Neural Network Models for Option Pricing

Abstract

There exist several different ways to evaluate financial derivatives but, in general, closed-form formulas, such as Black and Scholes, tend to provide unsatisfactory results. Therefore, nowadays, thanks to the increased computational capability of machines numerical methods are commonly used. The aim of this dissertation is to develop a nonparametric supervised machine learning method, namely a Multilayer Perceptron Feedforward Artificial Neural Network, to price financial options written on the FTSE MIB index. It means we try to implement a data-driven approach which, by exploiting the architectural structure of a multi-level neural network, is able to correctly identify the value of the analyzed derivative. In particular, the function used to train the algorithm is the Levenberg-Marquart backpropagation and the performance is evaluated by relying on the Root Mean Square Error (RMSE).