O uso da distribuição beta T generalizada nos erros do modelo Garch: uma aplicação no preço do café

Abstract

This work study some flexible distribution of Generalized Beta class, actually the Generalized Beta Generated, study some of its properties and show how can they be used as assumption to the residuals of regression models in the way to supplant the no normality in some situations. The likelihood method is used to find the parameters estimative by R software. The Nelder-Mead algorithm was chosen to objective function optimization. Some distributions has been used as base to ARMA-GARCH regression models applied to a coffee price series. They were Student t Distribution, Beta t Distribtution, Generalized Beta t ditribution, and the Normal Distribution. The hessian matrix for estimated models were numerically computed, evaluated on the estimated values and again by support of R software. The results shows a improvement when compared with the classical approach. The likelihood test was done to compare nested models.