| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 828.58 KB | Adobe PDF |
Orientador(es)
Resumo(s)
O presente trabalho tem como objetivo analisar o comportamento de um conjunto de estimadores do índice de cauda, num contexto de dependência dos dados. O índice de cauda mede o decaimento da cauda das distribuições com caudas pesadas e permite determinar os momentos finitos de uma variável aleatória. Os estimadores analisados foram o estimador de Gabaix and Ibragimov (2011), Hill (1975) e Nicolau and Rodrigues (2015).
Esta análise foi efetuada através de Simulações de Monte Carlo (SMC), onde foram considerados processo geradores de dados com e sem dependência, e com erros provenientes de distribuições com caudas pesadas, i.e. as caudas da distribuição tem uma massa probabilística significativa.
Os resultados apontam para que o estimador de Hill, considerando o nível de truncagem ótimo (k), tenha o melhor desempenho sob os processos de dados considerados, contudo o k, em geral, é desconhecido e de difícil estimação. Por sua vez, o estimador de Nicolau and Rodrigues (2015) apresenta uma robustez face ao desconhecimento do k, quando comparado com os outros estimadores.
The main goal of this paper is to analyse the behaviour of tail index estimators in a context of data dependence. The tail index estimator measures the distribution tails decay, and allows to set the finite moments of a random variable. The estimators studied was Gabaix and Ibragimov (2011), Hill (1975) and Nicolau and Rodrigues (2015). To do this analysis, we will use a Monte Carlo Simulation (MCS), where the data generating process can have dependence, or not, and the errors follow a heavily tailed distribution, i.e. a significant mass of the probability distribution is located in the tails of the distribution. The results showed that, when truncation level (k) is known, the Hill estimator has the best results under the specific data generating processes chosen. However, k most of the times is unknown and hard to estimate. On the other hand, the Nicolau and Rodrigues (2015) estimator seems to be robust to absent estimations of k, when compared with others estimators.
The main goal of this paper is to analyse the behaviour of tail index estimators in a context of data dependence. The tail index estimator measures the distribution tails decay, and allows to set the finite moments of a random variable. The estimators studied was Gabaix and Ibragimov (2011), Hill (1975) and Nicolau and Rodrigues (2015). To do this analysis, we will use a Monte Carlo Simulation (MCS), where the data generating process can have dependence, or not, and the errors follow a heavily tailed distribution, i.e. a significant mass of the probability distribution is located in the tails of the distribution. The results showed that, when truncation level (k) is known, the Hill estimator has the best results under the specific data generating processes chosen. However, k most of the times is unknown and hard to estimate. On the other hand, the Nicolau and Rodrigues (2015) estimator seems to be robust to absent estimations of k, when compared with others estimators.
Descrição
Mestrado em Econometria Aplicada e Previsão
Palavras-chave
índice de cauda dependência estimador GI11 estimador NR15 estimador Hill tail index dependence GI11 estimator NR15 estimator Hill estimator
Contexto Educativo
Citação
Nascimento, Bruno Miguel da Silva (2016). "Estimação do índice de cauda num contexto de dependência". Dissertação de Mestrado, Universidade de Lisboa. Instituto Superior de Economia e Gestão.
Editora
Instituto Superior de Economia e Gestão
