Logo do repositório
 
A carregar...
Miniatura
Publicação

Inferência estatística em modelos max-semiestáveis

Utilize este identificador para referenciar este registo.
Nome:Descrição:Tamanho:Formato: 
ulsd054581_td_Sandra_Dias.pdf7.79 MBAdobe PDF Ver/Abrir

Resumo(s)

As leis max-semiestáveis surgem como limite de sucessões de funções de distribuição da forma Fkn(anx+bn), onde {kn} é uma sucessão de crescimento geométrico com razão r ≥ 1, e podem ser caracterizadas pelo parâmetro r, por um índice de valores extremos e por uma função y definida em [0, 1]. Nesta dissertação analisam-se as relações existentes entre as sucessões subjacentes, no caso da existência de limites não degenerados G e G’ tanto para Fkn(anx+b) como para Fkn (anx+b n), sendo estabelecidas condições sobre as sucessões normalizadoras e sobre as sucessões de crescimento geométrico, as quais são suficientes para que ambos os limites existam. Com o objectivo de estimar o parâmetro r, identificador do modelo, é estudada a consistência e a normalidade assintótica de uma sucessão de estatísticas {Zs} (quociente de certas diferenças entre estatísticas ordinais). Alguns métodos de estimação dos parâmetros do modelo que usam Zs (ou funções apropriadas de Zs) são apresentados conjuntamente com os estudos de simulação correspondentes. O problema da estimação de quantis elevados é também abordado. Propõe-se uma estatística de teste que permite distinguir suficientemente bem as leis pertencentes ao domínio de atracção de leis max-estáveis das que pertencem ao domínio de atracção de uma max-semiestável, propriedade que não é verificada por outros testes já existentes para a condição de valores extremos. Deduz-se a lei limite da estatística de teste proposta e são apresentados estudos de simulação. A metodologia exposta é aplicada a duas amostras de dados reais. A primeira é constituída pelos tempos de recorrência entre sismos de magnitude elevada que ocorreram em todo o mundo e a outra pelos momentos sísmicos de sismos que ocorreram na Região do Pacífico.
The max-semistable laws arise as the limit of sequences of distribution functions in the form Fkn(anx + bn), where {kn} is a geometrically growing sequence with ratio r ≥ 1. The relations between the underlying sequences, in case of existence of non degenerate limits G and G* for both Fkn(anx + bn) and Fkn(a nx + bn) is analysed. Conditions are established between the norming and centering constants and between the geometrically growing sequences that are sufficient for both convergences to occur. Max-semistable models can be characterized by a parameter r, by the extreme value index and by a real function y defined in [0, 1]. The parameter r identifies each family of models and must be estimated before the estimation of the extreme value index and the function y. The consistence and asymptotic normality of the sequences of statistics {Zs} (that are ratios of certain differences of order statistics) is studied. A new sequence of statistics used in the estimation of r is proposed and its consistence and asymptotic behaviour is analysed. Some methods of estimation using Zs (or appropriate functions of Zs), are also presented along with the corresponding simulation studies. The problem of high quantiles estimation in max-semistable laws is treated. Because studies on the behavior of tests for the extreme value condition have showed nonrobustness for models in the domain of attraction of a max-semistable law, a statistical test is proposed that distinguishes well the laws belonging to the max-stable domain of attraction from the ones in the max-semistable domain. The limit law is deduced and the results from a simulation study are presented. The methodology presented is applied to two real data sets, one consisting in major worldwide earthquakes recurrence times and the other consisting in seismic moments of earthquake in the Pacific Region.

Descrição

Tese de doutoramento em Estatística e Investigação Operacional (Probabilidade e Estatística), apresentada à Universidade de Lisboa através da Faculdade de Ciências, 2007

Palavras-chave

Estatística e investigação operacional Teses de doutoramento - 2007

Contexto Educativo

Citação

Projetos de investigação

Unidades organizacionais

Fascículo

Editora

Licença CC