The Millennium Project Futures Research Methodology—V3.0
PREDICTION MARKETS
From papers by Justin Wolfers and Eric Zitzewitz1
with notes and additional references by Theodore Gordon
I. History of the Method
II. Description of the Method
III. How to Do It
IV. Strengths and Weaknesses of the Method
V. Use in Combination with Other Methods
VI Frontiers of the Method
Appendix: Samples of Applications
Bibliography
1 Justin Wolfers is Associate Professor of Business and Public Policy, The Wharton School, University of Pennsylvania; Eric Zitzewitz is Associate Professor of Economics at Dartmouth College. Professors Wolfers and Zitzewitz kindly allowed the Millennium Project to prepare this chapter using their previously published work; the editor arranged the material in the format of this CD ROM, and added several other references and comments along the way. The comments are identified by the parenthetical initials (TG). The final product was reviewed by Professors Wolfers and Zitzewitz. The Millennium Project Futures Research Methodology—V3.0
Acknowledgments
This chapter has benefitted from helpful comments and insightful remarks provided by peer reviewers Charles M. Perrottet, Principal, The Futures Strategy Group LLC, USA; Peter Bishop, director, Program for the Study of the Future, University of Houston; Theodore Gordon, Senior Fellow of The Millennium Project; and Jerome Glenn, Director, The Millennium Project. Elizabeth Florescu and Kawthar Nakayima provided excellent project support and John Young proofread. Thanks to all for your contributions.
Prediction Markets ii The Millennium Project Futures Research Methodology—V3.0
I. HISTORY OF THE METHOD
Prediction Markets, sometimes referred to as ―information markets,‖ ―idea futures‖ or ―event futures‖, are markets where participants trade contracts whose payoffs are tied to a future event, thereby yielding prices that can be interpreted as market aggregated forecasts. For instance, in the Iowa Electronic Market, traders buy and sell contracts that pay $1 if a given candidate wins the election. If a prediction market is efficient, then the prices of these contracts aggregate dispersed information about the probability of each candidate being elected. Markets designed specifically around this information aggregation and revelation motive are the focus of this chapter.2
The most famous prediction markets are the election forecasting markets run by the University of Iowa (Berg, Forsythe, Nelson and Rietz, 2001), but prediction markets have also been used to forecast movie revenues, corporate sales, project completion, economic indicators and Saddam Hussein’s demise. New corporate applications have emerged as firms have looked to markets to predict research and development outcomes, success of new products, and regulatory outcomes. In the public sector, the Pentagon attempted to use markets designed to predict geopolitical risks, although negative publicity stopped the project (Hanson, 2005). An intriguing attempt to apply prediction markets to forecasting influenza outbreaks is detailed in Nelson, Polgreen and Neumann (2006). Rhode and Strumpf (2004) have detailed the existence of large-scale election betting as far back as the election of Washington.3
Not only have prediction markets been used in numerous fields, but the market mechanisms have varied as well; including continuous double auctions (both with and without market-makers), pari-mutuel pools, bookmaker-mediated betting markets, and implementations as market-scoring rules.
