Tail asymptotic expansions for \(L\)-statistics (Q477271)

For \(X_1,\dots, X_n\) independent identically distributed positive random variables, denote by \(X_{1,n}\leq \dots \leq X_{n,n}\) the corresponding order statistics. Assuming that the right tail of \(X_1\) is eventually \(\lceil\alpha\rceil\) times continuously differentiable and that its derivative of order \(\lceil\alpha\rceil\) is regularly varying at infinity of index \(-(\alpha+\lceil\alpha\rceil)\), the authors provide a high-order asymptotic expansion for \(R(n,x):=\mathbb{P}\{c_1X_{n,n}+c_2X_{n-1,n}+\dots+c_nX_{1,n}>x\}\) as \(x\to\infty\), where \(c_1, c_2\) are positive and \(c_3,\dots, c_n\) are nonnegative constants. Also, a second-order asymptotic expansion of \(R(n,x)\) is pointed out under the assumption that the right-tail of \(X_1\) is second-order regularly varying of indices \(-\alpha\) and \(\rho\) (\(\alpha>0\), \(\rho\leq 0\)), i.e., \[ \lim_{t\to\infty}{\mathbb{P}\{X_1>tx\}/\mathbb{P}\{X_1>t\}-x^{-\alpha}\over A(t)}=x^{-\alpha}\int_1^x u^{\rho-1}\,\mathrm{d}u \] for an appropriate function \(A(t)\). Applications are given to ratios of two kinds of risk measures and evaluation of premium under stop-loss and `excess return on capital' rules. These are further illustrated with several examples and a Monte Carlo simulation study.