摘要
测度误差普遍存在于现实数据中。本文提出了无需额外信息的线性测度误差模型的两步估计法,该方法可以得到一致和渐进正态的估计量。在基本估计量的基础上,本文对两步估计法进行了拓展,得到了更有效和更稳健的估计量,并且将这一估计方法推广到了时间序列数据和面板数据模型中。本文进一步对比两步估计法和工具变量法,发现前者在一定条件下严格优于后者。蒙特卡洛模拟验证了这些估计量在有限样本中的良好性质,并且说明了两步估计法相对于工具变量法的优势。
Measurement errors are pervasive in data sets. This paper proposes a two-step estimation approach for linear measurement errors models without extra information, and argues that the approach can generate consistent and asymptotic normal estimators. The study derives more efficient and more robust estimators besides the baseline one, and applies the estimation approach to time series and panel data models. Comparison between the two-step estimation and instrumental variable estimation is exhibited, and under some assumption the former strictly dominates the latter. A Monte Carlo simulation corroborates the performance of the estimators in the ease of finite sample, and demonstrates also the comparative advantage of two-step estimation over its instrumental variable counterpart.
出处
《统计研究》
CSSCI
北大核心
2015年第11期103-112,共10页
Statistical Research
关键词
测度误差
线性模型
两步估计法
工具变量法
Measurement Error
Linear Model
Two-Step Estimation
Instrumental Variable Estimation