Regarding many topics, private information seems unlikely to be a significant concern, and few players will have sufficient financial incentives to seek to manipulate prediction markets anyway. In those cases, the media might consider publishing forecasts from prediction markets or perhaps sponsoring prediction markets. Media organizations, after all, have a long tradition of being involved in polling, so they can provide their readers and viewers with information that historically was not immediately available to those who view other news organizations’ coverage. Legal restrictions might limit the ability of media organizations in the United States to sponsor prediction markets (see Chapter 2), but media organizations in other countries could consider creating such markets, and U.S. media might create prediction markets using play money. Elections are among the most obvious subjects for media coverage of prediction markets, in part because many such markets already exist and in part because of the intense interest of the public in elections. Sports and entertainment markets (predicting winners of the Grammy Awards, for example) also seem like reasonable candidates. But the media in theory could create prediction markets on virtually any topic of interest to readers. Within the political arena, the media might create innovative new markets. For example, a market might predict what the president’s approval rating will be at the end of the term. This would help counteract a tendency for approval ratings to be based largely on recent events. At some times, the purpose of approval ratings is to assess the president’s recent performance, but at others, it might make sense to seek a long-term perspective. Or a market might predict whether particular legislation that is of interest to readers is likely to be passed. At the very least, prediction markets could supplement explicitly predictive coverage. For example, the National Journal runs a congressional insiders’ poll that routinely shows vast discrepancies in predictions between Democrats and Republicans.58 Outside the political arena, there are many possibilities for prediction markets that could offer information that journalists ordinarily cannot easily provide with objectivity. For example, instead of simply reporting on what politicians and academics predict will be the number of troops that the
News You Can Use
Policy Analysts
The number of prediction markets that media organizations can be expected to sponsor in the future will in all likelihood be limited. Although it is possible to imagine a great number of topics regarding which prediction market predictions might be of interest to readers, a limited number of individuals may wish to bet on prediction markets. This is especially so if prediction markets are zero-sum games, in which sophisticated players systematically earn profits from those who are less sophisticated. At some point, one might expect the latter to wise up, though the number of sports gamblers in this country might suggest otherwise. At least, relatively few seem likely to participate in prediction markets on esoteric topics. Without these unsophisticated bettors to take advantage of, many sophisticated players will have little interest as well. This is unfortunate, because if prediction markets are to have the potential to improve democratic governance, it will not be by revolutionizing predictions about which team will win the World Series. There are, however, many organizations committed to improving public policy through improved analysis, and some of these do so largely for free. These organizations include think tanks and even universities, particularly those specializing in fields such as economics, international relations, and law. I will refer to the group of scholars and other thinkers who engage in analysis relevant to public policy as “policy analysts.” The world of policy analysts is a strange one. The analysis produced might be seen largely as a public good; if a professor publishes a new theory, it will be hard to exclude anyone from finding out about it, despite the availability of copyright protection, and the fact that one person has learned about the theory will in general not make the theory any less valuable to someone else. There is thus a case for governmental funding of some policy analysis, and indeed the government funds some analysis via grant programs and via contracts awarded to organizations such as the Rand Corporation. To a great extent, however, analysis is subsidized by private donors such as the sponsors of think tanks and wealthy alumni after whom chaired professorships are sometimes named. These donors indirectly provide a valuable resource to the public. But how valuable is this resource really? Even many policy analysts who believe that they are making important scholarly contributions might not be confident in concluding that they are providing sufficient value to the world to justify from a social standpoint the portion of their salaries that might be attributable to research. Some might nonetheless take some satisfaction in the conclusion that their research would make the world a better place, if only others would adopt their suggestions. Others might acknowledge that their individual contributions will not have direct consequences but take pride in working out details of a broader intellectual movement that might have ramifications in the longer term. And still others might point out that a few policy analysts do produce breakthrough findings that produce gains for public policy, but because we cannot figure out who those analysts will be in advance, the system must subsidize a broader policy analysis community that produces mostly work of little value. There might, however, be another factor that contributes to the nagging suspicion that policy analysts might in the end not contribute much to policy. The problem is that we generally do not have good mechanisms with which to aggregate the views of different policy analysts and to produce consensus recommendations. This is particularly problematic in a world in which disagreement is pervasive. The basis of the humor in many jokes about economists, after all, is that economics is an inexact science. (Consider, for example, the economist who, asked what two plus two equals, answers, “What do you want it to equal?”) Indeed, formal surveys of economists identify numerous questions concerning which economists disagree.1 Yet, presumably, regarding any given economics question, it might be valuable for policy makers to have available to them some numbers aggregating the conventional wisdom, in spite of the reality that the conventional wisdom can be wrong. The policy analysis community might contribute greater value to policy analysis if somewhat less effort were paid to developing policy arguments and somewhat more effort were paid to identifying consensus positions. The problem, of course, is that we need institutions that can produce consensus positions of policy analysts. Surveys provide one means of identifying specific questions concerning which large numbers of policy analysts agree, but they are much less useful when applied to discrete questions for which only a relatively small number of individuals have done research that equips them to give good answers. There is rarely an objective metric for determining who is qualified to answer a particular question. Even if such a metric existed, the people surveyed might not be a random sample of the broader community of potential analysts. Perhaps liberals or conservatives, or simply people with strong prior beliefs about a particular issue, are more likely to obtain expertise about that issue. Moreover, surveyed policy analysts might not give honest answers, particularly if their answers might have some effect on policy. An analyst might favor a particular policy because it has one effect that he or she believes is beneficial, and he or she might therefore claim that the policy has a different effect that is the subject of the survey. Prediction markets might help produce estimates of consensus. This chapter considers only the ways in which prediction markets might be used to make predictions about factually verifiable matters, though much of the disagreement among policy analysts presumably reflects differences in values. (In Chapter 6 we will consider whether prediction markets might be useful ways of aggregating normative commitments, but presumably the normative commitments of policy analysts will not be of any special interest.) Policy analysis organizations can play a critical role in this process in two ways. First, they might sponsor prediction markets on objective questions that are of interest to public policy. Second, they might encourage the policy analysts who work for them to participate in prediction markets. It will be a long time before an academic coming up for tenure will boast that she has published four articles and earned a hundred thousand dollars playing prediction markets on the university account. If a goal of policy analysis is not merely to create new knowledge but to aggregate knowledge in an accessible form, then participation in prediction markets should be encouraged, however. This is especially important for some of society’s most divisive issues, such as global warming, concerning which objective predictions could be quite useful. This chapter elaborates the argument that prediction markets can help aggregate information, noting that at times what prediction markets aggregate might more accurately be understood as individual assessments of information. Prediction markets’ aggregative powers will be greater if markets are subsidized, and this chapter l explains how the market mechanisms described so far might be subsidized. It provides a partial description of the Policy Analysis Market, which was to have been a governmentally subsidized set of prediction markets. The collapse of that program should give caution to those who are optimistic about the short-term possibilities of prediction markets. This chapter also describes current regulatory obstacles and ways they might be avoided. It then discusses an existing application of prediction markets, the Foresight Exchange, which avoids those obstacles by using play money, and then imagines two potential government applications of prediction markets.
