Extreme value theory in mixture distributions and a statistical method to control the possible bias

In this paper, extreme behaviors of a mixture distribution are analyzed. We investigate some cases where the mixture distributions are in the proper domain of attraction so that the extreme value of mixture distributions converges to the proper Generalized Extreme Value distribution (GEV). However, in general, there is no guarantee that the distribution of the data is in the proper maximum domain of attraction. Furthermore, since the convergence rate can be slow even with guaranteed asymptotic convergence, GEV estimation method might provide a biased estimation, as shown in Choi et al. (2014). The paper provides a safe method to control the quality of the quantile estimator in extreme values.