Stock Representation And Quantitative Trading Based On Machine Learning Methods

Quantitative trading refers to a securities investment method that uses mathematical and statistical methods and computer technology to conduct transactions[1].Compared with subjective trading,quantitative traders can effectively reduce irrational trading de-cisions through trading procedures,thereby achieving a more stable profit.In recent years,how to use machine learning technology in the field of quantitative trading has become a central topic of financial technology[2].However,to apply artificial intelli-gence technology to quantitative trading faces many challenges,such as how to represent noisy high-frequency financial data,and how to automate the development of quantita-tive trading strategies.Both valuable and noisy information coexist in the high-frequency data.Con-cretely,the learning process of high-frequency factor extractor would be easily over-whelmed by noise,tending to cause over-fitting.Moreover,common tricks of preventing over-fitting lead to poor performance on this task,since they usually roughly restrict the model capacity and thus can hardly model complex trading signals in high-frequency data.That is,when designing high-frequency factor extractors,we face a tough dilemma-a high-capacity model would easily over-fit to noise,while a simple but robust model could not capture complex high-frequency patterns.To address these problems,we pro-pose to maintain the model capacity while preventing over-fitting by constructing two components and balancing the information and noise through interactions between them.Specifically,we propose a novel learning framework,named Digger-Guider,to extract informative stock representation from noisy higher-frequency data.We develop a high-capacity model,called Digger,to work on information-digging of the high-frequency data by extracting local and detailed features.We also design a robust model,called Guider,to capture global tendency features and help the Digger overcome the noise.The Digger and Guider enhance each other by mutual distillation during training.Ex-tensive experiments on real-world datasets demonstrate that our framework can produce powerful high-frequency stock factors,which can significantly improve stock trend pre-diction performance and our understanding of the finance market.Complete quantitative trading includes not only price trend forecasting,but also trading strategy design.In order to realize the design of automated trading strategies,we propose an adaptive trading model namely iRDPG,to automatically develop QT strategies by an intelligent trading agent.Our model is enhanced by deep reinforcement learning(DRL)and imitation learning techniques.Specifically,considering the noisy financial data,we formulate the QT process as a Partially Observable Markov Decision Process(POMDP).Also,we introduce imitation learning to leverage classical trading strategies useful to balance between exploration and exploitation.For better simulation,we train our trading agent in the real financial market using minute-frequent data.Ex-perimental results demonstrate that our model can extract robust market features and be adaptive in different markets.