thought is captured in his great (and short) 1945 essay, “The Use of Knowledge in Society.”¹

In that essay, Hayek claims that the central advantage of prices is that they aggregate both the knowledge and the tastes of many people, incorporating far more information than could be assembled by any central planner, group, or board. Hayek emphasizes the unshared nature of information—the “dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess.”² This knowledge certainly includes facts about products, but it also includes preferences and tastes, and all of these must be taken into account by a well-functioning market. Hayek stresses above all the “very important but unorganized knowledge which cannot possibly be called scientific in the sense of general rules: the knowledge of the particular circumstances of time and place.”³

For Hayek, the key economic problem is how to incorporate that unorganized and dispersed knowledge. No one person or group can possibly solve that problem. Central planners cannot have access to all of the knowledge held by diverse people. Taken as a whole, the knowledge that they hold is far greater than that held by even the wisest and well-chosen group or experts. Whether or not Aristotle was right in his hopes for deliberation, Hayek shows that the possibility of surpassing the quality of the few best thinkers certainly does hold for free markets.

Hayek’s central point is that if the problem is how to aggregate information, the best solution often comes from the price system. In a system in which knowledge of relevant facts is dispersed among many people, prices act as an astonishingly concise and accurate coordinating and signaling device. They incorporate dispersed knowledge and in a sense also broadcast it, because the price itself operates as a signal to all.

Even better, the price system has a wonderfully automatic quality, particularly in its ability to respond quickly to changes. If fresh information shows that a product—a television, a car, a tablet, a watch—doesn’t work, people’s demand for it will rapidly fall, and so too the price. And when a commodity suddenly becomes scarce, its users must respond to that event. In Hayek’s account, the market works remarkably well as a whole, not because any participant can see all its features, but because the relevant information is communicated to everyone through prices.

Hence, Hayek claims that “[it] is more than a metaphor to describe the price system as a kind of machinery for registering changes, or a system of telecommunications which enables individual producers to watch merely the movement of a few pointers.”⁴ Hayek describes this process as a marvel and adds that he has chosen the word marvel on purpose so as “to shock the reader out of the complacency with which we often take the working of the mechanism for granted.”⁵ In Hayek’s account, the price system is an extraordinary collective intelligence mechanism, partly because it collects what everyone knows and partly because it imposes the right incentives.

To be sure, there is a real question of whether Hayek was right or whether markets might also encode mistakes and confusion (as behavioral economists Robert Shiller, Richard Thaler, and others have emphasized).⁶ In particular, Shiller argues that informational cascades and herd behavior affect stock prices, with investors following one another in a process of the blind leading the blind. In a fundamental challenge to Hayek’s optimism about market processes, Shiller contends that “the same forces of human psychology that have driven the stock market over the years have the potential to affect other markets.”⁷

We will return to the behavioral challenge—that markets can incorporate errors and illusions as easily as valid information—in a moment. For our purposes now, the central, practical question is this: Can people use the market mechanism to improve group judgments?

Incentives, Again

How might a company predict whether its own products will do well or poorly? A few years ago, Google adopted an innovative method.⁸ The firm created a prediction market, in which its employees could place “bets” about a variety of outcomes of importance to the company. Participants made forecasts about when products would launch, their likely success, and many other prospects. They invested virtual money, which could be redeemed for various prizes. The investments, or bets, set a price. If, for example, most people believed that Google would sell two million units of a new product in the next year and they bet accordingly, the price would reflect that belief.

By and large, the outcomes of Google’s prediction markets turned out to be stunningly accurate. If the price suggests that a product is likely to sell two million units, it is likely to sell two million units. If the price suggests that the product is not likely to launch by July 1, the product is unlikely to launch by that date. Dispersed knowledge within the company has been accurately aggregated in this way. Google’s prediction markets have worked because many employees, with private information, have offered their own opinions, and the sum of those opinions, the market price, has usually been right. The most striking finding is that in prediction markets, prices generally have operated as probabilities.⁹ When the price suggests that an event is 90 percent likely to happen, the event will happen 90 percent of the time.

