Skip to main content

Swivel vision: developing a more rounded view on long-term mortality improvements

A new framework for future mortality – part 5 of a 7-part series

Insurance Consulting and Technology
Insurer Solutions

By Richard Marshall and Momtchil Iliev | November 4, 2019

Part 5 in our series discusses a rounded long-term improvements approach and looks at how our views of short-, medium- and long-term improvements can be combined to give a single consistent improvement surface.

Long-term improvements are conventionally a matter of judgement. Considering combinations of prospective and retrospective mortality improvement mechanisms will help refine that judgement.

Characterising the ‘long-term’ for improvement modelling

Back in the second part of this series, we characterised the long term as the period beyond the point up to which current information can (directly) be used to calculate probable future improvement rates.

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. In response to these issues, we introduce a new framework for the development of mortality improvement assumptions.

By about 15 years into the future, any information we currently hold about pharmaceutical pipelines, actual health and social care spending plans, population prevalence of disease by age, or likely lifestyle trends will probably have ceased to be reliable and predictive. Consequently, we might suggest that the ‘long term’ starts around 15 years into the future.

Whereas the medium-term assumption-setting process considers what is going to change in the future with a reasonable degree of confidence, the long-term equivalent considers only a plausible mid-point for the range of possible future changes. We can’t know enough about what is going to change, but we have various sources of information which can guide us as to how much things might change in aggregate over an extended period.

Fundamental drivers of mortality and horizon scanning

A useful starting point is a set of fundamental drivers of mortality. These are drivers that have a strong causal impact on mortality improvements and that are expected to remain relevant and influential into the long term. These may include various measures of economic growth, which in turn influence the level and distribution of wealth in the population, as well as the tax revenue that could be reinvested into healthcare, research and long-term care services. We could equally look directly at growth in health expenditure on a per-capita basis, adjusting for both the increasing cost of care with age and inflation (weighted towards medical cost inflation). Comparing expenditure with a demand model would be one way to approach this.

From there, further prospective views of mortality improvement mechanisms would also help enrich the consideration of established drivers. A ‘horizon scanning’ approach could be employed, where we attempt to identify the most likely and impactful new drivers of mortality variations that might be influential in the long-term. We could evaluate medical practices such as personalised and regenerative medicine, or environmental factors such as pollution, recognising that the potential long-term impacts of these drivers could be similar in nature and strength to other current drivers of mortality. We would also need to form a view of the range and distribution of health conditions that would constitute the backdrop to the long-term drivers of improvement. A consistent approach would be to use the projections that have been carried out to inform our medium-term views.

International trends

Beyond that, we could also examine the causes of trends in mortality improvements in other countries. Some such causes may not yet have affected domestic mortality experience. Their potential to influence domestic mortality could be considered, allowing for uncertainty around whether they would potentially take hold in the short, medium or long term as well as whether their impacts would be sustained. An example could be the adoption of medical healthcare systems and funding structures that have proven to be successful in other countries.

Historical improvements

Another source of information could be the average annual mortality improvements that have been observed over a sufficiently long period of time. However, whilst this data may be available at a national population level it is unlikely that companies have data on their portfolios going back far enough – and even then, it is arguable that even going back beyond 50 years yields too few data points to develop a credible extrapolation.

However, there are more significant issues with this approach, namely the assumption that the drivers of past mortality improvements will continue to have the same impact on the same causes of death into the long term. Given the changing dynamics of mortality improvement mechanisms, this seems unrealistic.

Some would argue that, in the very long term (over 25/50 years?), the idea of reasonable and ‘precise’ causal forecasts of any type become too mired in parameter and model risk anyway (i.e. ‘we don’t have a clue’). Here, the argument goes, we may as well adopt an assumption in line with an equivalently long-term historical period. But this approach remains vulnerable to the problem noted above; it is also particularly vulnerable – if we are hazarding very long-term assumptions – to the problem of diminishing returns, whereby every extra longevity ‘victory’ is increasingly hard and costly to achieve (see examples below).

Limiting factors

We can obtain a different perspective by considering factors that could limit the potential for mortality improvements in the long term. Some, for example, have argued that diminishing marginal returns apply to the relationship between healthcare funding and mortality improvements . That is to say, the amount of money required to improve life expectancy by one more year would be considerably greater than that required to achieve the previous extra year of improvement. However, the real value of this consideration is that it provides a natural sense check to see if assumed long-term mortality improvements would require levels of investment that may not be feasible.

As an example of a possible practical constraint on mortality improvements, one could consider the guidance of the UK’s National Institute for Health and Care Excellence regarding the ‘cost-effectiveness’ of public health interventions.

Other limiting factors that could play a part are associated with established physiological limitations which are unlikely to be overturned in what we are referring to as the long term. Examples include the mortality differential between males and females and the mortality convergence for different socio-economic groupings of lives at older ages. Another example of which work has recently been carried out is the limit implied, at least in part, by comparing the mortality of ‘super-healthy’ lives with that of normal lives.

Blending of views on improvements

With a more rounded view on long-term improvements and the analysis of short- and medium-term improvements discussed in previous posts, we then need to blend those views to build a single improvement surface for use in projections.

Overlap between the assumptions for the adjacent periods is helpful, as it allows us to blend the improvement assumptions more smoothly by applying weights to the improvements from each period.

For example, if we extend the short-term view to five years from now and the approach used to determine the medium-term view up to (say) two years from the present, this gives four years over which weights could progressively shift from the short-term view to the medium-term view (e.g. 80:20 in year two, 60:40 in year three, and so on).

Close to the boundary the two views should ideally not differ materially, but in reality – given the very different modelling approaches – they are likely to. A practical approach here is simply to blend as above; if the differences are severe, it may be necessary to apply expert judgement as to which view is better supported by the evidence, and whether any other data points might be found to inform the decision.

The following diagram presents a stylised view of how each period might contribute to the evolution of the mortality reduction factors over time.

A diagram showing a stylised view of how each time period might contribute to the evolution of the mortality reduction factors over time
Figure 1: Blending insights from different improvement models

Next time

In the next instalment, we will consider how our framework can support a more consistent relationship between the best-estimate improvement assumptions and an insurer’s internal model (for Solvency II firms) or the longevity trend component of an economic capital model. We will discuss:

  • New information risk and its interpretation within this framework
  • Model risk
  • Use in ORSA scenarios

Upcoming post

  • Potential impacts on financial reporting

1 Nolte, E; McKee, M; (2004) ‘Does health care save lives? Avoidable mortality revisited’. The Nuffield Trust, p. 139. ISBN 1902089944
King’s Fund (2006) ‘Spending on health care – How much is enough?’
2NICE (2013) ‘How NICE measures value for money in relation to public health interventions’,



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.

Momtchil Iliev
Lead associate

Momtchil Iliev is a Lead Associate in Willis Towers Watson’s Insurance Consulting and Technology business and an expert in bulk annuity pricing and risk modelling.

Related Capabilities

Contact Us