Some alternatives to exponential smoothing in demand forecasting

Abstract

The report contains a study devoted to a comparison of exponential smoothing with other alternatives to demand forecasting. Special attention is paid to the stock-out risks assumed whenever reorder levels are set using the various methods being compared. Models presently used by NavSup are employed in order that the results be applicable to the system in use. Simulation techniques are used for drawing comparisons. For constant mean, normal demand, it is shown that exponential smoothing does not produce as accurate results as ordinary maximum likelihood techniques. For the case of a linear mean changing with time, it is shown that the two methods are about comparable. Finally, a sequential Bayes forecasting method is defined and found to compare quite favorably with exponential smoothing. The need for additional study of Bayesian methods is established. (Author)