Strategies and Heuristics Used by the UMBCTAC Agent in the third Trading Agent Competition.pdf

Strategies and Heuristics Used by the UMBCTAC Agent in the third Trading Agent Competition ? Li Ding, Tim Finin, Yongmei Shi, Youyong Zou, Zhongli Ding, and Rong Pan Computer Science and Electrical Engineering Department University of Maryland, Baltimore County {dingli1, finin, yshi1, yzou1, zding1, panrong1}@ Abstract The UMBCTAC agent was one of the top ranked agents in the third international Trading Agent Com- petition (TAC’02). This paper describes and evalu- ates the key heuristics used by UMBCTAC, includ- ing the early bird heuristic, the balance heuristic, and the separation heuristic. We developed a simple gain- risk model to search safe and profitable allocations for hotel rooms and airline tickets. We also used a novel probabilistic approach to dynamically allocate entertainment tickets and bid in entertainment auc- tions. We conclude with a description of ongoing and planned work. 1 Introduction The trading Agent Competition (TAC) is a market simu- lation game proposed by Wellman and Wurman [1999] with the first competition held in the summer of 2000 [Stone and Greenwald, 2001]. The second and third TACs [Wellman et al., 2002a; Greenwald, 2003], which were held in the subsequent years, maintained research issues in simultaneous interrelated auction context, and had minor modifications for further research. The fourth competition initiates new research issues in supply chain management context [Raghu et al., 2002] and keeps the original TAC framework under the name “TAC Classic”. TAC Classic focused on automated strategies for soft- ware trading agent. A trading agent assembles a round- trip travel package for each of its customers by trading goods in multiple concurrent and interrelated auctions. TAC runs in client/server mode: a game server generates eight customers for each trading agent, and runs twenty- eight simultaneous auction instances which supply travel goods. On the client side, a human implemented trading agen