An application of recurrent artificial neural networks (RTRL) and ARIMA-GARCH processes for predicting the soybean series of prices

Abstract

This paper describes a comparative study to measure forecasting efficiency between ARIMA-GARCH process and artificial neural networks (ANN) using the real time recurrent learning algorithm (RTRL). An experiment, applying this two techniques, is performed to compare the forecast of the soybean price series. The forecasting window choice is arbitrary, in this work are used 1 to 10 steps ahead. Both methods, ARIMA-GARCH and ANN, requires data transformation of the original series (or level series), but final forecasting results are presented in terms of level series. According to the obtained results, can be verified a superior performance generated by artificial neural networks when compared with the traditional econometrics volatility models