Comparative study on early risk warning for corporate bond defaults

Abstract

Accurate prediction of default risk in corporate bonds is essential for maintaining market stability. This paper briefly introduces the decision tree and random forest (RF) algorithms. The RF algorithm was applied to predict default risk in corporate bonds. Moreover, simulation experiments were conducted. Firstly, the feature indices’ effectiveness for predicting default risk in corporate bonds was verified. Secondly, the impact of the number of decision trees on the RF algorithm was tested. Finally, the performance of the support vector machine (SVM), decision tree, back-propagation neural network, extreme gradient boosting, extreme learning machine, and RF algorithms was compared. The results indicated that the selected feature indicators could be utilized for bond default risk prediction. When using 120 decision trees, the RF algorithm exhibited the optimal prediction performance and efficiency. Furthermore, compared to the SVM and traditional decision tree algorithms, the RF algorithm demonstrated superior prediction performance with minimal computational time.