Robust and Explainable Deep Hedging with Linearized Neural Network
Abstract: Deep hedging is promising for risk management for financial derivatives through deep learning, yet it remains hindered by complex, resource-intensive training and the challenge of effectively integrating deep neural networks with hedging optimization. To overcome these issues, we introduce a robust and efficient linearized neural network architecture, seamlessly integrated with Black-Scholes' Delta, to streamline deep learning-based hedging optimization (DHLNN). Our approach enhances both the efficiency and interpretability of hedging strategies in derivative markets. The proposed model shows strong resilience to market fluctuations, effectively addresses action-dependence challenges, and achieves faster convergence compared to existing methods. Extensive simulations confirm the superior performance and cost-effectiveness of our method, under varying market conditions, when compared to state-of-the-art deep hedging models. These findings underscore the potential of DHLNN to significantly improve both convergence and hedging performance in derivative markets.
Submission Length: Long submission (more than 12 pages of main content)
Assigned Action Editor: ~Yan_Liu1
Submission Number: 3225
Loading