Multi-Agent Bidding and Contracting for Non-Storable Goods D.J. Wu, Yanjun Sun LeBow College of Business, Drexel University, Philadelphia, PA 19104, USA {Wudj, ys45}@drexel.edu Abstract We study electronic bidding and contracting for non- storable goods such as electric power using multi-GA agents and game theoretical approaches. In this framework, there is a long-term contract market as well as a back-stop spot market. Seller agents bid into an electronic bulletin board their contract offers in terms of price or capacity, while Buyer agents decide how much to contract with Sellers and how much to shop from the spot market. The problem is modeled as a von- Stackelberg game with Seller agents as leaders. We investigate if artificial agents will be able to discover equilibrium strategies if such an equilibrium exists; and if the agents can discover good and effective strategies when playing repeated non-linear games where there does not exist any equilibrium. This study is a companion of our earlier theoretical characterizations on optimal bidding and contracting strategies for non-storable goods, now adopting an agent-based approach. 1. Introduction In recent years, we have seen a growing interest in the study of artificial agents [7, 9] and their applications in electronic commerce, such as buying and selling goods over the Internet, automated bargaining and negotiation [4, 20], dynamic pricing in agent- mediated knowledge marketplace [24], multi-agent supply chain management [3, 5, 23] and multi-agent enterprise modeling [6, 12, 13, 15 21], restructuring the electric power industry [2, 10, 11, 12, 13, 14, 22], and off-exchange trading [8]. This paper is in this “tradition”. We study a general multi-agent bidding, auction and contracting support system (called “eBAC”) and apply it to the selling and buying of non-storable goods such as electric power. eBAC is implemented using an earlier theoretical framework developed by Wu and his colleagues [2, 16, 17, 18, 19, thereafter WKZ for short]. In this framework, there is a long-term forward/contract market as well as a back-stop spot market. Seller agents bid into an electronic bulletin board their contract offers in terms of price or capacity, while Buyer agents decide how much to contract with Seller agents and how much to shop from the spot market. The problem is modeled as a von-Stackelberg game with Seller agents as leaders. Thus, our eBAC auction has some unique properties different from traditional auctions in the following ways: First, traditional bids are discrete choices while our agents’ bids are continuous; Second, goods in traditional auctions are storable and those in ours are non-storable; Third, traditional auctions only consider a single Seller while ours considers multiple Sellers; Lastly, traditional auctions consist of only one source of procurement (the spot market), while ours consists of two sources: in addition to the back-stop spot market, we allow Buyers and Sellers sign long-term contracts in a forward market. The goal of this study is as follows: First, we investigate if artificial agents will be able to discover equilibrium strategies if such an equilibrium exists; Second, we investigate if the agents can discover good and effective bidding, auction and contracting strategies when playing repeated non-linear games where there does not exist any equilibrium; Third, we explore the emergence of trust in the sense of what kind of mechanisms induce cooperation in the above setting. We test eBAC in the multi-Seller electronic marketplace. In this paper, we focus only on myopic bidding strategies, while in a separate study, we extend to non-myopic bidding with an extensive focus on the emergence of trust [20]. The rest of the paper is organized as the following. Section 2 provides a brief literature review. Section 3 describes our methodology and implementation. Section 4 experiments on multi-agent price or capacity bidding. Section 5 reports results of further experiments. Section 6 summarizes our findings. 2. Literature Review In this section, we briefly review the auction of non-storable goods in WKZ papers and explain why an agent-based approach is interesting in this arena. In the WKZ model (which might be thought of as the “month ahead” market), Sellers and Buyers 0-7695-0981-9/01 $10.00 (c) 2001 IEEE 1 Proceedings of the 34th Hawaii International Conference on System Sciences - 2001