Testing extreme value models (Q1848515)

For the extreme value distribution (EVD) and the generalized Pareto distribution (GPD) with scale and location parameters asymptotically uniformly optimal tests for one-sided and two-sided hypotheses on the shape parameter are considered. Using the local asymptotic normality (LAN) property the author derives the asymptotic power of the tests under local alternatives. For the EVD the scale and location parameters are considered as nuisance ones. For the GPD the location parameter is assumed to be fixed (at zero) whence the scale is a nuisance parameter. The reason for this is the fact that the full GPD family is not LAN.