Analysis Of The Securities Market Based On The EEMD-MARS-SVR Model

In this paper,we propose an analysis and prediction model composed of Ensemble Empirical Mode Decomposition(EEMD),Multivariate Adaptive Regression Spline(MARS)and Support Vector Regression(SVR),and apply it to modeling and forecasting of financial time series.Among them,EEMD is an improved version of the Empirical Modal Decomposition(EMD),which introduces white noise into EMD,thus effectively solves the defects of one IMF in EMD.It is generally applied to data processing in communication and IT field.The advantage of this method is that there is no need to set any of the base functions in advance.MARS uses the spline function to fit the local complex nonlinear relationship,and can get the weight of the explanatory variable through the pruning process,which can be used for the selection of variables.SVR is a predictive model commonly used in machine learning and is widely used in IT and finance fields to train suitable models and predict them for a given training set.This paper first reviews the development of modern stock pricing theory,the commonly used forecasting model and the macroeconomic variables on the securities market,defines the research framework of EEMD-MARS-SVR as the analysis model and introduces 10 groups,84 categories of public economic variables indicators.This paper argues that these indicators of public economic variables are likely to have an impact on the stock market and are left to be screened.Then,this paper analyzes the time series of the five index of the China stock market and the 300 constituent stocks of the HS300,respectively,for the period from June 1,2010 to November 30,2016,for a total of 66 months,1579 trading days of data.In the process of analysis,EEMD decomposition of the target time series is carried out,and several IMFs are decomposed,and then the IMFs are reconstructed into high frequency sequences,low frequency sequences and trend sequence.Then,using the MARS method,the variables index related to the high frequency sequences and the low frequency sequences were screened out among the 84 common economic variables.These variables can be used to prediction,respectively.Finally,the public economic variables and target sequences of the screens are introduced into the SVM analysis framework to make prediction.In order to evaluate the prediction effect,this paper introduces the mean absolute percentage error(MAPE)and the directional symmetry metrics(DS),and compares it with other prediction models.The results show that the forecasting method has significant difference between the intra prediction and the prediction advantage.In the process of simulated stock choosing,this method has value of investment in theory.