SME default prediction using random forest including nonfinancial features: An empiricial analysis of German enterprises

ABSTRACT

Having accurate company default prediction models is vital for both financial institutions and businesses, especially small and medium enterprises (SMEs). Although these firms play a significant role in the economy of every country, they have a different risk profile than large companies. However, as these companies are not required to publish financial information and, further, display unique market characteristics, related financial data are scarce, which underlines the necessity of a precise and easily adaptable risk model. Our study applies random forests (RFs) to a sample of approximately three million German SMEs. Results show that compared to traditional methods, RFs can make a greater contribution to SME credit-risk evaluation in terms of prediction power and bridge the gap between statistical strength and business interpretability. They also show that, due to the lack of financial data, nonfinancial features play a significant role in risk modeling for these companies.

KEYWORDS:

• Small and medium enterprises
• default probability
• random forest
• nonfinancial features
• feature engineering
• machine learning

Acknowledgments

We express our gratitude to Munich Re and Talaria for sponsoring this research project and providing valuable input during numerous discussions. Further, we thank Frank Sieslack for supporting and advising us with the project. Special thanks also goes to Prof. Matthias Scherer for initiating the collaboration and sharing his profound fundamental and mathematical knowledge in the guidance of the project.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The status “not operational” also includes regular company liquidations that are not linked to insolvency and hence the figure shown for our sample overestimates the actual default ratio.