This paper considers two different strategies for a sales component of an intelligent agent designed for the supply chain management (SCM) game (www.sics.se/tac), which is a new addition to the 2003 Trading Agent Competition (TAC). TAC-SCM game involves six software agents attempting to maximize profits by manufacturing computers in a simulated market economy. TAC-SCM intelligent agents participate simultaneously in two simulated markets: a market of raw materials (RM) and a consumer market. This paper introduces two strategies for competing in the consumer market, without considering issues such as acquiring components and assembling products. The primarily goal of the sales strategies we introduce is to maximize revenue by selling out the existing finished goods (FG) inventory built by other components of the intelligent agent. The automation of the sales strategy is an interesting problem because it is one of the most important factors affecting the overall performance of an agent participating in an electronic market. Successful automation of the sales strategy requires a synthesis of several fields, including economics, game theory, artificial intelligence, machine learning, multiagent systems, statistics, and probability.
The primary contribution of this paper is its proposed approaches to the automation of the sales strategy and its extensive experimental analysis and comparison of the strategies in the TAC-SCM simulation environment. We describe our initial implementation of the sales algorithms for a TAC-SCM agent, and present experimental results that seem to indicate that this implementation is at least as sophisticated as those of other teams. These experiments also illustrate areas where our agent's performance could be improved. We conclude with a discussion of future research directions.