Inefficient Markets
An important argument against increasing reliance on prediction markets in the corporate context is that if prediction markets are inaccurate, then their misleading forecasts might lead to bad decision making. Of course, all proposals to use prediction markets depend at least implicitly on the proposition that the markets are accurate, at least relative to plausible alternatives. Ultimately, empirical work is needed to establish how accurate prediction markets are. The evidence that we have seen so far is encouraging. It is hard to look, for example, at the trading patterns in the Iowa Electronic Markets and detect significant shifts in prices that seem unrelated to the underlying dynamics of the various election races. Nonetheless, accuracy is a matter of degree, and further experiments may help to show the degree to which increases in market subsidies improve accuracy. Skeptics, moreover, might claim that a consideration of only evidence of the accuracy of prediction markets is too narrow, for our broader experience with the accuracy of market pricing is relevant to the issue of prediction markets. Indeed, these skeptics are likely to argue that evidence suggests that there are inefficiencies in securities markets generally and in the stock market in particular, and that prediction markets might inherit these inefficiencies. Not so long ago, claims that securities markets could be inefficient seemed to many observers to be unfounded. One reason is that there seemed to be a strong theoretical case that markets would be self-correcting. In 1953 Milton Friedman, arguing for flexible rather than fixed exchange rates, offered two simple reasons why “noise traders” acting on the basis of pet theories rather than evidence would not influence securities prices.22 First, other traders would have incentives to function as arbitrageurs and bet against the noise traders. Indeed, the presence of noise traders should increase the overall incentives to participate in the market because they make it possible to make money by trading. This rationale is, of course, similar to the simple theoretical case for prediction markets: that they give incentives for people with relatively good information to profit from that information by in effect betting against those with worse information. Second, the noise traders should tend to lose money and thus disappear from the market over time. By the same token, this Darwinian theory suggests that the most successful traders will tend to increase the money they have available to invest. In addition to this simple theoretical argument, until recently it seemed that at least some versions of the Efficient Capital Markets Hypothesis had a robust empirical foundation, for example, in studies indicating that it was not possible to predict the future direction of stock prices from past prices.23 In its strongest form, the hypothesis states that markets incorporate all information, both public and private. That view long ago proved untenable for at least two reasons. First, there are instances in which insider trading laws do succeed at preventing private information from being incorporated into securities prices. This will not always be so, and there is substantial evidence that insider trading occurs; one interesting study shows that members of Congress earn significantly higher returns than one would ordinarily expect, suggesting that they have inside information.24 But sometimes, public release of information previously known to private parties has an effect on market prices. Second, it is often expensive to find and analyze information. Thus, defenders of the hypothesis acknowledge that information will be incorporated into market prices only to the extent that it is cost-effective to trade on the information.25 In spite of these caveats, acceptance of the Efficient Capital Markets Hypothesis would tend to limit concerns about inaccuracy in prediction markets to concerns that there might be too little incentive in some contexts to find and analyze information. Recent developments in the finance literature, however, suggest that even this weaker version of the hypothesis might be vulnerable. In particular, a large literature has developed concerning “behavioral finance,” elaborating how human cognitive heuristics and biases might lead to systematic distortions in market pricing. Models suggest that in some circumstances well-informed traders might recognize that a market is in disequilibrium yet not have adequate incentive to engage in arbitrage to correct the disequilibrium. If such effects are sufficiently large, then markets sometimes might produce predictions that are not equivalent to what the best-informed individuals believe. The behavioral finance literature seeks to explain phenomena including “bubbles” such as the alleged technology bubble of the late 1990s that peaked in early 2000. If asset markets in general can be subject to bubbles, or for that matter to panics leading to depressions, then perhaps the same psychological factors can influence prediction markets as well. The behavioral finance literature provides an important qualification to optimism about the power of markets to make predictions, but from a practical standpoint, the concerns do not seem to be of sufficiently high magnitude to bring into question most prediction market proposals. The ultimate question, after all, is not whether a prediction market is a perfect predictive tool but whether it is the best available predictor. Virtually no one suggests that the literature would justify replacement of our securities markets with a radically different institution. Although there are some who advocate consolidating investment decisions in a centralized government decision maker,26 it is generally thought that the incentives of government officials to make decisions about which business ventures are likely to succeed would be inadequate. Private markets, whatever their flaws, seem likely to allocate scarce social resources more efficiently. Similarly, though prediction markets have imperfections, for some but not all predictive decision making proposals, these imperfections may be small relative to the flaws of the governmental decision-making approach that they seek to replace. Occasional gross mispricing, if such a phenomenon could be identified with confidence, would not necessarily