Optimal Buyer Negotiation Strategy On Agent Based Internet Markets

Present development of Internet leads us to imagine a near future in which electronic transactions between companies and consumers will be performed by electronic agents [1]. As these new markets will develop, interaction between buyers and sellers on the Internet will become more complex and more flexible evolving from mere search of information to communication and negotiation. For an agent to be able to adequately represent her user in a negotiation process, a fundamental point to be considered is her personalization by some set of parameters (as, for example, reservation value, utility function or time-constraints). On electronic markets, the user' agents will be self-interested in the sense that they will seek to maximize only their interests. In terms of economic theory, such an agent should have, not only some insights about what the user's interests are, but she should also be able to identify, by herself, an optimal negotiation strategy within the protocol which governs her interactions with other agents. Usually, computer scientists move away from such optimal design methods while programming negotiation strategies for agents based on rules of thumb [2] [3]. Although such approach can be partially justified by technical and theoretical difficulties related to the design of an agent capable to optimize on the behalf of her user, a great number of problems arise. Indeed, in a complex and uncertain environment like an agent-based market on Internet, the choice of a successful heuristic is rather difficult. Moreover, as agents will act in an open architecture, an agent using an unfitted heuristic could be taken advantage of by an opponent conceived to exploit this weakness. In this article, we propose a different prescriptive approach inspired by negotiation analysis [5]. We study the optimal decision-making process of an agent which enters on a market where agents negotiate, in a sequential way, the price of a product with an opponent who uses a rule of thumb. In this study, we are mainly concerned by the following two questions : how to compute the optimal strategy of the entrant agent ? and What price announced at the beginning of the negotiation can lead the entrant to obtain the best agreement ? In our analysis, we chose a buyer agent as entrant on the market. We defined seller agents by three characteristics: an utility function, a reservation price and a negotiation behavior. We limited the possible negotiation behaviors at three (conceder behavior, boulware behavior and imitative behavior) and we expressed the corresponding strategies of negotiation in term of probabilities. For example, if the seller agent adopts a conceder behavior, the probability for him to concede is relatively high during the first periods of negotiation and decreases while the time passes. Then, we studied the optimal strategy of the entrant buyer which has a good description of its opponent (i.e. he knows the reservation price and the probability associated with his concession decisions). His decision process was treated as a stochastic control problem. At a given stage, we assumed that the buyer agent can represent the negotiation as a state vector containing the observed decision of his opponent at previous period and the actual positions of the two agents on the price scale [4]. The use of dynamic programming techniques allowed us to compute an optimal strategy which maximizes the entrant's expected utility relative to the random decisions of her opponent in the next periods. In order to prescribe the best price to announce at the beginning of the negotiation, repeated rounds of negotiation were performed with different seller populations and several buyer's parameters were varied from one simulation to another (as reservation price or impatience).