基于贝叶斯MCMC方法的我国人口死亡率预测

Mortality Projection of Chinese Population Using Bayesian MCMC Approach

鉴于我国人口死亡率统计数据质量不高的实际和传统Lee-Carter死亡率预测模型两阶段方法存在的误差累积问题,本文采用贝叶斯Markov Chain Monte Carlo方法来预测我国人口死亡率。通过Win BUGS编程,文章在一体化框架下一次性给出模型的参数估计和未来死亡率的预测值。对研究结果的比较分析表明,贝叶斯方法不仅有效减少了数据质量问题的不利影响,提高了参数估计的稳健性,而且有效克服了参数估计和预测分开进行的弊端,在BIC值和残差项方差等模型选择标准上明显优于传统方法。

In view of the low quality mortality statistics of China＇ s population and error accumulation of the tradi- tional Lee-Carter model methodology that uses the separated two-period parameter projection, this article utilized the Bayesian Markov Chain Monte Carlo approach to project Chinese mortality rates. With the help of WinBUGS pro- gramming, we got the parameter estimations and their predicted value under an integrated framework. The study showed the Bayesian method not only effectively reduced the adverse influence of low quality mortality statistics on the robustness of parameter estimation, but also overcomed the problems of two-period parameter projection as well In terms of model selection criteria such as BIC and error variances, the Bayesian MCMC method had an obvious ad-vantage over the traditional ones.