Optimal Policy Choices Under Uncertainty

Policymakers often make changes to policies whose benefits and costs are unknown and must be inferred from statistical estimates in empirical studies. The sample estimates are noisier for some policies than for others, which should be adjusted for when comparing policy changes in decision-making. In this paper I consider the problem of a planner who makes changes to upfront spending on a set of policies to maximize social welfare but faces statistical uncertainty about the impact of those changes. I set up an optimization problem that is tractable under statistical uncertainty and solve for the Bayes risk-minimizing decision rule. I propose an empirical Bayes approach to approximating the optimal decision rule when the planner does not know a prior. I show theoretically that the empirical Bayes decision rule can approximate the optimal decision rule well, including in cases where a sample plug-in rule does not.