Time series in forecasting and decision: an experiment in elman nn models

Abstract

The paper examines the role of analytical tools in analysis of economic statistical data (commonly referred to as econometry) and artificial neural network (ANN) models for time series processing in forecasting, decision and control. The emphasis is put on the comparative analysis of classical econometric approach of pattern recognition (Box-Jenkins approach) and neural network models, especially the class of recurrent ones and Elman ANN in particular. A comprehensive experiment in applying the latter modeling has been carried out, some specific applications software developed, and a number of benchmark series from the literature processed. This paper reports on comparison findings in favor of Elman ANN modeling, and on the use of a designed program package that encompasses routines for regression, ARIMA and ANN analysis of time series. The analysis is illustrated by two sample examples known as difficult to model via any technique.