Modeling Of Group-specific Mortality In China Based On A Multi-population Mortality Model

Mortality prediction is an important data base for measuring longevity risk,scientifically and accurately predict the mortality is of great significance to the economic development of our country.Due to the factors such as transport links,political and trade exchanges,technology development and medical advances,the mortality of the populations that are closely related should have a trend of convergence in the long run.However,the mortality predicted by the Lee-Carter model applied to the single population respectively showed a divergent trend,resulting in the contrary to the reality.The multiple population stochastic mortality model can effectively solve the divergence problem and make the resulting forecasts more reasonable.Based on the latest population mortality data from 1995 to 2018（refer to “China census data”,“Statistical Yearbook of Chinese Population and Employment statistics”）,the mortality rates of the national men,national women,urban men,urban women,men in town,women in town,rural men and rural women were grouped into a multi-population group.We fit and forecast to the age-specific mortality of different population by applying Joint-K model,ACF model,extension model I of ACF model,extension model II of ACF model and CAE model,then we compare the results forecasted by these multiple population mortality models with the results of traditional Lee-Carter model.We found that the mortality prediction of ACF model and its extended models show a tendency of convergence,in contrast the mortality prediction of Lee-Carter model has crossed and diverged.In addition,the RSSE index also indicates that the ACF model has the best fitting effect among all models,besides ACF model passes all the tests.Therefore,the mortality prediction results obtained by ACF model are selected to calculate the life expectancy,the actuarial present value of a continuous annuity for life（APV）,two types of Capital requirements for longevity risk（LCR）and the longevity-risk-adjusted（L-Ra）age of different populations from 2019 to 2040.The longevity-risk-adjusted age further shows the multiple population stochastic mortality models are more reasonable: it is estimated that by 2040,the maximum difference of L-Ra age calculated by the LC model for different populations will reach 16.69 years,while the maximum difference calculated by the ACF model is only 6.25 years old.In addition,we show that the life expectancy and longevity risk of different gender-specific populations of country,cities,towns and counties have an upward trend over time,in addition,there are gender differences and regional differences among the populations.According to the results of L-RA age based on ACF model,the pension age of urban males,urban females,town’s males,town’s females,country males and country females can be adjusted to 62,52,60,51,59,48 in turn by 2025.