Investigation of alternative methods of option pricing

ABSTRACT This dissertation comprising part of a Master Course in Computational Finance investigates the performance of binomial method and neural networks when they mimic Black-Scholes function in pricing European call option. A binomial option pricing calculator with GUI was written in java and an artificial data set was created by using Excel based on Black-Scholes formula. The data set was divided into three subsets for the purpose of training, validating and testing artificial neural networks. The experiments of artificial neural networks architecture were carried out by using MATLAB neural network toolbox. I conclude that binomial converges to B-S model as the number of time steps increases. But there is round-off error should be aware of. Also I find that training function Resilient Backpropagation (trainrp) and Levenberg-Marquardt (trainlm) are capable of perform the task of European option pricing. From my experiments, the network using trainrp with eight neurons in the middle layer and the network using trainlm with ten neurons in the middle layer can produce option prices most closet to B-S results