The Nobody-Loses Prediction Market
A variation on the market scoring rule might help with the fraudulent information problem, and more important, with the type of manipulation that we have seen can pose problems for prediction markets that rely entirely on points that can be exchanged for dollars. In this variation, a central computer can limit the amount by which some participants can move the market consensus prediction. For example, if the current prediction of some variable of interest is twenty, and a participant enters a new prediction of forty, the system might announce a new consensus prediction of only thirty. If it turns out that the variable ultimately predicted (with the deliberative market, the market prediction at some later point in time) is forty or higher, the market participant’s profit on the transaction will be limited to thirty. The market would make larger changes in consensus market predictions for traders who have shown themselves to be accurate predictors over time. The system might maintain a statistical model, for example, using regression analysis. Given a new prediction and variables reflecting both the predictor’s past success and the current state of the market, the model would calculate a new market consensus prediction. Ordinarily, this would be somewhere between the old consensus prediction and the new one. For the purpose of determining the predictor’s profit or loss on the transaction, this new prediction would control. A possible exception would be cases in which the computer’s change in the consensus prediction overshoots the change recommended by the predictor, as might happen for predictors who have proved conservative about changing the market consensus. This system greatly reduces the ability of market participants to enter random predictions in the hope that they might happen to earn significant numbers of points. (An alternative approach is for prediction market designers simply to create an automated trader account that trades immediately after each participant trades, but this does not reduce the incentive to create accounts to make random predictions.) If individuals attempted to do this, then the system would learn not to trust them, and the amount of money that they make initially would be very small. At the same time, traders could earn a good reputation both by winning points and by entering predictions that turn out to be valid. This could happen even within a prediction market if a prediction successfully forecasts a later stage of the market, but it would be more likely across a large number of prediction markets. The system should improve over time; at first, it should be subject to some manipulation, but as it learns which participants are reliable, opportunities for manipulation would decline. This approach reduces the fraudulent information problem, because market participants’ primary incentives are to preserve their reputation. At times a participant with a good reputation might seek to make a killing with fraudulent information, but at least if many future prediction market opportunities exist, this should seldom be a significant problem. At the same time, if the currency is points rather than dollars, the market limits the ability of participants to open phantom accounts from which they make bad predictions, enabling real accounts to take advantage of those predictions. Because the phantom accounts should have only small effects on market predictions, it will be hard to channel points from phantom accounts to real accounts. The result is that this system can facilitate the creation of a prediction market system in which points can be exchanged for dollars and thus no market participant faces a risk of losing money. In part, this may be useful as a means of attracting risk-averse individuals to the system, but it also is particularly important as a method of avoiding regulatory impediments to prediction markets. Although participants who have established themselves as reliable predictors take a risk with every prediction, courts seem highly unlikely to condemn this system as illegal gambling. After all, anyone who participates in a performance-based reward system faces the risk that a bad performance will lower the eventual reward. At the same time, the market scoring rule is not likely to be subject to securities and commodity futures laws because there are no trades. An alternative approach that similarly can help aggregate information held by members of a group is to use a prediction market simply to gauge the predictive accuracy of different individuals and their relative tolerance for risk. Then, one can simply ask these individuals for predictions, promising to reward them according to a simple scoring rule, and aggregate the predictions based on the performance in the initial prediction market. Kay-Yut Chen and two coauthors conducted an experiment with this approach,60 and they found that they achieved greater accuracy than with a prediction market alone. In some settings, particularly where it is feasible to request simultaneous probability assessments from each of a number of participants, this approach therefore might be better than a prediction market alone because it helps isolate prediction market participants whose bad predictions tend to distort market outcomes. If prediction markets are run for long enough, however, underperforming participants may tend to leave the market or correct their errors, so for long-term projects, conventional prediction markets may be superior. Both of these approaches show that it should be possible to implement prediction markets on a large scale despite existing regulatory limitations. This system, admittedly, might not be as strong as one that allows individuals to place their money at risk. It might not sufficiently take into account the information of genuine predictors, and the need to accrue a good reputation over time might discourage some individuals from participating. Thus, someone who might have information that is relevant to a single short-lived prediction market seems unlikely to participate. The possibility of nobody-loses prediction markets might make policy makers less eager to relax restrictions on prediction markets, though it is also possible that success might lead to a broader appreciation of the possible benefits of prediction markets and thus regulatory relaxation.
2 Responses to “The Nobody-Loses Prediction Market”
Leave a Reply
January 30th, 2008 at 1:55 pm
[…] incentives provided by two of my technical proposals (the decentralized subsidy approach and the nobody-loses prediction market) are sufficiently straightforward to me that math seems superfluous to me, though I agree that […]
February 13th, 2008 at 3:05 pm
[…] predicto placed an interesting blog post on The Nobody-Loses Prediction MarketHere’s a brief overview… might make policy makers less eager to relax restrictions on prediction markets, though it is also possible that success might lead to a broader appreciation of the possible benefits of prediction markets and thus regulatory relaxation. […]