Research On Supporting Technique For Speculative Behavior Analysis In Multi-Agent Based Futures Trading Platform

For complex financial markets like real Chinese stock index futures market, the relationship between the price fluctuations and investors, such as speculative trading, conveys information that is of primary importance to understand financial markets and help policy design. However, it still remains unclear how heterogeneous traders with different characteristics affect the market. And researchers can use an agent-based method to analyze and study price dynamics from the angle of traders' micro-behavior. But the existing multi-agent platforms encounter the following problems: Firstly, those platforms are very old, and the agents are not extensible. Take the most famous SWARM platform which is proposed in 1999 for an example, it could not meet the current financial research needs of multi-agent simulation. The agent of SWARM is not smart enough to mirror the actual traders and hardly extensible. Secondly, most of the platforms is a standalone application which leads to a bottleneck: a limited number of agents. The simulation is not convincing without a sufficient number of agents. Thirdly, the experimental results are very abstract, and the more important detailed historical transaction data are not stored in a perfectly design database for future analysis.Aiming at the deficiencies of the existing platform, this paper covers supporting technique for speculative research based on multi-agent platform. First of all, the multiagent platform, proposed by this paper is supporting agent type extension. The agent is abstracted as agent module and trading strategy module. The agent module is handling tasks like the agent life cycle, while the trading strategy module is responsible for agent diversity. Users can add new types of agents and trading strategies in a method similar to spring injection. And then, the platform is taking advantage of C/S architecture which makes the platform more flexible. So the new nodes could be easily added and overcome the bottleneck of limited agents. The client side and the server side are communicating through network, which makes the simulation more realistic. Meanwhile, Java NIO technique is used in managing massive concurrent I/O requests from clients which reduce the communication latency and thus enhance platform performance. At last, hybrid storage architecture model is proposed in this paper. The hybrid storage architecture makes use of both relational and NoSQL technologies. The architecture is very flexible as it allows for the addition of a different storage solution without changing client software. This can be done easily accomplished which benefits from using the spring injection framework. The massive historical data stored in the hybrid database could reveal the business laws and the trader trading futures through further research.In order to verify the multi-agent platform, several experiments are designed and performed to test the availability and efficiency of Non-Blocking IO and hybrid storage architecture. In conclusion, the test result shows that more agent could be handled and the response time shortened by using Non-Blocking IO. The stress test using benchmark reflects that hybrid storage architecture is more suitable than relational databases in supporting the platform.