The role of alternative data in micro-enterprises’ credit risk assessment in China — Empirical evidence based on machine learning

Applying alternative data in credit reporting can enhance financial institutions' ability to assess the credit risk of micro-enterprises. This paper uses data from an internet bank serving micro-enterprises and categorizes alternative data into historical credit data (HCD) and behavioral data—the latter including economic transaction data (ETD) and social stability data (SSD). The random forest method is employed to compare these data types in credit risk assessment. The findings reveal that multi-dimensional alternative data holds significant credit value for micro-enterprise risk assessment and can improve the predictive performance and stability of the models. The behavioral data-based models demonstrate superior risk identification capability compared with the HCD-based one. ETD generally outperforms SSD in assessment, though SSD is more stable under external shocks. While external environmental shocks may reduce the model's precision in detecting potential defaults, the model's overall ranking stability remains intact. Notably, the integration of multi-dimensional alternative data can mitigate data volatility through feature complementarity, thereby enhancing model robustness.