Unquantified Benefits and Bayesian Cost-Benefit Analysis

Abstract

Regulatory agencies are required to perform cost-benefit analysis of major rules. However, in many cases regulators refuse to report a monetized value for the benefits of a rule that they issue. Sometimes, they report no monetized value; at other times, they report a monetized value but also state that not all benefits have been quantified. On occasion, regulators also refuse to monetize or fully monetize costs. These practices raise a puzzle. If a regulator chooses not to monetize all the benefits or all the costs, it is not doing cost-benefit analysis. If it is not doing cost-benefit analysis, what is it doing? To investigate this question, we compiled a data set consisting of all major regulations issued by agencies from 2010 to 2013. We come to three conclusions. First, there are countless examples where agencies fail to fully monetize the benefits and costs of regulations. Second, in most cases, agencies could easily monetize or partially monetize those benefits and costs. Third, even where monetization would be difficult, the agencies could and should have made explicit the implicit valuations they relied on and supported those valuations as much as possible with empirical evidence. We then proceed to explain how agencies could engage in cost-benefit analysis even when they do not have a reliable basis for estimating valuations. Even where they lack complete data, agency regulators may be able to make reasonable guesses about the harms or benefits from regulations. In many cases, these guesses will be based on the experience and latent knowledge of the agency staff. These preliminary guesses constitute Bayesian prior probabilities. While agencies should be permitted to “guess” — that is, supply a subjective prior probability — they must also be required to update their estimates as they gain new information.