The Application of Deep Learning Models in Capital Structure Optimization

Abstract

This study investigates the application of deep learning models in optimizing corporate capital structures, integrating firm-level financial indicators and macroeconomic variables to enhance predictive accuracy and decision-making. Using a mixed-method approach, the research employs recurrent neural networks (LSTM), convolutional neural networks (CNN), and transformer-based architectures to forecast optimal debt-to-equity ratios that minimize the weighted average cost of capital (WACC). The models are trained on a large-scale panel dataset, evaluated through root mean squared error (RMSE), mean absolute error (MAE), and robustness checks under macroeconomic shocks. Results reveal that transformer-based models consistently outperform traditional econometric and machine learning approaches, providing more accurate and adaptive forecasts.Key findings highlight profitability, firm size, and tax shields as the most influential determinants of capital structure, consistent with classical financial theories but dynamically quantified through SHAP-based interpretability techniques. Industry-specific analyses demonstrate significant variation in optimization outcomes, suggesting that tailored capital structure strategies are more effective than universal benchmarks. The study further shows that deep learning models maintain predictive stability under volatile economic conditions, such as inflationary pressures and GDP contractions, thereby enhancing financial resilience. Importantly, the integration of explainable AI ensures that recommendations are transparent and interpretable for corporate managers, investors, and regulators.Overall, the findings underscore the potential of deep learning as a paradigm shift in capital structure optimization, bridging theory and practice with a data-driven, adaptive, and explainable framework. The study provides both academic contributions and practical implications, positioning deep learning as a critical tool for modern corporate finance in uncertain global environments.