Skip to main content

Reality check: why it’s time to flip the back-to-front mortality improvement assumption process

A new framework for future mortality - part 2 of a 7-part series

By Richard Marshall | September 23, 2019

Part 2 in our series introduces our new framework for setting mortality improvement assumptions and contrasts it with an alternative decision-making process which might feel a bit closer to current industry practice.
Insurance Consulting and Technology
Insurer Solutions

Unlock more

About the series

In this series of articles, we discuss the impediments to understanding mortality improvements and the dangers of herding and group-think in this process.

The current approach to setting mortality improvement assumptions in insurance tends to be a bit of a backwards process.

Are the impacts of a basis change tolerable? How much do we need to hold back in reserve so that we can release something from technical provisions just when the CFO needs an extra, say, £100 million? How does a change in the best-estimate affect the capital that the regulator will expect us to hold? It’s tempting to focus so much on the outcomes from setting improvement assumptions that the realism of the assumptions is only a question for validation of choices.

While there are clearly significant differences in the approaches taken by individual insurers, in some instances the decision-making process can be largely reduced to a combination of a benchmarking exercise, possible adjustments for basis risk, and questions about the impact of any proposed changes on the balance sheet. Figure 1 shows an example for UK firms using the CMI model, but the same process would apply more widely, substituting the CMI model with whatever projection model is preferred locally.

Typical decision-making process
Figure 1. Typical decision-making process

Such homogeneity results from an assumption that the only option is to use the locally preferred model (such as the CMI model) and to use parameters set by academics or “core parameters” for those models (except for period smoothing in the case of the CMI model). Insurers may even reduce the assumption setting exercise to a decision about the variant/version of the locally preferred model to use and a limited number of parameters, such as the smoothing parameter to adopt and the long-term rate of improvement to target in the CMI model.

We believe that insurance as an industry can and should do better.

Moving towards an “ideal” approach

That’s why we’ve developed a framework within which insurers can develop their mortality improvement assumptions without assuming a model or parameters in advance, effectively inverting the decision-making process.

This approach involves determining expectations for future mortality improvements on the basis of all of the available information and then assessing whether there is a model and choice of parameters which reflects those expectations.

This is no insignificant task – drivers of mortality will vary over time and our ability to predict changes in some drivers will be short-lived for some, longer-term for others. This is tackled by dividing up the future into three broad time-periods:

  • The short term (say three to five years): where we’re confident in our predictions, short of extreme one-off events, because we see continuity in the effects of (recent) key drivers of mortality.
  • The medium term (10 to 15 years): where we can reasonably forecast a range of drivers of mortality (with increasing uncertainty over time), based on, for example, pharmaceuticals in development, technological innovations that support lifestyle changes, or academic studies of population health trends.
  • The long term: where expert judgement is key, though more fundamental drivers such as overall economic growth, or per-capita health expenditure growth could help to inform such judgement. No current information will likely be relevant or have a predictable effect by that time.

Our alternative framework (shown in Figure 2, below) uses the analysis of these separate views of improvements for each time-period to recommend a model and parameters to the relevant body governing best-estimate assumptions; benchmarking and movement analysis are used for validation only.

Typical decision-making process
Figure 2. Willis Towers Watson alternative framework

What has changed?

Key differences between this approach and the prevailing one are:

  • Understanding of why mortality is improving (or otherwise) leads to a choice of model and parameters which reflect this understanding.
  • No model (CMI or otherwise) is selected a priori; assumption setting is not reduced to a choice of parameters.
  • Even if a CMI model can reflect the expressed views1 , core parameters are not assumed – parameters are selected to fit the insurer’s views, rather than shoehorning the views into a core model specification.
  • Benchmarking is not used in generating a recommendation, but it is available for use in validation of the recommended assumptions, for example to explain differences between own views and those used by peers.
  • The target is a genuine best-estimate set of improvements; balance sheet stability isn’t considered to be more important than realism.
  • The assumptions lend themselves to alignment with an internal model or other model of economic capital. For example, ‘new information risk’ can be investigated by considering the actual changes which would be made to the best-estimate improvements in light of emerging information.

Making the best use of your time

Since this alternative approach requires a substantial time investment, it makes sense to focus on trends that are most material to your portfolio. For example, the relative materiality of improvements in each period differs depending on age, as shown below. In general, the older an individual, the greater the relative importance of shorter-term improvements. This is intuitively sensible – only those who survive into the middle- or long-term periods can benefit from those improvements.

Figure 3. Relative importance of different periods of improvements
Increase years Base EoL % increase
From To Increase. EoL (M65) EoL (M75) EoL (M65) EoL (M75)
N/A N/A 0.00% 22.165 13.657
1 4 0.50% 22.303 13.764 0.62% 0.78%
1 4 1.00% 22.441 13.87 1.24% 1.56%
5 14 0.50% 22.435 13.81 1.22% 1.12%
5 14 1.00% 22.706 13.965 2.44% 2.25%
15 30 0.50% 22.33 13.686 0.74% 0.21%
15 30 1.00% 22.498 13.716 1.50% 0.43%

Of course, this table is a simplification – illustrating a flat uplift at all ages and in all relevant future years (relative to PMA08 with CMI_2017_M[1.5%] core improvements). In practice, insurers would benefit from understanding variations in the ages at which improvements occur as well as the periods in which they occur.

…we’ve developed a framework within which insurers can develop their mortality improvement assumptions without assuming a model or parameters in advance, effectively inverting the decision-making process.”

Why flip?

Mortality improvements happen for a reason. Using a bottom-up analysis of the factors leading to improvements over various periods in the future will help develop a bespoke projection of mortality and inform the choice of improvements model used in an insurer’s best-estimate basis. It also concentrates minds on those improvements that matter the most to an individual insurer.

Improved realism, improved consistency with the internal model, and the ability to relegate benchmarking to a validation tool. All good reasons, we believe, why insurers stand to benefit from flipping how they approach mortality assumptions.

Next time

In the next instalment in this series, we’ll discuss how firms can develop their views of mortality improvements in the short-term, focusing on:

  • Past changes in drivers of mortality with a lagged impact.
  • Likely (immediate) future changes in drivers of mortality with a near-immediate impact.

Upcoming articles

  • Developing a medium-term view of mortality improvements
  • Developing a long-term view and blending between time-periods
  • Consistency with an Internal Model or economic capital model
  • Potential impacts on financial reporting

1 Here, the choice of the CMI model would not be UK specific – it is simply a tool used to reflect the improvement assumptions developed by the insurer.



Richard Marshall is a Director in Willis Towers Watson’s Insurance Consulting and Technology business and leads the development of mortality and demographic risk models for our UK business.

Contact Us