Modeling Probability of Default in the Construction Sector: Factors of Corporate Governance

Abstract

In this paper we estimated the probability of default of large construction companies in Russia using the classic method for this purpose - logistic regression. Our task incorporates testing corporate governance factors and analysis of the predictive power of the model with regularization (Lasso and Ridge). As a dependent variable we tested four definitions of default and compared them. The modeling was performed on the basis of information from the SPARK, Rosstat and the Bank of Russia database for the period 2007-2015 - final sample after elimination of outlying observations consist of 4761 construction companies. The added value of the corporate governance factors is verified on the basis of comparison of ROC-curves (AUC) and I and II errors. Based on international and domestic experience to assess the influence of the corporate structure on the company's stability were formed seven hypotheses, some of which were statistically significant. In particular, other things being equal, the default probability of the company below, if it's CEO is co-owner; the default probability the company above if the company is a subsidiary. Note also that in fact the companies with small board of directors better overcome financial distress (negative return on assets) in the Russian construction business. There was no confirmation of the hypothesis that older companies are less likely to default. Confirmed hypotheses give a new perspective look at a comprehensive assessment of risk large construction companies in the country. According to our estimates corporate governance factors really improved the predictive ability of the models, and regularization methods confirmed stability of these models. Using cross-validation the robustness of the coefficients of the final specification was confirmed. This result may be of interest to a greater extent for banks, commercial investors and partners-contractors.