Public external debt sustainability assessment: towards a machine learning based approach

This study addresses the challenge of sovereign external debt sustainability by employing a cointegration test, machine-learning classifiers, and explainable models. Focusing on 22 middle-income countries during the period 2000-2021, our study aims to provide accurate insights into debt positions and capture the complex dynamics between a set of economic and fiscal indicators. Unlike conventional econometric methods, which categorize debt situations as either sustainable or unsustainable over specific periods and often have limitations in generalizing the influences of public policies on debt positions, our machine-learning approach reveals a more nuanced perspective. The results indicate that some countries have encountered episodes of debt unsustainability. These results underscore the substantial role of macroeconomic indicators in shaping a country's financial position in conjunction with outstanding debt. Furthermore, our findings demonstrate that the impact of each feature varies based on its specific threshold, emphasizing the critical role of exchange rates in straining debt sustainability. This paper redefines sovereign debt sustainability analysis for middle-income countries by applying machine-learning techniques to reveal the influence of key economic indicators, including exchange rates, inflation, GDP growth, and foreign reserves. The findings demonstrate that debt sustainability is shaped by complex interactions between macroeconomic factors rather than debt outstanding, offering policymakers a practical framework to assess debt position with greater accuracy.