基于测量误差面板数据的两步分位回归估计方法

Two-step quantile regression estimation of panel data model with measurement error

该文考虑了含测量误差面板数据模型的参数估计问题.首先通过因子得分法消除自变量中测量误差的影响,获得固定效应面板模型的自变量修正值;随后分别采用一阶差分分位回归法和固定效应分位回归法求得估计量,结合Bootstrap算法得到估计置信区间.蒙特卡罗模拟结果显示,新提出的两步分位回归估计方法在同方差、异方差两种固定效应面板模型下的相对偏差和均方误差均优于传统分位数回归法.最后,基于美国各州实际香烟销售面板数据的实证分析表明,该方法在消除测量误差及提高估计准确性等统计应用分析上具有一定优势,可以为解决实际的复杂动态问题提供可靠依据.

In this paper,the problem of parameter estimation for panel data models containing measurement errors is considered systematically.The corrected values of the independent variables of the fixed-effects panel model are obtained by eliminating the effect of measurement error in the independent variables through the factor score strategy.Subsequently,the first-order difference quantile regression method and fixed-effect quantile regression method are employed to obtain the estimators,respectively.Meanwhile,the estimated confidence intervals are obtained by the Bootstrap algorithm.The Monte-Carlo simulation results indicate that the newly proposed two-step quantile regression estimation methods outperform the traditional quantile regression method in terms of relative deviation and mean square error under the two fixed-effects panel models of homoskedasticity and heteroskedasticity.Furthermore,the empirical analysis based on the panel data of actual cigarette sales in each state of the U.S.demonstrates that the proposed method has advantages in eliminating measurement errors and improving the accuracy of estimates,and could provide reliable basis for solving practical and complex dynamic problems.