Abstract
In this article I explore small sample properties of the three basic estimators used in the simultaneous equations model including the maximum likelihood estimator (LIML), the instrumental variable estimator (TSLS) and the ordinary least squares (OLS) estimator. Properties of estimators are studied by the conventional large sample asymptotic expansions, and by a new kind of the large sample expansion. Accuracy of expansions of the distributions of LIML, TSLS, and OLS estimators are examined by measuring deviations from empirical distributions made by experiments. An improvement of the maximum likelihood estimator is also referred to.
