Wavelet Kernel Function for Stock Index Forecast

Abstract

Stock index forecast plays an important role in finance.One of the challenging problems in forecasting the stock index is that general kernel functions in support vector machine (SVM) can't capture nonstationary characteristic of stock time series accurately. While wavelet function yields features that describe of the stock time series both at various locations and at varying time granularities, so this letter constructed a multidimensional wavelet kernel function and proved it meeting the mercer condition to address this problem. The applicability and validity of wavelet support vector machine (WSVM) for stock index forecasting were analyzed through experiments on real-world stock data. It appeared that the wavelet kernel is more accurate and performs better than the Gaussian kernel.