| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.89 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Neste trabalho analisam-se os principais modelos de regressão para dados de
contagem, na perspectiva habitual de especificação, estimação e avaliação. Após a
descrição detalhada dos modelos tradicionais, Poisson e Binomial Negativo, explora-se
pormenorizadamente o Poisson Generalizado, um modelo recentemente introduzido
nesta área. Os três modelos são depois adaptados a casos em que a amostra recolhida é
truncada em zero. De seguida, abordam-se alguns testes de especificação, começandose
por avaliar o modelo de Poisson face às duas alternativas referidas. Para tal,
deduzem-se testes score, os quais provam que, para determinadas parametrizações, os
modelos Binomial Negativo e Poisson Generalizado constituem alternativas localmente
equivalentes. De modo a testar a adequabilidade destes últimos modelos, consideramse
ainda diversos testes de hipóteses não encaixadas, para alguns dos quais é realizado
um estudo de simulação de Monte Carlo com o objectivo de examinar o seu
comportamento quando aplicados a modelos truncados. Os resultados obtidos
mostram que o desempenho dos testes é satisfatório, embora alguns deles sejam
claramente afectados pelo nível de sobredispersào e pelo grau de truncagem admitidos.
In this study, the specifícation, estimation and evaluation of some count data regression models are discussed. After describing the most common models, Poisson and Negative Binomial, the Generalized Poisson model, recently included in this area, is analysed in detail. The three models are then extended to cases where the sample is truncated at zero. Furthermore, some specifícation tests are considered. First, the Poisson model is tested against the other two models. Some score tests are derived, and it is possible to prove that, with some parametrizations, the Negative Binomial and the Generalized Poisson are locally equivalem altematives. In order to test the other count data models, some non-nested hypothesis tests are also discussed. Finally, a Monte Carlo simulation is done, with the objective of comparing the performance of non-nested tests, when they are applied to truncated models. The results obtained show that these tests perform satisfactorily, although some of them clearly depend on the magnitude ofthe overdispersion and truncation considered.
In this study, the specifícation, estimation and evaluation of some count data regression models are discussed. After describing the most common models, Poisson and Negative Binomial, the Generalized Poisson model, recently included in this area, is analysed in detail. The three models are then extended to cases where the sample is truncated at zero. Furthermore, some specifícation tests are considered. First, the Poisson model is tested against the other two models. Some score tests are derived, and it is possible to prove that, with some parametrizations, the Negative Binomial and the Generalized Poisson are locally equivalem altematives. In order to test the other count data models, some non-nested hypothesis tests are also discussed. Finally, a Monte Carlo simulation is done, with the objective of comparing the performance of non-nested tests, when they are applied to truncated models. The results obtained show that these tests perform satisfactorily, although some of them clearly depend on the magnitude ofthe overdispersion and truncation considered.
Descrição
Mestrado em Matemática Aplicada à Economia e à Gestão
Palavras-chave
dados de contagem Poisson Generalizado modelos truncados sobredispersào testes para hipóteses não encaixadas count data Generalized Poisson truncated models overdispersion nonnested hypothesis tests
Contexto Educativo
Citação
Ramalho, Joaquim José dos Santos (1996). “Modelos de regressão para dados de contagem”. Dissertação de Mestrado. Universidade Técnica de Lisboa. Instituto Superior de Economia e Gestão
Editora
Instituto Superior de Economia e Gestão
