Machine Learning–Enhanced Identification Strategies for Causal Inference in Economics

Abstract

Causal inference is central to economic analysis, enabling researchers to identify the effects of policy interventions, market shocks, and behavioral responses. Traditional econometric approaches, such as instrumental variables, difference-in-differences, and regression discontinuity designs, often rely on strong assumptions and linearity constraints. Recent advances in machine learning (ML) offer powerful tools to enhance identification strategies by capturing complex nonlinear relationships, high-dimensional interactions, and latent confounding factors. This paper proposes a framework that integrates machine learning techniques with classical econometric identification strategies to improve causal effect estimation. We demonstrate how ML methods—such as random forests, gradient boosting, and neural networks—can be used for flexible covariate adjustment, heterogeneity detection, and instrument selection. Through simulations and empirical applications in policy evaluation, labor economics, and macroeconomic interventions, the proposed approach shows improved precision, robustness, and interpretability. The results highlight the potential of ML-enhanced methods to advance causal inference in economics while maintaining theoretical rigor and policy relevance.