Experience Rating in the Setup of Multivariate Correlated Frailties - Using a Class of Conjugate Hierachical Priors
Specialeforsvar ved Jacob Juhl Sørensen
Titel: Experience Rating in the Setup of Multivariate Correlated Frailties - Using a Class of Conjugate Hiearchical Priors
Abstract: We consider experience rating, i.e. shrinkage estimation in a multi-state life insurance setting of Markovian jump processes and multivariate frailties. In particular, we study and compare a variety of empirical Bayes methodologies for the purpose of accurate risk estimation. To cope with inter-risk dependencies, induced latent risk characteristics, we resort to a class of conjugate hierarchical priors which adds significant complexity to the original (inferential) problem. In this way, a hierarchical Bayes modelling approach allowing for dependent transition rates in the classic Markov insurance setting is acquired and may serve as an inspiration for future work to come. As for maximum likelihood estimation, we derive an expectation conditional maximization (ECM) algorithm which to our belief is new. Ultimately, we compare the applicability of different estimation procedures in a simulation case study by varying the strength of dependency over multiple scenarios.
Vejledere: Mogens Steffensen, Christian Furrer
Censor: Kristian Buchardt, PFA