Financial Prediction Using Manifold Wavelet Kernel

This paper constructs an admissible manifold wavelet kernel (MWK) for support vector machine (SVM) to forecast the volatility of financial time series based on generalized autoregressive conditional heteroscedasticity (GARCH) model. The MWK is obtained by incorporating the wavelet technique and manifold theory into SVM. Unlike Gaussian kernel in SVM, the MWK can approximate arbitrary nonlinear functions. The applicability and validity of MWK for volatility forecast are confirmed through experiments on simulated data sets.