Comparative Analysis of Bankruptcy Predicting Models for Small Business

The problem of a credit risk assessment is topical and very important because of varied reasons, including a possibility for bank activity optimization and a danger of a financial crisis analogous to 2007-2008. Bankruptcy is one of the most significant part of the credit risk. The controversial statistics on bankruptcy among Russian companies is a good incentive for further research of this phenomenon. There are a lot of models predicting a bankruptcy risk. However most of them are foreign and do not fit for contemporary Russian economy. This justifies the novelty and importance of the current paper. The objective of this work is to identify best model predicting bankruptcy of small firms. Along with this task in paper were performed next tasks: - Overview of theoretical aspects of bankruptcy; - Analysis of statistics on Russian companies’ bankruptcies; - Comparative analysis of traditional approaches predicting bankruptcy risk. This research analyses 195 companies during the period from 2011 to 2015. 56 indicators, including financial and non- financial variables were calculated for these firms. All developed models were compared against the benchmark model of Altman. This work considers such parametric and non- parametric methods of classification as logistic regression, linear and non- liner discriminant analysis, decision tree and k nearest neighbors approach. Results of analysis show that the boosting option of the decision tree model possess the greatest bankruptcy prediction capacity among all applied methods. Considering all accuracy features it also outperforms Altman model. Boosting correctly classifies 79% of real bankrupts and 100% of companies with sustainable financial position. The overall accuracy is 96%. The Altman model properly identify 72.7% of defaulted firms and 67% of companies without any financial problems. The overall accuracy for the Altman model is 69.3%.