The effect of information on the performance of negotiation models

Negotiation is an everyday task in economic processes; ranging from corporations to markets, from agents to nations, always there is the need to mediate between conflicting intents or expectations. Therefore, in the scientific literature and in practice, several negotiation procedures have been developed, each one elaborated to perform in a specific context and under particular assumptions, as, for example, voting systems, auction and fair division mechanisms, negotiation protocols, and so on. All these may seem disjoined from each other, as their contexts are; nevertheless they all share the same assumption that each subject pursues his own best utility. Moreover, despite their specialization, they all have some problems in practical use like, for example, the presence of a dictator, the lack of truthfulness, or the possibility of being manipulated by fictitious declarations. In this contribution we focus on the negotiation over continuous issues and in particular we analyze the jointly Improving Direction Method (IDM), which is known for its generality, since different other negotiation protocols can be seen as an its particular subclass, and also because it has the nice theoretical property of being Pareto efficient. Nevertheless it is easy to implement, which makes IDM the perfect candidate for an automated negotiation support system. Despite its theoretical properties, we show the practical inefficiency of this method (even in the simple case of just two negotiating parties), which reduces significantly its performance in the operative context. In particular we show that the main drawback of IDM is due to the possibility to retrieve information about other ones utilities during the negotiation steps and to exploit it to manipulate the negotiation itself. For better explaining this phenomenon we show the deep connection between negotiation and the social choice problem. The bridge we build allows to carry in this context the Arrow’s Impossibility Theorem and the Gibbard-Satterthwaite Theorem, thus implying that each step of IDM (and of all the methods which it generalizes) may be affected by a dictatorial or a manipulatory party who can deviate the efficient Pareto frontier to get a better gain during the negotiation. In order to avoid the operative inefficiency of IDM, we propose a different negotiation paradigm, where the hypothesis that agents maximize their own utility is not modified, while the way they pursuit maximum satisfaction is substantially different, since each agent has to express a sub optimal choice, rather than his optimal one. In this context, the constraint of a sub optimal declaration by one side protects from information retrieval and by the other side it forces each party to leave to the others the possibility to improve their own gains in order to pursue his own best. The comparison of the performances with the IDM ones in different negotiation domains, both in terms of Pareto efficiency and manipulation resistance, shows the effectiveness of the proposed approach.