The Trend Of Economic Time Series Analysis And Empirical Research

Economic time series data reflects social economic operation status, it has greatsignificance to analysis economic time series correctly. Economic time series is unstable,complex and hard to predict. There are several methods to analysis economic time series,including traditional econometric methods, statistical methods developed in recent yearsand the machine learning method. These methods have certain advantages, and also someshortages. Econometric methods requires tough assumptions, neural network method is easyto be over-fitting.In order to improve the accuracy of the economic time series trend analysis further, thispaper will combine ARIMA model and support vector machine (SVM) model, becauseARIMA model has high accuracy in linear time sequence analysis, and support vectormachine model has good ability on nonlinear time series model, so this paper combine themto form mixture model to analyze economic time series.This paper selects total retail sales of consumer goods (RSCG) time sequence forempirical research, social total retail sales of consumer goods reflect social commoditypurchasing power and the size of the retail market situation. In the econometrics model,according to the characteristics of a disturbance with variance，this paper choose ARIMAmodel, and then use support vector machine (SVM) model to study RSCG, finally, thispaper combines ARIMA and support vector machines together, use the mixed model whichget the ARIMA model estimation value and support vector machine (SVM) model ofresidual modeling estimated value to get together and get hybrid model results. Compare thethree model, the hybrid model has best effect.In order to verify the accuracy of the hybrid model and application areas, this paperalso choose Shanghai stock composite index time series analysis, also take three kinds ofmodel were analyzed. The conclusion is that hybrid model has good fitting effect andestimated accuracy.