BP Neural Network Stock Selection Model Optimized By Genetic Algorithm

As finance and the rapid development of computer technology,combining computer technology and finance fields become a kind of trend,the current when this model can bring effective high profits for the investors,the people will choose to use this way,the quantitative way to build a portfolio,such as quantitative stock selection,high-frequency arbitrage trading and quantitative timing quantitative investment way,etc.In fact,quantitative investment is to classify and extract all kinds of factors that have an impact on stock price.Based on a large amount of data,it adjusts the parameter weight of the influence of each factor on stock price repeatedly and rationally allocates the portfolio in a refined way to achieve the purpose of obtaining excess returns.By constantly adjusting the link weight in the network,the neural network algorithm is constantly approaching the optimal result,and is constantly applied by the majority of investors in the quantitative field.However,the neural network model also has its own limitations,among which it is easy to fall into the local optimal solution,which makes the neural network may not reach the expected effect in practical application,resulting in investment failure.Therefore,how to improve the neural network easy to fall into the local optimal solution,improve the actual effect of the neural network in quantitative stock selection becomes very important.To make neural network model can jump out of local optimal solution,make the model of stock selection effect in practice were promoted,so as to help investors gain excess profits,this paper combines the genetic algorithm and BP neural network method,by using genetic algorithm to optimize BP neural network will adjust link weights in the network model,built into GA-BP model,The model can have more choices when updating parameters and avoid falling into the local optimal solution.In this paper,the sample objects are selected from the constituent stocks of CSI 300 index,and the monthly data are extracted.The stock data between January 2012 and June 2021 are intercepted,and a total of 53 relevant market indicators,valuation indicators,risk analysis,financial analysis and technical indicators are selected.Then the sample data sets are divided into training sets and test sets.The period from January 2012 to December 2019 was divided into sample training sets,and the period from January 2020 to June 2021 was divided into sample test sets.Through the empirical analysis and performance measurement,the following conclusions:(1)In this paper,another machine learning method is found to optimize the defect of neural network,that is,genetic algorithm is used to optimize and adjust the link weight of BP neural network,so that the BP neural network model can get rid of its own limitations,that is,jump out of the local optimal solution,so as to improve the classification effect of the model.But GA-BP model show that the performance of returning the measure effect produce larger retracement,reasons for after improve classification accuracy,makes choose "buy" label of the shares on quantitative change is little,makes the influence of outliers,when investors applying GA-BP model to pick stocks recommended long-term strategy.(2)By elaborating GA-BP model construction process,this paper provides a set of effective model construction methods for stock selection in quantitative investment,including factor screening process,data pretreatment method,model parameter selection,model building process,stock selection strategy,etc.(3)As a machine learning method with good optimization effect,genetic algorithm is combined with BP neural network to construct GA-BP model,so as to improve the classification effect of the model and contribute to the optimization method of machine learning algorithm.