Surrogate Data Models: Interpreting Large-scale Machine Learning Crisis Prediction Models

IMF Working Papers, February 24, 2023

Source details

Canonical URL: Surrogate Data Models: Interpreting Large-scale Machine Learning Crisis Prediction Models

Machine learning models are becoming increasingly important in the prediction of economic crises.
Large numbers of predictors (features) in these models impair model interpretability and their ability to provide adequate guidance in the design of crisis prevention and mitigation policies.
This paper introduces surrogate data models as dimensionality reduction tools in large-scale crisis prediction models and assesses the appropriateness of this approach by applying surrogates to large-scale crisis prediction models developed at the IMF.
The results are consistent with economic intuition and validate the use of surrogates as interpretability tools.

Surrogate data models function as dimensionality reduction tools that improve interpretability of large-scale machine learning crisis prediction models.
Application to IMF-developed large-scale crisis prediction models shows results that are consistent with economic intuition.
Surrogates help bridge the gap between complex, high-dimensional predictive models and the actionable guidance needed for crisis prevention and mitigation policies.

Subjects: Deposit rates, Early warning systems, Financial crises, Financial services, Machine learning, Technology, Yield curve
Keywords: Annex I. IMF ML, Crisis prediction, Deposit rates, Early warning systems, explainable models, Global, IMF ML crisis, machine learning, ML crisis prediction, model interpretability, surrogate data models, surrogates, Yield curve