Prior research in coherent mortality projection has been concentrated on the design of different mortality model structures while little attention has been paid to the decision which time series models should be employed to project the model’s parameters into the future. This thesis illustrates that the selection of the projecting time series model has a substantial impact on the projected mortalities’ characteristics including their coherence or incoherence. Therefore four different time series models are evaluated based upon the Poisson Common Factor Model (J. Li 2013) using qualitative and quantitative measures. Two different ordering approaches are used to determine the models’ orders: purely statistical and statistical plus a convergence check for the parameters’ projections. It is illustrated that the convergence check supports the mortality projections’ coherence without any clear negative effect on projection properties or performance. For vectorautoregressive and moving average models, the convergence check is shown to be vital to avoid empirically unjustifiable mortality projections. Strengths and weaknesses of the time series models in the context of coherent mortality projection are pointed out and the results’ robustness to selected cohort extensions of the underlying mortality model is shown.     «

Prior research in coherent mortality projection has been concentrated on the design of different mortality model structures while little attention has been paid to the decision which time series models should be employed to project the model’s parameters into the future. This thesis illustrates that the selection of the projecting time series model has a substantial impact on the projected mortalities’ characteristics including their coherence or incoherence. Therefore four different time series...     »