Machine Learning and Causality: The Impact of Financial Crises on Growth

IMF Working Papers, November 1, 2019

Source details

Canonical URL: Machine Learning and Causality: The Impact of Financial Crises on Growth

Machine learning tools are well known for their success in prediction, but prediction is not causation; causal discovery is central to most economic policy questions.
The paper introduces leading work on combining machine learning with causal inference using a concrete example: assessing the impact of a hypothetical banking crisis on a country’s growth.
Machine learning enables consideration of a rich set of potential nonlinearities and allows individually-tailored policy assessments, providing an invaluable complement to economists’ skill sets.

Emphasizes the distinction between prediction and causal inference and surveys methods that address causality within the machine-learning literature.
Uses a banking-crisis → growth counterfactual framework to illustrate causal discovery and treatment-effect estimation.
Highlights machine-learning strengths for:
- exploring nonlinear relationships,
- tailoring policy evaluations at the individual (country) level,
- complementing traditional econometric approaches such as instrumental-variables and randomized-experiment frameworks.

Machine learning can expand the toolkit for policy evaluation at the IMF and beyond by improving counterfactual prediction and treatment-effect heterogeneity analysis.
Individually-tailored assessments can inform country-specific policy responses to financial crises.
Integration with established causal-inference methods (for example, instrumental-variables approaches and randomized-experiment logic) is emphasized as a practical way to strengthen policy-relevant conclusions.

Subject: Exchange rate flexibility, Financial crises, Foreign exchange, Machine learning, Technology
Keywords: B. machine learning, banking crisis, causal inference, confidence interval, counterfactual prediction, Exchange rate flexibility, financial crisis, Global, instrumental-variables approach, Machine learning, machine learning tool, machine-learning literature, machine-learning model, machine-learning modification, ML technique, policy evaluation, randomized experiments, RF algorithm, Supervised machine learning, treatment effects, treatment variable, WP