Prediction Of Stock Indexes In Different Maturity Markets Based On Recurrent Neural Network Model

The stock market is a barometer of the economic performance of the country and has always been highly valued by the government and researchers.However,the high noise and non-linear characteristics of the stock market make stock prediction an academic problem.The prediction effect of traditional time series models is often unsatisfactory.With the rise of neural network research again,the reduction in computational cost makes it possible to train deep models.Among them,cyclic neural networks can not only solve complex nonlinear problems,but also introduce the concept of time series and thus have memory,so they have been widely used in financial time series prediction in recent years.In terms of real stock trading: the performance of models in markets of different maturity levels is very different,so it is of greater practical significance to differentiate and model the maturity levels.In terms of model improvement: In recent years,the improvement of the recurrent neural network model applied to the financial field is limited to the combination of linear and nonlinear models,and the improvement in prediction performance is not obvious.Therefore,this paper combines the idea of compression and noise reduction in the field of image processing,connects the autoencoder to the recurrent neural network model,and improves the quality of the input features to improve the prediction performance of the model.This article uses data related to Dow Jones Index,Hang Seng Index and Shanghai Composite Index as experimental data.First,establish a traditional time series model and a recurrent neural network model in these three different maturity markets for comparison.Secondly,the recurrent neural network model is improved,and the automatic encoder noise reduction method(AEs-RNN)is compared with the linear smoothing method(ARIMA-RNN).Finally,the signal function is used on the basis of the traditional time series model and the improved recurrent neural network model,and it is converted into a classification model for use in trading strategies and the accuracy is compared.The experimental results show that: first,the prediction accuracy of the recurrent neural network model is always higher than that of the traditional time series model,and the performance of the LSTM model is better than GRU in the market with lower maturity;second,the improvement based on automatic encoder noise reduction The algorithm has better prediction accuracy than the commonly used linear improvement algorithm,and the model is more robust.Third,the price rise and fall judgment of the recurrent neural network model is more accurate than the traditional time series model,and it can assist investor decision-making in the more mature market.