Leave No One Undermined: Policy Targeting with Regret Aversion

2025-06-19

AI Summary

This paper addresses policy targeting under observable-variable constraints and decision-makers’ concern for inter-group regret inequality. Recognizing the limitation of conventional methods in jointly accommodating regret aversion and fairness, we introduce the regret-averse criterion into personalized intervention rule design—first such incorporation in the literature—and propose a class of score-based optimal rules that simultaneously account for heterogeneous treatment effects and group-level fairness. Within a bias-corrected empirical risk minimization framework, we integrate weighted least squares modeling with asymptotic statistical analysis to establish a statistical theory achieving $1/n$ convergence rate. The method demonstrates high estimation efficiency and practical effectiveness on the NLS and IST datasets; under partial identification, it attains asymptotic efficiency. Our approach provides a novel paradigm for fair and robust policy targeting.

Technology Category

• Machine Learning: Causal Learning
• Game Theory and Economic Paradigms: Adversarial Learning
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments

Abstract

While the importance of personalized policymaking is widely recognized, fully personalized implementation remains rare in practice. We study the problem of policy targeting for a regret-averse planner when training data gives a rich set of observable characteristics while the assignment rules can only depend on its subset. Grounded in decision theory, our regret-averse criterion reflects a planner's concern about regret inequality across the population, which generally leads to a fractional optimal rule due to treatment effect heterogeneity beyond the average treatment effects conditional on the subset characteristics. We propose a debiased empirical risk minimization approach to learn the optimal rule from data. Viewing our debiased criterion as a weighted least squares problem, we establish new upper and lower bounds for the excess risk, indicating a convergence rate of 1/n and asymptotic efficiency in certain cases. We apply our approach to the National JTPA Study and the International Stroke Trial.

Problem

• Policy targeting under regret aversion with limited assignment rules
• Addressing regret inequality due to treatment effect heterogeneity
• Debiased empirical risk minimization for optimal rule learning

Innovation

• Debiased empirical risk minimization approach
• Regret-averse criterion for policy targeting
• Weighted least squares problem analysis