Designing efficient bidding strategies for sequential auctions remains an important, open problem area in agent-mediated electronic markets. In existing literature, a variety of bidding strategies have been proposed and have been shown to perform with increasing degrees of efficiency. However, most of strategies proposed so far do not explicitly model bidders' attitudes towards risk which, in mainstream economic literature, is considered an essential attribute in modeling agent preferences and decision making under uncertainty. This paper studies the effect that risk profiles (modeled through the standard Arrow-Pratt risk aversion measure), have on the bidders' strategies in sequential auctions. We model the sequential bidding decision process as an MDP and we analyze, for a category of expectations of future price distributions, the effect that a bidder's risk aversion profile has on her decision-theoretic optimal bidding policy. This analysis is performed separately for the case of first-price and second-price sequential auctions.Next,we simulate the above strategies, and we study the effect that an agent's risk aversion has on the chances of winning the desired items. We conclude the paper with an experimental study of how the presence of risk-averse bidders in a market affects allocation efficiency and expected seller revenue.

We present a visualization of bargaining processes between a seller and a buyer (or buyers) where negotiations take place over the composition and price of bundles of goods. The negotiations themselves are simulated by software specific for the kind of negotiations. The visualization collects information from the simulation steps in a simulation-specific way and visualizes those steps in a general way. We demonstrate the visualization for two cases: a seller who offers bundles of linearly dependent items to many customers and a seller who repeatedly offers a bundle of non linearly dependent goods to a specific customer meanwhile learning this customers preferences.

Graphical utility models represent powerful formalisms for modeling complex agent decisions involving multiple issues. in the context of negotiation, our previous work [3] has shown that using utility graphs enables agents to reach pareto-efficient agreements with a limited number of steps in high-dimensional negotiations involving non-linear dependencies. this paper considerably extends the results of [3], by proposing a method for constructing the utility graphs of buyers automatically, based on previous negotiation data. the proposed method is based on techniques inspired from item-based collaborative filtering, used in on-line recommendation algorithms. experimental results show that our approach is able to retrieve the structure of utility graphs on-line, with a relatively high degree of accuracy, even if a relatively small amount of data about concluded negotiations is available. furthermore, we provide a rigorous method for determining the cut-off number of edges for a wide category of random graphs.

This paper presents an agent strategy for automated negotiation over a set of issues with interdependent valuations. First, we show how complex utility functions over a set of binary issues (or bundle of items) can be concisely represented using the formalism of utility graphs. Next, a heuristic is proposed for automated learning of an opponent's preference function, starting from a given (maximal) factorization of this function as a utility graph. The contribution of our approach is that it speeds up learning considerably compared to other techniques proposed in existing literature for this problem. An extended version of this paper is given in [1], while [2] provides a further extension of this work, in which collaborative filtering is used to learn the starting structure of such graphs.