A Diagnostic Plot for Estimating the Tail Index of a Distribution

Abstract

The problem of estimating the tail index in heavy-tailed distributions is very important in many applications. We propose a new graphical method that deals with this problem by selecting an appropriate number of upper order statistics. We also investigate the method's theoretical properties are investigated. Several real datasets are analyzed using this new procedure and a simulation study is carried out to examine its performance in small, moderate and large samples. The results suggest that the new procedure overcomes many of the shortcomings present in some of the most common techniques—for example, the Hill and Zipf plots—used in the estimation of the tail index, and it performs very competitively when compared with other adaptive threshold procedures based on the asymptotic mean squared error of the Hill estimator.