Optimal Policy Choices Under Uncertainty

Abstract

Policymakers often face the decision of how to allocate resources across many different policies using noisy estimates of policy impacts. This paper develops a framework for locally optimal policy choices under statistical uncertainty. I show that posterior mean benefits and net costs are sufficient statistics for an oracle planner who knows the distribution of policy impacts. Since this distribution is unknown, I propose an empirical Bayes approach to estimate posterior means and approximate the oracle. I derive rates of convergence to the oracle's decision and show that, unlike empirical Bayes, plug-in methods can fail to converge. In an application to 127 policies, empirical Bayes rules have positive estimated local welfare effects, while the plug-in rule has negative estimated effects.