New methodology for modelling mortality by amounts

The Scandinavian Actuarial Journal today published an article on modelling mortality by continuous covariates, such as pension size.  The final published article can be viewed here (there is an earlier preprint here).

Mortality levels vary by benefit amount, and a common simplification is to create a handful of non-overlapping ranges of varying widths. However, this ignores the continuous nature of benefit amounts and leads to discretisation error, i.e. heterogeneity within benefit ranges and step jumps at range boundaries.  Another drawback of discretisation is that fitted parameters are not easily extrapolated to values outside the range of the experience data.  To address these shortcomings it is often better to model mortality continuously.  The new methodology avoids discretisation error and extrapolates to amounts outside the range covered by the calibrating data set, thus making it useful for setting pricing bases.  The approach is illustrated using seven international data sets of pensioners and annuitants, and is implemented in the Longevitas survival-modelling software.

References

Richards, S. J. (2022) Modelling mortality by continuous benefit amount, Scandinavian Actuarial Journal, doi 10.1080/03461238.2022.2025891.