Research On Quantitative Trading Based On Deep Reinforcement Learning

In the era of mobile internet,the diversification,security and convenience of financial transaction market make financial market produce massive data every day.Traditional financial transaction analysis methods are hard to deal with the massive financial data.There is no effective technical strategies for short-term and medium-term market analysis.The existing methods rely on the analysis of the stock trend chart and fundamental analysis,or national financial policies and quarterly report of enterprises.Moreover,the trading strategies are influenced by the negative emotions of human traders.In the face of massive historical data,human traders are uncapble to explore the hidden pattern in data.Nowadays,as the development of big data and artificial intelligence,more and more companies in the financial field employ artificial intelligence technology to realize the quantitative trading.Ant Financial Services Group,Jingdong Finance,Internet Crowd-funding Company and Virtu Financial have developed financial technologies based on artificial intelligence,which makes the artificial intelligence play an increasingly important role in the field of investments.Based on the existing methods and technologies of financial market analysis,this paper aims to the establishment the research of quantitative trading which combines the technologies of deep learning and reinforcement learning.This research focuses on finding an optimal trading strategy from a large number of financial historical data to reduce the cost and risk of transaction for investors.To find an optimal trading strategy,we employ a new artificial intelligence method,Deep Reinforcement Learning(DRL),to extract features from the historical stock price series and the traditional trading indicators based on the mathematical and the statistics methods.The selected features are fed into the deep neural network to extract useful features.Then,reinforcement learning method is used to optimize the parameters of deep neural network,so that the model can find the optimal trading strategy.The experiment are implemented on the data of the blue-chip sotcks in Shanghai and Shenzhen stock markets.The back-testing results shows that the market in the downturn trend,reduce the loss of investors,and the market in the upturn trend,properly expand the rate of return of investors.In these models,according to the annual return rate of stock,the market goes down for a long time to help reduce some investment losses.The market goes up,it helps investors increase the return by 5%~13%.To sum up,we aim to explore the models and methods of quantitative trading based on deep reinforcement learning.This methed tries tomine the hiding patterns from the high dimensional data and find the best trading points in real-time stock market,in order to reduce the trading risks of investors and provide investors with good decision-making tools.The system fully combines the advantages of traditional financial quantitative trading method and artificial intelligence technology and embodies its uniqueness and superiority in the field of financial investment.It has extremely important practical value and research significance.