A unified model for demand prediction in the context of provisining and replenishment

Abstract

An improved demand prediction model is presented which incorporates features of two earlier models. The unified model pools usage data classified by repair part class and by component class. The performance of the model is evaluated in a provisioning and replenishment context and compared with that for the current procedure which employs technicians' usage estimates. INTRODUCTIONThe problem of demand forecasting for military repair parts is of special importance in the design of efficient inventory systems. This special importance derives from the characteristic that for most items usage is sporadic, i.e., occurs infrequently, if at all, over long periods of time, and low, i.e., demand quantities are generally small when usage is observed.* This suggests the need for designing logistics systems which accommodate to the demand process. This is not to say that other factors, such as those affecting political and military strategy, can be ignored in the structuring of an inventory system. What is implied is that such factors need to be considered within the framework imposed by the demand process.For example, one consequence of the characteristics of sporadic and low usage is that preferred supply decisions should be made in a probabilistic rather than a deterministic context. With zero or low usage, the application of a deterministic model generally requires that usage estimates be greatly inflated if a sufficient range of repair parts is to be provisioned. Thus, the use of a deterministic model often leads to large over-buys even over the expected design life of the system to be supported. Another illustration in the supply context pertains to the end-consumer echelon. Although the range of items demanded in a given time period is small, it will generally be necessary to stock a large proportion of the installed items if adequate effectiveness is to be achieved. This is so because of the difficulty of predicting which of the large number of items subject to wear will actually fail. Thus, the often followed policy of off-loading stocked items after, say, 1 or 2 years of no observed movement of an item can lead to undesired reductions in readiness. The problem here is not that review and reduction of carried stock at the end-consumer echelon is unwarranted, but that additional information of an item's characteristics, such as its expected usage rate, unit price, unit cube, and military worth, are required before a rational determination can be made of the proper set of items to be off-loaded. Whereas at the end-consumer echelon, sporadic and low usage generally implies that a wide range *Studies documenting these usage patterns for military repair parts are reviewed in [I]. 29