Multi-echelon stochastic inventory systems with centralized stock information

As information technology (e.g., Electronic Data Interchange or EDI) proliferates, more and more companies now possess centralized stock information for their inventory systems. This dissertation examines how the centralized stock information can be used to improve inventory management. We consider several multi-echelon networks with centralized stock information and stochastic demand. These include serial, assembly, and one-warehouse multi-retailer systems. A natural replenishment strategy based on centralized stock information is to replenish each stage of a multi-echelon system according to its echelon stock level. The echelon stock of a stage is the aggregate inventory level of the subsystem consisting of the stage and all its downstream stages. Such replenishment strategies are in a sharp contrast with those based on installation stock information under which each stage replenishes its inventory according to its local stock level. We study a class of control policies--the so-called (R,nQ) policies--that operate according to echelon stock levels. An echelon stock (R,nQ) policy at a stage orders an integer multiple of Q, the base lot-size, once its echelon stock level falls to or below R, the reorder point. Such policies are easy to implement, and are similar to the widely used reorder-point order-quantity policies in practice. Our examination of the cost effectiveness of echelon stock (R,nQ) policies consists of three main steps: (1) performance evaluation of any given policy in the class, (2) optimization within the class, and (3) benchmarking. The first two steps are based on observations of the operation details of echelon stock (R,nQ) policies, and the third step is carried out by comparing the minimum costs of echelon stock (R,nQ) policies with lower bounds on the minimum achievable costs (by any policy) of the systems. Our benchmarking studies for serial and one-warehouse multi-retailer systems suggest that echelon stock (R,nQ) policies are close to optimal. We also compare echelon stock (R,nQ) policies with their installation-stock counterpart