AI-system, correlation-regression model and VaR-model for prediction of overdue debt of commercial banks of the Russian Federation and analysis of financial risk

Abstract

The article examines the theoretical foundations for the emergence of overdue debts on loans and forecasting financial risk in Russian banks in modern conditions. The relevance of the study is that the growth of bad debts of commercial banks on loans is currently one of the most acute problems. The collected material made it possible to analyze the dynamics of the volume of overdue debt on loans in commercial banks of the Russian Federation for the period 2013–2021. In the course of the study, it was revealed that many factors influence the volume of bad debts. In order to study the influence of factorial signs on the effective sign — the amount of overdue debt, an attempt was made to use such models as: correlation-regression, AI and VaR. A hypothesis has been put forward and proved that with the help of models: correlation-regression, AI and VaR, it is possible to obtain a forecast of the volume of overdue loans in the portfolio of commercial banks of the Russian Federation. The correlation-regression model, in addition to the resultant sign Y — the growth rate of overdue debt, included such factorial signs as: X1 — the growth rate of GDP per capita; X2 — the growth rate of the average per capita income of the population; X3 — the growth rate of foreign trade surplus; X4 — inflation index; X5 — growth rate of capital outflow; X6 — growth rate of cash; X7 — interest rate on loans; X8 — US dollar exchange rate; X9 — price of a barrel of oil URLS dollars; X10 — wage growth. The study showed that the use of various forecasting tools provides different amounts of arrears, for example, for 2022: the correlation-regression model — 7159.9 billion rubles, the neural network — 4466.251 billion rubles, and the VaR model 5426 .56 billion rubles overdue loans.