VitaMins: Interpreting the new CMI_2018 model

Club Vita take a look at the recent changes to the CMI projections model and discuss the drivers of these changes and what this could mean for liabilities.

The Continuous Mortality Investigation (‘CMI’) have released the latest version of their model for projecting mortality improvements, known as CMI_2018.

This new version has been updated to reflect experience of the England & Wales population up to 2018, and also incorporates a significant change to the ‘core’ calibration of the model. These two changes have had similar impacts on projected life expectancy produced by the model, and the value placed on liabilities using the unadjusted model as a result.

The main headline is that projected life expectancies using this model will be lower than corresponding projections using the previous version of the model - someone currently aged 65 would be projected to live for around 6 months less than before. In isolation, this could lead to a reduction in liabilities of around 2.5% from moving to the latest model (dependent on financial assumptions and maturity of scheme).

The unadjusted CMI model is intended to be suitable for projecting the life expectancy of the general population of England and Wales. Club Vita analysis, and other investigations, have shown that more comfortable pensioners are experiencing notably higher longevity improvement than the general population and that these members make up the bulk of the liabilities for most schemes.

It is important to consider the drivers behind the headline figures, and how the model can be adjusted to be more appropriate for use with pension schemes. Making these adjustments is likely to lead to smaller reductions in life expectancies (and so liabilities), than using the ‘core’ settings of the model.

Is life expectancy falling?

It’s important here to make the distinction between historical life expectancy and projected life expectancy.

Historical life expectancy is based on observing actual death rates, so can be objectively measured, and has continued to increase at the population level, albeit at a slower rate recently than seen over the 2000s.

Projected life expectancy is an estimate based on how mortality rates are expected to evolve in the future, so is inherently subjective. This is where the CMI model comes in – it uses historical data to estimate ‘current’ improvement rates, and then projects them into the future, blending to an assumed long-term rate. 

It is this projection for life expectancy that reduces under the new model, which effectively assumes slower current rates of improvement than under the previous model.

What has changed in the new model?

The new CMI_2018 model has made two significant changes to the ‘core’ parameterisation, relative to the previous version (CMI_2017).   

  • Firstly, it allows for 2018 experience data from the England & Wales population.  It has been well publicised that, after some relatively rapid improvements in life expectancy in the 2000s, this rate of improvement has slowed somewhat in recent years. 2018 was another year where improvements were fairly limited. Therefore, incorporating this data in the model has acted to reduce expectations for future improvements.
  • