A Bayesian Method for Forecasting Mortality Rates by Health State: LONGEVITY 13: International Longevity Risk and Capital Markets Solutions Conference 2017 A Bayesian Method for Forecasting Mortality Rates by Health State: with Rising Life Expectancy Atsuyuki Kogure Keio University, Japan Shinichi Kamiya Nanyang Technological University, Singapore Takahiro Fushimi Stanford University, USA September 21-22, 2017 1
Aging and mortality forecasting heavy burdens on long-term care cost 2
Japan has been and will be aging very fast ! Population Pyramid of Japan from 1920 to 2050
Subpopulation mortality forecasting 4
Our objectives 5
Mortality forecasting for total population Death numbers for age x at time t Exposures (population sizes) for age x at time t Dxt Ext Force of mortality
Subpopulations by health state Death numbers for age x at time t in state j Exposures (subpopulation sizes) for age x at time t in state j Dxt0 Health state 0 (no problem) Ext0 Health state 1 (least severe) Ext1 Dxt1 ... ... ... Health state J (most severe) ExtJ DxtJ
Lee-Carter structure by health state
Mortality forecasting for subpopulations
Force of mortality for total population
Mixture Lee-Carter model
Identifiability of the mixture LC model
Bayesian estimation: parameter uncertainty
Priors for observation equation 14
Priors for health factors 15
Priors for State Equation 16
Hyperparameters
Application: Public Long-term Care Insurance System in Japan
Japanese Public Long-term Care System
Source: Monthly Report on the Status of Long-term Care Insurance Trends of Persons Certified As Requiring Long-term Care Total number of certified persons in 2015 is 608 (in 10, 000’s) increased by a factor of 2.79 for the past 15 years. total In 10,000’s Care levels Transitional Care levels levels Support 2000 2005 2010 2015 Source: Monthly Report on the Status of Long-term Care Insurance
Health states
Sizes of LTC subpopulations
Bayes Computation: MCMC
Posterior distributions for η,γ65,β65,κ2001: male
Summary statistics of posterior distributions: male
Changes in posterior means of γx,βx,κt over x or t male
Posterior distributions for η,γ65,β65,κ2001: female
Summary statistics of posterior distributions: female
Chanes in posterior means of γx, βx ,κt over x or t female
Gender difference in health effects male ηj health effect femae j=health state
Future mortality rates by health status
Future mortality rates by health status j=5 j=4 j=3 j=2 j=1 j=0 Male Female
Survival rates by health status
Future survival rates by health status Male Female
Conclusions (1)
Conclusions (2)
References
References