The Problem Of Tuning Parameter Selection In Change Point Estimation And Its Application In Practice

The problem of change-point,which has been widely used in the fields of economy,finance,meteorology communication and so on,is one of the hot topics in the research of statistics in recent years.This thesis mainly researches the tuning parameter selection of CUSUM change-point estimator and the parameter change-point estimation of Weibull distribution,which are applied to the actual data analysis later.The choice of a more appropriate tuning parameter of CUSUM change-point estimation can make the result better.In the research of the influence of tuning parameter's value on change-point estimation,it turns out that with different real locations of change-points,tuning parameters have different effects on CUSUM change-point estimations.On this basis,a data driven algorithm for parameter selection is proposed.Through the simulation analysis,it is proved that the parameter selected by this algorithm can make the estimation of CUSUM change-point a steady performance.Then,the method is applied to the CUSUM change-point estimation of the Dow Jones weekly closing index from July 1971 to August 1974,and the result is accordant with the fact.In the second part of this thesis,based on Seguro(2000)three parameters Weibull distribution change-point maximum likelihood estimator,the non weather change-points are estimated for the detrended annual maximum wind speed data of the A weather station.The results show that the change-point of A weather station's maximum wind speed data is in 2003,which coincides with the actual situation of the A weather station.