Microfoundation Inference for Strategic Prediction

2024-11-13

arXiv.org

AI Summary

This work addresses performative prediction, where predictive models alter the distribution of target variables by influencing stakeholders’ strategic behavior. We propose a novel method to learn long-term distributional mappings induced by model deployment. Our core contribution is the first formulation of individual responses as utility-maximization problems subject to cost constraints, coupled with a theoretically grounded distribution-mapping estimation framework built upon optimal transport theory—providing provable convergence and explicit convergence-rate guarantees. The method jointly integrates utility modeling, cost-parameter estimation, and distribution alignment. Empirical evaluation on credit-scoring data demonstrates its ability to accurately capture structural distributional shifts in populations before and after model deployment. Compared to prior approaches, it significantly enhances interpretability and controllability of predictive models’ societal impact, enabling more responsible algorithmic decision-making.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Decision/Utility Theory

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Often in prediction tasks, the predictive model itself can influence the distribution of the target variable, a phenomenon termed performative prediction. Generally, this influence stems from strategic actions taken by stakeholders with a vested interest in predictive models. A key challenge that hinders the widespread adaptation of performative prediction in machine learning is that practitioners are generally unaware of the social impacts of their predictions. To address this gap, we propose a methodology for learning the distribution map that encapsulates the long-term impacts of predictive models on the population. Specifically, we model agents' responses as a cost-adjusted utility maximization problem and propose estimates for said cost. Our approach leverages optimal transport to align pre-model exposure (ex ante) and post-model exposure (ex post) distributions. We provide a rate of convergence for this proposed estimate and assess its quality through empirical demonstrations on a credit-scoring dataset.

Problem

• Model influence on target variable distribution
• Learning long-term impacts of predictive models
• Estimating agent responses via utility maximization

Innovation

• Models agents' responses via cost-adjusted utility maximization
• Uses optimal transport to align ex ante and ex post distributions
• Estimates social impact costs of predictive models