Prediction of Stock Pledge Financing Default Based on Machine Learning

Abstract

. Stock pledge has become a widely used financing method among controlling listed companies, while its default causes significant challenges to financial stability and risk management. This study investigates default prediction in stock pledge financing by identifying key indicators for early warning signals using machine learning. The dataset contains information on the default status of pledge financing of controlling shareholders of Chinese listed companies from 2017-2020 are used to construct training and test sets for modelling. The performance of several models is compared, including Logistic Regression, Random Forests, and Gradient Boosting. The findings suggest that the Gradient Boosting model performs the best, achieving an AUC of 0.99 and an accuracy of 0.96. The feature importance analysis further points out key indicators related to default risk, providing valuable insights for risk monitoring and preventive interventions.