Support Vector Machine Combined with GARCH Models for Call Option Price Prediction

The traditional BS evaluation model for options assumes volatility as a constant, and is unable to explain phenomena such as leptokurtic distribution and volatility clusters. In order to supplement this shortcoming, scholars have begun to use linear and non-linear GARCH models to estimate volatility. However, a consistent result has not been achieved with empirical analysis of various different volatility model estimations. Therefore, this study seeks to use symmetrical and asymmetrical GARCH models, with the Taiwan Electronic Sector Index Options (TEO) call as research target to estimate the volatility of each options target, and compare the results obtained from each model. Empirical results show that on average, TBGARCH has the smallest volatility, with most even volatility trends. Then, seven different types of volatility are used as the input variables for the support vector machines (SVM) and back-propagating artificial neural network (BPN), to establish the options prices predictive model. Prediction results show that overall, estimation models such as EGARCH, GJR-GARCH, and TBGARCH have better predictive performance; the performance of the artificial intelligence tools SVM and BPN are clearly better than the traditional BS model, and SVM has better predictive results than BPN.