Теоретичні та прикладні питання економіки. Збірник наукових праць.

Випуск 1 (50)

УДК 336.711/.717+336.77:330.567.22]:004.85(477)

JEL G32, C38

ORCID ID https://orcid.org/0009-0008-7017-0175

DOI https://doi.org/10.17721/tppe.2025.50.7

Krasovytskyi D. O., PhD student

Taras Shevchenko National University of Kyiv

daniel.krass2112@knu.ua

ANALYSIS OF THE IMPACT OF INDICATORS AVAILABLE IN THE CREDIT REGISTRY ON THE PROBABILITY OF DEFAULT OF INDIVIDUAL BORROWERS

The article examines the effectiveness of using the indicators available in the Credit Registry of the National Bank of Ukraine to assess the probability of default among retail borrowers. The study's relevance is driven by the need to improve the accuracy of credit risk forecasting in Ukraine's banking sector, especially during the ongoing war. The study aims to evaluate the predictive power of variables from the Credit Registry, including age, the number of days overdue, the debt-to-service-income ratio (DSTI), and other characteristics, as well as to determine their impact on default probability. The methodology uses multifactor regression analysis, logistic regression, and modern machine learning methods. This study emphasizes the benefits of combining traditional econometric models with machine learning approaches to improve model robustness and predictive performance. Statistical approaches such as Information Value (IV) and Variable Importance (VI) were applied to identify the key determinants of credit risk. The results confirm that data from the Credit Registry are a reliable source for assessing default probability. The most significant predictors of borrower default are the DSTI and the number of overdue days, while the influence of age and income is less pronounced. The application of the IV method revealed that most variables have moderate to high predictive value, emphasizing their importance in building models. The findings highlight the considerable potential of using Credit Registry data to optimize credit risk management processes. The models demonstrated high classification quality, indicating the suitability of the Credit Registry's indicators for modeling the default probability of retail borrowers. Future research prospects include expanding the analysis to assess the impact of macroeconomic factors, such as GDP growth or inflation, on default probability. The approach can be further enhanced through real-time data updates and the incorporation of systemic risk indicators to increase resilience in times of economic stress. Additionally, integrating supervised and unsupervised machine learning methods could enable the segmentation of borrowers by risk level, facilitating more precise management of credit portfolios.

Keywords: machine learning, credit risk, default probability, credit registry.

Full Text: PDF

DOI: https://doi.org/10.17721/tppe.2025.50.7