An extensive review of the current literature on supply chain management research shows that, traditionally, a supply chain is typically viewed as a hierarchical structure in which the information and materials flow in a vertical point-to-point fashion and each individual at a level is independent and does not interact with others horizontally. The methodologies often found are mathematical deduction or induction with, sometimes, numerical simulation or empirical data analysis to support the theories. These traditional approaches are very powerful and efficient in solving many problems. However, they are inadequate in the areas that the agent based modeling is capable of. These areas include capturing emergent phenomena; providing a natural description of a supply chain system; and being flexible (scalable and adaptive).
This dissertation investigates the supply chain system in a way different from the traditional approaches. Our view of a supply chain is a hierarchical structure with horizontal layers, in which exist interactions between the members (retailers, in our case, who exchange information and stocks). We no longer treat the system as point-to-point, but point-to-layer relationships. We use the agent based modeling to represent in a more natural way the individual interactive entities in the supply chain system and investigate the emergence due to the interactions between these entities.
In this dissertation we have demonstrated that interactions between retailers have plausible consequences on the supply chain performance. In particular, we look at the total average costs at the supplier side and find significant improvement in cost reduction as the interactions among the retailers increase. Through the external validation, we confirm that the new agent based modeling method produces the exact results as those from the previous research with traditional approaches. We use agents to represent the firms in the supply chain and program the behaviors for each agent by applying the existing mathematically proved functions. We also use the agents to demonstrate the bullwhip effect. We have shown that without information sharing, the supplier needs to periodically update the mean and variance of demand based on observed demand data through orders received from retailers, then the variance of the orders placed by the supplier will be greater than that by the retailer and in turn the demand. More importantly, providing sufficient information of the underlying demand to the supplier can significantly reduce this increase in variability.