This study proposes a hybrid model based on Long Short-Term Memory (LSTM) networks for corporate loan default prediction. Leveraging multidimensional data from China’s A-share listed companies (2000–2023), the framework innovatively incorporates secondary market dynamics—including stock price fluctuations, trading volumes, and market indices—alongside traditional financial ratios and macroeconomic indicators to assess credit risk. Addressing challenges of temporal dependencies and class imbalance, synthetic minority oversampling (SMOTE) is applied to enhance data representativeness, while hyperparameter optimization balances model sensitivity and stability. Empirical results demonstrate the model’s efficacy in identifying default risks, with improved early warning capabilities for financial institutions. Key contributions include: 1) establishing a cross-market predictive framework that empirically validates the informational linkage between stock market efficiency and corporate credit health, 2) demonstrating LSTM’s superiority in capturing nonlinear, time-dependent patterns in financial data, and 3) advancing systemic risk mitigation by integrating capital market signals into banking risk management. The study highlights the complementary role of secondary market indicators to conventional financial metrics, offering actionable insights for loan decision-making and resource allocation. This work provides a scalable methodology for financial fragility analysis, bridging capital market dynamics with credit risk assessment in emerging economies.