This study combines wavelet-based feature extractions with kernel partial least square (PLS) regression for international stock index forecasting. Wavelet analysis is utilized as a preprocessing step to decompose and extract most important time scale features from high dimensional input data. Owing to the high dimensionality and heavy multi-collinearity of the input data, a kernel PLS regression model is employed to create the most efficient subspace that keeping maximum covariance between inputs and outputs, and perform the final forecasting. Compared with neural networks, pure SVMs or traditional GARCH models, the proposed model performs best. The root-mean-squared forecasting errors are significantly reduced.