Combining Nonlinear Additive Autoregression with Multiscale Wavelet for Monthly Anchovy Catches Forecasting

In this paper, a nonlinear additive autoregressive model combined with multiscale stationary wavelet transform is used to improve the accuracy and parsimony of one-monthahead forecasting of monthly anchovy catches in northern Chile (18° 21´S-24° S). The general idea of the proposed forecasting model is to decompose the raw data set into trend and residual components by using SWT. In wavelet domain, the trend component and residual component are predicted with a linear autoregressive (AR) model and nonlinear additive autoregressive (NAAR) model; respectively. Hence, the proposed forecast is the co-addition of two predicted components. Data on monthly anchovy catches are available for a period of 44 years, from 1-Jun-1963 to 31-Dec-2007. We find that the proposed forecasting method achieves 99% of the explained variance with reduced parsimony and high accuracy. Besides, the wavelet-autoregressive forecaster proves to be more accurate and performs better than the multilayer perceptron (MLP) neural network model and NAAR model.