Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)

First Advisor

Linguo Gong


We consider a sequential decision process in a periodic review production system. At the beginning of each period, marketing decides how much money to be invested in demand forecasting. After its decision, marketing obtains the forecast and passes the forecast to production, which then determines production quantity. Centralized and decentralized models are formulated. In the centralized model, the two divisions make their decisions by considering the total profit of the company. In the decentralized model, marketing pays production a unit transfer price for every product sold. Three types of policies about the unit transfer price are considered. In the company with the marketing-oriented policy, the unit transfer price is decided by marketing; in the company with the production-oriented policy, it is decided by production; and, in the company with the coordination policy, it is determined by considering the overall profit of the company. We find that, in the decentralized companies, production favors a larger unit transfer price, but marketing prefers a lower unit transfer price. In the company with the marketing-oriented policy, marketing tends to spend more money on forecasting, but production becomes less likely to start a run. However, in the production-oriented company, production is more likely to produce, but marketing spends less money on forecasting. The results in the company with the coordination policy can be considered as outcomes of cooperation, and it has the best performance among the decentralized companies. Furthermore, we find that the centralized company has a larger total profit than do the decentralized companies. Most of the time, it spends more money on forecasting, and it is more likely to produce. We develop two multi-period models. For the multi-period centralized model with a finite planning horizon, we find that the learning effect can induce the system to adopt a forecasting method with high cost and precision. As to the other multi-period model, we study a system with a fixed batch size, a fixed lead time, and an infinite planning horizon. The important result is that there exists a threshold inventory level, which property is similar to that of an (s, Q) inventory management system.