The Usage of State Space Models in Mortality Modeling and Predictions

Abstract

In demography, mortality modeling with respect to age and time dimensions is often associated with the traditionally used Lee-Carter model. The Lee-Carter model considers a constant set of parameters of agespecific mortality change for forecasts, which can lead to the problem of overcoming the biodemographic limit. The main motivation of this paper is the use of more flexible models for mortality modeling. The paper explores the use of state space models for modeling and predicting mortality in a form not typically used in demography. In this context, it is a generalized Poisson state space model with overdispersion parameters. Concerning the empirical results, a comparison is made between the predictive abilities of the Lee-Carter and the generalized Poisson state space model with overdispersion parameters. The state space Poisson model with overdispersion parameters led to better results with respect to the comparison of modeled and historical observations. However, when comparing the predictions in the cross-validation area, both models were represented with similar overall mean squared error.

Keywords

• generalized state space models
• extended kalman filter
• exponential smoothing
• lee-carter model
• mortality
• prediction comparison