23º SINAPE - Simpósio Nacional de Probabilidade e Estatística

Página Inicial » Inscrições Científicas » Trabalhos

Dados do Trabalho


Título

OUTLIERS AND ROBUSTNESS IN SKEW-T MODELS

Resumo

Linear regression models with errors in the skew-t family, which includes normal, Student-t and skew normal distributions as particular cases, have been considered a robust alternative to the normal model. However, presence of outliers can affect estimates. The main goal of this work is to identify situations in which skew-t models are more vulnerable, robustness proprieties related to skew normal and skew-t models and to adapt a robust estimator for this class of models. We could fi nd that outliers in the right tail have stronger impact on the estimates of shape parameters. By a simulation study, it was possible to see that the effect on parameters depend on the proportion of contaminants and how far they are from the mass of the distribution. At last, we tested an adapted robust estimator that proved to have good behavior to outliers in datasets with moderate skewness.

Palavras-chave

Statistical Inference, Robustness, MLE, M-Estimators, Skew-t models

Área

Inferência Estatística

Autores

Simone Bega Harnik, Márcia D'Elia Branco, Marc G Genton