A key assumption that actuaries are responsible for on behalf or within organisations that are exposed to longevity risk is the exact degree of allowance for future improvements in mortality.
The UK appears to be leading the way on this front with arguably the most internationally developed methodology.
The widely adopted approach currently in use to set the allowance for future improvements in the UK is largely predicated upon the projection of past UK population experience.
This is not without foundation for the purpose of evaluating trends in the near term. Over longer time horizons, however, this will impose effects that result from one-off developments that are not expected to continue or recur.
The environment and conditions that fostered previous improvements is by definition unique as the resulting improvements bring about ever changing scenarios.
The imperfections of the projection methodology are widely acknowledged but the debate thus far has centred more around issues such as
smoothing of past empirical data
different projection methodologies
the relevance of improvements in UK population data for use against a subset thereof
the historic UK cohort effect
imperfections in UK population data as provided via the ONS, particularly at older ages
rather than more fundamentally questioning the use of such methods given the
long delay between drivers and their effects in mortlaity data (see Mortality Improvement Framework)
use of aggregate data traversing a number of fundamentally distinct homogeneous subgroups for his purpose
limited to scope for drivers of past improvements to present themselves repeatedly in future
The CMI model is the usual common language for the expression of such improvements (whether used directly or not).
One possible way in which a hybrid of differing approaches can co-exist is given below.
Instead of simply setting best estimate assumptions, it may be more complete to set out
(a) any longevity catalyst(s) whose occurrence are / is consistent with this initial best-estimate view
It is generally good practice to supplement best-estimate trend assumptions with an explanatory (causal) framework. By furnishing this with the longevity catalysts that are consistent with this / these pathway(s) a clearer and arguably more robust position emerges. In particular, this leads to easy identification of those catalysts which are already allowed for and implicit and those which are not.
(b) how these assumptions are expected to react to certain carefully considered longevity catalysts (not implied within the best-estimate under (a).
Adding (b) above also then gives a best estimate assumption "approach" or "policy" setting out anticipated reactions to real world events rather than a single, infrequently changed point estimate.
"Following the occurrence of Longevity Catalyst A during calendar year 20XY, current best estimate improvement assumptions would change to CMI_20XY_M [1.75] and CMI_20XY_F [1.50]"
This is, of course, quite a crude example using the common language of the CMI model to express how assumptions might react (one may also wish to impose a maximum on 20XY) but the general principle is clear.
This would mean that the occurrence of these pre-defined catalyst events is not ignored, to the extent they have already been anticipated. Indeed, this may lead to adoption of a new assumption before the occurrence of the specified event if its occurrence is deemed to be inevitable (to those responsible for providing and adopting the assumption) perhaps based on associated measurable proxies or indicators.
This is analogous in some ways to a share price already reflecting certain information.
This can also lead to clearly defined differing levels of prudence between longevity stakeholders which may add to regulatory transparency.
For example “Life Office Y is already ‘pricing in’ Longevity Catalyst A whilst Life Office Z is not”
As mentioned earlier, the current de facto position is that mortality improvement assumptions generally only change
• within an organisation: at an annual [say] review for an additional data point
• within an organisation: at an annual [say] review of their model driven by regulatory requirements but still fundamentally based on past data
• among organisations as a whole: once the CMI publish or advocate a new set of tables be that for baseline mortality or mortality improvements. Each of these are typically derived purely from historical data.