Managing a supply chain has been shown to be prone to systematic errors that often lead to spectacularly dysfunctional outcomes. For example, the drilling for oil and gas fluctuates three times more than the actual petroleum production, the demand for machine tools is at least twice as variable as automobile sales (the auto industry being the main consumer of machine tools), and the production of semiconductors is much more variable than industrial production as a whole (reported in Sterman 2000, pp. 666-667).
This pattern of "oscillation, amplification, and phase lag," known collectively as the bullwhip effect, has been also noted in individual firms' supply chains (see Lee et al. 2000), as well as in the laboratory Sterman (1989). Sterman's results demonstrated that individuals do not act optimally even in the relatively simple laboratory setting, but significantly underweight inventory they already ordered that has not yet arrived (called the supply line). He concluded, "the key to improved performance lies within the policy individuals use to manage the system and not in the external environment. Even a perfect forecast will not prevent a manager who ignores the supply line from over ordering." (Sterman 1989, p. 336).
The bullwhip effect is costly because it causes excessive inventories, poor customer service, and unnecessary capital investment. We separate explanations of the causes of the bullwhip effect into those that rely on participants' own cognitive limitations, their inability to determine the optimal way to adjust their inventory position, and those that rely on coordination, participants' beliefs about others' abilities or actions. The experiments we propose to conduct will be using both traditional supply chains with four participants and supply chains where a human participant plays one of the roles, and the other three roles are played by computerized agents programmed (and known) to act optimally. If deviations from optimal behavior are caused by an individual's cognitions, behavior should be the same in the treatments with all human subjects as with one human subject. If instead, behavior is being caused by an individual's beliefs about the decisions his supply chain partners make, behavior should be worse in treatments with all human subjects.
The results from our study can have potentially broad implications on the type of decision support tools that should be included in Enterprise Resource Planning (ERP) software. If behavioral causes of the bullwhip effect are important than the recent focus on reducing operational causes provides at best an incomplete solution. The question of what information or tools can be effective in mitigating behavioral causes of the bullwhip effect has not been addressed in the literature, and it is our hope that this study will open the door to this new research agenda.
