Research Of Multi-Agent-based Artificial Stock Market In A Distributed Environment

Represented by SFI-ASM (Santa Fe artificial stock market), modeling andsimulation of the stock market based on multi-Agent technology is one of the mainmethods to study the nature and regular of the stock market, and made a series ofimportant research results. But these ASM still have some insufficient parts, whichmainly including: Firstly all the agents in these ASM are serially running on a singlecomputer, and they are scheduled based on a queue, while the investors in real stockmarket are definitely parallel. Secondly the agents in these systems cannot achievereal-time interaction, which severely constrains the research on the effect brought tostock market behavior by interaction network among agents.This paper aiming at these problems, and present a Distributed Parallel ArtificialStock Market (DPASM) by applying the distributed parallel technology andMessage-oriented middleware technology to the artificial stock market based onmulti-Agents, and solve the above problems, wherein the main research and workincludes three parts:First, we study Agent and the run control structure of Multi-Agents, and designthe decision-making algorithms library and message exchange interface of Agentwhich realized respectively intelligent decision-making capability and messageexchange capability, adopt fully distributed run control structure betweenmulti-Agents, with failure independence between the Agent which increase systemstability, and Multi-Agents is completely parallel, in line with the relationshipbetween the real stock market investors.Secondly, we study the trade mechanism and the structure of stock trading market,in a distributed environment, the Agent may submit an order at any time, and any timethere may be more than one Agent orders too, so stock trading market need thecontinuous double auction trading mechanism, while taking advantage of the dynamicthread pool to achieve high load and high concurrent capacity.Finally, we study structure and method of message exchange betweenmulti-Agents, while taking into account that Multi-Agents interaction network is fixedduring simulation, we adopt centralized communication structure and message queuetechnology to achieve a lightweight and effective multi-agents Message-oriented exchange middleware-ASMMQ, it used fixed publish/subscribe method to achievereal-time and asynchronous message exchange between multi-Agents, and messageexchange process is transparent for all Agents.In addition, this article also implements a human-computer interaction terminal,which is used to initialize the settings of the market property, intuitive shows the priceand volume of stocks traded in simulation process, and you can control the simulationrun.