Private Decision Maker Information
When private decision makers can convey their information to the market, even in summary form, there is no reason to expect that prediction markets should fail to give it adequate weight. Until they convey their views, however, their private information presents another danger: that participants in conditional prediction markets will evaluate possible decisions in part by assessing the circumstances that might lead a decision maker to make them. Suppose, for example, that Action Abramowitz seems likely to be a bad movie if it stars Ray Romano but a successful movie if it stars Ben Stiller. Prediction market participants might then reason that the movie studio will make the movie only if the studio finds that it can sign Stiller. The prediction of the movie’s success will then reflect its success only if it stars Stiller. In this case there might be a straightforward solution: creating separate conditional markets for the cases in which the movie stars Romano and Stiller. At times, however, private information might not be so easily identified. In general, this should lead prediction markets to be biased in favor of being optimistic about every decision. If knowledgeable decision makers choose to make Action Abramowitz, that would indicate that the project is a bit better than market participants would have thought, but the same will be true for every other concept. Market participants will think not about their intrinsic views of particular concepts but about how each concept would likely fare in a world in which someone actually chose to produce it. This should not produce biased evaluations of the movies that are selected for production, but it should mean that the market will be overvaluing the movies that are not selected. In the movie context this should not generally be a serious problem, because what is most important is the relative evaluation of the success of different possible projects, though it does suggest that decision makers should not necessarily produce every concept that is forecast to have positive value. There might, however, be situations in which the amount of private information is expected to be different for different decisions. If, for example, a particular studio executive seems to have an uncanny ability to pick the best comedies but not much skill at picking dramas, the market should rate all comedies relatively highly. A higher price for a comedy than a drama would not necessarily reflect a market prediction that the comedy would be more successful than the drama. A related problem is the danger of future information. Suppose, for example, that conditional markets are used to predict a corporation’s stock price, contingent on decisions to build or not build a particular production facility. If that decision will not be made for another year, participants in the market now will anticipate that the eventual decision maker will have information not available to them. For example, it might be the case that the corporation will choose to build another production facility only if other unrelated projects are quite successful. The conditional market prediction might then seem to indicate that the production facility will raise the corporate stock price, when really the prediction simply means that if the corporation builds the facility, that might be because the corporation performed well in the intervening year. These considerations suggest that care is needed in assessing predictions of conditional prediction markets in cases in which private information will likely be significant. There may also be some ways of reducing the risk of problems. For example, conditional prediction markets might be established relatively near in time to when a decision will be reached. Another promising, if counterintuitive, approach may be to remove decision-making power from the individuals most likely to have inside information. Individuals with private information can contribute to a decision by announcing their views of how the decision should be made or perhaps by trading directly on the market. The decision maker would then be someone without inside information who would consider the market predictions along with the content of the experts’ stated opinions. Using less well-informed but still prudent decision makers may make the conditional prediction markets work better and thus perhaps improve the overall quality of decision making.
Leave a Reply