Modelling And Forecasting Of Mortality In Our Elderly Population

Modelling and forecasting of mortality among the elderly population is the basis for pension management and longevity risk measurement.Given the small size,low quality and high volatility of mortality data for the elderly population at retirement age and above in Chinese mainland,there is a realistic research value in selecting a mortality projection model that fits the characteristics of mortality data for the elderly population in China and in constructing a dynamic mortality projection model that can capture the changes in mortality rates of the elderly population with age,year and gender.Based on a summary of domestic and international studies on dynamic mortality models for the elderly population,a logistic multi-population parameter stochastic mortality model is introduced and combined with a single-population parameter stochastic mortality model to study the modelling and prediction of mortality for the elderly population in China.Firstly,the seven dynamic mortality models are compared and selected using high-quality mortality data of the elderly population in Taiwan and Hong Kong.Secondly,the model structure is secondly validated and selected based on the mortality data of the elderly population in Chinese mainland,and the best model is applied to model and predict the mortality of the elderly population in Chinese mainland,and the fit and robustness of the model parameters are tested.Then,a comparative analysis of the fitting effect of the selected model and the short-term and long-term forecasting effects is carried out.Finally,the mortality model with the most outstanding performance in all aspects is selected to forecast the mortality rate of the elderly population in China over the next 30 years.The experimental results show that:(1)Model M1 in the single-population parameter stochastic mortality model and Model M5 in the multi-population parameter stochastic mortality model have the most prominent structure and the best fitting performance in both Taiwan,Hong Kong and Chinese mainland.(2)In terms of the fitting effect of the historical data,model M1 and model M5 fit the mortality data between the ages of 60-80 best for both males and females in Chinese mainland,but for mortality data between the ages of 80-89,model M5 has a more significant fitting effect,possessing a smaller MAPE value.Overall,model M5 has significantly better fitting accuracy than model M1 for both males and females.(3)From the prediction results,the mortality rates of men and women in Chinese mainland are consistently predicted under model M5,and there is no significant crossover or deviation in the long-term prediction of male and female mortality rates,which is a good way to overcome the phenomenon of divergence in the prediction of male and female mortality rates modelled separately by model M1,and is more in line with the general regularity of human mortality changes.In addition,the gender-consistent prediction results show that model M5 is superior to model M1.Therefore,model M5 in the multi-population parameter stochastic mortality model is recommended for longevity risk analysis and actuarial assessment of pensions.