the jackknife and the bootstrap for general stationary observations（重叠和引导一般固定观测）.pdfVIP

The jac kknife and the b o otstrap for general stationary observations Hans R Kunsc h ETH Zuric h Summary We extend the jac kknife and the b o otstrap metho d of estimating standard errors to the case where the observations form a general stationary sequence We do not attempt a reduction to iid values The jac kknife calculates the sample variance of replicates of the statistic obtained by omitting each blo ck of consecutive data once In the case of the arithmetic mean this is shown to b e equivalent to a weighted covariance estimate of the sp ectral density of the observations at zero Under appropriate conditions consistency is obtained if n and nn General statistics are approximated by an arithmetic mean In regular cases this approximation determines the asymptotic b ehavior Bo otstrap replicates are constructed by selecting blo cks of length randomly with replacement among the blo cks of observations The pro cedures are illustrated by using the sunsp ot numb ers and some simulated data AMS sub ject classication Primary G G Secondary M Key Words and Phrases Variance estimation jackknife b o otstrap statistics dened by functionals time series inuence function Intro duction The jackknife Tukey and the b o otstrap Efron have b ecome well established as nonparametric estimators of the variance of a statistic However the assumption of indep endence of the observations is crucial It is easily seen that they give incorrect answers if dep endence is neglected compare Remark of Singh Recently the two metho ds have b een extended to ARMAmo dels by reducing to innovations which are iid see Davis Freedman Efron and Tibshirani Section Still