All Times ET
Keywords: Bayesian Poisson model, Mortality projection, Spike and slab algorithm, Spatio-temporal approach
Mortality projection is a crucial topic in demography since it is applied to wide social areas, for instance, public policy, public health, and finance. Lee and Carter contributed a pioneering approach for extended mortality research. Motivated by the progress of related Lee-Carter models, we consider a hierarchical bilinear model along with variable selection methods, and spatio-temporal structure under Bayesian frameworks in the estimation of mortality rate. This new model features information borrowing among populations and properly reflects variations of data. It also provides a solution to a long-time overlooked problem: model selection for dependence structures of population-specific time parameters. By introducing a novel spike and slab approach that hierarchically follows the conditional autoregressive (CAR) model via the probit link function, simultaneous model selection and estimation for population-specific time effects can be achieved without much extra computational cost. Additionally, this selection procedure can leverage spatial information to inform about the geographic correlation in adjacent areas. Markov Chain Monte Carlo (MCMC) is implemented to carry out parameter estimation and prediction for mortality in multiple populations. Via the CAR approach and data augmentation steps, a computationally efficient MCMC sampling algorithm is also developed. In the empirical study, we use the Japanese mortality data sets from Human Mortality Database to illustrate the desirable properties of our model.