| Abstract [eng] |
67 pages, 21 tables, 10 figures, 45 references. The aim of this Master’s thesis is to develop and empirically evaluate a trade credit risk assessment model for small and micro enterprises in Lithuania by applying a hybrid methodological framework combining logistic regression and decision tree (CART) methods. The study focuses on forecasting the probability of corporate insolvency (bankruptcy) and assessing the practical applicability of the model in trade credit decision-making processes. The thesis consists of three main parts: a review of scientific literature, the empirical research and its results, and conclusions and recommendations. The literature review examines the main theories of credit risk and bankruptcy prediction methods, with particular emphasis on the specific characteristics of small and micro enterprises. The economic importance of this group of firms in the Lithuanian context, their higher risk exposure, and credit-related challenges arising from information asymmetry are discussed. The analysis provides a detailed overview of financial and non-financial indicators commonly used in assessing the insolvency risk of small enterprises, as well as the possibilities and limitations of decision automation. Traditional and modern credit risk assessment methods are critically evaluated, substantiating the choice of a hybrid model based on logistic regression and a decision tree (CART) in this study. Following the literature review, an empirical analysis is conducted in which the credit risk of small and micro enterprises is assessed using financial statements obtained from the Lithuanian Centre of Registers. Model development is based on a hybrid methodology combining logistic regression and a decision tree (CART), implemented using the R programming environment. The study includes a comprehensive data diagnostics procedure encompassing missing data analysis, outlier assessment, correlation analysis, Mann–Whitney tests with Benjamini–Hochberg correction, chi-square tests of independence, multicollinearity assessment (VIF), ROC/AUC analysis, and model calibration techniques (Hosmer–Lemeshow, Brier score, ECE). These methods enable the construction of a statistically sound and practically applicable credit risk assessment model for the segment of small and micro enterprises. The results of the empirical analysis indicate that the developed credit risk model for small and micro enterprises is empirically robust and effective. Logistic regression reveals that firm age and the number of employees are the strongest factors reducing bankruptcy risk, while financial ratios provide additional explanatory power. The CART decision tree identifies clear threshold values for ROA, profitability, and indebtedness, allowing for accurate differentiation of risk profiles. Both models demonstrate adequate discriminatory power (AUC ≈ 0.72–0.75), while calibration procedures improve the accuracy of probability forecasts. In summary, logistic regression and the decision tree form a complementary system that enhances decision objectivity and improves the identification of high-risk enterprises. The findings suggest that the developed models can be practically applied in trade credit risk assessment, while the insights obtained are valuable for institutions seeking to implement data analytics and automated risk evaluation technologies. |
