The Research Of Recurrent Neural Network In Stock Index Prediction

With the continuous development of Chinese economy and the improvement of economic strength,people pay more and more attention to the financial field.Among them,the stock market,as an investment market with both high risk and high investment,has become the field that many researchers pay more and more attention to predict the stock so as to make better stock selection and obtain the maximum retum.The higher risk of stock investment is precisely due to the instability of stock prediction,and its influencing factors include economic factors,market factors,political factors,etc.,in addition to the value of the stock itself.Therefore,the prediction of stock index needs to integrate macroeconomic variables and technical indicator variables,and there is no obvious linear relationship between these variables,which also brings great difficulties to the prediction of stock index.At present,the commonly used stock forecasting methods include the fundamental analysis method and technical analysis method for individual investment,as well as the analysis and prediction method that professional investment institutions use statistics,artificial intelligence and other relevant theoretical knowledge to combine with fiundamentals and technical aspects.In contrast,neural network method in the field of artificial intelligence has unique advantages in the processing of nonlinear problems.Based on this,we have completed the following work:Firstly,the LSTM based stock index prediction model is proposed according to the two characteristics of the temporal sequence and the non-linear correlation.LSTM is an improved model of RNN,introducing the concepts of timing and directional loop,which can solve the problems related to the input data before and after,and has long-term memory.Secondly,two methods of stock index prediction based on feature parameter selection and LSTM model are proposed,which are clustering algorithm combined with principal component analysis and genetic algorithm for feature combination method respectively.In the past,the prediction process often paid more attention to the construction of the model,trying to improve the prediction accuracy of the model by means of data preprocessing,parameter optimization,combining with a variety of network structures and other methods,but ignoring the important aspect of the stock influencing factors.For this reason,we improved the selection of characteristic parameters.In the empirical demonstration,this paper conducted an experiment with NASDAQ index and S&P 500 and analyzed the advantages of LSTM model compared with BP model.At the same time,two methods of stock index prediction based on feature parameter selection and LSTM model were used to carry out experiments,demonstrating and analyzing that the accuracy and speed of the model were improved after feature parameter selection.