考虑模型不确定性的中国死亡率预测——基于贝叶斯模型平均方法

发布时间：2018-06-01 12:18

本文选题：模型不确定性 + 长寿风险　；参考：《保险研究》2017年05期

【摘要】：作为长寿风险研究领域的基础,死亡率预测近些年获得快速发展,诸多模型的提出使得历史数据的信息得以最大程度的挖掘,但也带来了模型不确定问题。本文对现有的死亡率预测模型进行了分析和整理,提出其中的模型不确定性问题,并针对死亡率预测的模型不确定问题,引入了贝叶斯模型平均方法。该方法以贝叶斯后验概率为权重,综合考虑"一揽子"预测模型的预测能力,并根据它们预测吻合程度进行加权,最终给出死亡率预测结果,结论表明,不但在理论上表现出超过单一模型的优势,也在实践中超过了任何一个单一模型。本文还给出了基于该模型的死亡率预测结果和预期寿命。
[Abstract]:As the basis of longevity risk research, mortality prediction has developed rapidly in recent years. Many models have been proposed to maximize the mining of historical data, but also lead to model uncertainty. In this paper, the existing mortality prediction models are analyzed and sorted out, and the uncertainty of the models is put forward, and the Bayesian model averaging method is introduced to solve the model uncertainty of mortality prediction. This method takes Bayesian posteriori probability as the weight, synthetically considers the forecasting ability of the "package" forecasting model, and weights it according to their prediction coincidence degree. Finally, the result of mortality prediction is given. The conclusion shows that, It not only has the advantage of surpassing the single model in theory, but also surpasses any single model in practice. The prediction results and life expectancy of mortality based on the model are also given in this paper.
【作者单位】：中央财经大学中国精算研究院;
【基金】：中国保险学会教保人身保险高校课题研究基金(2017年度) 北京市哲学社会科学基金项目(编号:15JGC153) 教育部人文社科项目(编号:16YJCZH148);教育部人文社会科学重点研究基地重大项目(编号:16JJD790060) 数据灯塔(Data Lighthouse)计划
【分类号】：C924.2

【参考文献】