Research On Multi-factor Stock Picking Strategy In A-share Market Based On Neural Network Methods

In recent years,computer big data and artificial intelligence technology have developed rapidly,thanks to the breakthrough of computer technology,quantitative investment has also beginning to explode.The essence of quantitative investment is to find investment strategies that can got stable returns by quantifying various economic indicators and establishing mathematical models.One of the effective methods is the multi-factor stock selection strategy.This paper attempts to use the method of neural network to explore the nonlinear relationship between effective factor and the return.Based on the characteristics of neural network self-learning,the weight of each factor can be optimized in repeated tests,and the loss function of the final model can be optimized.So that the neural network has a higher accuracy.Through that we get a model which can analyze the stock return,and use the neural network stock selection model to find stable return.In this paper,we try to find out the effective factors in the Chinese stock market by analyze trading information data and the financial information data from 2010 to 2018.Then establish a multi-factor neural network model based on TensorFlow framework.Use the effective factor of individual stocks as input data,the profit rate of individual stocks as output data,the parameters of the model are tested and adjusted repeatedly,including hidden layer number,number of hidden neurons,learning rate,etc.,and finally established a multi-factor stock selection model based on deep neural network.Through the experimental verification in this paper,we found a series of parameters for constructing multiple factor neural network stock selection model,and use it to predicted the return.The conclusion proves that the model established by neural network model can not only obtain high returns in historical data,but also predict future data well,and has the very good stock selection ability.