Estimating the quantile function of a location-scale family of distributions based on few selected order statistics (Q789845)

Let \(X_ 1,...,X_ n\) be i.i.d. random variables with d.f. F where \(F(x)=F_ 0((x-\mu)/\sigma)\) with \(F_ 0\) being a known d.f. and \(\mu\) and \(\sigma\) being unknown location and scale parameters. Given a positive integer k and numbers \(0<p_ 1<p_ 2<...<p_ k<1\) denote by \(X_{(n_ 1)}\leq...\leq X_{(n_ k)}\) the order statistics of \(X_ 1,...,X_ n\) with indices \(n_ i=[np_ i]+1\). By making use of the asymptotic joint normality of \(X_{(n_ 1)},...,X_{(n_ k)}\) one can establish an asymptotically best linear unbiased estimator (ABLUE) of the underlying \(\xi\)-quantile for every fixed vector \((p_ 1,...,p_ k).\) The main purpose of the present article is to find those \(p_ 1,...,p_ k\) which minimize the asymptotic variance over the class of ABLUE's in the particular case of exponential r.v.'s. Further results are obtained for Pareto and double exponential r.v.'s.