Estimadores Corrigidos para Modelos SUR Não-Lineares

Abstract

In this paper we derive general formulae for second-order biases of maximum likelihood estimators of the parameters in nonlinear seemingly unrelated regression models (SUR), which have contemporaneous correlation between the errors in different equations af a set of regression equations. These biases can be easily obtained as vectors of least squares estimators of suitable weighted linear regressions. They are simple enough to be used algebraically to obtain closed-form bias corrections in special cases where the inverse of the information matrix has a closed-form. The practical use of the bias corrections is illustrated by simulation, suggesting that the bias-corrected estimators are closer to the true parameters than the classical maximum likelihood estimators.