Abstract: Longevity risk calculation is central to both public pension plans and life insurance companies. The essential work in quantifying longevity risk is to model the dynamics of mortality rates. Based on the goodness-of-fit with respect to the available mortality rates at various ages, the dynamic mortality rate models mainly deal with two issues, namely, age extrapolation and trend prediction. In addressing the gaps in the literature, this paper integrates the extreme modeling method for oldest-old mortality rate and the hierarchical modeling technique, and proposes a dynamic mortality rate modeling approach. The paper further demonstrates the dynamic evolutions of old-age and oldest-old mortality rates by place of residence and gender from 2000 to 2010 in China, with better measurement of the tail risks of the life span distribution, and thus improves the quantitative method on China's longevity risk research.

Keywords: Longevity Risk, Oldest-old, Extreme Modeling Method, Non-linear Hierarchical Models, Tail Risks

摘要： 长寿风险作为公共养老金、寿险公司关注热点，其量化的基础工作是构建动态死亡率模型。动态模型在保证可获得的各年龄死亡率拟合效果的基础上，涉及年龄外推和趋势预测两个问题。针对这两个问题，目前研究仍存在很多不足，主要包括：第一，对超高龄死亡率的分析尚不够充分，或者说缺乏对生存分布的尾部风险特征的合理量化。第二，缺乏对死亡率的性别、人群、区域和国别差异的系统研究。文章将超高龄人口死亡率的极值建模方法和分层建模技术纳入到动态死亡率建模中，弥补已有研究存在的不足，诠释近10年来我国城镇乡男性和女性高龄乃至超高龄人口死亡率的动态演变，更好地度量寿命分布的尾部风险特征，完善中国长寿风险的量化研究。

关键词: 长寿风险, 超高龄, 极值建模方法, 非线性分层模型, 尾部风险