A full heteroscedastic one-way error components model allowing for unbalanced panel: Pseudo-maximum likelihood estimation and specification testing

This paper proposes an extension of the standard one-way error components model allowing for heteroscedasticity in both the individual-specific and the general error terms, as well as for unbalanced panel. Onthe grounds of its robustness to distributional misspecification, its robustness to possible misspecification of the assumed scedastic structure of the data, its computational convenience and its potential efficiency, we argue for estimating this model by Gaussian pseudo-maximum likelihood of order two. Further, we review how, taking advantage of the powerful m-testing framework,the correct specification of the prominent aspects of the model may be tested. So are surveyed potentially useful nested, non-nested, Hausman and information matrix type diagnostic tests of both the mean and the variance specification of the model. Finally, we illustrate the practical relevance of our proposed model and estimation and diagnostic testing procedures through an empirical example in the production analysis field.