Robustness Analysis of Negotiation Strategies using PAT

The evaluation criterion reflects the negotiation performance of different strategies in terms of average scores achieved within a fixed tournament setting. However, it does not reveal much information about the robustness of the negotiation strategies in different negotiation scenarios, since it assumes that each agent’s strategy is fixed beforehand. For example, we may be interested to know whether an agent adopting our strategy ABiNeS has the incentive to unilaterally deviate to other strategies under a particular negotiation tournament. Would any agent adopting other strategies be willing to switch to our strategy under different tournaments? To answer the above questions, we adopt the game-theoretic approach to analyze the robustness of our strategy under different negotiation settings. Since there exists an infinite number of possible negotiation strategies that the agents may take, we cannot apply the standard game-theoretic approach to perform such an analysis by explicitly considering all possible strategies. Therefore, in this paper we adopt the tool of empirical game theoretic (EGT) analysis to achieve this goal instead, which is originally developed by [1] to analyze the Trading Agent Competition. EGT analysis is a game-theoretic analysis approach based on a set of empirical results. It handles the problem of the existence of infinite possible strategies by assuming that each agent only selects its strategy from a fixed set of strategies and the outcomes for each strategy profile can be determined through empirical simulations. This technique has been successfully applied in addressing questions about robustness of different negotiation strategies in ANAC 2011 [2]. However, previous robustness analysis is based on the principle of best singleagent deviation, where only the agent with the maximum deviation profit among all participating agents can deviate from its current strategy. However, this assumption may not be able to reflect the realistic scenarios, since each participating agent usually represents different parties and makes decision independently. Each participating agent usually has its own freedom to choose whether to deviate from its current strategy or