Using Machine Learning for Prediction and Policy Analysis in Economics

Abstract

Machine learning is now part of the normal empirical toolkit of economics, but its contribution differs across tasks and must be judged through economic criteria. This review examines how machine learning contributes to prediction and nowcasting, measurement from new data, causal inference, policy learning, structural and equilibrium computation, financial applications, and operational deployment.

Across these settings, flexible algorithms can improve the use of high-dimensional administrative data, text, images, digital traces, and labour-market sequences. In causal research, machine learning is especially useful when it estimates nuisance functions within orthogonal, doubly robust, or sample-split procedures. In structural, financial, and market-design applications, it expands the feasible computational frontier, while identification and counterfactual reasoning continue to depend on economic structure.

The article argues that predictive accuracy is only one consideration in policy and institutional applications. Welfare, fairness, transparency, governance, and robustness to distributional change are also essential when evaluating whether machine learning improves economic evidence and decision-making.

The manuscript has been accepted for publication in The Review of Mathematical Economics. It is not hosted on this website due to copyright considerations.

Authors

Jiajing Sun, Michael Cole, and Wolfgang Karl Härdle

Recommended citation: Sun, J., Cole, M., & Härdle, W. K. (2026). "Using Machine Learning for Prediction and Policy Analysis in Economics." The Review of Mathematical Economics, accepted for publication.