Application Of Multi-Population Mortality Model Based On Neural Network

In recent years,the improvement of the mortality level has extended the life expectancy of the population,which is the direct cause of longevity risk and brings great challenges to the social endowment resources and pension system.Accurate prediction of future mortality is an important data base for measuring longevity risk and plays an important role in the development of national economy.With the deepening of the research on mortality model,the classical LC model has been extended to the mortality prediction of multi-population.The multi-population mortality model has good robustness,considering the common trend of mortality changes of different populations,and can effectively solve the divergence problem of mortality prediction of different populations separately,so as to obtain more reasonable results.In this paper,based on the data of male mortality in the Mainland,Taiwan and Hong Kong from 1994 to 2019,we applied the Joint-K model,ACF(0)model and its two extension models-ACF(1)and ACF(2)to fit and forecast the agespecific mortality of each population.It was found that the ACF(0)model had good robustness and fitting ability.On the other hand,thanks to the improvement of computing power,deep learning can accurately learn the data rules in samples,opening up a new direction for mortality prediction research.Therefore,this paper also combined the LSTM neural network model and the traditional ARIMA model which are both good at predicting time series to model and predict the common time effect factor of the multi-population model.The data from 1994 to 2015 was used as the training set,and the data from 2016 to 2019 was predicted.The absolute mean error and root mean squared error were calculated according to the predicted value and sample data as the evaluation criteria of the model.The results show that,to some extent,combining LSTM neural network model and ARIMA model can improve the prediction accuracy of common time effect factor,and then improve the prediction accuracy of mortality.By predicting the future age-specific mortality rate of males in the three regions,it was found that the age-specific mortality rate of males in the three regions will be significantly improved in the future,and the gap of mortality in the three regions will gradually narrow,without crossover,which accords with the biological rules.Finally,based on the predicted age-specific mortality of men in the Mainland,this paper calculated the life expectancy of 60-year-old men and measured the longevity risk of the pension system through three indicators: the actuarial present value of continuous annuity(APV)and two types of Capital requirements for longevity risk(LCR),and analyzed the gap of the pension insurance personal account in the future..