Mathematical Physics
Juncai Pu, Yuqi Xiang, Yong Chen
Nonlinear option pricing models are of significant research value as they better reflect the realities of financial markets, yet their numerical solution remains highly challenging. On the one hand, such models typically involve strong nonlinearity, multi-scale features of small parameters, and high sensitivity to initial data, which often make it difficult for traditional numerical methods to maintain stability and accuracy. On the other hand, many deep learning methods rely on boundary data, while in real financial markets boundary conditions are often unavailable, thereby limiting their applicability. Thus, deep learning methods that rely solely on initial data still face significant challenges in efficiently solving nonlinear option pricing models and achieving effective numerical predictions. To address these difficulties, this work employs the respecting causality physics-informed neural network (RCPINN), which depends solely on initial data and respects the inherent spatiotemporal causal structure of system evolution, enabling it to effectively handle the complex characteristics of nonlinear financial models. In the Ivancevic nonlinear option pricing model, the RCPINN successfully predicts the formation and evolution of both low-order and high-order financial rogue waves, revealing the underlying dynamical mechanisms of extreme localized waves. In the nonlinear Black–Scholes transaction cost model, the RCPINN effectively captures the dynamical evolution of European call and put option prices. This work not only validates the effectiveness and applicability of RCPINN in predicting financial rogue waves and European option prices within the framework of nonlinear option pricing models, but also demonstrates its advantages in handling initial data sensitivity, multi-scale features, and strong nonlinearity in financial problems. More importantly, this approach provides a novel technical pathway for both early warning of extreme financial risks and the simulation of European option price evolution, while the findings offer valuable insights for applications in derivative pricing, exchange rate forecasting and financial market risk monitoring and early warning.