基于CIR特性的高龄死亡率预测方法

Prediction method for elderly mortality based on the characteristics of CIR model

死亡率预测可以为人寿保险和养老基金提供重要参考,因此对死亡率预测的精确度要求尤为迫切.本文基于CIR模型自身的特性,将其应用于随机死亡率的预测建模,提出了一种基于已有模型的修正方案,旨在提高预测精度.首先选取台湾地区1970—2014年和香港地区1986—2017年的死亡率数据,以经典的CBD模型为例,对新的修正模型预测性能进行检验;在获得良好的修正效果后,将该方法应用于中国大陆1995—2017年全人口死亡率数据,使用样本外数据方法进行回溯测试,检验模型预测能力,提高预测质量;最后,对中国人口死亡率进行预测,结果表明修正后的预测值更为准确.

Mortality prediction provides an important reference for life insurance and pension funds,so the accuracy of mortality prediction is particularly urgent.Based on the characteristics of CIR model itself,this paper applies it to the prediction modeling of random mortality,and proposes a modification scheme based on the existing model to improve its prediction accuracy.Firstly,the mortality data from 1970 to 2014 in Taiwan and Hongkong from 1986 to 2017 are selected,with classical CBD model as an example,the forecasting performance of new correction model is tested.After obtaining good correction effect,this method is applied to whole mortality population data in China from 1995 to 2017,the out-of-sample data method is used for backtracking test to verify prediction ability of the model and improve the quality of prediction.Finally,the paper predicts the mortality rate of Chinese population,and the results show that the revised prediction is more accurately.