Finite-Sample Simulation-Based Tests in Seemingly Unrelated Regressions

In this paper, we propose finite and large sample likelihood based test procedures for possibly non-linear hypotheses on the coefficients of SURE systems. Two complementary approaches are described. First, we propose an exact Monte Carlo bounds test based on the standard likelihood ratio criterion. Second, we consider alternative Monte Carlo tests which can be run whenever the bounds are not conclusive. These include, in particular, quasi-likelihood ratio criteria based on non-maximum-likelihood estimators. Illustrative Monte Carlo experiments show that: (i) the bounds are sufficiently tight to yield conclusive results in a large proportion of cases, and (ii) the randomized procedures correct all the usual size distortions in such contexts. The procedures proposed are finally applied to test restrictions on a factor demand model.