Online auctions have become a popular method for business transactions. The variety of different auction rules, restrictions in supply or demand, and the agents’ combinatorial preferences for sets of goods, mean that realistic scenarios are very complex. Using game theory, we design trading strategies for participating in a single auction or group of similar auctions. A number of concerns need to be considered in order to account for all the relevant features of real-world auctions; these include: budget constraints, uncertainty in the value of the desired goods, the auction reserve prices, the bidders’ attitudes towards risk, competition and spitefulness between bidders, etc. To design a realistic agent, it is necessary to analyze the multi-unit auctions in which a combination of these issues are present together and we have made significant progress towards this goal. Furthermore, we use a principled methodology, utilizing empirical evaluation, to combine these results into the design of agents capable of bidding in the general real-world scenarios; we decompose the problem into sub-problems and use empirical evaluation to decide the final agent design.
In the remainder of the talk, the dual problem of mechanism design is examined. The fundamental difference of our work to related work is that the design is restricted to mechanisms immediately applicable in internet auctions; this means that designing a completely new mechanism is not allowed and only alterations of existing mechanism are examined. Despite this significant restriction, mechanisms with highly desirable properties are possible. This is demonstrated for the case of a seller with multiple heterogeneous items for sale, where a new mechanism alteration based on bundling improves significantly the seller’s revenue to the point that in most cased it is possible to outperform even the so-called “optimal auctions”. Concluding, future work and directions, e.g. applications in service procuring, energy markets, logistics and resource allocation, will be presented.