Recall that in a deliberating group, members may have little incentive to say what they know. By speaking out, they provide benefits to others while possibly facing high private costs. By contrast, prediction markets realign incentives in a way that is precisely designed to overcome this problem. Because investments in such markets are generally not disclosed to employers (let alone the public), the bettors, or traders, need not fear any harm to their reputation if, for example, they have predicted that their own company’s sales will be low or that a certain candidate, not much liked by their friends and colleagues, will be elected president.

Because participants stand to gain or lose from their investments, they have a strong incentive to use (and, in that sense, to disclose) whatever private information they hold. In fact, managers who have installed prediction markets tell us that the chat groups and information postings that accompany the market are as informative to the managers as the outputs of the markets themselves. And even if the details of the information remain private, it will be reflected in the price signal. In these crucial ways, the problems that afflict deliberating groups are largely eliminated in prediction markets.

Prediction markets also impose strong incentives for traders to ferret out accurate information. Traders are unlikely to trade blindly, and they are able to cease trading, for a moment or more, to seek better information that will give them an advantage. In many deliberating groups, by contrast, participants cannot leave; they must continue deliberating, and the necessary information is often dispersed and locked within individual participants.

Like everyone else, investors are, of course, subject to the informational pressure imposed by the views of others. But because people stand to gain from revealing (through investments or “bets”) what they hold, a competitive market creates strong incentives for the revelation of whatever information people actually hold. And indeed, prediction markets have usually been found not to amplify individual errors but to eliminate them; the prices that result from trading prove reliable even if many individual traders err. In recent years, prediction markets have done more than just provide valuable information. In countless domains, their forecasts have proved extremely accurate, with prices generally operating as probabilities.

Some Successful Prediction Markets

Prediction markets, aggregating diverse views, are flourishing in numerous domains. Consider the Hollywood Stock Exchange, in which people predict (among many other things) Oscar nominees and winners as well as opening-weekend box-office figures. The level of accuracy has been impressive, especially considering the fact that the traders use virtual rather than real money.¹⁰ Among the most noteworthy achievements of the exchange to date is its uncanny accuracy in predicting Oscar winners, with the most probable winner bringing home the Oscar in the overwhelming majority of categories for which trading has occurred. If you want to know who will be thanking the Academy, you will do quite well if you consult the Hollywood Stock Exchange. And in fact, prediction markets that use virtual money often do as well as markets that rely on real money.¹¹

There are plenty of other examples. Many people believe that you can’t predict the weather, but the National Weather Service does quite well, and orange juice futures do even better.¹² Another large prediction market focuses on the likelihood that economic data released later in the week will show specific values; the market has performed even better than the consensus forecasts of about fifty professional forecasters.¹³

Many companies are now using prediction markets to aggregate diverse views. Managers at the consumer electronics company Best Buy established a prediction market called TagTrade (named after the yellow tag in the company logo). In this market, employees can wager on the outcome of outside events (like presidential elections and sporting contests) and company events (like quarterly sales projections or whether a new branch would open on time). Although the market runs on token credits, participation is encouraged through prizes: the top investors receive a $200 gift certificate and, even more coveted, a special embroidered shirt.¹⁴

Remarkably, the TagTrade market has outperformed Best Buy’s official sales team in forecasting quarterly results, has bested construction managers in predicting whether new offices would open on time, and has been credited with helping the company outlast longtime rival Circuit City.¹⁵ By mid-2011, TagTrade had run 240 predictions and engaged twenty-one hundred of the company’s US employees.¹⁶

A number of years ago, Hewlett-Packard (HP) and the California Institute of Technology initiated a project to study prediction markets as an information aggregation mechanism involving product sales.¹⁷ The experimenters selected people who worked in different parts of HP’s business operation. Because of its small size, the market was a very thin one, meaning that there were few participants and hence the market was far less liquid than ideal. Participants were chosen with the belief that each person could contribute information from his or her department.

The markets were organized so that securities existed for intervals of sales. For example, one security would pay off if sales were between one and ten printers; another would pay off if sales were between ten and twenty. In most of the experiments, the possible range of sales was divided into ten intervals of equal size. On the basis of the prices of each security, the experimenters could guess how many units HP would sell that month. Prediction markets were expected to have large advantages over internal projections that involve deliberation. Employees involved in sales have an incentive to understate projected outcomes, so as to ensure that the projections do not fall short of expectations; this bias, or a competing bias in favor of excessive optimism, can be reduced through market incentives.

The results showed that the markets’ predictions were a real improvement over HP’s official forecasts. In no fewer than six of the eight markets for which official forecasts were available, the market prediction was significantly closer to the actual outcome than the official forecast. Many other companies, including Ely Lilly and Company and Microsoft, have used prediction markets as well to supplement deliberation about future courses of action.

Since 1988, the University of Iowa has run the Iowa Electronic Markets (IEM), which allow people to bet on the outcome of presidential elections. Before the 2004 elections, the IEM did far better than professional polling organizations, outperforming polls 451 out of 596 times.¹⁸ In the week before the four elections from 1988 to 2000, the predictions in the IEM showed an average absolute error of just 1.5 percent—a significant improvement over the 2.1 percent error in the final Gallup polls.¹⁹ In 2004, the Iowa market did remarkably well. On midnight of November 1, it showed George W. Bush with 50.45 percent of the vote and John Kerry with 49.55 percent—very close to the final numbers of 51.56 percent for Bush and 48.44 percent for Kerry.

In 2008, the IEM predicted that Senator Joe Biden would be Senator Barack Obama’s vice presidential nominee well before political commentators considered Biden the front-runner. In the hours before the nomination was announced, the IEM reflected an 80 percent likelihood that Biden would be the pick.²⁰ In 2008, the IEM offered a final prediction of Obama at 53.55 percent and McCain at 46.45 percent—very close to the ultimate tally of 52.93 and 45.65 percent, respectively. The IEM was similarly accurate during the 2012 presidential election than it was in 2008. The market’s final forecast predicted 50.9 percent of the national vote for President Obama and 48.4 percent for Mitt Romney; the final tally was 50.6 percent for Obama, 47.8 percent for Romney.²¹ Notably, the IEM has proved accurate not only on election eve but also in long forecasting horizons.²²

Biases

What about biases? Might they affect prediction markets? Just as in group deliberation, investors in a market might be subject to predictable heuristics and biases. Of course, people are subject to these influences. For example, social scientists have found that people overestimate the likelihood that their own preferred candidate will win an election—a form of optimistic bias. At a certain point in the 1980 campaign, 87 percent of Jimmy Carter’s supporters believed that he would win, while in the same election 80 percent of Ronald Reagan’s supporters believed that their candidate would win.²³ Obviously, Carter’s supporters greatly overestimated their candidate’s probability of victory.

Is it shocking to hear that some gamblers in New York are particularly likely to bet on the New York Yankees?²⁴ IEM traders show the same bias. In 1988, Michael Dukakis supporters were more likely to hold IEM futures in the Massachusetts governor’s ill-fated presidential bid than were supporters of George H. W. Bush.²⁵ More strikingly still, Dukakis supporters were more likely to view the candidates’ debates as helpful to the Democratic candidate and accordingly bought significant additional futures in his campaign after each debate.²⁶ Supporters of Bush displayed precisely the same pattern.

People usually assimilate new information in a way that confirms their view of the world—confirmation bias—and those who invest in prediction markets show the same bias.²⁷ Undoubtedly, many investors lose money in the stock market for the same reason. In general, traders buy and sell in a way that fits with their party identification.²⁸

Despite all this, the IEM proved extremely accurate, even more so than polls, in predicting the outcome of the 1988 presidential election. No less than three weeks before the election, the market provided an almost-perfect guess about the candidates’ shares of the vote.²⁹ The simple finding is that while many people are biased, the markets as a whole are not. How is such accuracy possible when many traders are prone to blunder?

The likely answer lies in the marginal-trader hypothesis, which emphasizes the behavior of a small group of traders who are less susceptible to biases. According to this hypothesis, certain traders, immune to the major biases, have a disproportionately large effect on prices. In election markets, these traders have been able to earn significant profits at the expense of other traders.³⁰ If marginal traders are active and able to profit from the errors and biases of other participants, then there will be no effect on the aggregate market price—a real testimony to Hayek.

