Estimating the tail index: another algorithmic method

Abstract

The tail index is a determinant parameter within extreme value theory. Under a semiparametric approach, one has often to choose the number of the largest order statistics to include in estimates. This is a hard task since it is not possible to know for sure where the tail of data really begins. This crucial topic has been largely addressed in literature and several methods were developed. In this paper we analyze, through simulation, a heuristic method and compare it with two very popular methodologies. It will be seen that the new method can be a good alternative.