Default Prediction for Auto Repair Services Firms in Russia Using Non-financial Data

Финансы и бизнес. 2024. Vol. 20. No. 3. P. 71–88.

Afanasev V.

The conventional approach to default prediction implies using financial ratios as predictors. This paper provides evidence for improvement in the quality of default prediction for auto repair firms if non-financial data is included in the models. The study uses a sample of more than 200 firms, which defaulted in 2018–2023 and 10 healthy peers samples (more than 2200 observations in total). Results revealed that using only financial ratios for auto repair firms default prediction results in low accuracy. However, adding non-financial data increases the accuracy by 7 p.p. Furthermore, this study pays specific attention to correct and transparent calculation of the values of the explanatory variables: two approaches to choose the theoretical forecast date are suggested. This study may be of interest to credit organizations and auto repair firm counterparties.

Keywords: машинное обучение, кредитный риск, machine learning, случайный лес, credit risk, random forest, default prediction, non-financial data, прогнозирование дефолта, нефинансовые данные, auto repair firms, авторемонтные фирмы