ommodity price forecasting using ARIMA-GARCH models and neural networks with wavelets: old technologies - new results

Abstract

The main objective of this study was to explore the possibility of applying a methodology capable of decomposing a time series through wavelets, in conjunction with econometric and neural network models, to forecast variables. The authors also compared the quality of the forecasts of chronological successions as applied to the study of a commodity, soy. The distinguishing feature of this study is based on the realization of the forecasts within the subseries decomposed by a wavelet and on obtaining estimates through reconstruction of the time series. From the analysis of the data for a 60 kg sack of soy, the results obtained were particularly satisfactory when using a wavelet filter in a recurrent neural network.