AN OBJECTIVE BAYES APPROACH FOR INVENTORY DECISIONS,
RAND CORP SANTA MONICA CALIF
Pagination or Media Count:
An approach to demand analysis, based on a mathematical technique called Bayesian inference, is described. Instead of trying to give a point estimate of the items true mean demand, this approach estimates the probability that the items mean demand is at various levels. An objective procedure is developed for determining the Bayesian prior distribution of an items true mean demand from data on the entire system of inventory items. Employing a multivariate log normal prior distribution, it is possible to refine the estimate of an items true mean demand by considering all relevant information, including unit cost and engineering estimates, as well as past demand data. Complicated models of the demand process, such as the compound Poisson, are readily accommodated. The advantages of Bayesian analysis in non-stationary problems are explored. It is expected that the Bayesian approach will be particularly important where item demand is low and erratic. The memorandum demonstrates that Bayesian analysis can lead to significant improvements in inventory management with only a slight increase in computation. The contention is illustrated with an example of recoverable item stock levels at a representative air base.