A method for choosing an optimum threshold if the underlying distribution is generalised Burr-Gamma : theory and methods
Abstract
In this paper the Generalised Burr-Gamma (GBG) distribution is considered to model data that includes extreme values. Very often in extreme value theory the tail of the distribution is the only interest and therefore the selection of an optimum threshold plays an important role. This paper shows that the tail of the Generalised Burr-Gamma (GBG) above an optimum threshold can be approximated by a type of Generalised Pareto distribution (GP) which is dependent on the threshold value. Our GP-type of distribution discussed here differs from the usual GP distribution because it depends on the chosen threshold value and it only has one parameter known as the extreme value index (EVI) and not two as is usually the case. This paper makes a valuable contribution in choosing the best threshold if the underlying distribution is GBG. We assume throughout the paper that the GBG parameters μ σ κ and Ɛ are known. A method for choosing the optimum threshold is by considering the posterior risk function as will be explained. The selection of a threshold is illustrated through a simulation study and in the case of a real data set.
