Empirical Comparison of Box-Jenkins Models, Artificial Neural Network and Singular Spectrum Analysis in Forecasting Time Series

Document Type : Original Paper

Authors

1 Department of Statistics, Payame Noor University, P. O. Box 19395-4697, Tehran, Iran

2 Department of Statistics, Bu-Ali Sina University, Hamedan, Iran

Abstract

The Box-Jenkins model is applied as a parametric method for time series analysis and fitting seasonal and non-seasonal autoregressive moving average models.  But this procedure is not useful for short length and non stationary time series data. To overcome these problems, two nonparametric methods i.e. Artificial Neural Network and Singular Spectrum Analysis are introduced.  These procedures do not require any statistical assumptions about normality of errors and could be used for short time series data. In this article, after introducing the above methods, their accuracy in forecasting sales of four types of food products, pharmaceutical and health care of a distribution Corporation are compared. Then using simulation studies, the effectiveness of these methods for short-term and long-term predictions are evaluated. The results show the superiority of Singular Spectrum Analysis ​​compared to the other two methods in terms of the root mean square error of forecasting.

Keywords

Main Subjects


Box, G.E.P., Jenkins, G.M. and Reinsel, G.C. (2008). Time Series Analysis: Forecasting and Control, 4nd ed., Hoboken, New Jersey: John Wiley & Sons.
 
Krose, B. and Smagt, P.V.D. (1996). An Introduction to Neural Networks, 8th ed.
University of Amsterdam.
Dunne, R.A. (2007). A Statistical Approach to Neural Networks for Pattern Recognition, Hoboken, New Jersey: John Wiley & Sons.
Rao, A.R. and Cecchi, G.A. (eds), (2012). The Relevance of the Time Domain to Neural Network Models, Springer.
 Dreyfus, G. (2005). Neural Networks: Methodology and Applications, Springer. 
 
Tirozzi, B., Puca, S., Pittalis, S., Bruschi, A., Morucci, S., Ferraro, E. and Corsini, S. (2005). Neural Networks and Sea Time Series: Reconstruction and Extreme-Event Analysis, Boston, Birkhauser.
Petridis, V. and Kehagias, A. (1998). Predictive Modular Neural Networks: Applications to Time Series, Springer, New York.
 Chatfield, C. (2000). Time Series Forecasting, Chapman & Hall/CRC.