A Wavelet Denoising Support Vector Regression Ensemble Model for Exchange Rate Prediction

Based on the nonlinear ensemble and level dependent denoising framework, a novel wavelet denoising support vector regression (SVR) ensemble forecasting model is proposed. The proposed model attempts to incorporate the level dependent denoising technique that utilizes the multi scale heterogeneous characteristics of data and noises into the modeling process. Forecasting results based on different wavelet parameters are firstly preprocessed by principle component analysis to reduce dimensionality and noise, then ensembled via SVR to further reduce forecasting biases and improve the forecasting stability. Experiment results reveal that the performance of the proposed approach is statistically superior to those more traditional methods presented in this study in terms of the same measurement.