摘要
文章针对面板数据在贝叶斯分析的框架下讨论了非参数分位回归建模方法。利用低秩薄板惩罚样条的展开,通过引入虚拟变量和非对称Laplace分布,建立贝叶斯分层分位回归模型,给出了未知参数估计的Metropolis-Hastings抽样算法。模拟结果显示,新方法在稳定性和无偏性方面都更优于4种传统方法。最后以消费支出面板数据为例,演示了新方法在实际建模中的应用,获得了一些有益的新结论。
The nonparametric quantile regression modeling method for panel data is discussed under the framework of Bayesian analysis in this paper.By using the expansion of punishment splines in low-rank thin plates,a Bayesian hierarchical quantile regression model is established by taking into account virtual variables and asymmetrical Laplace distribution,with a metropolisHastings sampling algorithm for unknown parameter estimation presented.The simulation results show that the new method is superior to the four traditional methods in terms of stability and unbiasedness.Finally,the paper takes the consumer expenditure panel data as an example to demonstrate the application of the new method in practical modeling,and obtains some useful new conclusions.
作者
张敏
罗幼喜
Zhang Min;Luo Youxi(School of Science,Hubei University of Technology,Wuhan 430068,China)
出处
《统计与决策》
CSSCI
北大核心
2020年第19期9-14,共6页
Statistics & Decision
基金
国家社会科学基金资助项目(17BJY210)。
关键词
惩罚样条
非参数分位回归
MCMC算法
蒙特卡罗模拟
penalty spline
nonparametric quantile regression
MCMC algorithm
Monte Carlo simulation