Information Aggregation and Assessment Aggregation
Markets are sometimes said to be good at aggregating information,2 but it may be useful to distinguish two types of information that prediction markets might aggregate. One type is the evidence that someone might use to make a particular prediction. For example, in a prediction market that forecasts election returns, it might include poll data, historical data, economic data, and the like. Secret news from campaign officials would also fall within this category. Another type of information is the predictive assessment that any particular individual makes on the basis of such evidence. For example, after reading data of the first type, you might share with me that you believe that there is a 65 percent chance that a particular candidate will win the election. Suppose that I had consulted the same data and had concluded, before you shared your assessment with me, that there was a 55 percent chance that the candidate would win. Once I hear your assessment, I will probably adjust my prediction to somewhere between 55 percent and 65 percent, depending on my judgment about our relative abilities as predictors.3 Ideally, we might hope that prediction markets would aggregate both kinds of information, which we might distinguish with the labels “evidence” and “assessments,” respectively. The relative importance of evidence and assessment aggregation might vary across prediction market contexts, but in many, the latter is more important than the former. In an election market, for example, different traders ordinarily have access to the same data, and the primary function of the prediction market is to aggregate different traders’ views of those data. To be sure, on occasion, some traders might trade on the basis of data not yet available to the public, such as poll results that have not yet been publicly released. The sponsors of a prediction market might even wish to encourage such insider trading. Nonetheless, in many contexts that are of relevance to public policy, evidence generally is widely available. The primary goal of many prediction markets is to provide a form of assessment aggregation. The distinction between evidence aggregation and assessment aggregation helps explain a common criticism of prediction markets: they do not seem to tell participants much more than they could figure out themselves by considering the underlying materials. For example, Orin Kerr commented on a TradeSports market used to predict the probability that each of various individuals would be chosen by President Bush as his nominee to replace Justice O’Connor. Kerr points out that neither the president nor any of his advisers seems likely to place bets on TradeSports.com, so “the people who are placing bets presumably are outsiders who are getting their predictions from newspaper articles, blogs, horoscopes, etc., and then placing bets.” As a result, “a site like TradeSports would seem to just mirror the collective common wisdom of newspapers and blogs on a question like this.”4 One possible answer to Kerr is that the Supreme Court selection market at least opened up the possibility of accomplishing the task of evidence aggregation. For example, someone might acquire some unique piece of information such as an observation of a particular candidate and his or her family driving into the gates of the White House. Perhaps the better response, however, is that Kerr’s observation is correct. The primary purpose of the market is assessment aggregation, but that can be important. For someone who is interested in assessing the probabilities of various candidates, it will probably take less time to look at the TradeSports prices than to read all of the underlying information. And although it will always be possible for someone skilled at reading blogs and tea leaves to beat the market, for the typical observer, the prediction market forecast will on average be more accurate than the prediction that the observer independently could derive, because the market will represent an aggregation of the views of a large number of observers. It might therefore in part cancel out random errors that individuals make in predictions by overweighing or underweighing particular pieces of evidence. Kerr’s critique emphasizes, however, that we should not blithely assume that prediction markets will succeed at perfectly aggregating information, whatever its type, and any claim that the Supreme Court market would incorporate inside information was clearly erroneous. Indeed, on the morning Bush made his announcement, the market followed many news reports’ rumors that Judge Edith Clement would be the president’s choice. The markets obtained the right answer only once the administration started notifying the press that the president had chosen Judge John Roberts instead.5 A market that perfectly performed the task of evidence aggregation would have confidently predicted Roberts’s selection as soon as the president had decided to pick him, but of course such perfection is impossible. That the trading price followed reports that many independent observers and journalists thought credible might seem to be an indictment of the market if its goal is seen as evidence aggregation. Such a market might have been even more useful, but the market’s performance was admirable if its task is seen as assessment aggregation. Although evidence and assessment are forms of information, aggregation of assessments often is an easier task than aggregation of evidence. Given the same public information, market participants often make similar assessments because they are considering the same underlying data. Individual assessments are not statistically independent, though for difficult estimation problems, individual assessments might vary considerably. Individual pieces of evidence, on the other hand, might be independent. Moreover, it may be very difficult for someone who possesses a piece of evidence to generate an accurate sense of what the broader distribution of pieces of evidence might look like. Suppose, for example, that a prediction market is used to predict the outcome of a criminal trial about which little public information has been released. One trader finds out about a piece of evidence that is not a smoking gun but militates toward guilt. It is very difficult to determine how important the piece of evidence is relative to other pieces of evidence. Even if each piece of information is known to one trader on the market, the prediction market might not do a great job of producing the probability estimate that someone who had all pieces of evidence might reach. The lines between evidence and assessments sometimes can be blurry, because analysis of data is somewhat like evidence and somewhat like assessments. Different participants in some prediction markets might analyze evidence in a relatively sophisticated way, for example, by using regression analysis. These individuals might take slightly different approaches and produce different results. On one hand, to the extent that each participant’s analysis determines that analyst’s prediction, the outcome of the analysis functions as a kind of independent assessment. On the other hand, if these participants do not release their analyses, then each analysis might be thought of as a piece of private evidence available only to that participant. The more sophisticated the individual analyses, and the greater the degree to which relatively sophisticated players effectively determine prediction market prices, the more the challenge of a prediction market can be said to be evidence aggregation rather than assessment aggregation and the more we might need to worry about whether the market does as good a job at information aggregation as would alternative possible institutions. It is difficult to know for sure where existing prediction markets are on the continuum between evidence aggregation and assessment aggregation. We cannot know for sure whether trading prices in the Iowa Electronic Markets or TradeSports are largely determined by sophisticated but secret statistical analyses, which themselves might be seen as representing a type of evidence. The overall level of market sophistication is less than the overall level of academic sophistication, given the academic studies showing profitable trading strategies. But there is some chance that the publicly released academic models, which were built on data that already exist, will ultimately be proved inferior to more sophisticated models being used by private traders. And there is a reasonable chance that this will be true in the future if it is not true today. Most existing prediction markets focus on questions of fairly general public interest. But it might also be useful to have prediction markets focusing on public policy questions that have policymaking relevance but are only interesting to small communities of people. Yet more arcane prediction markets might produce too few people to provide for a meaningful level of trading; we will see how market design can address this concern in Chapters 3 and 4. If there were a sufficient number of policy analysts who have expertise in the relevant area and engage in trading, however, would these public policy prediction markets primarily perform assessment aggregation, or would they perform evidence aggregation? The greater the extent to which a prediction market for a relatively arcane issue incorporates sophisticated models, presumably, the more accurate it will be. And of course, if the market encourages the creation of sophisticated models for questions of public interest, so much the better. That still leaves the question of how we can improve the market’s success at performing evidence aggregation if these models are created, a question I address below. The more immediate question, however, is whether prediction markets could draw in a sufficient number of policy analysts to perform even rudimentary assessment aggregation.