In July 2003, press reports began to surface of a project within the Defense Advanced Research Projects Agency (DARPA), a research think tank within the Department of Defense, to establish a Policy Analysis Market that would allow trading in various forms of geopolitical risk.4 Proposed contracts were based on indices of economic health, civil stability, military disposition, conflict indicators, and potentially even specific events. For example, contracts might have been based on questions like ―How fast will the non-oil output of Egypt grow next year?‖ or ―Will the U.S. military withdraw from country A in two years or less?‖ Moreover, the exchange would have offered combinations of contracts, perhaps combining an economic event and a political event. The concept was to discover whether trading in such contracts could help to predict future events and how connections between events were perceived. However, a political uproar
2 The material in this section is from: Justin Wolfers and Eric Zitzewitz, ―Prediction Markets in Theory and Practice, NBER Working Paper No. 12083, March 2006 3 In fact Rhode and Strumpf (2004) provide references to historical accounts of early uses of betting (prediction markets) in elections as dating back to the Italian City States in the 1500-1700’s and in the anticipation of the succession of Catholic popes. In Great Britain, it can be traced back to the 18th century. (TG) 4 This material from Wolfers, Justin and Eric Zitzewitz, Prediction Markets, Journal of Economic Perspectives— Volume 18, Number 2—Spring 2004—Pages 107–126
Prediction Markets 1 The Millennium Project Futures Research Methodology—V3.0 followed. Critics savaged DARPA for proposing ―terrorism futures,‖ and rather than spend political capital defending a tiny program, the proposal was dropped.5
Ironically, the aftermath of the DARPA controversy provided a vivid illustration of the power of markets to provide information about probabilities of future events. An offshore betting exchange, Tradesports.com, listed a new security that would pay $100 if the head of DARPA, Admiral John Poindexter, was ousted by the end of August 2003. Early trading suggested a likelihood of resignation by the end of August of 40 percent, and price fluctuations reflected ongoing news developments. Around lunchtime on July 31, reports started citing credible Pentagon insiders who claimed knowledge of an impending resignation. Within minutes of this news first surfacing (and hours before it became widely known), the price spiked to around 80. These reports left the date of Poindexter’s proposed departure uncertain, which explains the remaining risk. As August dragged on, the price slowly fell back toward 50. On August 12, Poindexter then issued a letter of resignation suggesting that he would resign on August 29. On the 12th, the market rose sharply, closing at a price of 96.
This anecdote illustrates a market where participants trade in contracts whose payoff depends on unknown future events.
II. DESCRIPTION OF THE METHOD
Much of the enthusiasm for prediction markets derives from the efficient markets hypothesis. In a truly efficient prediction market, the market price will be the best predictor of the event, and no combination of available polls or other information can be used to improve on the market- generated forecasts. This statement does not require that all individuals in a market be rational, as long as the marginal trade in the market is motivated by rational traders. Of course, it is unlikely that prediction markets are literally efficient, but a number of successes in these markets, both with regard to public events like presidential elections and within firms, have generated substantial interest.
Although markets designed specifically for information aggregation and revelation are the focus of this chapter, the line between these kinds of prediction markets and the full range of contingent commodities—from owning stock in your employer’s company to betting on the Super Bowl—can become blurry. Most contingent commodity markets involve some mix of risk sharing, fun, and information transmission, so these distinctions are not impermeable.
The design of how the payoff is linked to the future event can elicit the market’s expectations of a range of different parameters. Practitioners speak as though the market is itself a representative ―person‖ with a set of expectations. However, the reader should be warned that there are important but subtle differences between, say, the market’s median expectation and the median expectation of market participants.
5 Looney (2003) provides a useful summary of both the relevant proposal and its aftermath. Further, Robin Hanson has maintained a useful archive of related news stories and government documents at http://hanson.gmu.edu/policyanalysismarket.html.
Prediction Markets 2 The Millennium Project Futures Research Methodology—V3.0
Table 1 below summarizes some of the on-line prediction markets, and in the aggregate describes the range of applications being made today:
Table 1. Prediction Markets Typical Turnover on an Market Focus Event Small-scale election markets. A Iowa Electronic Markets Tens of thousands of dollars similar market is run by http://www.biz.uiowa.edu/iem/ (Traders limited to $500 UBC (Canada) Run by University of Iowa positions.) http://esm.ubc.ca/CA08/
Trade in a rich set of political InTrade futures, financial contracts, Hundreds of thousands www.intrade.com current events, sports and of dollars For-profit company entertainment. Economic Derivatives Large-scale financial market www.economicderivatives.com trading in the likely outcome Hundreds of millions Run by Goldman Sachs and of future economic data Deutsche Bank releases Virtual currency redeemable Political, finance, current for monthly prizes (such as a NewsFutures events and sports markets. television set). Also player www.newsfutures.com Also can use real money and For-profit company technology and pharmaceutical designate winnings to go to a futures for specific clients. charity. Political, finance, current Foresight Exchange events, www.ideosphere.com Virtual currency science and technology events Nonprofit research group suggested by clients Success of movies, movie stars, Hollywood Stock Exchange awards, including a related set www.hsx.com Virtual currency of complex derivatives and Owned by Cantor Fitzgerald futures. Data used for market research HubDub, The World’s News Popular prospective news Forecaster events in sports, politics, Virtual currency; game-like http://www.hubdub.com/ entertainment, science, competition for winners technology, etc.
