面板数据复合分位数回归模型的估计及应用

发布时间：2018-04-26 09:08

本文选题：复合分位数回归 + 分位数回归　；参考：《重庆工商大学》2017年硕士论文

【摘要】：面板数据的统计分析越来越引起人们的关注,其主要原因是该类数据融合了截面数据和时间序列数据的特点和优点,涵盖了研究对象丰富的信息。在进行模型的参数估计时,传统的最小二乘估计方法(LS)对于正态分布的数据表现良好,但在实际问题中,数据往往不满足正态分布的要求。当模型误差不服从正态分布或存在异常点时,经典的均值回归将会失效。本文引入了复合分位数回归方法(CQR),用以实现个体固定效应面板模型的参数估计,它综合了多个分位数处的分位数回归得出回归系数更有效的估计。该方法既保留了分位数回归的稳健性,又通过复合的方式提高了估计的效率。本文针对个体固定效应面板数据模型,提出了回归参数的复合分位数回归估计。首先引入一个特定的幂等矩阵,消除个体效应项,避免了参数祸根问题,将面板数据模型转化成线性模型;然后采用复合分位数回归方法构造回归系数的目标函数。在一些正则条件下,证明了所得估计的渐近正态性。为了讨论CQR估计量的有效性,结合LS估计和中位数回归QR_(0.5)估计,对误差项分别服从标准正态分布、T分布和柯西分布的面板数据模型进行了Monte Carlo模拟研究。模拟结果表明CQR的估计精度最优,特别是模型误差服从非正态分布的情形。最后将所提出的估计方法应用于探究我国行业工资差异的影响因素,根据我国各行业2005-2014年的年平均工资及相关因素的数据进行建模分析,进一步验证了CQR估计在实际案例中的有效性,研究结果显示人力资本水平、行业垄断程度和劳动生产率是影响行业间工资差异的主要因素。
[Abstract]:The statistical analysis of panel data has attracted more and more attention. The main reason is that this kind of data combines the characteristics and advantages of cross-section data and time series data, and covers the rich information of the research object. In the parameter estimation of the model, the traditional least square estimation method (LSs) has good performance on the normal distribution data, but in the practical problems, the data often do not meet the requirements of the normal distribution. When the model error is not satisfied from normal distribution or the existence of abnormal points, the classical mean regression will fail. In this paper, a compound quantile regression method is introduced to estimate the parameters of the individual fixed effect panel model. It synthesizes the quantile regression at multiple quantiles to obtain a more effective estimate of the regression coefficient. This method not only preserves the robustness of quantile regression, but also improves the efficiency of estimation by compound method. In this paper, the compound quantile regression estimation of regression parameters is proposed for individual fixed effect panel data model. Firstly, a specific idempotent matrix is introduced to eliminate the individual effect term, and the problem of parameter curse is avoided, and the panel data model is transformed into a linear model, and then the objective function of the regression coefficient is constructed by using the compound quantile regression method. Under some regular conditions, the asymptotic normality of the obtained estimates is proved. In order to discuss the validity of CQR estimator, the panel data model with standard normal distribution T distribution and Cauchy distribution is studied by using LS estimation and median regression QRK 0.5) estimation. The simulation results show that the estimation accuracy of CQR is optimal, especially in the case of model error from non-normal distribution. Finally, the proposed estimation method is applied to explore the factors affecting the wage gap in China's industries, and modeling and analysis are made according to the data of the average annual wages and related factors of each industry in China from 2005 to 2014. The results show that the level of human capital, the degree of industry monopoly and labor productivity are the main factors that affect the wage difference between industries.
【学位授予单位】：重庆工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F249.24

【参考文献】