?
Как прогнозировать дефолты банков: эволюция методов, моделей и факторов риска
Predicting bank defaults is an important task for the entire economy. Early identification of troubled banks helps to prevent impending bank failures or minimize the losses associated with them. The paper discusses the state of the art of instrumental methods and data used for this purpose. The theoretical background, the evolution of methodological approaches used to predict bank defaults, the specifics of data handling, and the lists of predictors that are included in early warning models are successively reviewed. We conclude that there is still considerable controversy in the literature regarding both the methods and the variables to be used in predictive models. Machine learning methods show a better ability than traditional statistical models to detect non-linear dependencies and to handle large samples. Their advantages are often offset by out-of-sample estimation. Other limitations of such methods are the risk of overfitting and the difficulty in interpreting the results. The lists of potential predictors of bank defaults also vary from country to country. Most commonly, predictive models use bank balance sheet data and financial ratios. However, there are studies that show that forecast accuracy improves when market, macroeconomic and non-financial indicators are included for special countries. Prospects for further research in this area include finding an optimal combination of parametric and non-parametric approaches, investigating the potential of non-financial indicators as factors in bank failures, and research on large samples including both developed and developing countries.