Abstract:
Agent-based modeling t (ABM) and system dynamics (SD) are two methodologies commonly used in the modeling of complex systems, which are conventionally placed at the two ends of the aggregation spectrum. Disaggregated ABM facilitates the analysis of the relationship between individual features, individual dynamics, and aggregate dynamics. Aggregated SD captures the aggregate dynamics without considering the individual scale. Its representative capability thus depends on the system considered. This research asks the general questions about the level of aggregation in the scope of a three-echelon, multiagent supply chain. First, an agent-based model is built, and the effects of individual firms’ features (in terms of demand forecasting, order batching, and dynamic pricing) on aggregate dynamics are analyzed. The analysis focuses especially on inventory/order variability and the bullwhip effect, which is the increase in variability as one moves up in a supply chain. It is shown that the bullwhip effect does not exist if none of these features exist in the agent schemata (Fixed S policy). On the other hand, demand forecasting, order batching, and dynamic pricing are shown to be the causes of variability and bullwhip effect in aggregate orders and inventories. It is demonstrated that these factors at the agent level affect aggregate dynamics through individual dynamics, which systematically depend on the supply chain topology and interactions among individuals. Given the ABM findings, an aggregated SD model is built. SD model is shown to capture the dynamics under Fixed S policy perfectly. It captures the general behavioral pattern under demand forecasting, but not under order batching. The variability amplification characteristics of both demand forecasting and order batching are also captured by the SD model. However, SD fails to capture the variability amplification characteristics of dynamic pricing; because SD does not distinguish individuals and thus does not consider the interactions of agents according to their price levels.
