A1787
Title: Multi-objective policy learning
Authors: Daido Kido - Otaru University of Commerce (Japan) [presenting]
Abstract: When treatment effects exhibit heterogeneity, allocating treatments to individuals who benefit most is essential for constructing efficient policies. However, treatments often influence multiple outcomes rather than a single one. In such cases, a treatment allocation optimized for one specific outcome may result in undesirable assignments from the perspective of another outcome. Therefore, understanding these trade-offs is crucial. Estimation and inference methods are developed for the Pareto frontier -the boundary where it is impossible to improve one outcome without worsening another. The estimator is shown to be consistent and confidence sets are shown to asymptotically achieve nominal coverage.
