The implementation of dynamic heteroskedasticity convertible SVM model in financial time series

In this paper, in order to overcome the deficiency of the traditional SVM, a positive mapping between price volatilities and sample periods of underlying financial time series has been assumed according to the theorems of behavioral finance. By embedding this mapping into the constraint equations of the classic SVM algorithm, an improved SVM model named DHC-SVM (Dynamic Heteroskedasticity Convertible SVM) is proposed for the financial price forecasting. Comparing to the classic SVM model, the experimental results on the real-time HS300 index data illustrates that the DHC-SVM has the advantage both in higher accuracy and better stability.