Predicting IMF-Supported Programs: A Machine Learning Approach

IMF Working Papers, March 8, 2024

Source details

Canonical URL: Predicting IMF-Supported Programs: A Machine Learning Approach

Applies state-of-the-art machine learning (ML) techniques to forecast IMF-supported programs.
Compares ML prediction results relative to traditional econometric approaches.
Explores non-linear relationships among predictors indicative of IMF-supported programs.
Evaluates model robustness with regard to different feature sets and time periods.

ML models consistently outperform traditional methods in out-of-sample prediction of new IMF-supported arrangements.
Key predictors align well with the literature and show consensus across different algorithms.
Importance of incorporating a variety of external, fiscal, real, and financial features as well as institutional factors like membership in regional financing arrangements.
Data processing choices—feature selection, sampling techniques, and missing data imputation—substantially influence ML model performance and point to the usefulness of a flexible, algorithm-tailored approach.
Models most effective in near and medium-term predictions may tend to underperform over the long term, indicating the need for regular updates or adoption of more stable – albeit potentially near-term suboptimal – models when frequent updates are impractical.

External features.
Fiscal features.
Real features.
Financial features.
Institutional factors, explicitly including membership in regional financing arrangements.

Feature selection materially affects which variables drive predictions across algorithms.
Sampling techniques influence out-of-sample performance and can interact with algorithm choice.
Missing data imputation methods alter model results, underscoring the need for careful data preprocessing tailored to the algorithm and prediction horizon.
Non-linear relationships among predictors are identified and analyzed, supporting the use of ML methods that capture non-linearities.

Short- and medium-term optimized models can underperform in long-term forecasts.
Trade-off highlighted between models that are frequently updated (near-term optimal) and more stable models suitable when updates are impractical.
Implication: establish procedures for regular model updates or prefer more stable models depending on operational constraints.