A0330
Title: Multivariate analysis of mortality data using time-varying copula state space models
Authors: Ariane Hanebeck - Technical University of Munich (Germany) [presenting]
Claudia Czado - Technical University of Munich (Germany)
Abstract: The aim is to model and quantify the dependencies between five causes of death, conditional on the weekly number of Covid deaths. Based on the given time series data, we propose to use the model class of copula state space models. The associated latent variable, which we assume to be independent of the number of Covid deaths, can be interpreted as a general driving factor of the causes of death. The dependence between the causes and the latent state however is modeled as varying with the number of Covid deaths. Using this approach, the data in the pre-Covid and the post-Covid time can be modeled within one setup. This leads to a very flexible model allowing for the time dynamics between the causes of death. For the inference, a Bayesian approach is chosen. Due to the high nonlinearity and non-Gaussianity, a Hamiltonian Monte Carlo algorithm is used to sample from the posterior density. After fitting the model, we are able to conduct scenario-based projections for future mortality levels and life expectancy.
