A Bayesian Generalized Additive Model Approach for Forecasting Mortality Improvement with External Information

Abstract

Mortality modeling is facing new challenges as historical mortality experiences are insufficient to foresee unprecedented changes, such as the impact of the COVID-19 pandemic. Expert opinion has become one important source of information that provides additional insights into the pandemic's possible future courses. In this paper, we develop a Bayesian generalized additive model where external information can be seamlessly integrated into the projection of future mortality improvement rates. A collection of spline functions over the age and period dimensions is utilized to construct a smooth transition of mortality improvement trends from recent changes to long-term rates. Our modeling approach is able to incorporate different types of external information and elicit expert opinions in a coherent probabilistic manner. Lastly, we use three case studies with COVID-19 mortality data to illustrate the applications of the proposed model in different modeling scenarios.

Keywords: Expert elicitation, Spline functions, Generalized additive model, Predictive imputation, Metropolis-Hastings-within-Gibbs

JEL Classification: G22