Bankruptcy Prediction Models of Small and Medium Construction Industry Firms of Russia

The subject of research are the small and medium firms of the Russian construction industry. The object of the research is the probability of default factors in one, two and three years prior to the default. The data was obtained from SPARK database and includes financial and non-financial variables. The sources of financial factors are Balance Sheet and Profit and Loss statements of the firms in 2015-2017. The target variable is the default in 2018, which is defined according to the methodology in Big4 and the banks. The applied methods are Logit with variable grouping and Random Forest algorithm. The novelty of the research is, firstly, inclusion into the analysis of the new factors, which were not considered previously for the Russian construction industry small and medium firms. The second novelty is applying grouping to the construction industry firms’ data, which makes possible to interpret Logit model factors’ impact on probability of default, as opposed to the regular Logit model specification. The findings of the present research are the following. The predictive power of both developed models monotonously decreases with the increase in the prediction horizon. The factor set, which optimally predicts the default probability of construction industry SMEs, differs for each prediction horizon. Moreover, it was found, even the strongest factors for each prediction horizon are dependent on the method being applied. Nevertheless, there are factors, which have high predictive power throughout all periods, but which are specific to the method. The results of the present research can be applied in practice for scorecard development for the Russian construction SMEs.