Potential spillovers from the banking sector to sovereign credit ratings

ABSTRACT

The global financial crisis and European sovereign debt crisis underlined the links between the banking sector and sovereign risk. This paper uses a machine learning technique (random forest regression) to examine whether sovereign ratings account for the potential spillovers from the banking sector to sovereign risk. To do so, we use a panel of sovereign ratings issued by the three main credit rating agencies (CRAs) (Fitch, S&P and Moody’s) for 30 European countries from 2002 to 2016. We find that in addition to the main economic indicators (macroeconomic, government and institutional quality factors), the soundness of the banking system is relevant in determining the sovereign rating. In fact, with the outbreak of the crisis, the importance of these banking sector characteristics (namely, liquidity, concentration and volume of non-performing loans) for sovereign ratings increased substantially. These results suggest a change in CRAs’ policies since the onset of the crisis, involving a re-appraisal of the structure of the banking sector when assessing countries’ sovereign risk.

KEYWORDS:

• Sovereign ratings
• credit rating agencies
• random forest
• crisis
• banking sector

JEL CLASSIFICATION:

• G01
• O11
• G24
• H63

Disclosure statement

No potential conflict of interest is reported by the authors.

Notes

¹ Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland and United Kingdom

² The latest version of the GFDD provides data up to 2017, however since there are some missing values for 2017 our sample ends in 2016. For robustness purposes, we re-run the model including 2017. The results are qualitatively similar.

³ The R-package ‘randomForest’ is used.

⁴ Being consistent with Laeven and Valencia (2018)’s crises database for the European countries.

⁵ We also employ alternative indicators: profitability (ROE), capitalization (capital-to-risk weighted assets ratio), size (deposits to GDP), risk (provisions to NPL) and concentration (assets held by the fifth-largest banks). The results (available upon request) are qualitatively similar.

⁶ For robustness purposes, we compute the changes in the variables’ relevance following the crisis onset using the extreme gradient boosting. This algorithm reveals that on average, the banking sector variables increased its importance by 42.20%. This result (available upon request) confirms our findings.

Additional information

Funding

This work was supported by the Spanish Ministry of Science and Innovation under Grant [PGC2018 – 099415 – B – 100 MICINN/FEDER/UE]; Junta de Andalucía under Grant [P18-RT-3571 Project and P12.SEJ.2463]; and FUNCAS Foundation.