Pca Based Support Vector Machine Technique for Volatility Forecasting

Conditional Volatility of stock market returns is one of the major problems in time series analysis. Support Vector Machine (SVM) has been applied for volatility estimation of stock market data with limited success, the limitation being in accurate volatility feature predictions due to general kernel functions. However, since Principal Component Analysis(PCA) technique yields good characteristics that describe of the volatility time series in terms of time varying risk estimates, this paper focuses designing PCA based SVM technique (PCASVM)to overcome the above limitations. Gaussian kernel trick has been invoked to build the learning mechanism of non-linear classifier.The applicability of this proposed combinatorial structure (PCASVM) for forecasting the volatility has been confirmed through experiments on real time bench mark data sets. The comparative study of PCASVM with SVM shows the superiority of the proposed technique.