Skip to main content
Article

Going back to go forwards: developing a short-term view on mortality improvements

A new framework for future mortality - part 3 of a 7-part series

Insurance Consulting and Technology
N/A

By Richard Marshall | October 2, 2019

Part 3 in our series examines the factors affecting short-term mortality improvements and considers the arguments for and against using an extrapolative approach to determine improvements over this time-period.

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.

It’s easy to mistakenly assume that the question of how mortality rates will evolve in the near future is a question about what is going to happen in the future. In reality, it can be equally (if not predominantly) a question of what has happened in the recent past.


Past choices affect future mortality

Take, for example, when people choose to stop smoking. Their risk of developing (or dying from) smoking-related conditions decreases, but not instantaneously. Consider these statistics1 :

  • One year after quitting smoking, a person's risk for coronary heart disease decreases by half.
  • After five years without smoking, arteries and blood vessels begin to widen again, lowering the risk of stroke. The risk of stroke will continue to reduce over the next 10 years.
  • After 10 years, a person's chances of developing lung cancer and dying from it fall by roughly half compared with someone who continues to smoke. The likelihood of developing mouth, throat, or pancreatic cancer also significantly reduces.
  • After 15 years of not smoking, the likelihood of developing coronary heart disease is the equivalent of a non-smoker.

Such figures illustrate that there is a significant lag between the decision to stop smoking and all of the (mortality-related) benefits of that decision being realised. Estimates for the time taken for an ex-smoker to become effectively equivalent in mortality risk to a never-smoker (or at least as close to this as possible, depending on the period over which they smoked) range from around 20 to 30 years.

This means that when we think about how population mortality will improve over the coming year, we might (theoretically) need to consider changes in the prevalence of smoking from 30 years ago. However, the run-off of the excess mortality risk from smoking is not linear – indeed a significant proportion of the excess risk runs off in the first few years and the decreases thereafter are more modest by comparison, so it’s probably reasonable to restrict attention to changes in smoking over the past few years.

Other behavioural changes leading to improvements in/worsening of mortality have similar lag periods, such as cessation of alcohol consumption (following persistent alcohol abuse), changes in the prevalence of sugar consumption and obesity, and potentially changes in exposure to air pollution (e.g. “PM <2.5” particulate matter from diesel exhausts).

Likewise, changes in the funding of health and social care services can have a lagged impact. With an increase in funding of a national healthcare service, it may take time for new members of staff to be recruited/trained and for treatment waiting times to be cut. Alternatively, with increased subsidies to medical insurance, demand for medical care may increase and private providers may take time to make the changes required to meet that increased demand.

In the event of a cut in funding, service providers will typically try to make do until it becomes impossible to do so; the impact of the cut may be felt some time later through the laying-off of staff, gradual increases in waiting times for treatment, longer waiting periods for transfers of elderly patients to social care settings and potentially a reduction in the provision of healthcare services (closures of wards and surgeries, or 'consolidation' or 'centralisation' of health service providers). Decreases in subsidies for medical insurance may result in a more immediate reduction in (new) demands for care, but existing treatments would likely continue.

A key part of understanding short-term mortality improvements is therefore developing awareness of:

  1. recent trends in mortality-relevant behaviours in society,
  2. political decisions which have an impact on the provision of healthcare services, and
  3. their respective impacts on population mortality, e.g. via relative risks for particular causes of death.

Arguments for (and against) an extrapolative approach

So, what might life insurers make of this?

If recent changes in (for example) smoking behaviour are going to drive next year’s mortality improvements, then by the same argument they would have driven this year’s mortality improvements and those seen previously. If we believe that the changes over time in mortality-relevant behaviours are reasonably consistent and their impact should be consistent over time, then there is an argument for using a purely extrapolative approach in the short-term.

This is an attractive argument, as it allows us to ignore the specifics of the changes in individual behaviours and simply assume that these changes are stable over time. It means that if there is a functional form which adequately describes the mortality in each period, and some way of relating successive annual instances of that functional form, then we can project best-estimate, short-term mortality improvements without having to know anything about what’s driving the changes.

We believe that this argument is flawed for a range of reasons:

  • Past changes in behaviour won’t have been stable over time and relationships between successive years’ mortality (as captured in, for example, the parameters in a time-series model fitted to a long data window) are unlikely to capture these variations in the contributions of different behavioural changes over the period of data used for calibration.
  • Changes in mortality linked to government policy changes (e.g. real-terms funding changes for healthcare and social care relative to demand for those services) are driven by discrete events which we would not expect to see repeated, but these events will influence time-series model parameters.
  • Mortality experience in some years may have been affected by particularly severe events or the absence of contributions from usually predictable sources of mortality. Examples are unusually harsh or mild winters, effective or ineffective influenza vaccines with unusually high or low take-up rates, and extreme heat-waves. Volatility, particularly in very recent years, can make purely extrapolative methods (without smoothing of annual changes) less reflective of the drivers of future mortality.
  • Not everything affecting next year’s mortality improvements is reflected in existing mortality data. We may have information which is relevant to near-term future improvements which has not yet affected population mortality – indeed, this is a key concept within internal models of longevity trend risk: “new information” affecting our view of future mortality. To be consistent with our internal models, it is imperative that we allow our views of mortality to reflect such new information that is available to us.

This is not to say that there is no benefit to using extrapolative methods, but rather to say that using them without appropriate application of expert judgement can be misleading. A good example of the potential pitfalls of a purely extrapolative approach is the significant change in improvements seen prior to and after 2012, perhaps felt most acutely in the very heavy mortality experience of 2015.

A blended view of short-term improvements

We believe that an understanding of the key drivers of recent mortality improvements allows the ongoing impacts of past behavioural and policy-linked changes to be distinguished from the impacts of one-off events. As in life in general, it often pays to take a step back before charging forward.

Our preferred approach to estimating short-term improvements is to extrapolate from recent experience and to overlay expert judgement, both removing the effects of those one-off events which are not relevant to future mortality experience and adding in the effect of any expected immediate future changes which are not reflected in recent mortality data.

If a CMI model is used to represent our short-term views on mortality improvements, then the initial rates of improvement and the direction of travel for those improvements will be the key inputs used to tailor the model projection to reflect our short-term views.

Footnote:

Within our framework for the setting of future improvement assumptions, the short-term view may contribute only three to five years’ worth of improvements; those at the end of this range will be influenced by views in the medium-term as well (and we will cover blending between short-term, medium-term and long-term views later in this series).


Next time

In the next instalment, we will consider how to develop views of mortality improvements in the medium-term, considering:

  • Multi-state medical modelling of a non-stationary population.
  • Future behavioural trends assessed via “driver-based” modelling.

Upcoming articles

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

1Taken from "What happens after you quit smoking?" (MedicalNewsToday.com; Fletcher, J.;19 Nov. 2018), though similar statistics are available from multiple sources.


Author

Richard Marshall
Director

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
Related content tags, list of links Article Insurance Consulting and Technology Insurance