Prediction Markets 3 The Millennium Project Futures Research Methodology—V3.0
III. HOW TO DO IT
Betting has probably been part of society’s way of dealing with uncertain futures since money and history began; prediction markets simply organize these activities into precise and generally interesting questions about future uncertainties.
There are three main types of contracts. First, in a ―winner-take-all‖ contract, the contract costs some amount $p and pays off, say, $1 if and only if a specific event occurs, like a particular candidate winning an election. The price on a winner-take-all market represents the market’s expectation of the probability that an event will occur (assuming risk neutrality).6
Second, in an ―index‖ contract, the amount that the contract pays varies in a continuous way based on a number that rises or falls, like the percentage of the vote received by a candidate. The price for such a contract represents the mean value that the market assigns to the outcome.
Finally, in ―spread‖ betting, traders differentiate themselves by bidding on the cutoff that determines whether an event occurs, like whether a candidate receives more than a certain percentage of the popular vote. Another example of spread betting is point-spread betting in football, where the bet is either that one team will win by at least a certain number of points or not. In spread betting, the price of the bet is fixed, but the size of the spread can adjust. When spread betting is combined with an even-money bet (that is, winners double their money while losers receive zero), the outcome can yield the market’s expectation of the median outcome, because this is a fair bet only if a payoff is as likely to occur as not. The basic forms of these relevant contracts will reveal the market’s expectation of a specific parameter: a probability, mean, or median, respectively.
Table 2, adapted from Wolfers and Zitzewitz (2004a), shows how different contracts can be designed to reveal various types of forecasts.
6 The price of a winner-take-all security is essentially a state price, which will equal an estimate of the event’s probability under the assumption of risk neutrality. The sums wagered in prediction markets are typically small enough that assuming that investors are not averse to the idiosyncratic risk involved, seems reasonable. However, if the event in question is correlated with investors’ marginal utility of wealth, then probabilities and state prices can differ. In what follows, we leave this issue aside and use the term probability to refer to risk-neutral probability.
Prediction Markets 4 The Millennium Project Futures Research Methodology—V3.0
Table 2: Contract Types: Estimating Uncertain Quantities or Probabilities Reveals Market More General Contract Details Examples Expectation of Application Defining many Contract costs $p; Event x: A Probability that events, x1, x2, Binary Pays $1 if and nominee wins the event x occurs, ..,xn reveals Option only if event x popular vote p(x). probability occurs. distribution F(x). Contract pays $1 Contract pays for every some function of Index percentage point Mean value of Contract pays $x. x: $g(x). Reveals Futures of the popular vote outcome x: E[x] specific moments, won by George E[g(x)] Bush Contract costs $1 Contract pays $1 contract pays Pays $2 if x>x* even money if $(1/q) if x>x*. Spread Median value of Pays $0 otherwise. Bush wins more Reveals specific Betting outcome, x. Bid according to than x*% of the quantile, F1-q(x). the value of x*. popular vote
Consider a market with two index contracts, one that pays in a standard linear form and another that pays according to the square of the index, y2. Market prices will reveal the market’s expectation of E[ y2] and E[ y], which can be used to make an inference about the market’s beliefs regarding the standard deviation of E[y], more commonly known as the standard error. (Recall that the standard deviation can be expressed as the square root of the mean of the squares less the square of the means.) By the same logic, adding even more complicated index contracts can yield insight into higher-order moments of the distribution.
In addition, prediction markets can also be used to evaluate uncertainty about these expectations. For instance, consider a family of winner-take-all contracts that pay off, if and only if the candidate earns 48 percent of the vote, 49 percent, 50 percent and so on. This family of winner- take-all contracts will then reveal almost the entire probability distribution of the market’s expectations. A family of spread betting contracts can yield similar insights. An even-money bet in a spread contract will define the median, as explained above, but,. for similar reasons, a contract that costs $4 and pays $5 if y achieves the value of y* will elicit a value of y* that the market believes to be a four-fifths probability, thus identifying the 80th percentile of the distribution.
There exist a variety of off-the-shelf and custom software for implementing prediction markets, (including open source applications) and consulting assistance provided by for-profit companies, such as: Inkling: http://inklingmarkets.com//index.html Newsfutures: http://us.newsfutures.com/home/home.html Crowdcast: http://www.crowdcast.com/
Prediction Markets 5 The Millennium Project Futures Research Methodology—V3.0
Mercury: http://www.mercury-rac.com/ Spigit: http://www.spigit.com/products/p_index.html
Open source software may be found at: Zocalo: http://zocalo.sourceforge.net/ Open Prediction Markets: http://openpredictionmarkets.org/
Prediction Markets 6 The Millennium Project Futures Research Methodology—V3.0
IV. STRENGTHS AND WEAKNESSES OF THE METHOD
Accuracy of Prediction Markets
Arguably the most important issue with these markets is their performance as predictive tools. In the political domain, Berg, Forsythe, Nelson and Reitz (2001) summarize the evidence from the Iowa Electronic Markets, documenting that the market has both yielded very accurate predictions and also outperformed large-scale polling organizations. In the week leading up to the election, these markets have predicted vote shares for the Democratic and Republican candidates with an average absolute error of around 1.5 percentage points. By comparison, over the same four elections, the final Gallup poll yielded forecasts that erred by 2.1 percentage points.
Figure 1 shows that predictions typically become more accurate as the event draws nearer, pooling data from various political markets run by the Iowa Electronic Markets. The horizontal axis shows the number of days before the election. The vertical axis measures the average absolute deviation between the prices of index contracts linked to the two-party shares of the popular vote for each party and actual vote shares earned in the election. The graph also shows how the accuracy of the market prediction improves as information is revealed and absorbed as the election draws closer.
Figure 1. Information Revelation through Time
That said, comparing the performance of markets with a mechanistic application of poll-based forecasting may not provide a particularly compelling comparison. A more relevant starting point might be to compare the predictions of markets with those of independent analysts. For an example along these lines, consider the ―Saddam Security,‖ which was a contract offered on TradeSports paying $100 if Saddam Hussein were ousted from power by the end of June 2003.
Prediction Markets 7 The Millennium Project Futures Research Methodology—V3.0
The price of this contract moved in lock step with two other measures: expert opinion as shown by an expert journalist’s estimate of the probability of the United States going to war with Iraq; and oil prices.
In a corporate context, the Hollywood Stock Exchange predicts opening weekend box office success, and these predictions have been quite accurate also. Further, this market has been about as accurate at forecasting Oscar winners as an expert panel (Pennock, Lawrence, Giles and Nielsen, 2001).
Some firms have also begun to experiment with internal prediction markets. An internal market at Hewlett-Packard produced more accurate forecasts of printer sales than the firm’s internal processes (Chen and Plott, 2002). Ortner (1998) described an experiment at Siemens in which an internal market predicted that the firm would definitely fail to deliver on a software project on time, even when traditional planning tools suggested that the deadline could be met. While the Hollywood markets have drawn many participants simply on the basis of their entertainment value, the HP and Siemens experiences suggested that motivating employees to trade was a major challenge. In each case, the firms ran real money exchanges, with only a relatively small trading population (20–60 people), and subsidized participation in the market, by either endowing traders with a portfolio or matching initial deposits. The predictive performance of even these very thin markets was quite striking.
Predicting Small Probability Events
Prediction markets do seem to display some of the deviations from perfect rationality that appear in other financial markets. There is substantial evidence from psychology and economics suggesting that people tend to overvalue small probabilities and undervalue near certainties. For example, there is a well-known ―favorite–long shot bias‖ in horse races in which bettors tend to overvalue extreme long shots and thus receive much lower returns for such bets, an effect that is offset by somewhat higher (albeit still negative) returns for betting on favorites. The ―volatility smile‖ in options refers to a related pattern in financial markets (Bates, 1991; Rubenstein, 1994), which involves overpricing of strongly out-of-the-money options and under-pricing of strongly in-the-money options (relative to their future values or their ex-ante values from the Black- Scholes option pricing formula). These experiences suggest that prediction markets may perform poorly at predicting small probability events.
Desire-Based Trading
Another behavioral bias reflects the tendency of market participants to trade according to their desires, rather than objective probability assessments. Strumpf (2004) provides evidence that certain New York gamblers are more likely to bet the Yankees, while Forsythe, Reitz and Ross (1999) provide evidence that individual traders buy and sell in political markets in a manner correlated with their party identification. Even so, as long as marginal trades are motivated by profits rather than partisanship, prices will reflect the assessments of (unbiased) profit motive. Thus far, there is little evidence that these factors yield systematic unexploited profits.
Prediction Markets 8 The Millennium Project Futures Research Methodology—V3.0
Speculative Bubbles
A further possible limitation of prediction market pricing arises if speculative bubbles drive prices away from likely outcomes. Traditional markets may be subject to bubbles because of constraints on short selling and because investors will be reluctant to commit a large share of their wealth to an arbitrage opportunity, since if the mispricing does exist, it may grow worse before it gets better (Shleifer and Vishny, 1997). Since prediction markets typically impose no restrictions on short selling, and the markets are sufficiently small-scale that it is unlikely that informed investors will be capital-constrained, the scope for bubbles might be more limited. It is impossible to make any serious attempt at describing the frequency of bubbles in the data we have so far. However, through September 2003, we suspected a bubble in the TradeSports security on whether Hillary Clinton would win the Democratic nomination. Our suspicions were based on her public statements that she was not a candidate and the tenor of discussion among traders, which seemed to indicate that trading was being driven by expectations of future price movements rather than by fundamentals. Equally, these high prices may have reflected those with access to campaign insiders who knew more about her state of mind than we did.
Empirically, the best that we can say is that the performance of past markets at predicting the future has been, on average, pretty good, whether or not specific markets were in some cases distorted by biases or bubbles. Laboratory experiments hold out the possibility of learning more about bubbles, as it is possible for the experimenter to know the ―true price‖ and, hence, to observe deviations. Plott and Sunder (1982, 1988) have set up extremely simple examples in which bubble-like behavior occurs in simple prediction markets. At the same time, bubbles in experimental markets often burst and give way to more rational pricing.
(TG) Berg and Rietz (2006) have studied the Iowa Electronic Market and have listed some additional factors to consider such as:
Traders are not a random sample of the electorate.
Some traders are occasionally irrational, sometimes leaving small amounts of money on the table.
Some traders are robots and while only a dozen or so have been identified, the volume of their trades was high
Large orders to trade can move prices
The strengths of prediction markets depend in large part on developing answers to five open questions:
1. How to attract uninformed traders? While Delphi studies require experts, prediction markets need traders to provide uninformed order flow in order to function. If only rational traders (those who have expected returns as their motivation) participate, the No Trade Theorem
Prediction Markets 9 The Millennium Project Futures Research Methodology—V3.0 binds and the market unravels (Milgrom and Stokey, 1982). Uninformed order flow can have a variety of motivations (entertainment, overconfidence, hedging), but with the exception of hedging, these motivations are usually non-economic, putting economists at a comparative disadvantage in predicting which markets will succeed.
2. How to tradeoff interest and contractability? A fundamental problem in the design of prediction markets is that the outcomes of interest are often impossible to write into contracts. Running financial markets on policy outcomes in and of itself does nothing to help with this problem. In fact, using a measure to set payoffs for financial contracts can turn what would otherwise be a good proxy for a policy outcome into a problematic one.
3. How to limit manipulation? In addition to manipulation of the outcomes on which prediction markets are based, one might worry about manipulation of prediction market prices themselves, particularly where high-stakes decisions are made based on the prices. Clearly there may be conflicts of interest (players can benefit from the implementation of policies based on the outcomes of prediction markets) that would lead some participants to distort their inputs in the hopes of later real world gains.
4. Are markets well calibrated on small probabilities? Many of the proposed uses of prediction markets will involve the evaluation of small probability events. A range of behavioral evidence suggests that people are quite poor at distinguishing small probabilities from tiny probabilities, and even when arbitrage is possible, frictions can cause this miscalibration to carry over into market prices.
5. How to separate correlation from causation? Contingent prediction markets allow one to estimate the probability of an event contingent on another event occurring. Thus contingent markets provide insight into the correlation of events. However, determining whether one event causes the probability of another to change is a separate and potentially more important question. Many of the proposed uses of decision markets presuppose that the direction of causality can be readily established.
Prediction Markets 10 The Millennium Project Futures Research Methodology—V3.0
V. USE IN COMBINATION WITH OTHER METHODS
Some of the lessons gained from prediction markets might be applied to other futures techniques. The reward of ―getting it right‖ is probably a powerful feedback mechanism that leads to deeper thought in providing an input to, say, a Delphi study. In prediction markets these rewards are couched in terms of ―pay-offs‖; in Real-Time Delphi, respondents are offered at least the psychic reward of being able to compare their responses to the group’s. What other reward feedback might be employed to bolster the incentive to ―get it right?‖ Prediction markets have used prizes of various sorts, including gifts to the charity of the winners’ choice. Most prediction markets use ―play money‖ to avoid conflict with local gambling laws, but there is a belief among those who construct such markets that real money rewards might provide a stronger incentive.7 Imagine adding a prize to Delphi studies which is awarded to the participant who comes closest to the correct answers.
One attractive possibility is using prediction markets to judge probabilities of possible future events considered for inclusion in a scenario set or as important to consider in the analysis of global challenges. The question might be asked, for example, ―Do you think this event would appear in a future scenario dealing with extreme revisions to the global financial system (or other topic or the moment)?‖
8 VI. FRONTIERS OF THE METHOD
(TG) An intriguing possibility arises: to compare the principal forecasting methods on their ability to accurately forecast short-term developments. The techniques that might be included in this face-off could be: a multi-round Delphi, Real Time Delphi, a popular statistically-based opinion poll, prediction markets, and if appropriate, a quantitative model. The topics might be events with an expected time horizon of six months to a year after which the methods could be compared on the basis of several criteria including: accuracy, cost, time required by participants, ability to furnish decision-relevant insights, etc.
(TG) Most expert-based studies ask direct questions, such as ―What is the probability that Bin Laden will be captured during Obama’s first term?‖ The question could be easily changed to: ―If you had $100 to bet on the outcome how much would you bet that Bin Laden would be captured during Obama’s first term if the odds were 3:1 against capture?‖ The structure could be enriched further if the participants faced a dozen such questions and had only the $100 to allocate; this could induce them to bet on the outcome of the questions about which they felt strongest.
7 See Hibbert, 2005, op.cit. 8 From: Wolfers, Justin, and Eric Zitzewitz,‖ Five Open Questions About Prediction Markets,‖ Working Paper 12060, http://www.nber.org/papers/w12060, National Bureau Of Economic Research, Cambridge, MA, February 2006; also Wolfers, Justin and Eric Zitzewitz, 2006. ―Five Open Questions About Prediction Markets,‖ in Information Markets: A New Way of Making Decisions in the Public and Private Sectors, ed. Robert Hahn and Paul Tetlock, AEI-Brookings Joint Center, Washington D.C.
Prediction Markets 11 The Millennium Project Futures Research Methodology—V3.0
(TG) Prediction markets produce time series that indicate the changing perceptions of the players about, for example, the probability of a future event. Futures studies, by contrast, almost always yield a simple snapshot of the probability of future events. The use of a time domain is quite useful since it shows how current information influences the perceptions of probability of a future event.
Another possibility for the future is the coupling of news reports with prediction markets so that, in reporting on current events, news media might ask the participating public to contribute their views on how the reported event might impact on future possibilities, as ad hoc polls do poorly now. This could be a new ongoing tracking mechanism that would allow policy makers to judge the consequences of change-in-process and to gain lead time in creating policy to defeat negative consequences.
Prediction Markets 12 The Millennium Project Futures Research Methodology—V3.0
APPENDIX: SAMPLE APPLICATIONS
Hewlett-Packard developed an internal market prediction system known as BRAIN. This is a conventional market system except for the HP introduction of a factor that ―calibrates‖ each participant based on their apparent risk-taking propensity and past forecasting successes. They described their application as follows:9
The ability to predict the outcome of future events quickly and accurately is critical in today's business environment. Agile sensing of climate changes can help us alter both spending and production without the inevitable lags inherent in the traditional marketplace. Forecasting is an integral part of the business management framework. However, month-to-month prediction is difficult in our business and data environment. Those reporting information, consciously or not, may under- or over-report numbers. Some people are more risk-loving than others, and this may be reflected in forecasts. Further, individuals are aware to whom they are reporting, and that can play an important strategic role in reports. Compounding this problem, given that the current system is hierarchically organized, it takes a lot of time to pass through the layers, and distortions may be further amplified.
To address this issue, we have developed a process that takes much of the bias, manipulation, and hierarchy out of prediction. The BRAIN Process is an information aggregation tool that harnesses all of the power and truth-telling properties of market mechanisms and implements it with all of the simplicity and robustness of a simple survey. It is designed to perform in environments where market mechanisms are either not useful (due to the environment) or non- viable for complexity reasons.
To customize our mechanism, we briefly run a market game, allowing participants to buy and sell assets whose values are tied to the outcome of some future event (such as revenues, the outcome of a deal, or the success of a project or product). We monitor the buying and selling patterns of the participants, the market prices of the assets, and the true outcome. We then use that data to construct a quantitative behavioral profile of each participant, known as a beta coefficient. This beta term summarizes each participant's risk attitude, and their predictive power.
Once we have calibrated a reliable behavioral coefficient for each participant, the BRAIN Process is implemented. One could think of it as a betting game with a behavioral twist. We ask each participant to place bets (from a finite pool of betting tokens) on the possibility that various outcomes might occur. We also have a mechanism to figure out which bets are being entered based on publicly available information, and which are based on private. We then aggregate all of the participants' bets, weighting them by their individual coefficients. This aggregate, weighted prediction has been shown in laboratory experiments to outperform the market, as well as the best individual in the room.
9 http://www.hpl.hp.com/research/ssrc/competitive/brain/?jumpid=reg_R1002_USEN
Prediction Markets 13 The Millennium Project Futures Research Methodology—V3.0
We have implemented a pilot of the BRAIN mechanism within an HP Division. We are making monthly predictions of Revenues and Operating Profits, with the participation of a small team of senior executives
Figure 2. Accuracy of Predictions, by Mechanism
.
(TG) The third example comes from NewsFutures. They explain the operation of their market system as follows:10
…Let's suppose that you want to invest in the outcome «The US will catch Osama bin Laden while Bush is President». The trading price for the corresponding "contract" is currently X$3 ("X$" is the game's play money). If the predicted outcome occurs, then each contract that you bought will yield X$100 and you will have earned X$97 per contract (100 - 3). That's a 3,233% return on your investment!
Trading price for The US will catch Osama bin Laden while Bush is President
In general, you invest in specific outcomes by buying contracts with prices between X$1 and X$99. While the outcome is undecided, supply and demand -- yours and the other traders' - determine the appropriate price for these contracts. Then, when the outcome is
10 http://us.newsfutures.com/guide.html#exemple
Prediction Markets 14 The Millennium Project Futures Research Methodology—V3.0
finally known, the winning contracts are cashed in for X$100 each, while the losing contracts are worthless.
All transactions take place directly among traders. When you buy a contract, it is because another player has sold it to you. When you sell a contract, it is because another player has bought it from you.
And that's the truly neat feature of this game: you can resell your contracts at anytime to other players. That means you do not have to wait for the outcome to make a profit! Indeed, some players find great excitement in repeatedly buying low and selling high, turning a profit each time by astutely anticipating market movements––that is, others' collective opinion.