justify abandonment of the technique. It would be a mistake, moreover, to infer that any market inefficiencies in securities markets generally will necessarily carry over to prediction markets. The concept of a prediction market, we have seen, does not refer to any single implementation but rather to a class of approaches that give financial incentives to individuals who sequentially make predictions about an event of interest. The probability estimate prediction market and numeric estimate prediction market introduced in Chapter 1 seem akin to classic securities markets, but even with these market designs, investors might behave differently. Prediction markets would not ordinarily be useful vehicles for retirement, because investments in them do not have a systematic upward trend, and so the investors choosing to place money into prediction markets on the whole might be a more sophisticated group. The systematic upward trend in the stock market means that many foolish investors will make some money, but it might be much harder for noise traders to make money on prediction markets. Even with subsidized prediction markets, potential noise traders should recognize at least that they should focus on prediction markets in which they might have some insight about the variable being predicted. Unsophisticated investors who choose to invest in prediction markets might be more sophisticated about their predictions than they are about their decisions whether to move funds into the stock market. Some people, for example, might try to “time” the stock market. An empirical study analyzing ninety-one thousand investors in an index fund shows that investors can be grouped into two rough categories: those who generally seek to follow trends and those who generally act against trends.27 Individuals who blindly follow stock market trends, potentially providing a feedback effect that can lead to a bubble, might not follow prediction market trends. Someone might mistakenly believe that if the stock market is going up it will keep going up but not believe that because George W. Bush’s probability of winning has risen from 50 percent to 60 percent in the past week in the Iowa Electronic Markets that inevitably his price will keep rising. For one thing, it will be obvious to participants that the trend is not sustainable, that Bush will never have a 110 percent chance of winning. Indeed, those who are most skeptical about the efficiency of the stock market might not be skeptical of the accuracy of individual prediction markets. For example, Robert Shiller, a leading behavioral finance theorist, has emphasized that “no one should conclude from any of my or others’ research on financial markets that these markets are totally crazy.”28 Rather, he believes that “the aggregate stock market in the Prediction market forecasts in many cases also might be simpler to make than stock market predictions. Shiller points out that “investing for the long term means judging the distant future, judging how history will be made, how society will change, how the world economy will change. . . . With such a confusion of factors, it is hard for anyone to make objective judgments without being influenced by the recent success behavior of the market and the recent success of investments.”30 Modeling the result of a presidential race might also involve many imponderables, and the most sophisticated prediction market participants will do a better job of assessing these than will others. But because these imponderables will have largely marginal effects, with basics such as poll numbers largely determining prices, investors are less likely to throw up their hands and focus solely on immediate past trends. Of course, some prediction markets might involve forecasts that are just as difficult as anticipating the future success of the stock market, and with those markets, psychological factors might have more of an effect. Another reason that anomalies in the stock market might not appear in prediction markets is that the mechanisms by which the markets operate might be different. Would anomalies observed in traditional stock markets also occur in the dynamic pari-mutuel market? In the market scoring rule, which seems more like a series of bets than a market in any meaningful sense? Some behavioral finance anomalies might persist in a wide variety of market designs, but in other cases, market design might make bubbles and other observable deviations from rationality less sustainable. Of course, prediction markets might introduce anomalies and imperfections of their own, but the literature identifying specific anomalies in stock market investing, such as the closed-end fund anomaly,31 might not be helpful in anticipating particular reasons that prediction markets might be inaccurate. It is possible that some pathologies of the stock market will have analogues in prediction markets, but prediction market design might limit or exacerbate these pathologies. One possible source of volatility in markets is people’s overreaction to the information of others. Kenneth French and Richard Roll point out, for example, that if the first ten trades on a particular stock all happen to be sell orders one morning, other traders “might draw the reasonable conclusion that something bad has happened to the company,” and their sales might induce others to sell as well.32 Colin Camerer and Keith Weigelt conducted a laboratory experiment in which they sought to identify such “information mirages.”33 The experimental design spanned a number of periods, and in each period all experimental subjects shared certain public information about the price of an experimental asset. In some cases, however, some randomly chosen subjects also received additional inside information about the price. Of a total of forty-seven periods in which no such information existed, in a total of four periods a mirage occurred in which traders acted as if some information did exist. “Sustained mirages do occur,” Camerer and Weigelt conclude, “but they are not common.”34 Something similar could occur in a prediction market. Suppose that different predictors in a market have different levels of information, with some predictors functioning as “noise traders” who have no information at all make predictions based on crackpot theories that well-informed traders would discount. Because each predictor has