Option Pricing: The empirical tests of the Black-Scholes pricing formula and the feed-forward networks

In this article we evaluate the pricing performance of the rather simple but revolutionary Black-Scholes model and one of the more complex techniques (neural networks) on the European-style S&P Index call and put options over the period of 1.6.2006 till 8.6.2007. Our results on call options show that generally Black-Scholes model performs better than simple generalized feed-forward networks. On the other hand neural networks performance is improving as the option goes deep in the money and as days to expiration increase, compared to the worsening performance of the BS models. Neural networks seem to correct for the well-known Black-Scholes model moneyness and maturity biases.