The Mechanism analysis of Corporate Financial Risk Influences via BiLSTM-MAttention Model

Abstract

Against the backdrop of increasing macroeconomic volatility and growing complexity in corporate operating environments, exploring the intrinsic mechanisms underlying financial risk factors is of great significance for enhancing firms’ capabilities in risk identification and prevention. Based on quarterly data of Chinese A-share listed companies from 2020 to 2024, this study proposes a BiLSTM-MAttention model constructed on quarterly financial indicators. Compared with prior studies relying on annual reports, quarterly data enable the model to capture the dynamic fluctuations of corporate financial risks with higher frequency, thereby improving its sensitivity and timeliness in risk detection. The BiLSTM structure effectively extracts temporal dependencies among financial indicators, while the Multi-Head Attention mechanism (MAttention) further achieves deep feature integration and adaptive weight allocation, strengthening the model’s ability to identify financial risks. Furthermore, the SHapley Additive exPlanations (SHAP) approach is introduced to enhance model interpretability by quantifying the marginal contribution of each financial indicator to the prediction results, thereby revealing the underlying mechanisms of risk formation. Empirical results demonstrate that the proposed model not only outperforms traditional interpretable machine learning models in predictive accuracy but also elucidates the heterogeneous predictive importance of various indicators at the feature level. These findings provide valuable insights for government regulators and corporate managers in improving financial risk prevention frameworks and enhancing overall risk governance capacity.