基于MCMC模拟和伪似然估计法的交叉分类信度模型费率厘定  

MCMC Simulation and GLMM for Experience Rate-making of Crossed Classification Credibility Model

作　　者：康萌萌[1] 孟生旺[2] 

机构地区：[1]山东财经大学保险学院,山东济南250014 [2]中国人民大学统计学院,北京100872

出　　处：《统计与信息论坛》2014年第2期34-39,共6页Journal of Statistics and Information

摘　　要：针对传统交叉分类信度模型计算复杂且在结构参数先验信息不足的情况下不能得到参数无偏后验估计的问题,利用MCMC模拟和GLMM方法,对交叉分类信度模型进行实证分析证明模型的有效性。结果表明:基于MCMC方法能够动态模拟参数的后验分布,并可提高模型估计的精度;基于GLMM能大大简化计算过程且操作方便,可利用图形和其它诊断工具选择模型,并对模型实用性做出评价。In this paper, we use MCMC simulation and GLMM to deal with traditional crossed classification credibility model, which is computationally complex and cannot get the unbiased posterior estimation of the parameters because of no sufficient prior information for the structural parameters. The paper indicates that MCMC method can dynamically simulate parameter posterior distribution and improve estimation precision. GLMM method can simplify the computational process greatly, and can be applied conveniently. By using graphs and other diagnostic tools, MCMC method makes the models selection and appraisal more easily.

关 键 词：交叉分类信度模型 经验费率 MCMC模拟GLMM 

分 类 号：F840[经济管理—保险]