Credit Risk Assessment of Listed Companies Based on Long Short-Term Memory Neural Networks

Abstract

Listed companies are vital to capital markets, but issues like information opacity and poor governance elevate credit risks, impacting economic stability. This study proposes a Long Short-Term Memory (LSTM) neural network model to assess credit risk by analyzing time-series financial indicators. Factor analysis reduces dimensionality of indicators, followed by LSTM training on sequential data to predict risk levels. Using CSI 300 firms’ data, the model achieves 87.5% accuracy, outperforming traditional methods like logistic regression (80.2%). The approach captures temporal dependencies, offering dynamic risk forecasts. Limitations include data quality reliance and computational complexity. Results support regulators and investors in enhancing risk management.