Subsidized Markets: Introduction
One reason that prediction markets sponsored by TradeSports do not necessarily reflect the predictions that could be made by the most sophisticated models is that TradeSports charges commissions. These commissions, we have seen, are small, but they compound the inherent risk of investing in prediction markets. The best strategy for prediction market investing could go awry in the short term, just as the best card counters in poker might lose money on a bad day or in a bad year. Thus, someone who has created a model and is confident that this model can beat the market by two or three percentage points on average might well decide not to invest in the TradeSports markets. Enough sports fans and gamblers exist to provide ample trading at many of the TradeSports markets anyway, but this is less likely to be the case for markets on esoteric topics aimed at policy analysts. Many policy analysts seem likely to invest only if they can expect to make money. For example, development of an improved account of how clouds work might improve knowledge about global warming, but it could still be risky to invest money in a global warming prediction market. The first step toward increasing incentives for prediction market participation would be to eliminate commissions. Of course, a for-profit institution such as TradeSports seems unlikely to take this step, though an alternative profit model would be to earn money by collecting interest on invested money.6 The nonprofit Iowa Electronic Markets does not charge commissions, and other nonprofit entities might follow suit. Even TradeSports has eliminated the per-transaction commissions charged to a market participant who places an order on the bid and ask queue that another trader later accepts. Presumably, this is because TradeSports recognizes that subsidizing the placement of these orders increases market liquidity and will make others more willing to consider trading. On the other hand, if it turned out that sophisticated players were more likely to place such limit orders, the subsidy of such participation might make rational third parties no more willing to play. So it is not clear how well the TradeSports strategy in general might translate to other contexts. Ultimately, reducing commissions might not be sufficient to encourage participation. Ideally, prediction markets should be subsidized. Policy analysis institutions might provide such subsidies if the government does not. Admittedly, subsidizing prediction markets would leave the policy analysis institutions with less control over the output than would directly subsidizing position papers and scholarship. But sometimes such an institution might genuinely be interested in a particular question and thus might be willing to fund a prediction market in the area. At other times, it might be confident that the market prediction will support its position regarding a particular public policy issue. Suppose, for example, that some politicians have called for limits on oil drilling because they want to make sure that oil will remain available in the distant future. It might then be in the economic interest of the oil industry trade association to persuade the public that the amount of oil produced twenty years from now will be relatively high, but observers will discount the claims and studies of oil industry executives. If the trade association truly believes its position, it might fund a prediction market to show that its position in fact is widely shared. The challenge is to devise a means of subsidizing the market that rewards participants who are acting in a way that enhances market accuracy. The danger is that various techniques might be used to obtain a portion of the subsidy at low risk. For example, a sponsor of a prediction market might promise to distribute a fixed amount of money–say, ten thousand dollars–in proportion to the amount individuals win in the market. A problem with this approach is that it might encourage individuals to enter wash transactions. For example, suppose the bid-ask spread is twenty-eight cents to thirty cents. I might simultaneously enter a large volume of buy and sell orders at 29, using separate accounts. Placing aside the subsidy, and in the absence of transactions fees, there would be no economic upside or downside to the transaction, but the subsidy would mean that one of the accounts is likely to make money. Of course, the result is not only that the subsidy might be distributed to parties who have provided no informational benefit but also that many trades would reflect no information and could distort the market. Distributing a subsidy in proportion to winnings is therefore too simplistic an approach. An alternative approach might be to use some type of market maker. In securities markets, a market maker is a trading firm, often called a specialist. A market maker maintains an inventory of securities of a particular type, posting offers to buy and sell the securities. The bid-ask spread provides the market maker with an opportunity to profit by buying a share at a low price and simultaneously selling a share at a high price. A market maker, however, can lose money on some transactions owing to unexpected changes in prices. In some exchanges, one or more market makers will commit to maintaining a bid-ask spread of no more than a certain size, thus promising to provide liquidity. These market makers receive access to the market as part of the bargain. In many exchanges, traders can execute transactions only with the market maker, rather than with one another, as in a continuous double auction. Independent market makers generally seek to maximize their own profit. In a laboratory experiment, Jan Krahnen and Martin Weber have shown that when there is a single market maker, it keeps a wide bid-ask spread and earns some monopolistic rents, which come at the expense of profits for informed traders.7 When there are competing market makers, informed traders can earn considerably greater profits, and the losses of uninformed traders are reduced. This is especially true if the market makers are just as informed as market participants, but often that is not the case in prediction markets. A continuous double auction design can improve the ability of informed traders to profit on their information by allowing them to enter into transactions with uninformed traders. This helps explain why the prediction markets we have considered so far rely on the continuous double auction without market makers. Uninformed traders may have little interest in some prediction markets, however, especially if they are not generally suitable as investment vehicles. To make it possible for informed traders to profit, the sponsor of the market might wish to add a market maker that is willing to lose money to the continuous double auction, thus subsidizing the market. One approach is for the sponsor of a prediction market to develop an automated market maker, which uses a computer algorithm to offer tradable contracts to buy and sell and provide liquidity to the market. The Hollywood Stock Exchange (HSE) uses such an automated market, and it has received a A disadvantage of the HSE approach is that it may be difficult for the sponsor of the prediction market to determine the extent of the subsidy that it is providing the market by adding liquidity. The subsidy might end up being higher or lower than intended, particularly when trading volume cannot easily be anticipated. The problem is particularly severe in prediction markets involving a very small number of traders, because slight differences in trading volume and in the dispersion of information might have large effects on the subsidy provided. The subsidy, for example, might turn out to be relatively low if all traders have approximately the same estimates and relatively high if a single trader has unexpected information that allows that trader to push the tradable contract price a great distance in the correct direction. In Chapter 4 we will consider the market scoring rule, which can function as an automated market maker that limits the maximum potential loss associated with the market, but even with this rule, the subsidy is not fixed. If the market sponsor wishes to offer a fixed, predictable subsidy, it might agree to pay the full amount of the subsidy to an independent firm that agrees to serve as the market maker. It could choose the firm by holding an auction, agreeing to provide the subsidy to the firm that commits to maintaining at all times the shortest bid-ask spread. That firm might in turn use an automated market maker or human specialists, but it would bear the risk. Placing this high level of risk on a single third party might mean, however, that the best offer the exchange receives will not be as generous as it might prefer.
