摘要
目的比较多种右偏、厚尾的多参数分布模型与非参数核密度估计在社会健康保险精算中的应用价值。方法应用Gamma分布、Weibull分布、Pareto分布、对数正态分布、Log-logistic分布、Pearson-V分布、逆高斯分布以及非参数核密度估计方法分别对2004年某市某社保人群全年住院医疗赔付额的年龄别、性别数据进行分布拟合比较。结果非参数核密度估计方法的拟合效果远优于各参数分布拟合方法。结论非参数核密度估计对数据依赖程度低,表现稳健,拟合效果理想,在社会保险精算中应用值得推广。
Objective To compare the performance between multi-parameters distribution fitting models with long thick right-side tails and non-parameter kernel density estimate in the actuary of social health insurance. Methods Gamma distribution,Weibull distribution,Pareto distribution,log-normal distribution,log-logisitc distribution,Pearson-V distribution,Inv-Gaussian distribution and Kernel Density Estimation( KDE) were used to fit the distributions of claim data of a Chengdu hospitalization medical insurance plan respectively,and chi-square test was used to test the goodness-of-fit. Results The results of goodness-of-fit tests demonstrated that KDE performed much better than the parameter distribution models. Conclusion Research of actual data proved that KDE was a robust good-performance method,and the application of KDE in social health insurance actuary was recommendable.
出处
《公共卫生与预防医学》
2015年第1期34-38,共5页
Journal of Public Health and Preventive Medicine
关键词
社会健康保险精算
分布拟合
参数分布
核密度估计
Social Health Insurance actuary
Distribution fitting
Parametric distribution
Kernel density estimate