Research On Quantitative Investment Based On Neural Network

With the rapid development of computer science and technology,especially the arrival of the era of big data and cloud computing,quantitative investment is highly sought after,because this investment method is based on economics,statistics,computer science and technology,data mining technology and other scientific branches,based on scientific methods and history,from seemingly disordered data,to explore the actual value of stocks.Because the nonlinear characteristic of neural network is consistent with the relationship between stock price and stock index,this paper mainly uses the neural network modeling method to analyze the stock price.The data input is the annual report of Listed Companies in the A-share market and the corresponding performance forecast table.Combined with the corresponding macroeconomic data and the stock market data,the corresponding preprocessing operation is carried out on the stock index.Thus,the BP neural network and RBF neural network models are established.In the model training,the inhibition over fitting mechanism and the corresponding super parameter adjustment are added to make the model The empirical study found that when two kinds of neural networks predict the stock price,the actual prediction accuracy is relatively small,and with the improvement of the stock index information over the years,the prediction accuracy of the stock will increase,up to about 36%.Then through the neural network model to predict the stock over the years,the first 50 stocks with the lowest loss value are selected as the stock basket,so as to realize the quantitative stock selection,and carry out the quantitative investment on joinquant platform.The results show that the quantitative strategy selected by the neural network model basically outperforms the stock market yield based on the CSI 300 index,while the actual stock selection,the plan can be based on the investor’s preference.The innovation of this paper lies in: it mainly relies on the basic indicators to predict the stock price,and adopts the way of classification to abstract the stock price through onehot,avoiding the problem that a large number of indicators are difficult to linear regression.The classification function of RBF neural network is applied to the classification of stock price,and the two neural network models are optimized by adjusting the super parameters,which makes the model more concise and effective.