Efficient estimation of the reciprocal of the density quantile function at a point (Q1079307)

Let \(F_ n(x)\) be the edf of a sample from df F(x) having derivative \(f(x)>0\) at \(F^{-1}(p)\). Then \[ D(x,n,p)=n^{1/(2k- 1)}\int^{x}_{0}[F_ n^{-1}(p+vn^{2k-1})-F\quad_ n^{- 1}(p)]h(v)dv \] is a consistent estimator for \(1/f(F^{-1}(p))\). h(y) serves as a substitute for \(1/(v\cdot n^{2k-1})\) and it is assumed that \(h(y)e^ y\) is a polynomial of degree \(\leq k\), \(k\geq 2\). D(x,n,p) converges in mean square sense with rate \(n^{(2k-2)/(2k-1)}\). If \(k\leq 6\), then D(log n,n,p) is efficient among the class \(\{D(x,n,p)| x>0\}\).