Quantitative modeling in inventory management with imperfect information.

摘要

In today's highly competitive business environment, minimizing inventory management costs and increasing operational efficiency are first order determinants of profitability. Improved inventory management policies in turn depend on optimally responding to changes in market conditions, understanding the implications of various constraints on policy decisions, and proactively integrating information technologies. In this dissertation, we address these issues by designing and building sufficiently general inventory/production models that capture the complexity of an operation and incorporate available information.; In the first part of the dissertation, we present a general formulation of a multi-item production problem addressing two important features observed in practice. First, we address the impact of fluctuations in the demand environment on replenishment and production decisions. Second, we investigate how constraints at different stages of production impact the policy decisions. In particular, we develop a lower bound and close-to-optimal policies for a two-stage multi-item system with Markov-modulated demands and production quantity requirements. Through a numerical study, we quantify the impact of production requirements and the nature of the demand environment on the benefits of postponement.; In the second part of the dissertation, we investigate the inventory record inaccuracy problem and the value of visibility in inventory management. In existing computerized inventory systems, the discrepancy between the inventory records and sales-available on-hand inventory distorts the replenishment process. The systems fail to order when they should or carry unnecessary inventory due to early replenishment. We show that the main sources of discrepancy between inventory record and physical stock can be grouped under three classes: shrinkage, misplacement, and transaction errors. We explicitly model how different error sources lead to inventory discrepancies. We then demonstrate how companies should manage their inventory in the presence of unobserved inventory discrepancies. Finally, we quantify the true value of inventory visibility, as well as the value of reduction of some of the error sources offered by the latest technologies such as RFID (Radio-Frequency Identification).