Natalia Nehrebecka

Predicting the Default Risk of Companies. Comparison of Credit Scoring Models: LOGIT vs Support Vector Machines

Abstract The aim of the article is to compare models on a train and validation sample, which will be created using logistic regression and Support Vector Machine (SVM) and will be used to assess the credit risk of non-financial enterprises. When creating models, the variables will be subjected to the transformation of the Weight of Evidence (WoE), the number of potential predictions will be reduced based on the Information Value (IV) statistics. The quality of the models will be assessed according to the most popular criteria such as GINI statistics, Kolmogorov-Smirnov (K-S) and Area Under Receiver Operating Characteristic (AUROC). Based on the results, it was found that there are significant differences between the logistic regression model of discriminatory character and the SVM for the model sample. In the case of a validation sample, logistic regression has the best prognostic capability. These analyses can be used to reduce the risk of negative effects on the financial sector.

Construction measures of bankruptcy: case study of Polish enterprises

The paper presents a model assigning a bankruptcy probability to a company, developed on the basis of individual data from balance sheets and income statements of Polish companies in the 2001-2010 period. The research was based on a logistic regression performed on categorised variables transformed using a weight of evidence approach. In the forecasting model of a possible bankruptcy in a years horizon the highest weight was assigned to the indicator for the ability to cover financial costs which explained the companys ability to meet the interest payments and capital costs. In the more generalised model which accounts for the macroeconomic situation the most important was the indicator for the ability to pay off debt. In the model forecasting bankruptcies three-years in advance - the early warning model - no dominant indicator was found.