A distinct bias that might be expected to affect prediction markets is the favorite-long-shot bias often seen in horse races. In horse racing, heavy favorites tend to produce higher returns than do other horses in the field, whereas long shots tend to offer lower-than-expected returns.³¹ The same bias is seen in tennis matches, where the best players get higher returns and the lower-ranked players attract a disproportionate share of bets.³² In these domains, bettors undervalue near certainties and overvalue low probabilities.

If this point holds in general, prediction markets might not be accurate with respect to highly improbable events.³³ The market should be expected to overestimate the likelihood that such events will occur. But in existing prediction markets, we see at best modest evidence of systematic errors in this vein.³⁴

In some markets, of course, biased traders do mean biased markets. The whole field of behavioral finance attempts to explain how biased markets can persist over time.³⁵ But prediction markets have generally been free from this problem. We will undoubtedly be learning a great deal more about this topic as it is an active area of research.

Bubbles and More

Everyone knows the stock market can contain bubbles, in which stocks trade well above their fundamental value. Bubbles often occur when people believe not that a stock is really worth a great deal, but that other people think that the stock is worth a great deal. Hence people invest with the expectation that value will increase because of the enthusiasm of other investors.³⁶

Can bubbles occur in prediction markets? Of course they can. Prediction bubbles are easy to imagine, with investors moving in a certain direction with the belief that many other investors are doing the same. And if bubbles are possible, crashes are possible too.

In any case, informational influences can lead people to make foolish investments in any market, including those involving predictions. Informational cascades play a large role in investment decisions, just as in choices about what products to buy. A fad might suddenly benefit a sneaker company, a book, a television program, a restaurant, or a movie. Many people might be attracted to one of these, not because they have independent information that the product is good, or even because they believe that it is much better than the alternatives, but simply because they are following the signals provided by the cascade.

Hayek did not much grapple with the risk that markets will suffer from herd behavior. If he had, his best response would be that smart investors will be alert to that risk and willing to take advantage of it. If herds of people ensure that a product or a stock is a bargain, then other people—first a few, then a lot—will purchase it, and the market will eventually correct itself.

But experience shows that this view can be too optimistic, at least for ordinary stocks.³⁷ As prediction markets develop, significant individual errors should be expected and will undoubtedly produce some errors in the price signal. Consider the 2004 presidential election. On the day of the vote, dramatic news of pro-Kerry exit polls produced not only a huge switch in the conventional wisdom, but also a great deal of volatility in election markets, with a wild swing in the direction of Senator Kerry at the expense of President Bush.³⁸ The rumors affected investors, not just onlookers.

Large-scale errors are always possible when apparently relevant news leads numerous investors to buy or sell. Indeed, election day in 2004 may well have been a cascade, with investors responding to one another’s judgments, even though the judgments were based on misleading information. But for those enthusiastic about prediction markets, there is some favorable evidence: the erroneous figures lasted only a few hours, after which the numbers returned to their previous level of accuracy.

Feasibility

For many groups, prediction markets face a pervasive problem of feasibility. They are not simple to create, and establishing them might prove confusing and time-consuming. If an employer or a local government needs to make a short-term decision, a prediction market is unlikely to make a lot of sense. When the relevant groups are small, effective markets may be impossible to create, simply because of the absence of sufficient numbers of investors. A certain number is necessary to ensure that prediction markets have enough information to aggregate. Nonetheless, some private and public institutions might well enlist such markets to resolve a number of questions, and ambitious efforts are under way to examine how public officials might use these markets.

In fact, governments might consider using prediction markets to help make projections about budget deficits, the risk of insolvency, and the probability of outbreaks of disease or natural disasters. In each of these cases, the results of prediction markets might provide a reality check on deliberative processes. Officials might take into account the markets’ predictions of the anticipated damage from a natural disaster, the number of annual deaths from an actual or anticipated disease, and much more. In all these cases, private or public institutions might create markets to provide information on crucial questions, and public institutions might take that information into account in making judgments about policy.