Options Pricing Using Neural Network Models and Variations on Volatility

Abstract

In this study, various configurations of options pricing neural network models were developed, trained and tested. Their pricing performances were evaluated against each other and against the standard Black-Scholes options pricing model in order to determine the optimal neural network model configuration for options pricing. In particular, both neural network pricing models and volatility models were combined and tested with three different volatility inputs: historical volatility, Black-Scholes implied volatility, and GARCH-derived volatility. Models were also tested with and without the risk-free interest rate parameter, and a 10-day stock price percent change parameter newly introduced in this study. In total, 48 different configurations of neural network models were developed and tested. Among other conclusions, this study found that the neural network models were able to adapt to the volatility grimace and significantly outperform the Black-Scholes model, and that inclusion of the 10-day stock price percent change parameter increased the pricing performance of most neural network models in the study.