Relative Performance of Bilateral Multiattribute Negotiation Strategies in Open Markets

Abstract

The long-running Automated Negotiating Agents Competition (ANAC) is comprised of various agent-agent and human-agent negotiation leagues. One such competition is the Automated Negotiation League (ANL) which involves repeated, bilateral negotiation over multiple issues. Researchers have investigated a tournament setting for this scenario involving a small, fixed number of agents. We are interested in automated agents participating in large and open marketplaces containing many instances of well-known agent types of varying sophistication. We experiment with four representative negotiation behaviors as agent types: Hardliner, Boulware, Conceder, and Tit-for-Tat. We simulate open markets with varying negotiation domain sizes, agent type distributions, and negotiation time available to evaluate the relative performances of different negotiation strategies. We analyze and report relative performances of the strategies on relevant performance metrics. We also extend this analysis using a head-to-head matrix.