Pricing path-dependent exotic options with flow-based generative networks

In this study, we aim to significantly reduce the computational time for pricing path-dependent exotic options using a flow-based generative model called RealNVP (Dinh et al., 2016). The flow-based generative network learns simulated large-scale two-dimensional random states based on two stochastic volatility (SV) models. As a result, the generative network can efficiently simulate the random states within a short time. Furthermore, they can provide explicit probability density functions for the SV models due to the unique advantage of flow-based generative models. These lead to fairly exact option prices being achieved by simulating random states with the network or integrating option payoffs for the network-based density. Finally, we compare the network-based prices with those of naive Monte-Carlo simulation in terms of accuracy and time cost to show the superior performance of the proposed method.