贝叶斯隐马尔科夫异质面板模型及应用研究  

Bayesian Heterogeneous Hidden Markov Panel Model and its Application

作　　者：朱慧明[1] 黄仁存 游万海[1] 

机构地区：[1]湖南大学工商管理学院

出　　处：《统计研究》2014年第7期97-104,共8页Statistical Research

基　　金：国家自然科学基金委员会创新研究群体科学基金项目"金融创新与风险管理"(71271075);国家自然科学基金重点项目"战略导入的投资决策与风险管理"(71031004);国家自然科学基金面上项目"基于贝叶斯分位回归的面板数据建模;算法及应用研究"(71171075);湖南省研究生科研创新项目"空间面板分位数回归模型及其应用研究"(CX2013B127)项目资助

摘　　要：针对不可观测异质性非时变假设导致的删失变量偏差及推断无效问题,本文构建了贝叶斯隐马尔科夫异质面板模型,刻画了截面个体间的动态时变不可观测异质性,对经济系统环境中可能存在的隐性变点进行了诊断,并设计了相应的马尔科夫链蒙特卡洛抽样算法估计模型参数,对中国各地区的金融发展与城乡收入差距关系进行实证分析,捕捉到金融发展与城乡收入差距间长期稳定关系的隐性变化,发现了区域个体不可观测异质性存在的动态时变特征。研究结果表明,各参数的迭代轨迹收敛且估计误差非常小,验证了贝叶斯隐马尔科夫异质面板模型的有效性。For timeinvariant unobserved heterogeneity assumption lead to censored variable bias and invalid inference problem, this paper constructed Bayesian Heterogeneous Hidden Markov Panel Model to characterize the dynamics time varying unobserved heterogeneity between sectional individuals and diagnose the hidden changes existing in economic system environment, designed the corresponding Markov Chain Monte Carlo sampling algorithm to estimate model parameters, and analyzed the relationships between financial development and urbanrural income gap. The empirical research captured hidden changes in the long term stable relationships between financial development and urbanrural income gap, found out that regional individual unobserved heterogeneity existed dynamic timevarying characteristics. Results showed that the parameters iterative trajectory converges and the estimated error was very small, it verified the effectiveness of the Bayesian Heterogeneous Hidden Markov Panel Model.

关 键 词：面板数据 异质性 变点 隐马尔科夫模型 贝叶斯分析 

分 类 号：O212[理学—概率论与数理统计]