Document Type : Original Article
Authors
1 Department of Mathematics, Allameh Tabataba'i University, Tehran, Iran
2 Department of Mathematics, Allameh Tabataba'i University, Tehran, Iran
3 Department of Mathematics, Allameh Tabataba'i University, Tehran, Iran
Abstract
In the calibration of a financial model, the process is an optimization task that can be viewed as an inverse problem. Solving this problem typically necessitates having an appropriate pricing function. With recent breakthroughs in machine learning, i.e., physics-informed neural networks (PINNs), a cutting-edge approach that combines artificial neural networks with fundamental physical principles, the task can be efficiently implemented by the inverse physics-informed neural network (iPINN). This paper centers on solving an inverse problem for a financial P(I)DE model by means of iPINN. Firstly, we model the bond price dynamics of catastrophe bonds (CAT bonds) and derive a partial integro-differential equation (PIDE) through a no-arbitrage strategy within the framework of an incomplete market. Thereafter, we employ the iPINN to estimate a specific parameter of the model, namely the market price of risk. The market price of risk is treated as a global learnable parameter and is embedded directly into the PIDE operator. The proposed iPINN is evaluated through three practically meaningful criteria: repricing error, parameter stability, and PIDE residual consistency. The outcomes demonstrate that iPINNs have the capability to solve the inverse problem effectively, and this technique could be applied broadly to real-world data.
Keywords
[2] Büchel, P., Kratochwil, M., Nagl, M. et al., Deep calibration of financial models: turning theory into practice., Rev Deriv Res 25, 109–136 (2022). https://doi.org/10.1007/s11147-021-09183-7.
[3] D. Ahmadian, O. Farkhondeh Rouz, K. Ivaz and A. Safdari-Vaighani, Robust numerical algorithm to the European option with illiquid markets, Applied Mathematics and Computation 366, 124-693, (2020).
[4] Dimitroff, Georgi and R¨oder, Dirk and Fries, Christian P., Volatility model calibration with convolutional neural networks (2018), https://ssrn.com/abstract=3252432 or http://dx.doi.org/10.2139/ssrn.3252432
[5] Engl, Heinz W. and Hanke, Martin and Neubauer, Andreas, Regularization of Inverse Problems, Springer, ( 1996).
[6] Jacot, Arthur and Gabriel, Franck and Hongler, Cl´ement, Neural Tangent Kernel: Convergence and Generalization in Neural Networks,dvances in Neural Information Processing Systems (NeurIPS), (2018).
[7] Jamalpour Malekabadi, Moslem and Zakeri, Ali and Salehi Shayegan and Amir Hossein, Application of Radial Basis Functions Meshless Method for Solving an Inverse Parabolic Problem, Journal of Mathematics and Modeling in Finance 6(1), 251-263, (2026).
[8] Liu, S., Borovykh, A., Grzelak, L.A. et al., A neural network-based framework for financial model calibration, J.Math.Industry 9, 9 (2019).
[9] Liu, S., Borovykh, C. W., and Bohte, S. M., Pricing options and computing implied volatilities using neural networks., Risks, 7(1), (2019)
[10] M. Raissi, P. Perdikaris, G.E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics, 378 , 686-707,(2019).
[11] Siddhartha Mishra and Roberto Molinaro, Estimates on the generalization error of physicsinformed neural networks for approximating a class of inverse problems for PDEs, IMA Journal of Numerical Analysis, 42(2), 981-1022, (2022).
[12] Raeisi Makiani, M.A., Neisy, A. & Safdari-Vaighani, A. Pricing Catastrophe Risk Bond Using the Physics-Informed Neural Network., Comput Econ, (2025), https://doi.org/10.1007/s10614-025-11180-z.
[13] Safdari-Vaighani, A., Ahmadian, D. & Javid-Jahromi, R. An Approximation Scheme for Option Pricing Under Two-State Continuous CAPM. , Comput Econ,57, 1373–1385, (2020).
[14] Tikhonov, A. N. and Arsenin, V. Y., Solutions of Ill-Posed Problems, Wiley, (1977).
[15] Wang, Shu and Yu, X, On the convergence and generalization of physics-informed neural networks, Journal of Machine Learning Research, (2022).
[16] Wang, Sifan and Teng, Yu and Perdikaris, Paris, Understanding and mitigating gradient flow pathologies in physics-informed neural networks, SIAM Journal on Scientific Computing, (2021).
)