A Comparative Monte Carlo Study of the Properties of Econometric Estimators

Abstract

Together with straightforward simulation, this study makes use of antithetic-variate and control-variate techniques in investigating the properties of econometric estimators. It is shown that when the covariances between the disturbances are small, the results on the relative standing of the estimators are indeterminate; but when they are sufficiently large, full-information methods are superior to single-equation methods. FIML gave the best performance and 2SLS was the best limited-information technique. The asymptotic standard errors underestimated the variations in the coefficients.