Credit rating prediction with supply chain information: a machine learning perspective

In this paper, we adopt an ensemble machine learning framework—a Light Gradient Boosting Machine (LightGBM) and develop an algorithmic credit rating prediction model by innovatively incorporating firms’ extra supply chain information both from suppliers and customers. By utilizing data from listed firms in North America from 2006 to 2020, our results find that the accuracy of the prediction improves by incorporating supply chain information in the previous year, compared to the inclusion of supply chain information in the current year. Besides, we identify the most important factors the stakeholders should pay attention to. Interestingly, we show that the models utilizing the current year’s information perform better after the strike of the COVID-19, indicating that the epidemics may have accelerated the spread of credit risk along the supply chain. Furthermore, supplier information is found to be more valuable than customer information in predicting the focal firm’s credit rating. A comparison of our framework with the existing methods vindicates the robustness of our main results.