2025, Vol. 12, No. 4. - go to content...

Permanent address of this page - https://resources.today/en/34ecor425.html

Метаданные этой статьи так же доступны на русском языке

DOI: 10.15862/34ECOR425 (https://doi.org/10.15862/34ECOR425)

Full article in PDF format (file size: Файл не найден)

For citation:

Shurzhunov N.V. Comprehensive methodology for assessing credit risk of real estate development companies based on open data and machine learning in accordance with IFRS 9 and Bank of Russia requirements. Russian journal of resources, conservation and recycling. 2025; 12(4). Available at: https://resources.today/PDF/34ECOR425.pdf (in Russian). DOI: 10.15862/34ECOR425

Comprehensive methodology for assessing credit risk of real estate development companies based on open data and machine learning in accordance with IFRS 9 and Bank of Russia requirements

Shurzhunov Nikita Valeryevich
Moscow University «Synergya», Moscow, Russia
E-mail: alotholl@gmail.com
RSCI: https://elibrary.ru/author_profile.asp?id=1180691

Abstract. In the context of transformation of housing construction financing mechanisms in Russia, the task of improving methods for assessing credit risk of real estate development companies becomes relevant. This work presents a comprehensive methodology for predicting the probability of default of developers, integrating traditional financial ratios with alternative data sources: escrow coverage indicators, construction dynamics and arbitration disputes. To validate the methodology, a synthetic dataset (500 companies, default rate 8 %) was used, generated by the Gaussian Copula method based on the empirical correlation structure of the Polish Companies Bankruptcy Dataset with calibration to the Russian market according to data from the Federal Tax Service of Russia, the Central Bank of the Russian Federation and DOM.RF. Comparative analysis showed the superiority of the Random Forest model (AUC-ROC = 0,763, Gini = 0,525) over logistic regression (AUC = 0,695) and XGBoost (AUC = 0,700). The inclusion of alternative data provided an increase in predictive power of 6,6–13,8 percentage points depending on the model. Feature importance analysis revealed the critical role of escrow coverage (8,1 %) and arbitration disputes (5,7 %) as early indicators of financial problems. The methodology complies with the requirements of IFRS 9 and regulatory acts of the Bank of Russia (Regulations No. 590-P, No. 845-P, No. 716-P).

Keywords: credit risk; developers; project financing; escrow; machine learning; Random Forest; XGBoost; IFRS 9; open data; synthetic dataset

Download article in PDF format