Modelling option prices using neural networks

Abstract

In this research, modelling of the European option prices of S&P 500 index options was carried out using Multi-layer Perceptron Neural Networks. The goal was to train the neural networks using historical data to accurately determine option prices, given the index price, strike price and time to expiry as inputs. There is no hard and fast formula for pricing options, with the exception of the Black Scholes model, which is only a theoretical model and often under-performs in practical applications. Therefore, developing a model for pricing real options is of great importance, and Neural Networks have the potential to be vital vehicles to that end. That is what motivated this study. Di erent results with respect to accuracy are achieved by partitioning the data according to moneyness of options, with the Neural Network performing exceptionally for in-the-money options, but poorly for out-of-the-money options. This suggest that in a volatile market the neural network outperforms the Black Scholes model for in-the-money options, however the Black Scholes model is still better for at-the-money options.