Predictive Cost-Benefit Analysis
A simple approach to deciding on the payouts for a normative prediction market would be to commit to asking the later decision maker whether he or she agrees with or disagrees with the particular decision. Assuming that the ultimate evaluation of the decision occurs a long time after the market has run its course, however, this prediction could be misleading. Suppose that there is a large probability that a particular regulation will on net provide a small social benefit but a small probability that the regulation will cause a very large social harm. If it will become clear before the ultimate evaluation which of these is true, then the prediction market would probably forecast that the decision maker will approve of the regulation. Conceivably, decision makers might be instructed to evaluate the regulation from the ex ante perspective, but the danger of hindsight bias makes this difficult. This normative prediction market design thus forecasts whether a decision will turn out to be justified but not the magnitude of benefit or harm caused by the decision. A solution to this problem is to require the ex post decision maker to announce not simply whether the decision was desirable, but how beneficial or costly it turned out to be. Although one might devise numerous scales for this assessment, perhaps the most natural is a scale of dollars (or other currency). The ex post decision maker would announce either a positive number, indicating that the decision had net benefits equal to a specified number of dollars, or a negative number, indicating that the decision had net costs. To the extent that the ex post decision might occur at a point before all of the benefits and costs of the decision have been felt, the decision maker would have to consider not only those realized already but also those projected for the future. Ideally, the decision maker should be instructed to aggregate benefits and costs by considering when they occur, using interest rates to discount them to the time at which the decision was made. This approach builds on a familiar institution of the regulatory state, cost-benefit analysis, and could be called “predictive cost-benefit analysis.”13 Traditional cost-benefit analysis can be subjective, inevitably depending on the assumptions in models of the future.14 A Democratic administration might find that a particular regulation would have net benefits, while a Republican administration might find that it would have net costs. Partisans of a position might do the cost-benefit analysis in ways that generally support their instinctive views. Though agency officials may conduct cost-benefit analysis act in good faith, liberals and conservatives may value some effects of regulations differently. Because predictive cost-benefit analysis would involve a prediction of a cost-benefit analysis to be conducted by someone of indeterminate party membership, it should largely succeed at removing the ideological component or at least averaging a broad spectrum of views. Predictive cost-benefit analysis is not the only way of seeking to ensure that cost-benefit analysis is relatively objective. An alternative approach is to craft detailed methodological rules for conducting the analysis, in order to ensure that the identity of the practitioner does not have an effect on the result. Indeed, the With predictive cost-benefit analysis, no guidelines are necessary, but ex post decision makers should be encouraged to provide detailed explanations of their methodologies. Prediction market participants will then need to forecast the proportion of future decision makers who would make particular methodological decisions, and the predictive cost-benefit analysis should reflect a balanced weighting of different methodologies. In traditional cost-benefit analysis, it is often tempting to omit certain considerations altogether, despite their obvious relevance, because there is too much room for debate about how to take the considerations into account. Predictive cost-benefit analysis makes it possible to include all recognized considerations, weighting them as average decision makers would be expected to weight them, without giving the actual decision makers the ability to manipulate the analysis. Robert Frank and Cass Sunstein have argued, for example, that cost-benefit analysis should take into account that people value some goods in part only because those goods allow them to improve their status relative to other individuals.17 Because they do not value health care this way, an assessment of the value of life based on willingness to pay for health care will be inaccurate. An employee who is unwilling to pay some amount of money for a health benefit might nonetheless benefit from a law requiring everyone to pay that amount for the benefit. Thomas Kniesner and W. Kip Viscusi, however, have argued that Frank and Sunstein overestimate the effect that they discuss.18 Given the difficulties of identifying an appropriate methodology for assessing such effects, they are best omitted altogether, according to Kniesner and Viscusi. This might be the best solution in a world in which methodological flexibility introduces the danger of ideological analysis and decision making, but predictive cost-benefit analysis makes this no longer necessary. Similarly, predictive cost-benefit analysis would permit ex post decision makers to take into account “soft variables” that are difficult to determine through any formal methodology. A recurrent debate about cost-benefit analysis concerns the legitimacy of contingent valuation studies, in which surveys are used to determine individuals’ existence values, such as the most that individuals would pay to prevent logging of a forest.19 Such studies are notoriously vulnerable to framing, with the size of a forest bearing little logical relation to individuals’ claimed valuations.20 The possibility that contingent valuation is a useful methodology means that results of studies would receive some weight in predictive cost-benefit analysis. But prediction market participants would recognize that lack of validity does not mean that existence value should be assumed to be zero. Different decision makers would assign different values based on their own calculations or hunches, and market participants would seek to predict the average of these numbers. As long as prediction markets are sufficiently well subsidized and thus accurate, predictive cost-benefit analysis should be preferable to cost-benefit analysis even for someone who believes that agencies should be allowed some degree of ideological freedom. The purpose of analysis is to produce a signal about the objective costs and benefits of a particular regulation. Ideology merely adds noise to the signal. Eric Posner, for example, has argued that the purpose of cost-benefit analysis is to offset the informational advantage that agencies have relative to the president and Congress,21 allowing other branches to determine whether they approve of what the agency is doing. That function, among others, cannot be performed as well if observers must guess the degree to which the agency’s decision is the result of idiosyncratic preferences. It would be better to allow administrative officials some flexibility to enact regulations with predicted net costs than to permit them to obfuscate the decision by manipulating a cost-benefit analysis.
Leave a Reply