| Abstract [eng] |
The objective of this final master's thesis is to assess credit risk prediction methods for their accuracy in predicting the probability of default for Small and Medium-sized Enterprises (SMEs) in Lithuania. The study is based on an analysis of scientific literature and research that involved the development of a new logistic regression model using financial and non-financial data from 2022 to evaluate the accuracy of credit risk for SMEs in Lithuania. The thesis is structured into four main sections: analysis of scientific literature, research methodology, research results, and conclusions and suggestions. The literature analysis explores various interpretations of the "credit risk" concept from different scientific sources, as well as highlighting distinctions between SMEs and large companies in terms of credit risk. Additionally, it examines credit risk models applied in Lithuania, with a focus on their suitability for local businesses. The analysis also includes a review of credit risk models specifically tailored to SMEs, emphasizing the importance of incorporating not only financial but also non-financial information, such as the sector, the company's years of operation, and the number of employees, to improve the accuracy and applicability of credit risk assessment models. Following the literature analysis, the research methodology was established. The methodology included a detailed discussion of the approach for creating a new logistic regression model for the assessment of SME credit risk, specifically tailored to SMEs operating in Lithuania, and identified that existing models suitable for analysing SME credit risk could not be applied, as the study aimed to rely exclusively on publicly available information for SMEs operating in Lithuania. Research results indicate that the new logistic regression model, developed using exclusively publicly available information, achieves high accuracy, reaching 93.11%, in predicting SME credit risk in Lithuania. Its adaptability and reliance on publicly accessible data enhance its practical value and broad usability. Furthermore, correlation analysis, assessment of multicollinearity, and the results of the Hosmer-Lemeshow test confirm the reliability and goodness-of-fit of the new model. Notably, the model's Area Under the Curve (AUC) reaches 0.96, indicating exceptional efficiency. Therefore, the model is identified as the most suitable for assessing the credit risk of SMEs operating in Lithuania. The conclusions summarize key findings from the scientific literature and provide suggestions on how to enhance the model’s accuracy, such as incorporating macroeconomic indicators like inflation and unemployment rates, testing the model across different economic cycles to ensure its robustness and adaptability, and segmenting the analysis by regions to capture potential regional economic disparities and improve the model's precision. |
