Deep learning approach to American option pricing

This study focuses on pricing the American put option by applying a deep learning-based algorithm under the double Heston model. The double Heston model is a multi-factor stochastic volatility model that offers more flexibility in modeling the volatility term structure and better empirical fit to option prices compared to one-factor models. The option price derivation under this model leads to a linear complementarity problem. To solve this problem, we utilize the deep Galerkin method (DGM), which is a method based on deep learning. Our numerical results show the efficiency and accuracy of the algorithm as evidenced by comparing it with the antithetic variable Least-square Monte Carlo (AV-LSM) method.