A Decentralized Subsidy Approach
A decentralized strategy for dispersing the risk assumed by market makers would borrow from the TradeSports policy of offering lower commissions to traders who make bid or ask offers that are later accepted. One possibility is to distribute a fixed subsidy to traders who offer the most generous bid and ask prices, in proportion to the time and volume of the traders’ exposure. (Higher subsidies might be offered for exposure at times of day in which trading is more likely to occur.) For example, suppose that the current bid and ask prices are 28 cents and 30 cents, respectively. If I offer to buy up to 100 shares at a price of twenty-nine cents, and if that offer stays at the front of the bid queue for ten minutes, I would receive 100 <mul> 10, or 1,000 credits.9 The same would be true if I offered to sell up to 100 shares at a price of twenty-nine cents, if that offer stayed at the front of the ask queue for the same length of time. If over the duration of the market 1,000,000 credits were awarded, then 1,000 credits would be worth 0.1 percent of the total fixed subsidy amount. This system would provide traders with incentives to make generous bid and ask offers, allowing those with information to profit on that information. In effect, this system improves the incentives for individuals to serve as market makers willing to buy and sell tradable contracts and for these individuals to maintain a small spread between the prices at which they are willing to buy and sell. The system is not easily manipulated. If, in the above example, I simultaneously offered to buy at twenty-nine cents and to sell at twenty-nine cents, those orders would be fulfilled immediately, and so, having exposed myself to no risk, I would receive no credit. This might not be the only subsidy system available, but it should help convert what otherwise would be a zero-sum (or, with commissions, negative-sum) game from the perspective of traders into a positive-sum game, and it therefore should help enhance market participation even if the topic is of intrinsic interest to only a few individuals. Someone who feels relatively confident in a prediction, regardless of the extent to which that prediction is the result of private information, may be able to profit in two ways. First, if the trader’s prediction is below the bid price or above the ask price, the trader can accept the corresponding offer. The subsidy should result in more favorable terms’ being available to that trader than otherwise would exist, thus increasing trading profits. Second, after any such transactions clear, the trader might seek to make more generous offers to buy and sell than those currently in the queue, thus directly earning a portion of the subsidy. This approach provides a reward not only for a trader who finds that a tradable contract is mispriced but potentially also for a trader whose efforts provide additional confidence that a tradable contract is correctly priced. Will subsidies be sufficient to allow a prediction market effectively to perform the task of information aggregation? The answer depends both on the amount of the subsidy and on the nature of the prediction market. A million-dollar subsidy, after all, would attract interest in virtually any topic, and though the number might seem impossibly large, some policy analysts or philanthropists might think such a subsidy would be well spent. Of course, for some issues, no subsidy at all is necessary, and indeed TradeSports shows that prediction markets can flourish with the opposite of subsidy: commissions. A relatively small subsidy, say, one thousand or one hundred dollars, might be enough to accomplish the task of assessment aggregation on an issue regarding which relatively few people have opinions. Many of these individuals ordinarily might not be interested in participating a game in which they might be as likely to lose as to gain money, but they might participate when making money is the more likely outcome for a skilled participant. Subsidies will be particularly helpful if a prediction market might aggregate not only assessments but also evidence. If private information is likely to be useful in generating the predictions, then individuals who do not have such information ordinarily might be unwilling to trade. The possibility of evidence aggregation makes it more difficult to aggregate assessments, because individuals who have assessments based on available evidence will be less likely to trade if others can take advantage of the assessors’ relative lack of knowledge. Where the goal of a prediction market is simply assessment aggregation, that goal will be more difficult to achieve when some participants might hope to profit from evidence aggregation. In general, the greater the degree of private information in a market, the larger the bid-ask spread in that market. Subsidies will tend to narrow the spread between the bid and ask prices, and thus if few trades occur, the bid-ask midpoint can serve as an effective prediction.
