Research On Deep Quantitative Trading Strategy Of Futures

In the financial sector,the trading market of securities has become increasingly mature.Although futures is one of the important investment objectives of the trading market of securities,the unpredictable changes in the futures market and the rapid growth of futures data have caused the traditional futures prediction and trading methods to waste a lot of human and material resources.In recent years,artificial intelligence has played an important role in the development of the financial field,and quantitative trading has attracted more and more investors due to its stable and efficient performance.Therefore,how to apply quantitative trading technology to the field of futures has become one of the hot research fields that domestic and foreign scholars pay attention to.First,a price prediction method of futures based on difference LSTM is proposed.This method introduces the difference in mathematics into the long-term memory cell state of the hidden unit of the long and short-term memory neural network(LSTM),and constructs a difference LSTM model.The difference LSTM model can not only learn the characteristics of futures price at a single point in time,but also learn the characteristics of non-linear changes in futures price between different points in time.In the empirical analysis,the short-term(1 day),mid-term(5days)and long-term(20 days)close prices of the historical trading data of 50 commodity futures involving agricultural futures,metal futures and energy futures were selected to verify the effects of the LSTM model,the first-order difference LSTM model and the second-order difference LSTM model in different time periods for futures price prediction.The results show that the difference LSTM model has a significant improvement in the accuracy and stability of the futures price prediction compared with the LSTM model.Then,a quantitative trading strategy of futures based on deep policy gradient is proposed.This strategy builds a trading agent based on the difference LSTM neural network and uses the policy gradient algorithm.First,the differential LSTM neural network is used to output useful features of futures trading,and then based on the features,the policy gradient algorithm is used to optimize the parameters of the difference LSTM neural network,so that the trading agent can explore the optimal trading strategy.In the experimental analysis,the historical trading data of five commodity futures,which are more concerned by the country and the general public,including fresh apples,crude oil,gold,yellow corn,and iron ore,are selected as the research objects to verify the profitability of the trading agent based on the fully connection layer network,the trading agent based on the LSTM network,the trading agent based on the first-order difference LSTM network and the trading agent based on the second-order difference LSTM network in quantitative trading of futures.The results show that the trading agent based on the difference LSTM network has a significant improvement in the return and the rate of return in the quantitative trading of futures compared with the trading agent based on the fully connection layer network and the trading agent based on the LSTM network.In general,this thesis applies deep learning and deep reinforcement learning to price prediction and quantitative trading in the field of futures respectively,so that when faced with the intricate futures market,it can capture the nonlinear changes of futures price data,accurately and effectively predict the futures price trend,and choose optimal trading strategy to maximize profits.Therefore,the methods proposed in this paper have strong research significance and practical value.