The Study of Option Pricing Problems based on Transformer Model

Abstract

Option pricing is an important topic in the field of quantitative finance. The traditional Black-Scholes model formulation requires a large number of assumptions, which often does not exist in practice, and the statistically-based regression analysis and time series methods have poor fitting ability for non-stationary data. Deep learning has advantages over traditional econometric models in identifying the structure and patterns of data, and can effectively learn the nonlinear and non-stationary characteristics of time series, which is more suitable for the study of option pricing problems. The Transformer model has greater advantages over the traditional recurrent neural network model in the processing of time series data, mainly in terms of performance and speed. In this work, we will compare different models and get the deep learning model with the strongest prediction ability. Based on the collected data related to 50 ETF options and stocks in the Chinese market for empirical analysis, it is demonstrated that the Transformer model outperforms the traditional deep learning model in time series prediction.