The Policy Analysis Market
This chapter has imagined that policy analysis organizations might subsidize prediction markets concerning issues relevant to public policy. Private sponsorship is more likely than government sponsorship, because the sole governmental foray into prediction markets ended in a public relations disaster. The plan, called the Policy Analysis Market (PAM), was to use prediction markets to aggregate expert assessments of military and policy instability. It was funded by the Defense Advanced Research Projects Agency (DARPA), but the military cancelled funding in August 2003 after Senators Ron Wyden and Byron Dorgan generated national press coverage by complaining in a press conference that the military’s plan was to allow gambling on terrorism. The program was originally conceived by Michael Foster of the National Science Foundation, who had been familiar with earlier studies of prediction markets.10 The research sponsored by DARPA concerned many speculative technologies that might have relevance to national defense. It solicited proposals from government contractors to “develop electronic market-based methods and software for decision analysis, to aggregate information and opinions from groups of experts.”11 The solicitation called for proposals that would cover multiple phases. In Phase I, the contractor would select events of interest to the Department of Defense to review and to construct and design an electronic market. In Phase II, the contractor would manage the markets previously developed and analyze their performance. Finally, the department envisioned that the markets developed might have nonmilitary applications, so that in Phase III, the technology could be used “in strategic analysis for business, technology prediction for engineering, and other analyses of decision outcomes.” The research program as a whole was named FutureMAP. The FutureMAP program led to multiple contracts, including one with a company called Net Exchange, which was founded by John Ledyard, an economist at the California Institute of Technology. Net Exchange subcontracted to two professors at Two decisions that might have appeared innocuous at the time contributed to the eventual termination of the FutureMAP program. First, DARPA placed FutureMAP within the Information Awareness Office. This office had already attracted considerable controversy for a plan titled Total Information Awareness that would use a statistical technology called “data mining” to help identify and stop terrorist activity.13 The office was run by Admiral John Poindexter, a Reagan Administration National Security Advisor who had been implicated in the Iran-Contra Affair.14 He was convicted of various charges in connection with that earlier controversy, but the convictions were overturned on the basis of a grant of immunity that Poindexter had earlier received.15 Second, the l PAM Web site included as examples predictions of the assassination of Palestinian leader Yasser Arafat and a missile attack from The two senators who criticized the PAM program, Ron Wyden (D-Ore.) and Byron Dorgan (D-N.D.), described the market incompletely as a market in terrorist attacks. Some of their and others’ criticism reflected concern that the market might be inherently repugnant. The theory might be that the market inappropriately commodified human lives and tragedy. There are many other important institutions, however, that do the same. Life insurance companies, for example, in effect make bets on how long individuals will live. Of course, a life insurer hopes that its clients will live long lives, but society also tolerates sellers of annuities and even viatical companies, which provide cash value for future life insurance proceeds.16 These companies in effect are betting that specific individuals will die, and yet because there are benefits to allowing them (for example, enabling the terminally ill to meet their medical expenses), they are widely tolerated. Presumably, critics of PAM thought the benefits of PAM too small to outweigh any discomfort that it might cause. Senators Wyden and Dorgan also suggested that the funding ($8 million had been requested for future extensions of the markets) would be better spent on more traditional intelligence activities. Other criticisms were more sophisticated. Consider, for example, the critique of Joseph E. Stiglitz, who won the Nobel Prize in Economics in 2001.17 Stiglitz offered reasonable concerns that the market might be manipulated, perhaps providing false comfort or unnecessary alarm, though he did not offer any analysis to defend the claim that manipulation might succeed. More central to Stiglitz’s analysis were the following rhetorical questions: “Did [Poindexter] believe there is widespread information about terrorist activity not currently being either captured or appropriately analyzed by the ‘experts’ in the FBI and the CIA? Did he believe that the 1,000 people ‘selected’ for the new futures program would have this information?” These questions seem to suggest that Stiglitz was unaware that PAM would focus on issues such as economic growth and political instability in addition to terrorism. But the questions also suggest that whatever the context, Stiglitz assumed that assessment aggregation would have relatively little value. Stiglitz noted that markets might not produce reliable information “where there are large asymmetries of information.” He won the Nobel Prize largely for his work on the implications of asymmetric information in markets, and his comment accurately reflects the point made above that private information can lead to very wide bid-ask spreads in prediction markets. Stiglitz’s comments, however, also appear to reflect an assumption that the purpose of PAM was to perform the task of evidence aggregation by allowing individuals with private information about particular terrorist attacks to profit by trading on that information. Robin Hanson has since argued that it might be possible to design markets to predict specific terrorist attacks based on concrete information,18 but PAM clearly was not designed to perform this task. These traders might not have had access to specific information available to the FBI and the CIA, but at least the market would have done a reasonable job of assessment aggregation based on public information. Such assessment aggregation, moreover, might have been useful in this context. Congress and the Department of Defense routinely have to make decisions about resource allocations, and the predictions made by PAM might have been relevant to these decisions. To be sure, government decision makers might rely on official governmental analyses of these questions, but there is always a danger that official analyses will reflect the predictions that the government believes will justify the policies that it prefers. For example, an administration with deep ties to the leadership in
Regulatory Impediments to Prediction Markets
In closing his critique of PAM, Stiglitz asks, “If this is such a good idea, why haven’t the markets created it on their own?”19 The question suggests that those who believe in markets ought to explain why markets have not already created market-based institutions. There might, however, be some straightforward answers. There is no incentive for private firms to create prediction markets for the purpose of selling the data that they produce, because they would be unable to prevent others from obtaining the facts for free. Under Perhaps the most serious obstacle to prediction markets is gambling law. Almost all of the states strictly regulate gambling, and some states have specifically prohibited gambling over the Internet.21 The Federal Wire Act of 1961 also prohibits using “wire communications” to place bets and has been used to prosecute In an analysis of the legality of prediction markets concerning scientific propositions, Tom Bell assesses whether prediction markets would be found to be gambling under the common-law definition.23 Two of the three elements of gambling under the common law–the existence of a prize and of consideration (in this case, payment for the right to the prize)–are present, but presence of the third element, chance, is unclear. Nonetheless, there are strong policy arguments that prediction markets, in particular those dealing with esoteric topics of policy interest, will not generally cause the social ills often attributed to gambling.25 The other obstacle is federal regulation of futures exchanges by the Commodity Futures Trading Commission (CFTC). The CFTC regulates markets such as the Chicago Mercantile Exchange, which allows for trading of securities based on commodities (such as frozen concentrated orange juice) and on economic variables such as interest rates and foreign exchange rates.29 Whether prediction markets are subject to CFTC regulation is a legal question that is not easily resolved. “Commodities” include “all services, rights, and interests in which contracts for future delivery are presently or in the future dealt in.”30 Prediction markets can seek to register with and operate under the authority of the CFTC, and at least one, HedgeStreet, appears to have taken this path.34 But such regulation is perhaps too cumbersome for relatively low-volume prediction markets. In particular, CFTC regulation currently requires clearance for individual contracts to be traded, including demonstrations that the trading will not be susceptible to price manipulation.35 An alternative is to register with the CFTC as an “exempt board of trade.”36 The Trade Exchange Network, which owns TradeSports, has registered via this route,37 but such exempt boards are subject to considerable restrictions; for example, trading is limited to certain types of organizations and individuals with a net worth of more than $10,000,000.38 A representative of TradeSports reports that TradeSports is