Seemingly unrelated regressions under additive heteroscedasticity. Theory and share equation applications (Q1260675)

The paper provides consistent, asymptotically efficient, and asymptotically normal estimators for Seemingly Unrelated Regression systems that have additive heteroscedastic contemporaneous correlation. Both the estimator for the location vector and the parameters of the covariance matrix possess these properties. The procedure is superior to other methods because generalized least squares is used to estimate the parameters of the covariance matrix. The method also permits the use of cross-equation parameter restrictions. The paper concludes by discussing how this type of heteroscedasticity arises naturally in share equation systems and random coefficient models, and how these models can be uniquely estimated with the presented two-step estimation technique.