An Adaptive Prediction-regret Driven Strategy for One-shot Bilateral Bargaining Software Agents

Abstract

Bargaining is a popular paradigm to solve the problem of resource allocation. Factors such as complexity of dynamic environment, bounded rationality of negotiators, time constraints and incomplete information, make the design of optimal automated bargaining strategies difficult. Currently, most bargaining strategies are designed under the assumption that opponents offer according to specific models. Therefore, most of them focus on modeling opponents or predict opponents’ private information such as reserve price, deadline, or the probabilities of different behaviors. Without model opponents, this paper presents an adaptive prediction-regret driven negotiation strategy for bilateral one-shot price bargaining, which extends the existing heuristic method of “looking forward” into “looking forward and reviewing the past” pattern by the regret principle in psychology. Four sets of experiments are designed and implemented to verify the general performance of this strategy. Results show that this strategy outperforms the strategies that model opponents and existing adaptive strategy when bargaining with multifarious opponents who offer according to pure consecutive concession strategies, sit-and-wait strategy, fixed mixture strategies, random mixture strategies, or even intelligent strategies.