considering creating markets to allow large entities subject to risk associated with events such as sports contests, for example, sports teams and broadcast networks, to hedge that risk. The only easily accessible real-money legal prediction market in the It is possible, of course, that the CFTC will seek to deregulate prediction markets or find some mechanism for providing a less onerous regulatory regime. Economists from the CFTC have attended at least one academic conference concerning prediction markets and appear to be interested in the topic. Robert Hahn and Paul Tetlock have suggested one regulatory approach: encouraging the CFTC to allow all prediction markets that pass an “economic purpose test.” For example, the creator of a prediction market could show that the particular contract would “provide significant financial hedging opportunities” or that the “prevailing price of the information market contract is likely to provide valuable information for improving economic decisions.”41 In order to overcome concerns about state gambling regulation, they suggest that the CFTC declare state law to be preempted for prediction markets with such contracts. In addition, they recommend broad exemptions from any regulation for markets that (for example) limit the maximum size of individual investments. If not exempt, prediction markets should be permitted to self-certify contracts, thus avoiding the necessity of obtaining CFTC approval before listing each contract. Such an approach would be a good start to legitimating prediction markets, but there is a strong case that they could be expanded further. Focusing on economic purposes rather than on a broader range of public purposes could unnecessarily restrict prediction markets. Election markets, after all, do not clearly fit within the scope of the definition. Conceivably, if “significant hedging opportunities” were interpreted broadly, some election markets might be allowed, because certain companies face risk associated with uncertain presidential elections. But this would cover only a few markets of great significance and presumably would not include, for example, election markets focusing on individual congressional elections. Election markets, of course, are not the only markets of public policy significance that have little economic resonance. The Policy Analysis Market itself would have predicted political instability, in addition to economic variables, and ideally the law would tolerate, if not encourage, such markets. The case for initially restricting legality to economically significant markets is that such markets are the only ones that are likely to be politically feasible. Yet most markets with public policy implications seem likely to incite no more opposition than economic markets. An alternative approach to regulation might be to authorize all prediction markets except those fitting within specific classifications. For example, markets for predicting the outcome of sports contests could be excluded, although they might present hedging opportunities, for example, for a broadcaster that stands to lose advertising revenue if the World Series lasts only four or five games. An exempt board of trade restricted to organizations such as hedge funds and sports teams might sponsor such a market, but because opposition to legalized prediction markets is likely to be based on concerns about gambling, individuals might be precluded from trading. Similarly, prediction markets concerning the entertainment industry (for example, the Oscars) might be excluded, as would those for purely random events (such as the spin of a roulette wheel). Regulations allowing prediction markets also might exclude any markets making predictions about numbers of deaths or terrorist incidents; such markets might well be useful, but the experience with PAM suggests that they are not yet politically feasible.
Legal Markets
As long as prediction markets remain illegal in the United States, U.S. policy analysis organizations’ options are considerably restricted, but there might still be some means of taking advantage of the information aggregation power of prediction markets. One approach is to enter into a contract with an organization abroad that runs prediction markets. I know of no examples of this type of arrangement, but sufficient demand would presumably bring supply. An organization such as TradeSports, after all, can add markets quite easily and presumably would be willing to host a market for a fee. TradeSports might even be willing to implement subsidized markets, provided the market sponsor pays the subsidy and some fee that would at least cover the hosting services that TradeSports provides. This approach reflects a form of regulatory arbitrage, taking advantage of the legality of TradeSports in the jurisdiction in which it exists. As long as the sponsor of a prediction market retains no control and no knowledge of who participates in the market, it seems unlikely that The prediction market sponsor could instead provide money to each of a number of selected participants with the stipulation that this money could be used to bet on a prediction market but that no participant would be allowed to bet any more money than initially provided. For example, a trader who receives one hundred dollars might lose that money but could not lose any more than that. Perhaps in part because of legal restrictions, PAM was to have been structured this way. Conceivably, prosecutors might claim that such an arrangement violates gambling laws, though the impossibility of losing any money from the market as a whole would seem to make such a prosecution unlikely. To improve appearances and decrease the chance of such a prosecution, the prediction market might run on the basis of points, with points convertible into dollars only after the market closes. It seems unlikely that the CFTC would seek to prevent such an arrangement, because the individual traders receiving subsidies would not likely be seen as ordinary investors. A danger in taking this approach to subsidizing a market is that some traders might not participate in evaluation and trading. Instead, they might simply pocket the money provided to them. Sponsors may be tempted to require that funds be traded at least a certain number of times, but there are hazards to this approach. In particular, traders might seek to evade the limitation by executing the same kind of wash transactions that, we saw above, could frustrate a simplistic subsidy design. A single trader presumably would not be allowed to open multiple accounts, but traders might work together to enter wash transactions. For example, one trader might agree with another to buy a number of shares from the other at a price somewhere in the bid-ask spread, with the traders then immediately executing the same set of transactions in reverse. The market sponsor would need to find some way of policing such transactions, as well as more subtle variations. If it were possible to do so, individual traders might respond to the requirements by executing only the minimum number of trades and hedging risk with offsetting transactions over time. An alternative approach might be to use the decentralized subsidy system (described in Chapter 2) to provide incentives for individuals to place their money at risk. For example, suppose that a would-be sponsor of a prediction market would like to invest ten thousand dollars to derive probability estimates from one hundred individuals who seem likely to have information or to be able to obtain information about a particular prediction. Instead of simply giving each of these individuals one hundred dollars that they might trade, the prediction market sponsor might give each trader fifty dollars and then offer five thousand dollars as a subsidy, to be distributed among those who place the money provided to them at risk by placing bid or ask orders. The market sponsor might run a points-only prediction market in which points would not be redeemable for dollars but then distribute the entire ten-thousand-dollar subsidy in proportion to the number of points that participants put at risk. Those who seek a portion of the subsidy by placing points at risk could expect to earn profits, and more knowledgeable participants should expect to earn higher profits than less knowledgeable ones. Although the points in this system have no direct cash value, they are valuable as means of obtaining subsidies, and so market participants will have incentives to be careful about the bid and ask offers that they make. Usually, however, it will be preferable to allow points to be cashed in at some value, in order to provide robust incentives for individuals to accept bid and ask offers when they believe that there has been mispricing. Although allocating a greater portion of a subsidy to encourage trading reduces the risk that individuals will simply pocket the trading funds, there is a significant drawback if too large a portion of the subsidy is allocated in this way. In a market in which points initially allocated are not redeemable for dollars or are redeemable for only a small portion of the subsidy, someone who does have a better probability estimate might not be able to invest any more than individuals with worse probability estimates. Such a market design is likely to be successful only if the appropriate probability estimate will tend to become more obvious to all participants over time. In that case, an individual who makes a sensible trade will earn points while the market is still operating and will thus have more points to place at risk in order to obtain a higher portion of the subsidy. If predictions are not likely to improve markedly over the life of the market, then someone who has good information will not accumulate points much faster than others and therefore will not receive a particularly large portion of the