Default forecasting in the Russian banking sector[Прогнозирование Дефолтов В Российском Банковском Секторе]

In this paper we consider various methods of default forecasting to identify а model with the greatest predictive power on actual Russian data for the period from 2015 to 2016. For this purpose we developed early warning systems using traditional and modern methods: the panel random effects logit model, the panel pooled logit model and the random forest algorithm. The latter algorithm of machine leaning has been applied for the analysis of the Russian banking sector for the first time. Our results show that the new machine learning forecasting tools have greater predictive power than the standard ones: а forecast based on the random forest model gives the lowest mean absolute error and correctly identifies the state of 99.62% of banks in the sample. Besides that, by analyzing estimated panel logit models, the paper determines the factors that make it possible to estimate the probability of bank's bankruptcy based on the dynamics of its financial statements. Moreover, the influence of perceived factors preceding а default is in agreement with previous empirical results and reveals particular features of the Russian banking sector. Thus the results of the paper allow the advancement of the early warning systems of bank defaults, which can be used by commercial banks and monetary authorities to improve their activities.