Predicting Decisions
Deliberative prediction markets substitute for the deliberation performed by committees by providing incentives for the production and release of information. Such markets do not by themselves produce decisions (though we will see beginning in Chapter 5 how prediction markets can be designed to make decisions), but they can do much of the work of decision support, providing a range of arguments and counterarguments plus the market’s assessment of the persuasiveness of those arguments. Of course, in order to use a market as a substitute for a committee, the market designer must find a suitable number to be predicted that will generate the relevant arguments. In some cases, this might initially seem quite difficult. Suppose, for example, that a committee is charged with selecting a piece of equipment for a children’s playground. It might seem difficult to design a market that would generate arguments about the pros and cons of each piece of equipment, given the absence of an objective correct answer that someday will become available. There is, however, a relatively simple antidote: A prediction market can be used to forecast what a decision will be. For example, one person might be charged with making the decision, and a deliberative prediction market would be used to gauge the probability that each of various pieces of equipment might be chosen. The decision maker could take advantage of the arguments produced by the prediction market. For example, someone might make a prediction based on information that a particular piece of equipment is unsafe, recognizing that the provision of this information ultimately would make the piece of equipment less likely to be chosen. In principle, a deliberative prediction market can be used to generate information for any decision that can be quantified ex post. That includes not only decisions among two or more choices but also decisions settling on a particular number, such as the number of students who should be admitted in a particular year to an undergraduate institution or the size of a fence that should be placed around company grounds. One aspect of these markets that a decision maker might find particularly attractive is that participants in the markets will seek to provide only information that the decision maker would find relevant. For example, if the decision maker has previously expressed an opinion that the aesthetics of playground equipment should not matter, participants will not have much incentive to produce information about aesthetics. Ordinarily, a decision maker who creates a committee to help with a particular decision takes the risk that the committee will base its recommendations on its own agenda. Even if committee members realized that they should not make arguments about the aesthetics of playground equipment because the ultimate decision maker would not care, they might secretly factor this information into their analyses. In a deliberative prediction market, there would be incentives for participants to identify such hidden agendas. Not everyone will celebrate the fact that a deliberative prediction market caters to a decision maker’s preferences. The decision maker may have idiosyncratic preferences, perhaps preferences that would be inferior to those of a hypothetical average group member. In some organizations, the idiosyncratic preferences might be desirable; perhaps the president of a business is more loyal to the owners than are the employees. In other contexts a more democratic approach to decision making may be appropriate. Where this is the case, however, the deliberative prediction market itself might be used to predict the decision of a more representative committee. There may be danger in allowing those who participate in the decision making also to be on the committee, because decision makers might make decisions in order to boost their profits from prediction rather than on the basis of the best interests of the group. A partial antidote would be to prohibit decision makers from participating only in the final stage of the prediction market, in which a strictly proper scoring rule is used to predict the actual decision. Nonetheless, there is a danger that particular decision makers might gain (and indeed seek) a reputation for making decisions that are consistent with their earlier predictions. Ideally, the decision makers should be separate from the individuals who are predicting their actions. If this separation is achieved, such prediction markets can help show how unpredictable a particular decision maker is. Suppose, for example, that a decision maker routinely ignores the forecasts of the prediction markets and the information that they produce. That might signal that he or she is not taking into account the considerations that the prediction market thought he or she would take into account. Once arbitrary decision making becomes expected, market participants will hesitate to produce new information and analysis, because they would not think that these considerations were relevant to the decision maker. Unpredictable decision making does not always indicate arbitrariness; perhaps the decision maker simply is smarter than any of the participants in the market and can be expected to produce better analysis than they can. The information produced in a deliberative prediction market provides a way of assessing what concerns, if any, appear to animate a decision maker. If such a market produces insights about a wide range of considerations, that is a sign that the decision maker has a reputation for taking these considerations into account. If information that might seem irrelevant appears to move the market, that would indicate that market participants, rightly or wrongly, expect the decision maker to take the information into account. It might be disturbing, for example, if the deliberative prediction market used in forecasting the playground equipment choice changed its prediction when a member pointed out that one type of equipment is manufactured by a friend of the ultimate decision maker.
One Response to “Predicting Decisions”
Leave a Reply
February 29th, 2008 at 10:51 am
Sprenger (2007) published in journal of prediction markets would fit nice into this chapter!