Statistical Policy Choice in Dynamic Environments

Brown University, Providence RI

Investigators

Abstract

This award will fund research to develop new methods to guide economic and other public policy making in rapidly changing environments. Policy makers, such as Central bankers, fiscal authorities, and foreign currency dealers, make decisions in rapidly changing environments in which they often have to learn and adapt on-the-fly. However, economists do not currently have methods that can effectively guide policy makers in these changing environments. This research will build on methods in statistical decision theory, frameworks for establishing causation from data, and research in machine learning to develop new methods to inform evidence-based policy choices while accounting for uncertainty about the environment. The results of this project will make important contributions to economic science by offering new methods to guide policymakers. The results will improve the quality of policy making, increase efficiency, productivity, and economic growth hence improve the living standards of citizens. There are many challenges inherent in studying optimal policy choices in dynamic environments, including nonstationary dynamic causal effects, non-iid observations, ensuring external validity, and agents' reactions to changes in policy rules. This award funds research that builds on recent development in time-series, statistical decision theory, and machine learning to develop new econometric methods for policy analyses in dynamic environments. The first part studies short-run policy choice problems where reactions to policy choices by agents and long-run equilibrium spillovers are ruled out. The second part focuses on learning in a long-run policy rule considering agents’ reactions to changes in policy (i.e. robust to Lucas critique). The third part considers an unsupervised learning setting where the policymaker interacts with agents and the environment over time to collect data, learn about parameters governing the causal structure and data generating process, and chooses policy to maximize welfare. The results establish foundations for statistical decision theory and policy learning in dynamic settings and provide new methods to guide policy making in a wide range of empirical contexts. The results of this research will improve policy making, increase efficiency, productivity, and economic growth hence improve the living standards of citizens. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.