Comparison of ACER and POT Methods for estimation of Extreme Values

Abstract

Comparison of the performance of the ACER and POT methods for prediction of extreme values from heavy tailed distributions. To be able to apply the ACER method to heavy tailed data the ACER method was first modified to assume that the underlying extreme value distribution would be a Fréchet distribution, not a Gumbel distribution as assumed earlier. These two methods have then been tested with a wide range of synthetic and real world data sets to compare their preformance in estimation of these extreme values. I have found the ACER method seem to consistently perform better in the terms of accuracy compared to the asymptotic POT method.