only a limited amount of information available, predictors should discount their own views to some extent and give weight to the predictions of other traders. Occasionally a predictor might misestimate the quality of information possessed by other predictors. Sometimes this might mean that the market insufficiently reflects good information, but at other times this might mean that the market places too much weight on a meaningless prediction. The eventual market prediction might thus be a mirage, a result of misinterpretation of activity in the market itself. In the situations in which a mirage occurs, other prediction mechanisms might have outperformed the market. There are at least two reasons that this type of flawed derivatively informed trading does not provide a strong argument against prediction markets. First, the reason that predictors pay attention to others’ predictions is rational, and so given a set of predictions in the market, predictors in settings outside the market would have an incentive to take the trades into account. Sometimes information signals that are valuable as a general matter will lead predictors astray. For example, I might decide to bet against a horse only because I find out that the horse did not eat well that morning. I conclude that the horse might be ill, and yet he just might not have liked his oats, and he might win the race. That does not mean that information about horses’ eating habits reduces my predictive accuracy or the overall market’s. Presumably, the overall accuracy of the markets that Camerer and Weigelt studied would have been reduced if there were no way for traders to divine other traders’ information. Relying on others’ predictions, like relying on information about a horse’s eating, will tend to make one’s own predictions and the market’s consensus predictions more rather than less accurate, though in some cases the information will lead to mistaken inferences. Second, many of the prediction markets imagined in this book would involve a relatively small number of participants with relatively high amounts of private information or private analysis about the event being predicted. This will not be true of all prediction markets; the presidential election prediction markets, for example, attract a large number of participants, and there is relatively little private information. When it is true, however, market participants will not simply look for oddball trading patterns but might also consider the reputation and track record of other predictors. Especially in contexts in which some outsiders will have incentives to manipulate the market, prediction market participants will be hesitant to place much weight on the actions of traders without strong track records. Moreover, with a deliberative prediction market design, market participants will have incentives to reveal their underlying reasoning, and others will have incentives to scrutinize that reasoning, so errors and arbitrary decision making might more easily be exposed. A more difficult scenario is one in which the noise traders, instead of posting arbitrary bets, make systematic errors. Perhaps the most prominent explanation of inefficient markets, developed by J. Bradford De Long, Andrei Shleifer, Lawrence Summers, and Robert Waldmann, relies on such a scenario.35 Suppose, for example, that a large number of noise traders irrationally think that technology stocks are worth much more than well-informed traders think they are likely to be worth. Rational arbitrageurs will bet against these noise traders by, for example, selling technology stocks short (that is, borrowing shares and selling them in the hope of making profit if the price goes down). But they might not do so sufficiently aggressively to counter the noise traders. One reason is that there might be some underlying fundamental risk in the securities being traded. The rational parties do not want to assume the risk that technology stocks will appreciate for reasons having nothing to do with the noise traders’ predictions. An additional reason is that the noise traders’ actions directly create some additional risk. The rational arbitrageurs may care not only about their long-term profits but also about their medium-term profits. For example, if they have borrowed stocks and then sold them in the hope that the price of the stocks will fall, a strategy known as “short selling,” at some point they will have to return those stocks. If the noise traders continue to overvalue technology stocks at that time, then the rational arbitrageurs will lose money. It is not enough to know that a market is in a bubble to profit from that bubble; one must also be able to guess when the bubble will pop, and that must occur sufficiently soon that it will be profitable to bet against the noise traders. They, meanwhile, can be expected to survive in the market, potentially for a long time. Indeed, their willingness to take on risk that the rational arbitrageurs seek to avoid, De Long and his coauthors point out, means that for long periods of time, they may earn higher returns than will the rational parties. This, of course, will encourage even more traders to follow the irrational theory. Similar dynamics could affect prediction markets, too. Suppose, for example, that a group of Green Party voters have deluded themselves into thinking that their candidate will win the presidency. Sensing a profit opportunity, they buy up TradeSports shares in the Green Party candidate. There might be some limit to the willingness of other market participants to seek to counter this activity (for example, by shorting shares, if TradeSports permitted this). There might be a small risk that the Green Party candidate would surge to victory and a larger risk that the price level could fail to fall for a sufficiently long period of time that arbitrageurs could lose money in the short term. The scenario might be more plausible if the Green Party candidate had some chance of winning, but that chance was considerably smaller than the deluded voters thought. This scenario seems much less likely than the bubble story that De Long and his coauthors tell, even putting aside the fact that there is no reason to suspect that there would be a group of voters with such an odd prediction. The presidential election will necessarily occur at a particular point in time, whereas stocks generally