Option Data Fusion Incorporating Moment-Matching and Dynamic-Matching Simulations

Pricing financial derivatives has been a prevalent topic since the 1970s, and with the development of deep learning technologies in recent years, this area has experienced significant innovation and transformation. However the quality and quantity of available data used for training deep learning models remain critical challenges. This paper proposes two robust simulation methods to generate data for training deep learning models for option pricing. By incorporating moment-matching technique and VAR model, we preserve the distributional characteristics and dynamic features of moneyness and implied volatility from the original data. A significant advantage of our methodologies is their model-free nature, which enables the generation of substantial amounts of high-quality data even from a small sample. These simulation methods effectively address the data quantity and quality challenges commonly encountered in deep learning applications, thereby enhancing the performance of option pricing models. Our approach ensures that the gener-ated data accurately reflects the underlying market dynamics, enabling more precise predictions and risk assessments.