Reducing Inventory System Costs by Using Robust Demand Estimators

Abstract

Applications of inventory theory typically use historical data to estimate demand distribution parameters. Imprecise knowledge of the demand distribution adds to the usual replenishment costs associated with stochastic demands. Only limited research has been directed at the problem of choosing cost effective statistical procedures for estimating these parameters. Available theoretical findings on estimating the demand parameters for (s, S) inventory replenishment policies are limited by their restrictive assumptions. The impact on total system cost of using the sample mean and standard deviation as compared to robust parameter estimators has not been tested. This paper explores the circumstances under which the cost due to statistical estimation can be substantially reduced by a better choice of estimators. Specifically, an exponentially smoothed average and a modified exponentially smoothed mean absolute deviation are shown to outperform the sample mean and standard deviation for a wide range of computer simulated and U.S. Air Force empirical demands when the (s, S) policies are calculated using Ehrhardt's Power Approximation. Those situations in which the method of demand parameter estimation has negligible impact on total system cost are also indicated.