Optimal multi-scale time series decomposition for financial forecasting using wavelet thresholding techniques

Wavelet analysis as a recently data filtering method (or multi-scale decomposition) is particularly useful for describing signals with sharp spiky, discontinuous or fractal structure in financial markets. This study investigates the optimal several wavelet thresholding criteria or techniques to support the multi-signal decomposition methods of a daily Korean won / U.S. dollar currency market as a case study, specially for the financial forecasting with a neural network. The experimental results show that a cross-validation technique is the best thresholding criterion of all the existing thresholding techniques for an integrated model of the wavelet transformation and the neural network.