Option pricing using a committee of neural networks and optimized networks

Abstract

The derivative market has seen tremendous growth in recent times. We look at a particular area of these markets, viz. options. The pricing of options has its roots in stochastic mathematics since option pricing data is highly non-linear. It seems obvious to apply the training techniques of neural networks to this type of data. The standard multi-layer perceptron (MLP) and radial basis functions (RBF) were used to model the data; these results were compared to the results found by using a committee of networks. The MLP and RBF architecture was then optimized using particle swarm optimization (PSO). The results from the 'optimal architecture' networks were then compared to the standard networks and the committee network. We found that, at the expense of computational time, the 'optimal architecture' RBF and MLP networks achieved better results than both unoptimized networks and the committee of networks.