subsidy. The remedy is to allocate a meaningful portion of the subsidy to initial funds, so that trading can directly lead to profits. This analysis suggests that careful design can improve traders’ incentives in a market in which no participant can lose money. That does not mean that such markets will be just as good as subsidized markets in which traders can put their own money at risk. The markets will be particularly poor at encouraging evidence aggregation. Someone who has a specific piece of information indicating that the market price is off by a significant amount will be able to push the market in the right direction using only whatever dollars or points are initially assigned to the trader’s account. Even if the market succeeds in encouraging traders to place large amounts of money at risk in the bid and ask queues, the increased liquidity will not greatly increase the amount of money that someone with information will be able to earn by trading. Suppose that the total subsidy to a market with one hundred participants is ten thousand dollars. It will be difficult for any participant to earn a significant portion of that amount, and the market will provide only very limited incentives for individuals to conduct complicated and time-consuming analysis. If the goal is simply to obtain a crude form of assessment aggregation, however, the market is likely to be more successful. In this case, though, the market will not weigh individual traders’ confidence in their predictions as effectively as a market in which traders can invest their own funds. One approach to making markets in which participants cannot lose more like real prediction markets is to provide participants with accounts that can be used in each of a variety of prediction markets. Suppose, for example, that a policy analysis organization is interested in sponsoring ten different prediction markets in related topics and that it has identified a group of individuals that it wishes to encourage to trade in these markets. It could provide each individual with ten separate accounts, each assigned to a separate market. But pooling the accounts gives each trader an incentive to decide which market or markets to focus on. A trader, for example, who has private information relevant to only one of the markets might invest all of the funds allotted to trading in that market. Other traders will tend to invest in the markets in which they have information. Thus, each trader is likely to move prices more in markets in which the trader might have information, providing some assurance that market prices will be weighted by the self-confidence each trader has for that market.
The Foresight Exchange
An alternative means of complying with legal restrictions on prediction markets is to sponsor a play-money market. As discussed in Chapter 1, some such markets, such as the Hollywood Stock Exchange, have shown remarkable predictive capacity. The HSE does not provide any cash subsidy, although market participants can cash in some of their play money for discounts at the HSE’s on-line store. (For example, a mere ten million dollars in HSE currency lowers the price of a T-shirt from twelve dollars in real currency to five dollars, perhaps still enough for the Exchange at least to break even on such sales.) It is difficult, however, to design prediction markets with substantial subsidies available to anyone who signs up for free on the Internet, because individuals might open accounts simply to take high-risk gambles. Given such possibilities for manipulation, absent alternative designs that are responsive to this problem, a market using only play money might produce as much accuracy or perhaps more than a market in which play money is convertible to real money. People may be less likely to try to engage in market manipulation when they are playing only for points than when they are playing for dollars. This is not necessarily true, however, as anyone who has observed cheating in a pick-up basketball game or a game of Monopoly knows. The theory behind a play-money game is that people will seek to earn the recognition associated with play money, and so if cheating is undetectable, people might cheat to obtain such recognition. Participants in multiplayer on-line games often genuinely care about their scores, so much so that there is a robust market in which people exchange their virtual earnings for real cash.42 Prediction markets, however, seem likely to be less addictive than graphically intense on-line games that allow for greater degrees of virtual interaction, and so play-money markets without any subsidy whatsoever might generate at least roughly reliable predictions. Indeed, someone interested in creating a play-money market need not host a server running that market, for an existing Web site has hosted play-money prediction markets. The Web site, called the Foresight Exchange,43 initially provides each participant with fifty dollars in virtual money plus an additional fifty for each week of continued participation. Participants can pay some of the money that they have earned–or, more accurately, been given–to sponsor new claims. A new claim can in effect be a probability estimate or a numeric estimate prediction market. On the Foresight Exchange, the latter are called “scaled claims,” paying off somewhere between nothing and one dollar according to a formula decided by the market creator. Both probabilistic and scaled claims have an expiration date, and a judge is assigned to each claim. The judge reports in advance on how he or she plans to address any apparent ambiguities in the claim and resolves the claim as soon as it is possible to do so. On casual analysis, the Foresight Exchange generally seems to do a reasonably good job of assessment aggregation. In mid-January 2006, for example, the market predicted a 42 percent chance that the Of course, this analysis emphasizes that objectively ascertaining market accuracy is difficult, given that individuals might have different views of the appropriate probability for a particular tradable contract. Looking at predictions in closed markets might not resolve the mystery, given the absence of an alternative set of predictions to which to compare the Foresight Exchange’s predictions. It is troubling, however, that some of the market’s predictions seemed to be at variance with those of real-money prediction markets. For example, also as of mid-January 2006, the Foresight Exchange predicted a 63 percent chance that a Democrat would be elected president in 2008, but the corresponding prediction in the TradeSports market was 48 percent. This suggests either that prediction markets generally might be wildly inconsistent or (more plausibly, in my view) that prediction markets that depend on points alone leave considerable room for mispricing that no one bothers to correct. The Foresight Exchange includes claims in a wide variety of topics. For example, it has a math category with a contract about whether the Poincaré Conjecture will be proved by 2031 (a 97 percent chance), a medicine category including a contract about whether cancer mortality will by 2010 have fallen 90 percent relative to 1994 (a 17 percent chance), and a physics category including a contract concerning whether a power plant will sell nuclear energy produced by fusion by 2045 (a 73 percent chance). Whatever the plausibility of these contract estimates, they cover relatively important issues in the relevant fields. It seems likely that the market would be less accurate if contracts concerned much more esoteric issues, because only a small number of people, who might have very little direct knowledge, would be interested in trading on those issues. Thus, although the Foresight Exchange provides an interesting example of working prediction markets in issues that in some cases have public policy relevance, it seems unlikely that the Exchange would perform well if taxed with much more specific claims. If the Foresight Exchange’s performance is disappointing, it might not require a large subsidy to improve it considerably. Many transactions in the Exchange appear to be due to a relatively small number of active traders, some of whom have amassed substantial virtual wealth (more than thirteen thousand dollars in one case). Some of these traders, however, appear to invest in only a few tradable contracts that are wildly mispriced, and so a subsidy scheme that encouraged these traders to reduce bid-ask spreads might improve market accuracy. A scheme rewarding provision of liquidity might help the market even if points were not redeemable for dollars. Adding a direct subsidy might give a small core of generalist traders an adequate incentive to push the market toward reasonable assessment aggregation. The more difficult question is how to provide such a subsidy in a way that minimizes the danger of manipulation from people who open free accounts and make either random trades or nonrandom trades designed to benefit other participants. An imperfect solution is to reward one or more of the top participants in a market or in a group of markets. Such a system, however, provides little incentive for someone unlikely to be among these top participants to take part. Nonetheless, in some contexts, it might be sufficient. NewsFutures, for example, has used this approach with success. According to Emile Servan-Schreiber, by the end of trading in a market in which ten prizes might be awarded, about one hundred people might be competing for these prizes. Servan-Schreiber also reports that massive fraud attempts can generally be detected.46 There is a danger, however, that as the value of prizes goes up, attempts to mask these schemes could become more sophisticated. In general, a broader prize base (in the extreme, a system that cashes out all points on a per-point basis) encourages more participation but also more gaming of the system.
