A new paper by Professor Maximilian Kasy, co-authored with Nicolò Cesa-Bianchi (Università degli Studi di Milano and Politecnico di Milano) and Roberto Colomboni (Politecnico di Milano), has been accepted for publication in Econometrica.
The paper, Adaptive Maximization of Social Welfare, develops a framework for policy design that learns and adapts over time, with the goal of maximising social welfare defined as a combination of individual well-being and public revenue. The authors consider situations where outcomes of a policy are not fully observable, such as when policymakers can see employment rates but not the underlying welfare of individuals. In such cases, determining the best course of action requires careful experimentation.
To address this, the paper proposes an algorithm that adaptively updates policies based on past results, while also ensuring a robust performance in uncertain or changing environments. Drawing on ideas from both economics and machine learning, the authors show that their method achieves the best possible improvement rate over time, even in challenging settings where outcomes are unpredictable or adversarial.
The results offer a new benchmark for sequential policy design, with applications in taxation, benefit systems, and beyond.
Find out more and read the paper: Forthcoming Papers | The Econometric Society
