Financial time series forecasting based on wavelet and multi-dimensional Taylor network dynamics model.

Presented in this paper is a new approach to establishment of the dynamics model of multidimensional Taylor network and its parameter identification, whereby a method based on the model and wavelet for financial time-series forecasting is proposed. The financial time series are decomposed into sub-series of a low frequency signal and several high frequency signals via Mallat algorithm, for each of which a multi-dimensional Taylor network model is established. Model parameters are trained by conjugate gradient method, and then the models are used for forecasting. All forecasting results are superimposed to obtain the predicted value of the original time series. As verified by our experiments, the proposed method works well in ensuring the accuracy of financial time-series forecasting. [ABSTRACT FROM AUTHOR]