A Powerful Neural Network Method with Digital-contract Hints for Pricing Complex Options
Jun Lu, and Hiroshi Ohta
Department of Industrial Engineering, Graduate School of Engineering, Osaka Prefecture University, Sakai, Osaka 599-8531, Japan
Many researches have proved that common neural network methods outperform parametric methods for option pricing. However, performance of the common neural network method usually suffers from the non-stationary and noisy properties of observed financial data. In this paper, we propose some parametric digital-contract (DC) hints, which can be utilized as auxiliary information to guide a neural network’s learning process about target pricing formula, and thus can be expected to get a better pricing performance in the case of observed data with noise. The DC hints are incorporated into a neural network with serial and parallel forms. Some Monte Carlo simulation experiments are performed and demonstrated that both the two forms not only have the nonparametric method’s advantages like generalization and superior accuracy, but also have the parametric method’s robust property to financial data with noise. The results also show that these two forms have their own strengths and limitations.