Semi-parametric estimation of the second order parameter in statistics of extremes (Q1424690)

The authors present a class of semi-parametric estimators for the second order parameter related to a probability distribution with regularly varying tail. Let the distribution function \(F\) satisfy the regular variation condition \[ \lim_{t\to\infty} [1-F(tx)]/[1-F(t)]= x^{-1/\gamma}, \quad\text{for all }x>0, \] where \(\gamma\) is the tail parameter. A condition for asymptotic normality of the Hill estimator of the tail parameter is the second order condition \[ \lim _{t\to\infty} a^{-1}(t) [\bigl(1-F(tx)\bigr)\bigl(1-F(t) \bigr)^{-1}-x^{-1/\gamma}]= x^{-1/\gamma} (x^{\rho/ \gamma}-1) (\rho/\gamma)^{-1}, \quad x>0, \] where \(a\) is a suitably chosen function of constant sign near infinity (positive or negative), and \(\rho\leq 0\) is the second order parameter. The aim of this paper is to present a class of estimators for \(\rho\) and to prove consistency and asymptotic normality under appropriate conditions. The proposed class of estimators is parameterized by a positive tuning parameter \(\alpha\) which we can choose appropriately in order to improve the performance of the estimator. The authors present the simulated behavior of the proposed \(\rho\)-estimators for distributions in Hall's class, like the Fréchet, the Student and the Burr models. A comparison with the maximum likelihood \(\rho\)-estimator is performed.