Structural changes in mortality rates: with an application to Dutch and Belgian data

We focus on a collection of stochastic mortality models, applied to two age buckets (20-89 and 60-89) of Dutch and Belgian mortality data. Recent literature relies on the standard ARIMA-framework (in particular: a random walk with drift) to project mortality rates. As a result the projections can be highly sensitive to the calibration period. We present a modelling strategy for the time-dependent parameters that allows for objective, statistical detection of one or more structural changes in the time series. By comparing projections based on different calibration periods and different time series specifications, we show that the proposed methodology leads to more robust mortality projections with respect to the calibration period used.