A flexible framework for selecting longevity models
Longevity models have a long history, and widely accepted models in the literature include the Lee-Carter model, the Age-Period-Cohort models and CMI models. These models play a central role for insurance and annuity providers in understanding and quantifying longevity risks and are sometimes used to value businesses worth several millions of pounds. This creates a need for actuaries to properly select their longevity models as well as understand the underlying model risks. Traditionally, model selection in the context of longevity modelling has been achieved by fitting all the models under consideration and ranking them using a model selection criterion, such as AIC. This has certain disadvantages; for example, this requires practitioners to fit many different models and manually compare them. Another disadvantage is that considering only one final model necessarily leads to loss of information since other models may also be informative. In this talk, we propose a general and flexible modelling framework for selecting longevity models. First, we build a general model which encompasses a wide variety of models considered in the literature. Next, we provide a novel inference approach where the model selection is part of the inference process itself. This eliminates the need for manual model selection and provides a comprehensive framework which includes a variety of models. We achieve this by using a Bayesian model selection procedure, which is an established framework for selecting different variables or different model variations. The approach allows the full model to tailor itself towards specific terms as informed by the data while discarding terms that are not relevant. We have implemented our model in an R package to allow practitioners to consider additional effects (either continuous or categorical) into their mortality modelling. The package will report the model variation which is more strongly supported by the data. We demonstrate our framework by using data from the Human Mortality Database.
Speakers
Aniketh Pittea- Grant Thornton UK,
Dr Alex Diana – Lecturer at School of Mathematics, Statistics and Actuarial Science (SMSAS),
Chair- Michelle Lister – UK Life Consulting Director- AON