Research On Stock Index Prediction Based On Nrural Network

The effective prediction of the stock price index helps to explore the inherent laws of stock prices,thereby avoiding financial risks in a timely manner and improving the stability of the stock market.Therefore,it is of great theoretical and practical significance to study an efficient stock price index prediction model.The powerful self-learning and adaptive features of artificial neural networks make their prediction results better than traditional time series models.With the rapid development of artificial neural networks,deep neural network algorithms have also been continuously applied to stock market research.As a type of deep neural network,DIDLP is a deep neural network.It is composed of linear operators and non-linear operators.Current research shows that the DIDLP model can effectively improve the problem of firstorder delay in stock time series prediction.Therefore,the paper introduces DIDLP neural network to study the prediction of stock price index.This paper first summarizes the research methods of financial time series forecasting,introduces traditional time series models and artificial neural network models,and focuses on the basic principles of deep and decreasing linear neural networks.The DIDLP neural network was optimized based on the results of the pre-analysis,and a DIDLP neural network model with regular terms was established.Finally,the model was applied to the prediction of stock price index series.The main research work is as follows:(1)In order to grasp the characteristics of the stock price index,use BDS,Hurst index,discrete derivative and other methods to perform statistical analysis on the stock index series.The analysis found that the Hurst Index of the CSI 300 Index was 0.6.This shows that the CSI 300 Index has a certain long-memory correlation,which has certain predictability,but because its Hurst Index is close to 0.5,it is less predictable.The difficulty of prediction is that the prediction results of most prediction methods have a strong first-order delay.Further analysis found that the reason why the CSI 300 index is difficult to predict is that the stock index series shows a short-term linear relationship,and the long-term non-linear relationship is accompanied by increasing and decreasing trends.(2)According to the characteristics of the stock price index series,DIDLP neural network is introduced.By using the early stopping method and regularization method to optimize the DIDLP neural network model,the risk of model overfitting is reduced.Finally,a DIDLPregularized neural network prediction model suitable for stock price index is proposed and established.(3)In order to measure the performance of the DIDLP-regularization model,ARIMA,ARIMA-GARCH,DNN,and DNN-regularization models are established as reference models,and multiple sets of experiments are designed to verify the model prediction effect.Using the closing price of the CSI 300 Index as a data set,the algorithm is implemented using MATLAB software.The results show that the DIDLP-regularization model has the best prediction effect,and the model shows obvious advantages in the prediction of the CSI 300 index.