Extreme value analysis of speeding data

Abstract

Is extreme value theory a suitable approach for modeling the behavior of speeding data? In the following thesis I will reply to this question by introducing the basic concepts of extreme value theory. For this purpose, I am going to analyze a set of speeding values recorded by the Berlin police from 2009 to 2011. First, I will approach this question by using basic statistic indicators. Afterwards, I am going to enlarge the understanding of the underlying distribution by using a quantile plot approach. Finally, I will calculate the extreme value index by using Hill's estimator and moment estimators as well as the endpoint. I am going to show that the resulting endpoint depends on the chosen estimator for the extreme value index. For several speed limits, the realized maximum will be close to the estimated endpoint, whereas for other speed classes an even more excessive speeding behavior should be expected.