| Abstract: | This article considers the problem of estimating parameters of the demand distribution in lost sales inventory systems. In periods when lost sales occur demand is not observed; one knows only that demand is larger than sales. We assume that demands form a sequence of IID normal random variables, which could be a residual demand process after filtering out seasonality and promotional nonstationarities. We examine three estimators for the mean and standard deviation: maximum likelihood estimator, BLUE (best linear unbiased estimator), and a new estimator derived here. Extensive simulations are reported to compare the performance of the estimators for small and large samples and a variety of parameter settings. In addition, I show how all three estimators can be incorporated into sequential updating routines. © 1994 John Wiley & Sons, Inc. |
