Prediction Analysis Of Shanghai And Shenzhen 300 Index Closing Price Based On Time Series

The stock market,which is at the center of the economic information center,has collected business information in many fields through a series of transactions.Predicting the future status of the stock has always been a hot spot for stock market participants.Among the many indicators of stocks,price is one of the most direct indicators.Predicting stock prices is of great significance to stock market participants.On the one hand,it enables investors to allocate resources reasonably for investment,on the other hand,it can promote reasonable stock price fluctuations and play the function of an economic barometer.However,the stock market is affected by many factors,and the price data exhibits characteristics such as non-stationary,non-linear,and highly noisy.How to use these characteristics reasonably to find a suitable method for forecasting is always a question worthy of continuous research.This article takes the daily closing price data of the CSI 300 Index as the research object and uses three methods to predict it.The first,establish ARIMA model for non-stationary closing price data.After the first-order difference,the closing price becomes a stationary time series,so an ARIMA model can be established for prediction.The second,Establish an LSTM model for non-stationary closing price data.No longer pay attention to whether the closing price data is stable,and choose a neural network model LSTM that does not require data for prediction.Since the research object is one-dimensional data,the selection of the prediction window length is determined by the autocorrelation coefficient of the closing price.The third,Decompose the closing price data and establish the LSTM-ARMA model.Using Singular Spectrum Analysis to decompose the closing price into independent trend and volatility terms.The trend term is non-stationary,and an LSTM model is established for it.The volatility term is stable,and an ARMA model is established for it.The sum of the two forecast results is used as the forecast value of the closing price.Analysis of the three prediction results,LSTM-ARMA has the best prediction effect,ARIMA takes the second place,and LSTM has the worst.The prediction effect of the combined model is better than that of the single model.The ARIMA model that theoretically does not meet the characteristics of the data has achieved good prediction results.Although the LSTM model has the worst prediction result,its prediction value is smoother.And from the LSTM prediction results of the smoother trend terms obtained from the time series decomposition,the LSTM model may be more suitable for the prediction of smoothed data.The combination model obtained the best prediction results,thanks to the decomposition of the time series into simpler,more characteristic sub-sequences.The time series decomposition algorithm,Singular Spectrum Analysis,is a non-parametric method for studying nonlinear time series.Set the window length to one-third of the data volume to construct a trajectory matrix,and perform singular value decomposition on the trajectory matrix to obtain multiple components that are independent of the original data.The singular values are different,and the corresponding components contain different information about the original data.This article combines the components based on their cumulative contribution rate to obtain two independent subsequences of the closing price data,the trend term and the volatility term.The empirical results show that the decomposed subsequences are easier to capture features and get better prediction results.Through the overall description of this article,you can understand which method is more applicable in stock index price prediction,and provide technical support for future stock index price prediction.The forecast in this article has achieved relatively satisfactory results,which can provide relatively objective data forecast support for market participants.