Approximation by penultimate extreme value distributions

Suppose that \(X_1\), \(X_2\),\dots are i.i.d. random variables with d.f. \(F\) and the sequence of sample maxima \(M_n=\max(X_1,\dots,X_n)\) (suitably normalized) converges to an extreme value distribution \( G_\gamma(x)=\exp(-(1+\gamma x)^{-1/\gamma})\) (\(\gamma\in R\), \(1+\gamma x>0\)). The authors investigate the situation when the approximation of the sample distribution (\(F^n\)) can be improved using a sequence \(G_{\gamma_n}\), i.e. when \[ \lim_{n\to\infty} { F^n(a_nx+b_n)-G_{\gamma_n}(x) \over f(\log n) }= G_\gamma'(x) M_\gamma(-\log(-\log G_\gamma(x))) \] for some \(f\), \(a_n\), \(b_n\) and \(\gamma_n\), where \( M_\gamma(x)= \frac 12\int_0^x u^2 e^{\gamma u}du \) if \(\gamma>0\) and \( M_\gamma(x)=\frac 12\int_x^0 u^2 e^{\gamma u}du \) if \(\gamma<0\).

Recommendations

Cited in

(32)

This page was built for publication: Approximation by penultimate extreme value distributions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1979092)