Strategic Sales Management in an Autonomous Trading Agent for TAC SCM

To operate successfully in a competitive trading environment such as the Trading Agent Competition for Supply Chain Management (Collins et al. 2004) (TAC SCM), an agent has to allocate resources and set prices in a way that maximizes its expected profit. This requires the ability to detect changing market conditions and act accordingly. In TAC SCM six agents buy parts, assemble personal computers, and sell them in daily auctions to customers. Sales decisions in our agent, MinneTAC, are driven by three different models: an automated characterization and prediction of market conditions, which we call economic regimes (Ketter 2005), a linear program that optimizes daily sales quotas, and a model of order acceptance probability. While economic regime models are commonly used at the macro economic level (Osborn & Sensier 2002), such predictions are rarely done for a micro economic environment. We focus on the sales decisions the agent has to make, where predicting prices and customer demand play an essential role. The strategies we present have been inspired, among others, by the work of Kephart et al. (Kephart, Hanson, & Greenwald 2000). MinneTAC makes sales decisions in two steps. The first step is a strategic decision, where resources are allocated over a planning horizon in a way that maximizes expected profit over the horizon. The second step is a tactical decision, which determines the offer prices that are expected to sell the quantities determined by the strategic decision, given the current demand and the pricing model.