Persuasion and Welfare

Information policies such as scores, ratings, and recommendations are increasingly shaping society's choices in high-stakes domains. We provide a framework to study the welfare implications of information policies on a population of heterogeneous agents. We define and characterize the Bayes welfare set, consisting of the population's utility profiles that are feasible under some information policy. The Pareto frontier of this set can be recovered by a series of standard Bayesian persuasion problems, in which a utilitarian planner takes the role of the information designer. We provide necessary and sufficient conditions under which an information policy exists that Pareto dominates the no-information policy. We extend our results to the case in which information policies are restricted in the data they can use and show that ``blinding" algorithms to sensitive inputs is welfare decreasing. We illustrate our results with applications to privacy, recommender systems, and credit ratings.