Using Machine Learning for Prediction and Policy Analysis in Economics

This working paper reviews the role of machine learning (ML) in economics, emphasizing that ML should be judged through economic tasks and validation disciplines rather than as a generic contest between algorithms and econometrics.

Authors:

Jiajing Sun, Michael Cole, and Wolfgang Karl Härdle.

Summary and Contribution:

The paper reads the recent ML-in-economics literature as a set of task-specific research designs. It distinguishes how ML contributes to:

• prediction and nowcasting
• measurement from new data
• causal inference
• policy learning and targeting
• structural and equilibrium computation
• financial applications
• operational deployment in policy and institutions

The central argument is conservative but constructive: ML improves economic evidence when flexible approximation is embedded in credible validation, identification, counterfactual, or welfare frameworks.

Key Themes:

In forecasting and measurement, ML helps economists use high-dimensional administrative data, text, images, digital traces, and labor-market sequences. In causal work, ML is most useful when it estimates nuisance functions inside orthogonal, doubly robust, or sample-split procedures. In structural and financial settings, ML expands the feasible computational frontier, but economic structure remains essential for identification and counterfactual reasoning.

The paper also emphasizes operational settings such as credit scoring, inspections, humanitarian targeting, central-bank communication, and social protection, where predictive accuracy is only one input into broader welfare, fairness, transparency, and governance problems.

Availability:

The manuscript is not hosted on this website due to copyright considerations.