For the third data set, the results are signi®cant at the 0.05 level. The results of the statistical tests for equality of MAPE suggest that the neural network approach to the model selection problem is indeed comparable to the best methods. While a direct comparison of the intelligent model selection approach in the context of various additional standard forecasting models is of some interest, this issue is not addressed in the current research.
All the networks in this research are developed using the software package, Brainmaker (California Scienti®c Software, 1993). The networks are developed on a 486-based computer.
This research explored the possibility of designing a decision support system as a means of providing insights into model selection and forecasting. The proposed system relies on times series characteristics as the forecast model selection criteria with a neural network structure developed for the model selection phase. The decision support system combines calibration, model selection and forecasting into an intelligent forecasting system. At selected time intervals both the model selection and parameter estimation are recalibrated, resulting in the most recent time series information to both select and estimate the quantitative model. Utilizing actual time series data the forecasting system achieved reasonable levels of accuracy in the testing phase of the research. As practitioners continue to rely on quantitative forecasting methods, it is believed the decision support system developed in this research will provide guidance in developing the most accurate forecasts for a variety of time series.
The authors wish to thank Professor Marvin Karson for his valuable assistance in revising the original paper and the referees for their thoughtful suggestions and insights that generally improved the quality of the research.
California Scienti®c Software, Brainmaker: User's Guide and Reference Manual, Nevada City, California, 1993.
Chu, C. and Widjaja, D., `Neural network system for forecasting method selection', Decision Support Systems, 12 (1994), 13±24.
Clements, M. P. and Hendry, D. F., `On the limitations of comparing mean square forecast errors', Journal of Forecasting, 12(8) (1993), 617±637.
Collopy, F. and Armstrong, J. S., `Expert opinions about extrapolation and the mystery of the overlooked discontinuities', International Journal of Forecasting, 8(4) (1992a), 575±582.
Collopy, F. and Armstrong, J. S., `Rule-based forecasting: development and validation of an expert system approach to combining time series extrapolations', Management Science, 38(10) (1992b), 1394±1414.
Dalrymple, D. J., `Sales forecasting practices', International Journal of Forecasting, 3 (1987), 379±391.
Diebold, F. X. and Mariano, R. S. `Comparing predictive accuracy', Journal of Business & Economic Statistics, 13 (1995), 253±263.
Donaldson, R. and Kamstra, M., `Forecast combining with neural networks', Journal of Forecasting, 15 (1996), 49±61.
Fildes, R. and Lusk, E. J., `The choice of a forecasting model', Omega, 12(5) (1984), 427±435.
Fletcher, D. and Goss, E., `Forecasting with neural networksÐan application using bankruptcy data', Information & Management, 24(3) (1993), 159±167.
Gorr, W., Nagin, D. and Szczypula, J., `Comparative study of arti®cial neural network and statistical models for predicting student grade point averages', International Journal of Forecasting, 10 (1994), 17±34.
Hill, T., Marquez, L., O'Connor, M. and Remus, W., `Arti®cial neural network models for forecasting and decision-making', International Journal of Forecasting, 10 (1994), 5±15.
Hill, T., O'Connor, M. and Remus, W., `Neural network models for time series forecasts', Management Science, 42(7), (1996), 1082±1092.
Iman, R. and Conover, W. J., Modern Business Statistics, New York: Wiley, 1983.
Lachtermacher, G. and Fuller, J., `Backpropagation in time-series forecasting', Journal of Forecasting, 14 (1995), 381±393.
Уважаемый посетитель!
Чтобы распечатать файл, скачайте его (в формате Word).
Ссылка на скачивание - внизу страницы.