Pricing call warrants with artificial neural networks: the case of the Taiwan derivative market

Abstract

In this paper, artificial neural nets are applied to pricing the call warrants in the Taiwan stock market. Warrants were initialized in Taiwan in 1997 and hence a still very new product. It, therefore, may provide us a chance to test whether artificial neural nets, as a data-driven tool, can be more effective than the model-driven tools in dealing with this emerging derivative market. The data employed in this paper are the two earliest listed stock call warrants, namely, Yageo`s and Pacific Electric Wire and Cable`s warrants, ranging from September 4, 1997 to September 2, 1998. 24 neural nets, covering different inputs, numbers of hidden nodes and transfer functions, were attempted. Each neural net was trained for 20 independent runs. Based on the average of the in-sample performance, the best neural net was selected to compete with the Black-Scholes model and binomial model in the post-sample data. The post-sample performance of each model was evaluated by statistics. We found that the neural net model outperformed both the Black-Scholes model and the binomial model in almost all criteria