An explained extreme gradient boosting approach for identifying the time-varying determinants of sovereign risk

Publicado en

  • Finance Research Letters

Resumen

  • We use a combination of Extreme Gradient Boosting and SHAP Additive Explanations to uncover the determinants of sovereign risk across a wide range of countries from 2002 to 2021. By considering numerous variables established in existing literature within a single framework, we identify year-by-year determinants of sovereign credit risk. To gage the liquidity and solvency aspects of sovereign risk, we utilize 5- and 10-year yield spreads as proxies. Our findings show that the key variables driving sovereign risk have remained relatively stable over time and exhibit similarities in both liquidity and solvency components. Among the prominent variables, various macroeconomic fundamentals play a crucial role, including the current account, GDP growth, per capita GDP growth, and the real exchange rate. Prior to the Global Financial Crisis, macroeconomic variables, particularly the current account, held the highest importance in explaining sovereign risk. However, following the GFC, the relative importance of these variables diminished, giving way to institutional variables, especially the rule of law.

fecha de publicación

  • 2023

Líneas de investigación

  • Explainable AI
  • Extreme gradient boosting model
  • Macroeconomic and institutional factors
  • Sovereign risk

Volumen

  • 57