Klüppelberg, Claudia; Lindner, A.
(2005):
Extreme value theory for moving average processes with lighttailed innovations.
Collaborative Research Center 386, Discussion Paper 432

Preview 

PDF
400kB 
Abstract
We consider stationary infinite moving average processes of the form $Y_n = \sum c_i Z_{n+i}$, where the sum ranges over the integers, (Z_i) is a sequence of iid random variables with ``light tails'' and (c_i) is a sequence of positive and summable coefficients. By light tails we mean that Z_0 has a bounded density $f(t)$ behaving asymptotically like $v(t) \exp (g(t) )$, where v(t) behaves roughly like a constant as t goes to infinity, and g(t) is strictly convex satisfying certain asymptotic regularity conditions. We show that the iid sequence associated with Y_0 is in the maximum domain of attraction of the Gumbel distribution. Under additional regular variation conditions on g, it is shown that the stationary sequence (Y_n) has the same extremal behaviour as its associated iid sequence. This generalizes results of Rootz\'en (1986, 1987), where $g(t) = t^p$ and $v(t)=c t^d$ for p > 1, positive c and a real constant d.