Multiscale wavelet decomposition based functional autoregression for monthly anchovy catches forecasting

In this paper, a multi-scale stationary wavelet decomposition technique combined with functional auto-regression is used to improve the prediction accuracy and parsimony of anchovy monthly catches forecasting in area north of Chile (18 21'S-24 S). The general idea behind this approach is to decompose the observed anchovy catches data into low frequency (LF) component and high frequency (HF) component by using stationary wavelet transform and to separately forecast each frequency component. The forecasting strategy was evaluated for a period of 42 years, starting from 1-Jun-1963 to 31-Dec-2007 and we find that the proposed forecasting method achieves a 98% of the explained variance with a reduced parsimony and high accuracy. Besides, is showed that the wavelet-autoregressive forecaster is more accurate and performs better than both multilayer perceptron neural network model and functional autoregressive model.