A principled study of the design tradeoffs for autonomous trading agents

  • Authors:

  • Ioannis A. Vetsikas;Bart Selman

  • Affiliations:

  • Cornell University, Ithaca, NY;Cornell University, Ithaca, NY

  • Venue:
  • AAMAS '03 Proceedings of the second international joint conference on Autonomous agents and multiagent systems
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper we present a methodology for deciding the bidding strategy of agents participating in a significant number of simultaneous auctions, when finding an analytical solution is not possible. We decompose the problem into sub-problems and then use rigorous experimentation to determine the best partial strategies. In order to accomplish this we use a modular, adaptive and robust agent architecture combining principled methods and empirical knowledge. We applied this methodology when creating WhiteBear, the agent that achieved the highest score at the 2002 International Trading Agent Competition (TAC). TAC was designed as a realistic complex test-bed for designing agents trading in e-marketplaces. The agent faced several technical challenges. Deciding the optimal quantities to buy and sell, the desired prices and the time of bid placement was only part of its design. Other important issues that we resolved were balancing the aggressiveness of the agent's bids against the cost of obtaining increased flexibility and the integration of domain specific knowledge with general agent design techniques. We present our observations in dealing with these design tradeoffs and back up our conclusions with empirical results.