Predictive Planning for Supply Chain Management: Adapting to Competitor Behavior∗

In today’s industrial world, supply chains are ubiquitous in the manufacturing of many complex products. Traditionally, supply chains have been created through the interactions of human representatives of the various companies involved. However, recent advances in planning, scheduling, and autonomous agent technologies have sparked an interest, both in academia and in industry, in automating the process (Kumar 2001). From a planning and scheduling perspective, supply chain management simultaneously requires long-range inventory management, mid-range customer negotiations, and short-term factory scheduling, all of which interact closely. One barrier to supply chain management research is that it can be difficult to benchmark automated strategies in a live business environment, both due to the proprietary nature of the systems and due to the high cost of errors. The Trading Agent Competition Supply Chain Management (TAC SCM) scenario provides a unique testbed for studying and prototyping supply chain management agents by providing a competitive environment in which independently created agents can be tested against each other over the course of many simulations in an open academic setting (Arunachalam & Sadeh 2005). In a TAC SCM game, each agent acts as an independent computer manufacturer in a simulated economy. The agent must procure components such as CPUs and memory; decide what types of computers to manufacture from these components as constrained by its factory resources; bid for sales contracts with customers; and decide which computers to deliver to whom and by when. One crucial challenge in supply chain management is that decisions must often be made in the face of considerable uncertainty. For instance, purchases of production resources may need to be negotiated long before accurate information about customer preferences becomes available. This challenge is particularly evident in TAC SCM, where