Research on the Time Series of Short Memory and Forecast
Xu Bin · Jisuanji fangzhen · 2007
AR series,as one of time series models,is applied broadly in system identification because its parameter estimation and rank decision are simple.On the basis of multi-dimension AR series modeled by least sequence criterion and the Kalman filtering technique,a method for estimating parameters of multi-dimensional AR series by Kalman filtering is developed in this paper.Because it is not necessary to keep historical data for this method,the estimated parameters of AR series can be updated real-time.The two step F-tested method is proposed in the decision of rank of AR series.The series of some A stock of Shanghai stock market are chosen as swatch.By analyzing the relativity of time series,this method is validated.By comparing the index of RMSE and MAD,the result shows that this method can decrease much modeling calculation work and has good forecasting capability.