Fejer-Korovkin Wavelet Based MIMO Model For Multi-step-ahead Forecasting of Monthly Fishes Catches

This paper proposes a Multiples Input-Multiples Ouput Autoregressive (MIMO-AR) model based on two stages to improve monthly anchovy catches forecasting of the coastal zone of Chile for periods from January 1958 to December 2011. In the first stage, the stationary wavelet transform (SWT) based on Fejer-Korovkin (FK) wavelet filter is used to separate the raw time series into a high frequency (HF) component and a low frequency (LF) component. In the second stage, both the HF and LF components are the inputs into a FK+MIMO-AR model to predict the original time series. The performance of the FK-MIMO-AR model is evaluated by comparing its prediction with MIMO-AR model based on SWT with Daubechies (Db) wavelet filter (Db+MIMO-AR). Results show that the FK+MIMO-AR model outperforms the Db+MIMO-AR model in terms of root mean square error, modified Nash-Sutcliffe efficiency and coefficient of determination for 15-month-ahead anchovy catches forecasting.