An Improved ARFIMA Model using Maximum Overlap Discrete Wavelet Transform (MODWT) and ANN for Forecasting Agricultural Commodity Price

Abstract

Autoregressive fractionally integrated moving average (ARFIMA) is widely employed model for long memory time series forecasting in divergent domain from several decades. One of the main pitfall of this model is the presumption of linearity. As many long memory time series data in real world are not purely linear, therefore there is an opportunity to enhance the prediction ability of ARFIMA models by fusing with other nonlinear models. With this reasoning, the present article attempts to estimate the parameters of ARFIMA model by maximum overlap discrete wavelet transform (MODWT) and long memory time series prediction was made by combining ARFIMA-MODWT and ANN for forecasting spot prices of mustard. Experimental study justied the superiority of the proposed hybrid model over ARFIMA model in terms of several measurement indices