A strategic decision model for multi-attribute bilateral negotiation with alternating

In this paper we present a strategic decision model for bilateral, alternating offers, multi-attribute negotiations under incomplete information. Theoretically, Bayesian-Nash equilibrium provides the correct strategy given the prior distribution of the agents’ types. But the computation of Bayesian-Nash equilibrium is generally intractable even for a single attribute. We use the powerful formalism of Bayesian inference to compute a tractable approximation to the ideal Bayes-Nash equilibrium strategy. The formalism underlying our general approach was originally published by Tesauro in [2], which proposed a negotiation strategy based on Bayesian updating and combinatorial search. An implementation of the formalism was studied in [2][1] using an asymmetric protocol, in which one agent (seller) makes proposals and the other agent (buyer) only accepts or rejects the offers. Strategic interactions were not considered as the buyer behavior was assumed to be non-strategic. We extend that approach to an alternating-offers protocol, and explicitly consider strategy interdependence. The strategy computes offers and offer acceptance to approximately optimize expected utility using depth-limited combinatorial search and Bayesian updating. The optimization is performed with respect to a model of the opponents’ presumed strategic behavior and a probability distribution on the opponent’s utility function. The probability distribution is updated based on observed offers and responses. We establish relationships, via simulation, relating the effects of ∗The authors are grateful to Jeff Kephart and Bill Walsh for many valuable discussions and comments.