针对工具变量线性回归模型的未知参数研究了模型平均框架, 该框架能同时适用于独立数据和相依数据. 在这一框架下, 文章推导了模型平均估计的渐近分布,构造了一个覆盖真实参数的概率趋于预定水平的置信区间, 并证明了该置信区间与基于全模型的置信区间渐近等价. 最后, 文章还进行了模型研究, 以考察模型平均估计在有限样本下的表现.

In this paper, we propose a model averaging scheme for the unknowns in IV linear regression models, which is useful for both independent data and dependent data. First, we derive the model averaging estimator’s asymptotic distribution. Then we develop a confidence interval procedure of the unknowns with an actual coverage probability that tends toward the nominal level in large samples. We further show that confidence intervals, based on the model averaging estimators, are asymptotically the same as those obtained under the full model. A simulation investigates the finite sample performance of the model averaging estimators.