represent income streams flowing into the indefinite future. This reduces the danger that a bubble will be prolonged and increases the ability of arbitrageurs confidently to combat the foolish trading. Support for this theory comes from a study finding that noise trader sentiment does not affect market price in futures markets that clear at some specific point in the future.36 Of course, some prediction markets might last for a very long period of time, so in these markets, the danger is somewhat greater. Also, deliberative prediction markets might make the problem worse, because participants would be betting on the price in the next time period rather than on the closing price. In these situations, however, the knowledge that the bubble will necessarily pop by a certain date might lead the bubble to pop well before that date. Arbitrageurs, for example, can seek to obtain loans to be repaid after that date. Another explanation for bubbles is that in some cases, it might be infeasible for arbitrageurs to short overpriced securities because there might be an insufficient number of such shares to sell. Robert Shiller points out that as arbitrageurs sought to attack the shares of Palm, Inc., they created so much demand for the shares that “the interest cost of borrowing Palm shares reached 35 percent by July 2000, putting a damper on the advantage to exploiting the mispricing.”37 Eli Ofek and Matthew Richardson have argued that restrictions on selling short played a significant role more generally in the alleged technology bubble.38 This thesis is controversial. Robert Battalio and Paul Schultz argue that more complete data suggest that options prices tracked stock prices so closely during the alleged bubble that possibilities for arbitrage existed for anyone who recognized the existence of the bubble.39 Nonetheless, if the thesis is correct, the bubble was largely an artifact of the mechanism (short selling) through which traders can profit on a belief that securities are overpriced. Market mechanisms that save arbitrageurs from the formalities associated with short selling might be able to help. For example, the Iowa Electronic Markets allows any trader to pay one dollar to receive a share in each candidate. An arbitrageur thus does not need directly to short shares or to pay interest charges in doing so but can simply buy shares in all candidates and then sell shares in an overvalued candidate or buy shares in an undervalued candidate. Similarly, the market scoring rule simply allows an arbitrageur to enter a new prediction, and if that prediction is better than the last one, the arbitrageur can expect to make money. With the market scoring rule, the ultimate question becomes which group–the noise traders or the fundamental values traders–will give up first, no longer willing in effect to take bets from the other group. It seems unlikely that there would be many contexts in which the noise traders with a bad theory would have more financial strength than all other traders combined. Arbitrageurs also tend to have more luck correcting mispricing in prediction markets when there are many prediction markets and the varieties of mispricing differ from one market to another. Andrei Shleifer and Robert Vishny note that one source of systematic deviations from underlying value in markets is that investors “rationally allocate money based on past returns of arbitrageurs,” and so arbitrageurs have some incentive to seek to ride a market bubble for some period of time.40 Survey evidence suggests that investors attempt to do precisely this.41 The problem is more serious when the mispriced securities form a large portion of the arbitrageurs’ portfolio. When mispricing occurs in a single prediction market, the risk from a sustained bubble is not great for arbitrageurs who make predictions in a variety of prediction markets. In such cases, arbitrageurs recognize that they may need to wait out some bubbles longer than others, but the happenstance of an occasional long wait will not prevent them from taking advantage of mispricing. Some of the problems that can beset stock markets, however, could be at least as serious with many prediction markets. De Long and his coauthors suggest that one underlying psychological mechanism leading to bubbles is that people suffer from a “tendency to underestimate variances and to be overconfident” about their predictions.42 Experts are vulnerable to this tendency, and prediction markets might tend to attract individuals who are particularly confident, perhaps overconfident, with respect to the particular question posed by the prediction market. This might seem likely to produce systematic errors in probability estimation prediction markets, with too few predictions in the middle of the probability spectrum. This claim is empirically testable, however, indeed more easily testable with prediction markets than in the stock market, because former eventually end. The preliminary evidence appears to suggest that large errors of this kind do not seem to occur in the prediction markets for which data are available so far (see Chapter 1). Whatever subsequent evidence reveals, analysis of this sort seems more likely to help assess the accuracy of prediction markets than does theorizing about inefficient securities markets in general. Nonetheless, if securities markets are inefficient, prediction markets that forecast stock prices might be inefficient even if prediction market mechanisms do not inherit the problems of securities markets directly. For example, if a prediction market is gauging the stock price of Google one year hence, and participants believe that Google is in a bubble that is unlikely to pop within a year, then the prediction market will reflect the bubble price. The market price itself would not be a bubble, but an accurate prediction would reflect the bubble in the underlying market. If bubbles are a sufficient concern, it might make sense for prediction markets to take a longer-term perspective. For example, they might forecast the effect of different possible decisions about stock price (controlling, of course, for possible stock splits and dividend payments) five years into the future. As long as a prediction market takes a sufficiently long-term perspective, then the only bubbles that we need to worry about are those within the prediction market itself.
Leave a Reply