Research Of The Platform For Stock Markets Simulation With Parallel Agents In Clusters

Stock market is an important part of financial market, and the simulation platform of stock market provides a reliable modeling method for financial researchers. With the simulation platform, researchers can propose innovating algorithms and verify whether the new algorithms fit the market, which provides effective means to achieve a healthy innovation environment in financial market.Participants in stock markets of real system are large amounts of investors, which initiatively pursue best market benefit. Based on this theory a stock market can be regarded as a complex adaptive system which consists of large amounts of self-adaptive agents that communicate with each other. The complex adaptive system shows a feature of evolution with simulation going on. Agent-based modeling(ABM) is an effective method to simulate such complex systems, and this paper introduces a simulation platform for stock market using ABM.In traditional ABM method, all the agents run in a sequential scheme and the simulation goes on in “round” scheme, in which a round is a moment that activities happening at a same round are concurrent activities in real systems. In this paper we introduced parallel agents to ABM, and then we can take the advantage of multi threads to mimic the concurrency of real investors, and parallel agents can make use of the computing resources of a cluster to enhance the performance of the simulation platform. On the other hand, we proposed a concurrent scheduling system, so that agents can run independently which express the relation of agents more exactly. At the same time, the macro phenomenon of complex system will emerge only when agents have interactions on each other, so that the distributed communication among agents is also an important part of the simulation platform. In the communication system we take a compromise between a centralized communication mode and a P2 P communication mode to support effective communication. We complete the communication system with message queue middleware to support cross platform, and shade the physical cluster environment from upper agents. What's more, to keep the consistent communication interface, we regard the exchange market as a complex agent.The simulation platform provides easy interface for financial researchers to customize their own financial algorithms or models to verify the validity. Those interfaces includes: to customize the decision-making strategy of an investor agent, to customize the communication network among agents, and to customize the auction mechanism of the market.At last this paper verify the validity of the platform with the stylized facts, and test the performance of the platform of stock market simulation with parallel agents based modeling, which also shows a good scalability.