If we need to estimate the extreme value index, we must estimate the threshold first, that is giving the optimal fraction for the order statistics of the sample which be observed. The extreme value operators are bewildered by the question how to get the threshold all along.In this paper, a new method for selecting the threshold from the other point of view is presented. As is well known, the data be analyzed should involve more information about the tail index. So , we researched the relation between the threshold and the Shannon entropy of order statistics based on the theory of discrimination information and Shannon entropy. The method how to select suitable threshold and fracture sample in generalized pareto models is put forward. Then, the generalized pareto distribution is simulated and ideal results are attained.
|