Predicting Fiscal Crises: A Machine Learning Approach

IMF Working Papers, May 27, 2021

Source details

Assesses the ability of econometric and machine learning techniques to predict fiscal crises out of sample.
Compares econometric approaches commonly used in policy applications with simple heuristic rules and with machine learning techniques.
Machine learning techniques used: elastic net, random forest, gradient boosted trees.
Expands the set of potential predictors and uses algorithmic selection techniques rather than relying on a small set of variables chosen by the literature.

Econometric approaches used in many policy applications cannot outperform a simple heuristic rule of thumb.
Machine learning techniques (elastic net, random forest, gradient boosted trees) deliver significant improvements in accuracy.
Performance of machine learning techniques improves further, particularly for developing countries, when the set of potential predictors is expanded and algorithmic selection techniques are used.
There is considerable agreement across learning algorithms in the set of selected predictors.
Results confirm the importance of external sector stock and flow variables found in the literature.
Demographics and the quality of governance emerge as important predictors of fiscal crises.
Fiscal variables appear to have less predictive value.
Public debt matters only to the extent that it is owed to external creditors.

Relying on a small set of variables deemed important by prior literature may limit predictive performance; algorithmic selection across a broader predictor set can improve results.
Machine learning approaches can provide materially better out-of-sample prediction of fiscal crises than standard econometric models and simple heuristics.
Particular gains from machine learning are observed for developing countries when broader predictor sets are considered.

Source: Predicting Fiscal Crises: A Machine Learning Approach (IMF Working Paper by Klaus-Peter Hellwig, May 27, 2021).