Jackknife estimator for an \(m\)-dependent stationary process (Q1210228)

Let \(\{\underset\widetilde{} X_ i,\;i\geq 1\}\) be a \(p\)-variate \(m\)- dependent stationary process and suppose \(\theta=T(F)\) is a parameter of interest where \(F\) is the unknown distribution function of \(X_ 1\) and \(T\) is a functional defined on \(\mathcal J\), a convex set of distribution functions containing \(F\) and all the degenerate distributions. An estimator of \(\underset\widetilde{}\theta\) is \(\underset\widetilde{}{\widehat\theta}=T(F_ n)\), where \(F_ n\) is the empirical distribution function based on \(\{\underset\widetilde{} X_ i,\;1\leq i\leq n\}\). It can be shown that \(n^{1/2}\) \((\underset\widetilde{}{\widehat\theta}_ n- \underset\widetilde{}\theta{)} N(0,\sigma^ 2)\) in distribution where \(\sigma^ 2\) is an unknown positive constant. It is of interest to estimate \(\sigma^ 2\) for statistical inference. The ordinary jackknife estimator is \[ \widehat\sigma^ 2_ 0=(n-1)\sum^ n_{i=1}(\widehat\theta_{-i}-\overline\theta)^ 2\quad\text{where}\quad \widehat\theta_{-i}=T(F^{(i)}_ n),\;\overline\theta=n^{-1}\sum^ n_{i=1}\widehat\theta_{-i}, \] where \(F^{(i)}_ n\) is the empirical distribution function of \(\underset\widetilde{} X_ j\), \(1\leq j\leq n\), \(j\neq i\). It is known that \(\widehat\sigma^ 2_ 0\) is a consistent estimator of \(\sigma^ 2\) when the observations \(\underset\widetilde{} x_ i\), \(i\geq 1\), are independent and identically distributed. The author proves that this estimator is inconsistent when \(\{\underset\widetilde{} X_ i,\;i\geq 1\}\) is an \(m\)-dependent stationary process with \(m\geq 1\). An alternative estimator is proposed which is strongly consistent. An example is presented.