MODELLING OF RUSSIAN BANKS’ PROBABILITY OF DEFAULT

This article discusses an approach to modeling the probability of banks default based on their internal reporting indicators using the random forest model. Figures from the annual reports of the Russian banks in the balance sheet section and cash flow statement section were used as indicators for the model. In order to assess the probability of default of Russian banks, Moody’s Investors Services’s long‑term international foreign currency rating was used. Credit ratings are opinions about the credit risk. The rating expresses the opinion of agencies on the ability and readiness of the issuer, for example, a corporation, a state or a municipality, to fulfill financial obligations in a timely and complete manner. Each rating agency has its own methodology of assessment of the credit rating; however, it has, as a rule, a descriptive nature, and as a result, it is impossible to say, which factor has a greater impact on the financial stability of the bank. Not all classification methods can be applied to the testing sample due to the fact that the number of banks that have a credit rating is limited. This imposes a restriction on the methods based on the learning algorithm, since many of them require the use of a large data array (more than 10,000 objects). The random forest model makes it possible to identify, which factors from the reporting internal indicators of the Russian banks have the greatest influence on the formation of a credit rating. The random forest model showed a high predictive ability. The accuracy of the random forest model for the validation sample is higher than the model constructed using the support vector machine.

bank’s default estimation, supervised training, classification random forest model, decision tree, financial sustainability, public Russian banks, private Russian banks, multivariable analysis, credit rating.

Chelyshev D. S. Modelling of Russian banks’ probability of default. Business. Education. Law, 2019, no. 2, pp. 262–266. DOI: 10.25683/VOLBI.2019.47.271.