Negotiation Dynamics: Analysis, Concession Tactics, and Outcomes

Abstract

Given that a negotiation outcome is determined to a large extent by the successive offers exchanged by negotiating agents, it is useful to analyze dynamic patterns of the bidding, what Raiffa calls the "negotiation dance". Patterns in such exchanges may provide additional insight into the strategies used by the agents. The current practice of evaluating a negotiation strategy, however, is to primarily focus on fairness and quality aspects of the agreement. There is a lack of tools and methods that facilitate a precise analysis of the negotiation dynamics. To fill this gap, this paper introduces a method for analysis based on a classification of negotiation steps.The method provides the basic tools to perform a detailed and quantified analysis of a negotiation between two agents in terms of dynamic properties of the negotiation trace. The method can be applied to well-designed tournaments, but can also be used to analyze single 1-on-1 negotiation.Example findings of applying the method to analyze the ABMP and Trade-Off strategies show that sensitivity to the preferences of the opponent is independent, respectively dependent, on a correct model of that opponent. Furthermore, the results illustrate that having domain knowledge is not always enough to avoid making unintentional steps.

A Multi-Agent Environment for Negotiation

“…For that purpose they combine the Bayesian learning technique as proposed in [14] with a Tit-for-Tat tactic, see e.g., [5], and the classification of negotiation moves as described in, e.g., [11]. As is typical for Tit-for-Tat, it avoids exploitation by a form of mirroring of the bids of the opponent.…”

“…A number of the state-of-the-art agents have found a place in SAMIN: ABMP [17], Bayesian agent [14], Bayesian Tit-forTat [12], FBM [29], Trade-off agent [6], QO agent [24], Random Walker [11]. As they were developed by different teams, their design, architecture, and implementation varies.…”

“…This type of knowledge, labelled shared domain knowledge, is modelled explicitly as a separate component that can be accessed by all negotiating agents. The analytical toolbox layer of the negotiation environment a set of statistical analysis methods to perform an outcome analysis on negotiation sessions as introduced and discussed in e.g., [11,30]. Furthermore, the toolbox contains methods for the analysis of dynamic properties of negotiation sessions as discussed in e.g., [11].…”

“…The interaction engine also feeds information to the advanced logging capabilities of SAMIN. Logs are used by the analytical toolbox to assess the performance of negotiation strategies and algorithms, see [11,13]. Interaction protocols are implemented in the negotiation environment as a separate component to allow the use of a variety of protocols.…”

“…The Random Walker strategy introduced in [11], also known as Zero Intelligence (ZI) strategy [8], randomly jumps through the negotiation space. It does not use own preferences or a model of opponent's preferences to make an offer.…”