Finite Sample Properties of the Two-Step Empirical Likelihood Estimator
研究了两步经验似然估计量在有限样本下的性质，发现其偏差可能很大且对矩条件数量敏感，尾部也可能很重，对使用该方法的实证研究者有参考价值。
ABSTRACT We investigate the finite sample properties of two-step empirical likelihood (EL) estimators. These estimators are shown to have the same third-order bias properties as EL itself. The Monte Carlo study provides evidence that (i) higher order asymptotics fails to provide a good approximation in the sense that the bias of the two-step EL estimators can be substantial and sensitive to the number of moment restrictions and (ii) the two-step EL estimators may have heavy tails.