(TG) Players on NewsFutures can play with fake money, or if they wish, they offer an alternative market called Bet2Give, in which one can bet real money, with all winnings donated to charities of the winners’ choice.
Prediction Markets 15 The Millennium Project Futures Research Methodology—V3.0
BIBLIOGRAPHY
Berg, Joyce and Thomas Rietz, ―The Iowa Electronic Markets; Stylized Facts and Open Issues,‖ a chapter in Information Markets, Robert Hahn and Paul Tetlock (ed.), AEI-Brookings, 2006 .
Berg, Joyce, Robert Forsythe, Forrest Nelson and Thomas Rietz, 2001. ―Results from a Dozen years of election Futures Markets Research,‖ in Handbook of Experimental Economic Results. Charles Plott and Vernon Smith, eds. Amsterdam: Elsevier, forthcoming.
Berg, Joyce, Forrest Nelson, and Thomas Rietz, 2006. ―Accuracy and Forecast Standard Error in Prediction Markets,‖ mimeo, University of Iowa.
Charette, Robert, ―An Internal Futures Market,: BI Review Magazine, February 28, 2007
Chen, Kay-Yut and Charles Plott, 2002. ―Information Aggregation Mechanisms: Concept, Design and Implementation for a Sales Forecasting Problem,‖ CalTech Social Science Working Paper No. 1131.
Hanson, Robin and Ryan Oprea, 2005. ―Manipulators Increase Information Market Accuracy‖, mimeo, George Mason University.
Hibbert, Chris, ―Zocalo: An Open-Source Platform for Deploying Prediction Markets,‖ CommerceNet Labs Technical Report 05-02, February 2005 : http://wiki.commerce.net/images/d/d3/CN-TR-05-02.pdf
Journal of Prediction Markets, The University of Buckingham Press, Buckingham, UK., 2009. http://www.predictionmarketjournal.com/index.htm
Looney, Robert. 2003. ―DARPA’s Policy Analysis Market for Intelligence: Outside the Box or Off the Wall.‖ Strategic Insights. September, :9; available at http://www.ccc.nps.navy.mil/si/ sept03/terrorism.asp.
Nelson, Forrest, George Neumann and Philip Polgreen, 2006. ―Operating With Doctors: Results from the 2004 and 2005 Influenza Markets‖, mimeo, University of Iowa.
Ortner, Gerhard, 1998. ―Forecasting Markets—An Industrial Application,‖ mimeo, Technical University of Vienna.
Pennock, David, Steve Lawrence, Finnrup Nielsen and C. Lee Giles, 2001. ―Extracting Collective Probabilistic Forecasts from Web Games,‖ in Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 174-183.
Pharmaceutical Industry webinar on Prediction Markets: http://www.eyeforpharma.com/virtual/forecastingpm/talk1.htm (2008)
Prediction Markets 16 The Millennium Project Futures Research Methodology—V3.0
Prediction Markets on News, Midas Oracle, http://www.midasoracle.org/predictions/ (2009)
Rhode, Paul W. and Koleman S. Strumpf, 2005. ―Manipulation in Political Stock Markets‖, mimeo, University of North Carolina.
Robert Woods Johnson Foundation, ―Influenza Prediction Market to Help Strengthen Early Warning System for U.S. Bird Flu Outbreak,‖ March 2007. http://www.rwjf.org/pr/product.jsp?id=21942
Wolfers, Justin and Eric Zitzewitz, Prediction Markets, Journal of Economic Perspectives— Volume 18, Number 2—Spring 2004—Pages 107–126
Wolfers, Justin and Eric Zitzewitz, 2006. ―Five Open Questions About Prediction Markets,‖ in Information Markets: A New Way of Making Decisions in the Public and Private Sectors, ed. Robert Hahn and Paul Tetlock, AEI-Brookings Joint Center, Washington D.C.
Wolfers, Justin and Eric Zitzewitz, ―Prediction Markets in Theory and Practice, NBER Working Paper No. 12083, March 2006
Prediction Markets 17