Aggregation of Secret Information
In all of these cases, the information that is relevant to the task is largely public, so the markets would be performing assessment aggregation. Some contexts, however, might lend themselves to sophisticated modeling more than would others, and to the extent that the predictions of sophisticated models count as private information, these markets might perform evidence aggregation as well. It might be possible to create markets specifically designed to aggregate evidence, in particular, evidence that otherwise would be secret. Prediction markets, for example, might be used to make forecasts that ordinarily would need to be based on information largely unavailable to the public. For such markets to have a chance of reflecting the secret information, individuals with access to that information will need sufficient incentive to participate. Recall the markets that were used to predict whether weapons of mass destruction would be found in Perhaps a heavily subsidized prediction market would have done so. In this context, however, the subsidy would have to be quite high. After all, the secret information was top secret in the legal sense, and individuals with access to such information might have feared the possibility of prosecution for revealing information if they traded on it. An interesting legal question would be whether trading on top-secret information violates laws against spying; more likely, it violates government ethics rules.47 It could be argued, however, that it ought to be legal. In this case, at least, there was no governmental interest in keeping secret the officials’ true estimates of the probability that weapons of mass destruction would be found. Indeed, top governmental officials were working hard to convince the public that they really did perceive a threat. Conceivably, had the officials with access to the information been able to participate in a prediction market, the market might have established that the officials genuinely believed this. Or it might have shown that the claimed confidence in fact was mere bluster. Prediction markets could have served as a kind of lie detector that either proved the government’s good faith or revealed a deception. Either way, by providing objective forecasts, prediction markets could improve democratic governance, but only if officials with the relevant information are allowed to trade. Given the aftermath of PAM, it is likely to be a long time before the government sponsors prediction markets. But it could do much to spur the development of markets by tolerating governmental officials’ trading on such markets, at least where public release of the prediction itself could not jeopardize national security. Admittedly, just as presidents and other officials seek to suppress leaks to the press, so, too, might they seek to suppress or otherwise discourage prediction market trading. But if government policy toward trading remains ambiguous, some government officials might invest anonymously in subsidized markets. This provides an opportunity for policy analysis organizations to sponsor subsidized prediction markets if they believe that the public is being misled. Prediction markets that encourage individuals to trade on secret information can, of course, be used for good or ill, and sometimes one might argue about which is which. Historically there has been an intense debate about the degree to which government should be open. A free press keeps the government more open than it might like, and a privately sponsored prediction market could be used as another weapon by those who seek more openness. Prediction markets also might be used to obtain information about corporate plans. Companies often have had difficulty keeping some of their plans secret; Apple Computer is a particularly prominent example of a company that has tried to suppress information release.48 At least those companies know that there are a sufficient number of false rumors to give the public trouble identifying true ones. Privately sponsored prediction markets in corporate plans, however, might make it easier to distinguish true rumors from false ones. Policy analysis organizations might sponsor some markets in order to improve their monitoring of corporations. For example, rumors that Wal-Mart was exploring the possibility of investing in robotic research so that it could replace human labor did not receive much traction after a heated denial by the company.49 It might have, however, if prediction markets had suggested that Wal-Mart in fact would deploy robots within the next twenty years. Independent organizations also might sponsor prediction markets that would track the effects of corporate activity. For example, the Recording Industry Association of America might sponsor a market that would predict the number of copyrighted songs that will be downloaded over some future time span on a particular file-sharing service, thus making clear that particular designs were viewed as likely to foster copyright violations. Conceivably, companies might also sponsor prediction markets to help reveal their competitors’ information and plans. Courts in the future might need to decide whether such actions should be seen as violations of laws governing trade secrets. Prediction markets sponsored by third parties to find information about an organization might not be as successful as those sponsored by the organization itself. One reason, of course, is that loyal employees might worry that they are betraying their employer if they make personal trades based on information obtained at the job or might even worry about legal liability–for example, for insider trading.50 A less obvious reason is that individuals with information will ordinarily have liquidity constraints and may not push prediction market prices as far in the appropriate direction as their information would justify. If other traders knew that the trades were coming from individuals who might have access to relevant information, then those traders might change their evaluations. But in a large market that does not identify traders, such derivatively informed trading will not occur. As this analysis suggests, if a sponsor of a prediction market wants to encourage such trading, it might sponsor two markets, one for people who might be expected to have private information (perhaps receiving a higher subsidy) and one for anyone. Presumably, trading on the private market would have significant ramifications for the public one. This strategy, however, might be difficult to accomplish without the cooperation of the organization in which the private information lies. Of course, policy analysis organizations will only have an interest in subsidizing such prediction markets if in fact they aggregate evidence well enough that the media, government, and public grow to have confidence in them. But public acceptance of prediction markets is unlikely to materialize in the absence of markets that inspire confidence. A particular danger is that unsubsidized prediction markets might not aggregate evidence very effectively, and the public will grow to distrust the markets before the market designs most likely to spur success have been tried. Chicken-and-egg problems, however, have a way of solving themselves eventually. In the short term, it is likely that only a few policy analysis organizations will sponsor prediction markets, and these markets will probably be unsubsidized. At some point, though, some organizations might try offering modest subsidies, and if these subsidies prove successful, prediction markets that provide strong incentives for individuals with information to trade might develop.