EFFECTS OF NON-ORTHOGONAL REGRESSORS ON THE DISTRIBUTION OF SEEMINGLY UNRELATED REGRESSION ESTIMATOR

Abstract

The Zellner's estimator for two-equation seemingly unrelated regression model is considered. We express the Edgeworth approximation of the distribution function of the “unrestricted” estimator in terms of the canonical correlation coefficients between regressors in both equations. Based on the resulting approximate formula, we examine analytically the effects of non-orthogonal regressors on the pdf of the unrestricted Zellner estimator. For a special case, the effect is examined numerically by a Monte Carlo experiment.