Recursive Least Squares Estimator with Multiple Exponential Windows in Vector Autoregression

(整期优先)网络出版时间:2002-01-11
/ 1
AbstractIntheparametertrackingoftime-varyingsystems,theordinarymethodisweightedleastsquareswiththerectangularwindowortheexponentialwindow.Inthispaperweproposeanewkindofslidingwindowcalledthemultipleexponentialwindow,andthenuseittofittime-varyingGaussianvectorautoregressivemodels.Theasymptoticbiasandcovarianceoftheestimatoroftheparameterfortime-invariantmodelsarealsoderived.Simulationresultsshowthatthemultipleexponentialwindowshavebetterparametertrackingeffectthanrectangularwindowsandexponentialones.