Conditional Prediction Markets
The Unwinding Approach Prediction markets might not seem to be a suitable alternative to standards because they predict what will happen, not what will happen contingent on some choices or on some other events. A rule of enterprise liability imposed on automobile manufacturers for all automobile accidents will lead those manufacturers to make conditional assessments, considering how many accidents are likely to occur for different possible configurations of safety features. A prediction market can forecast how many accidents will occur for a particular model, and so in theory the government might require a prediction of relatively few accidents before a car could be sold. The prediction market result would be unstable, however. If it anticipates many accidents, then the car cannot be sold, so then there will be few accidents, but then the car can be sold. What we need are conditional prediction markets, in this case assessing how many accidents will occur if a particular model is allowed on the market. The conceptually simplest approach to implementing conditional prediction markets is to unwind all transactions in the market if the condition does not occur. For example, if A bought a prediction market share in an outcome from the market sponsor for one dollar, then A would be refunded one dollar if the condition were not met. If A later sold the share to B for eighty cents, then A would refund the eighty cents to B. If, for example, a prediction market forecasts the attendance at a baseball game in the event of rain, but it does not rain, then participants will have lost time but not money investing in the market. This approach will not always be simple in practice, especially if the market takes place over a long period of time. The market sponsor will need some mechanism for locating all of the market participants and ensuring that they pay or receive money as appropriate. Refunding money to a party might well be easier than collecting money that a party previously received. One way of accomplishing this is by placing any money received from a market participant in an escrow account, perhaps an interest-bearing one, with amounts to be paid back dependent on whether the condition is met. This approach has, I admit, a significant drawback: the market participant will have no control over funds received in market transactions until the conclusion of the market. In any event, the unwinding approach has proved commercially feasible, because TradeSports regularly specifies that it will unwind contracts in the event of certain contingencies. For example, for National Basketball Association games, TradeSports notes that games become official after forty-three minutes, and “[g] In some contracts, the unwinding condition is of greater significance. For example, when President Bush nominated Harriet Miers to the Supreme Court, TradeSports offered both a contract predicting whether she would be confirmed and a contract predicting whether she would receive at least a specified numbers of votes in the Senate. Once TradeSports clarified that the latter set of contracts would be unwound if Miers did not receive a Senate floor vote, the trading prices straightforwardly indicated that Miers’ probability of receiving fifty votes in the Senate if she received a floor vote was greater than her chance of being confirmed. In the end, Miers withdrew her name from consideration, and so the condition was not met. A significant drawback of the unwinding approach to conditional markets is that it might encourage relatively little estimation in the case of conditions that are unlikely to hold. Consider, for example, a prediction market forecasting the probability that various potential nominees to the Supreme Court would be confirmed. The incentive to trade shares of someone who only has a small chance of being nominated would be low. A trader’s expected profits from improving the probability prediction by a set amount for someone who has only a 1 percent chance of being nominated would be only one-fiftieth the expected profits from doing so for someone who has a 50 percent chance of being nominated. Such a contract could be subsidized, but the potential subsidy, which ordinarily will not be paid, would need to be roughly fifty times higher to compensate for the low probability. A potential way around this problem is to use a design that rewards market participants for making predictions at a later stage of the market. For example, in the deliberative prediction market proposal described above, each prediction is evaluated on the basis of the degree to which it the market prediction at some interval thereafter. With this approach, it would not be necessary to unwind all transactions. Rather, the market sponsor could assign a score of zero (meaning no profit or loss) to any transactions that took place after it became apparent that the condition would not be met or when the probability of its occurrence, as determined by a separate prediction market, fell permanently below some threshold, such as 1 percent. The market sponsor would then also need to assign a score of zero to any earlier predictions that otherwise would have been evaluated on the basis of the market consensus in this time period. Suppose, for example, that this market were established to determine the probability of confirmation of potential Supreme Court nominees. The market sponsor could create the market well in advance of any expected nomination and provide for a prediction interval of about a month for the conditional probabilities. At the same time, the sponsor would create a market predicting the probability that each of various potential nominees in fact would be nominated by a particular date. If I believed that Judge Jones had only a 5 percent chance of being nominated, I still could trade on information about her confirmation chances. As long as Judge Jones’s nomination probability remained above the 1 percent (or other designated) threshold about a month after I entered my prediction, then the fact that the condition might not occur would have no effect on my ability to make money on the basis of my prediction. If Judge Jones were nominated, then the market sponsor would have no need to unwind the market. The reason that this mechanism would provide appropriate incentives to make predictions is straightforward. Anyone investing in this market would recognize that announcing a conditional prediction can lead to three different possible outcomes when the prediction interval elapses. It might be the case that the condition has become true, that the condition has become false, or that the condition remains uncertain. If the condition becomes true, then market participants will be predicting the unconditional probability of confirmation, and so the original predictor’s incentive is to predict this probability. If the condition becomes false, then the prediction will receive of pay nothing, and so that possibility should not affect the prediction. If the condition remains uncertain, future predictors will continue to look toward the possibility of the condition’s becoming true, and so once again the best strategy is to announce the unconditional probability of confirmation. Although one could argue about whether the chance of the condition’s occurrence provides sufficient discipline in cases of extremely low probabilities (say, one in a billion),10 it should work for conditions that have some reasonable chance of becoming true.
Leave a Reply