Research And Implementation Of High Frequency Stock Trading System Based On Neural Network

With the transformation of the financial industry by big data,cloud computing and artificial intelligence,quantitative trading has become an investment tool that investors pay more and more attention to.As a branch of quantitative trading,high-frequency trading focuses on obtaining highfrequency financial data in the market through programmatic trading,and using the analysis of high-frequency financial data to quickly complete transactions.In order to explore how to use high-frequency financial data to capture the micro-change information of the stock market and predict the trend of stock price changes in the short term,this thesis researches and implements a high frequency stock trading system based on neural network,so as to judge the trading timing,avoid risks and obtain ideal returns.The main work of this thesis consists of three parts.First,the SSA-GA-BP stock price prediction model is proposed.Based on the BP neural network model,a set of preprocessing flow for high-frequency stock market data is designed.Aiming at the problem that the BP neural network model is easy to fall into the local extreme value and the convergence speed is slow,this thesis optimizes the initialization process of the weights and thresholds of the BP neural network neurons,that is,when initializing the weights and thresholds,a genetic algorithm is used to search.After the first optimization,the weights and thresholds are obtained,and then the second optimization is carried out according to the sparrow search algorithm.Finally,the BP neural network model is trained according to the weights and thresholds after the second optimization,so as to establish the SSAGA-BP model.Second,this thesis designs and proposes two highfrequency stock trading strategies,including the ChanLun 1-min trading point strategy and the improved version of the ChanLun 1-min trading point strategy.Based on the basic theory of ChanLun morphology and ChanLun dynamics,this thesis designs and implements the ChanLun 1min trading point strategy that conforms to the ChanLun theorem and definition.Based on the SSA-GA-BP stock price prediction model,this thesis optimizes the trading strategy of ChanLun 1-min trading point,and proposes an improved version of the ChanLun 1-min trading point strategy,which improves the accuracy of identifying trading points,and optimizes the stop-loss ability and stability of the trading strategy.Based on the above high-frequency trading strategy,this thesis designs and implements a highfrequency trading system.In order to ensure the stability and robustness of the system,this thesis comprehensively tests the function points of each module of the system.In this thesis,the experimental comparison method is used to verify the optimization effect of the genetic algorithm and the sparrow search algorithm.The results show that the SSA-GA-BP model proposed in this thesis has a good prediction effect on the 1-min level stock price,and has stronger fitting ability for price turning points.This thesis conducts backtesting experiments on two high-frequency stock trading strategies.The experimental results show that the two strategies have a higher Sharpe ratio,a smaller maximum drawdown rate,and stable returns.The improved version of the 1-min trading point strategy has a higher accuracy of trading points,and reduces high-risk operations.It also has better stop loss capabilities,and stable income performance under complex market conditions.The research work of this thesis combines the neural network model and quantitative investment strategy to provide investors with valuable trading advice.The results obtained in this thesis have certain reference significance for the research on stock price forecasting based on neural network and the research on quantitative investment theory.