Multi-strategy Integration for Actionable Trading Agents

Trading agents are very useful for developing andback-testing quality trading strategies to support smarttrading actions in the market. However, the existingtrading agent research mainly focuses on simple andsimulated strategies. As a result, there exists a big gapbetween academia and business when the developedtrading agents are deployed in the real life. Therefore, theactionable capability of developed trading agents is oftenvery limited. In this paper, we introduce approaches foroptimizing and integrating multiple classes of strategiesfor trading agents. Five categories of trading strategies,including 36 types of trading strategies are trained andtested. A strategy integration and optimization approachis proposed to identify golden trading strategy in eachcategory, and finally recommend positions associatedwith these golden strategies to trading agents. Test in fiveinternational markets on ten years of data respectivelyhas shown that the final strategies recommended totrading agents can lead to high benefits while low costs.Concurrent execution of positions recommended by allgolden strategies can greatly enhance performance.