Limited information estimator of a dynamic structural equation with autocorrelated arrors: a correction and new evidence

摘要

There are two parts to this paper. In this first part we report corrected MSE values for a set of Monte Carlo experiments evaluating the properties of a number of estimators for a structural equation with autocorrelated errors that were originally published in Moazzami and Buse (1986). We find that the corrected MSE values do not change the rankings of the estimators. The second part reports the results of a Monte Carlo evaluation of the finite sample performance of Theil's generalized two state least squares (G2SLS), an estimator which has largely been neglected in this literature. We find that G2SLS (and a variant of it) not lonly dominates all estimators in terms of bias but its MSE is almost as low as that of the asymptotically efficient Sargan 2SLS and unambiguously superior to the widely used Fair estimators. Because the G2SLS estimator uses the autoregressive transforms of the predetermined variables as instruments it provides a workable solutin to the degrees of freedom problem without the need to resort to ad hoc selection procedures for instruments. We conclude that the neglect of G2SLS is quite unjustified.