Markets without Makers - A Framework for Decentralized Economic Coordination in Multiagent Systems

Most electronic marketplaces are derived from a client/server model, with a central coordinator institution in the middle and a closed group of market participants submitting bids and asks to that institution. In contrast, technical views on electronic commerce often envision ad-hoc cooperation between market participants in open and decentralized IT environments, where software agents negotiate for their human principals. Such environments will naturally form unregulated market-coordinated multi-agent systems with selfish agents negotiating for utility maximization, and they need concepts for decentralized economic coordination a mechanism for distributed resource allocation that works without a market maker, with maximum privacy, security and coherent coordination as result. This article describes a framework used for the realization of a multiagent system which coordinates a supply chain using a decentralized economic approach. 1 Markets Are Needed, Market Makers Are Not The short history of electronic marketplaces has already seen different business models appearing and vanishing again. The recent paradigm connected with all these business models seems to be the ubiquity of market makers, central institutions in the center of a electronic marketplace whose sole purpose is to bring supply and demand together. These are either catalogs, arbitrators, matchmakers, or auctioneers, and they gain money by either access fees or transaction fees. But are they necessary? Are centralized institutions mandatory to control economic coordination? Or is that view just another occurrence of the “centralized mindset” [27], which says that where is coordination, there has to be someone who controls? We should note that our physical world does not have „the“ market for steel, beverages, paper, not even for commodities like stocks. The multitude of marketplaces available creates a dynamic network of economic opportunities, interconnected supply and demand bids which create the dynamic pattern of constantly changing market situations. This emergent network is what we regard as „the market“ for one good, and the dependencies between the different goods and markets lead to our understanding of „the economy“. But what is a market, anyway? From an economic perspective, it is “paradoxical how variously and vaguely defined the concept of market is" [18]. The mechanism of price generation in a market is mostly connected to Adam SMITH’s “invisible hand” [33], but until today there exists no commonly agreed-upon description of this market mechanism. The dominant neo-classical equilibrium theory, for example, has been criticized that “to answer the question, which mechanisms effectuate an equilibrium state, it has contributed only little and unsatisfyingly” [4]. Generally spoken, "markets describe the exchange and production of commodities in the private property regime” [20]. In other words, the purpose of the market is the realization of a dynamic, distributed resource allocation mechanism. Economics is thus essentially all about the coordination of systems consisting of utility-maximizing agents, who satisfy their needs using some mechanism for solving a distributed resource allocation problem. The effect of this mechanism is a state where prices are set so that supply and demand is perfectly balanced: the general equilibrium. The proverbial invisible hand of Adam SMITH was a first concept of how a decentralised mechanism, without a central (thus visible) co-ordinator, leads to that desired effect, but SMITH gave no implementation of that mechanism. A century later, Leon WALRAS proposed the tâtonnement process therefore [37]. WALRAS made the invisible hand visible by introducing a central auctioneer, who iteratively solved the allocation problem out of his total knowledge of supply and demand. Most of today’s economic research now investigates into the equilibrium state, and relies on WALRAS’ tatônnement process as a valid picture of the mechanism. Starting with KEYNES, neoclassical economics even went one step further: the homo oeconomicus market participant is completely rational, has total knowledge of the market situation and processes this information in infinitesimal speed. A transition from one market state to another virtually takes no time – a dynamic picture of the mechanism is not needed at all.. In contrast, economic research on decentralized self-organization still aims at explaining the mechanism of the invisible hand. Not being satisfied with either the homo oeconomicus or the tâtonnement process, the research fields of evolutionary economics [21], econophysics [19] and neo-austrian economics (HAYEK’s spontaneous order) [12] uses a bottom-up approach to explain economic coordination. Multiagent system implementations of the coordination mechanism can be distinguished according to the types mentioned above. Computable general equilibrium models [31] apply no dynamic mechanism at all, but calculate results out of total, global knowledge. The WALRASIAN auctioneer is directly implemented in marketoriented programming [38]. In this article, we show how the third possibility, the decentralized self-organization, can be implemented and leads to coherent economic coordination with predictable results. In the remainder of this article, we will first clarify some definitions and introduce a framework for building agent-based electronic marketplaces. Section 3 describes the software architecture of a prototypical multiagent system that coordinates by decentralized automated negotiation and price signals, and shows the results of some experimental runs. The article closes with an outlook to possible application areas and further research. 2 A Framework for Economic Coordination in Multiagent Systems To distinguish between the different possible types of agent-supported electronic marketplaces, we modify the coordination framework described by MALONE [17] and KLEIN [15] to describe coordination in multiagent systems. To characterize different coordination processes more precisely, they are described here in terms of successively deeper levels of underlying processes, each of which depends on the levels below it. These layers are analogous to abstraction levels similar to the protocol layers for network communications. The advantages of layered frameworks are reduced complexity, hierarchical ordering of terms and functions, and the identification of dependencies [34].