Mixture of extreme-value distributions: identifiability and estimation |
| |
| Authors: | C E G Otiniano C C Y Dorea |
| |
| Institution: | 1. Department of Statistics, University of Brasilia, Brasilia, DF, Brazil;2. Department of Mathematics, University of Brasilia, Brasilia, DF, Brazil |
| |
| Abstract: | In this paper, the mixture model of k extreme value distributions is investigated. Using the Laplace transform of extreme value distributions given in terms of the Krätzel function, we first prove the identifiability of the class of arbitrary mixtures of extreme-value distributions of type 1 and type 2. We then find the estimates for the parameters of the mixture of two extreme-value distributions, including the three different types, via the EM algorithm. The performance of the estimates is tested by Monte Carlo simulation. |
| |
| Keywords: | Extreme-value distributions finite mixture identifiability EM - Algorithm krätzel function |
|
|
