Long Memory Test And Prediction Of Shanghai Securities Composite Index

In recent years,China's financial industry has continued to grow and develop.As a part of the financial market,the securities market has naturally becomes one of the research objects.Studying the memory characteristics of the Shanghai Securities Composite Index will not only help to understand the development of the securities market,but also provide practical guidance for investors.Based on the time series period and related variables of the Shanghai Securities Composite Index,the predictable models and effective estimates are obtained to avoid the external risks.The main contents and results of this paper are as follows.1.The closing price of the Shanghai Securities Composite Index from 1991 to 2018 was selected as the original sample,and Logarithmic and logarithmic difference samples were obtained by logarithmic and logarithmic difference processing.The memory of samples was tested by the method of R/S analysis,modified R/S analysis,V/S analysis,DFA analysis and MFDFA analysis.It is found that the logarithmic difference processing will affect the memory of the Shanghai Securities Composite Index.However,the results of the five analysis methods indicate that the Shanghai Securities Composite Index has long memory characteristics.2.Taking into account the impact of time and ups and downs on the Shanghai Securities Composite Index,the original sample is divided into three different time periods,and each time period has also been removed the large fluctuations.Then,the memory test is performed on the obtained samples.The research results show that the long memory test of the Shanghai Securities Composite Index is indeed affected by time and ups and downs.The memory index obtained in different time periods are different and sometimes the Hurst index is less than 0.5,which explains the problem that the memory of the Shanghai Securities Composite Index is different in many studies.Compared with the time factor,the change of the memory index obtained by removing large fluctuations of samples is more obvious,that is,the impact of ups and downs on the long memory test of the Shanghai Securities Composite Index is greater.3.The parametric model and nonparametric model are used to predict the closing price of the Shanghai Securities Composite Index.Taking the closing price of Shanghai Securities Composite Index in recent years as the original sample.In order to achieve the research purpose,the sample was normalized and removed the large fluctuation,and the normality and unit root test were performed to obtain samples in which the sample was non-normally unstable time series.The ARFIMA model and the GRU model are used to predict the processed samples.The results show that the GRU model is better than the ARFIMA model,whether it is the original sample or the up-and-down sample,long-term prediction or short-term prediction.4.When using the same model to predict the original sample and the sample of remove the large fluctuation,it can be clearly found that the predicted result of the sample of remove large fluctuation is better than the original sample.This shows that the data processing method of removing large fluctuations can reduce the predictable error of the closing price of the Shanghai Securities Composite Index.The long memory characteristics of the Shanghai Securities Composite Index provides favorable condition for its predictable research.In the predictable results of ARFIMA model and GRU model of the original sample and the up-and-down sample,it is found that the GRU model with the up-and-down sample has the best predictable results.This shows that data preprocessing and model selection will affect predicted results of the Shanghai Securities Composite Index.Therefore,the focus of predictable research of the Shanghai Securities Composite Index is that effective data processing methods and adaptive model selection.