Andrew Byde

Decision procedures for multiple auctions

This paper presents a decision theoretic framework that an autonomous agent can use to bid effectively across multiple, simultaneous auctions. Specifically, our framework enables an agent to make rational decisions about purchasing multiple goods from a series of auctions that operate different protocols (we deal with the English, Dutch, First-Price Sealed Bid and Vickrey cases). The framework is then used to characterize the optimal decision that an agent should take. Finally, we develop a practical algorithm that provides a heuristic approximation to this ideal.

Economic dynamics of agents in multiple auctions

Over the last few years, electronic auctions have become an increasingly important aspect of e- commerce, both in the business to business and business to consumer domains. As a result of this, it is often possible to find many auctions selling similar goods on the web. However, when an individual is attempting to purchase such a good, they will usually bid in one, or a small number, of such auctions. This results in two forms of inefficiency. Firstly, the individual may pay more for the good than would be expected in an ideal market. Secondly, some sellers may fail to make a sale that could take place in an ideal market. In this paper, we present an agent that is able to participate in multiple auctions for a given good, placing bids appropriately to secure the cheapest price. We present experiments to show; (i) Current auction markets on the web are inefficient, with trades taking place away from equilibrium price, and not all benefit from trade being extracted. (ii) Our agent is able to exploit these inefficiencies, resulting in it making higher profits than the simple strategy of bidding in a small number of auctions. (iii) As more participants use our agent, the market becomes more efficient. When all participants use the agent, all trades take place close to equilibrium price, and the market approaches ideal behaviour.

Applying evolutionary game theory to auction mechanism design

We describe an evolution-based method for evaluating auction mechanisms, and apply it to a space of mechanisms including the standard first-price and second-price sealed bid auctions. We replicate results known already in the auction theory literature regarding the suitability of different mechanisms for different bidder environments, and extend the literature by establishing the superiority of novel mechanisms over standard mechanisms, for commonly occurring scenarios. Thus this paper simultaneously extends auction theory, and provides a systematic method for further such extensions.

Agent-based service composition through simultaneous negotiation in forward and reverse auctions

Service composition is the act of taking several component products or services, and bundling them together to meet the needs of a given customer. In the future, service composition will play an increasingly important role in e-commerce, and automation will be desirable to improve speed and efficiency of customer response. In this paper, we consider a service composition agent that both buys components and sells services through auctions. It buys component services by participating in many English auctions. It sells composite services by participating in Request-for-Quotes reverse auctions. Because it does not hold a long-term inventory of component services, it must take risks; it must make offers in reverse auctions prior to purchasing all the components needed, and must bid in English auctions prior to having a guaranteed customer for the composite good. We present algorithms that is able to manage this risk, by appropriately bidding/offering in many auctions and reverse auctions simultaneously. The algorithms will withdraw from one set of possible auctions and move to another set if this will produce a better-expected outcome, but will effectively manage the risk of accidentally winning outstanding bids/offers during the withdrawal process. We illustrate the behavior of these algorithms through a set of worked examples.

AutONA: a system for automated multiple 1-1 negotiation

The AutONA (Automated One-to-one Negotiation Agent) system was conceived as a means of reducing these operational procurement costs, enabling procurement departments to automate as much price negotiation as possible, thus creating the option of reducing direct costs and/or redeployment of operational effort into strategic procurement requiring high human involvement. The problem domain has been limited to the automation of multiple 1:1 negotiations over price for quantities of a substitutable good subject to the organisations procurement constraints of target quantity, price ceiling and deadline.

A Dynamic Programming Model for Algorithm Design in Simultaneous Auctions

In this paper we study algorithms for agents participating in multiple simultaneous auctions for a single private-value good; we use stochastic dynamic programming to derive formal methods for optimal algorithm specification; we study a number of algorithms of complementary complexity and effectiveness, and report preliminary tests on them. The methods and analysis in this paper extend naturally to more complicated scenarios, such as the purchase of multiple complementary goods, although different problem areas bring their own challenges with respect to computational complexity.

A comparison between mechanisms for sequential compute resource auctions

This paper describes simulations designed to test the relative efficiency of two different sequential auction mechanisms for allocating compute resources between users in a shared data-center. For each of two possible auction types we apply a genetic algorithm to a broad class of bidding strategies to determine a near-optimal bidding strategy for a specified auction type, and use statistics of the performance of these strategies to determine the most suitable auction type for this domain.

Evolved hybrid auction mechanisms in non-ZIP trader marketplaces

A previous paper by D. Cliff (see ibid., 2002) demonstrated that a genetic algorithm could be used to automatically discover new optimal auction mechanisms for automated electronic marketplaces populated by software-agent traders. Significantly, the new auction mechanisms are often unlike traditional mechanisms designed by humans for human traders; rather, they are peculiar hybrid mixtures of established styles of mechanism. This previous work used software agents running the ZIP trader algorithm (recently shown to outperform human traders). We provide the first demonstration that qualitatively similar results (i.e., non-standard hybrid mechanism designs being optimal) are also given when similar experiments are performed using a different trader algorithm, namely Gode & Sunders (1993) ZI-C traders. Thus, the paper is the first to offer significant evidence that evolved hybrid auction mechanisms may be found that out-perform traditional market mechanisms for many styles of trader-agent.

Using a Genetic Algorithm to Design and Improve Storage Area Network Architectures

Designing storage area networks is an NP-hard problem. Previous work has focused on traditional algorithmic techniques to automatically determine fabric requirements, network topology, and flow routes. This paper presents work performed with a genetic algorithm to both improve designs developed with heuristic techniques and to create new designs. For some small networks (10 hosts, 10 devices, and single-layered) we find that we can create networks which result in savings of several thousand dollars over previously established methods. This paper is the first publication, to our knowledge, to describe the successful application of this technique to storage area network design.

On the Impact of Strategy and Utility Structures on Congestion-Averse Games

Recent results regarding games with congestion-averse utilities (or, congestion-averse games--CAGs) have shown they possess some very desirable properties. Specifically, they have pure strategy Nash equilibria, which may be found in polynomial time. However, these results were accompanied by a very limiting assumption that each player is capable of using any subset of its available set of resources. This is often unrealistic--for example, resources may have complementarities between them such that a minimal number of resources is required for any to be useful. To remove this restriction, in this paper we prove the existence and tractability of a pure strategy equilibrium for a much more general setting where each player is given a matroid over the set of resources, along with the bounds on the size of a subset of resources to be selected, and its strategy space consists of all elements of this matroid that fit in the given size range. Moreover, we show that if a player strategy space in a given CAG does not satisfy these matroid properties, then a pure strategy equilibrium need not exist, and in fact the determination of whether or not a game has such an equilibrium is NP-complete. We further prove analogous results for each of the congestion-averse conditions on utility functions, thus showing that current assumptions on strategy and utility structures in this model cannot be relaxed anymore.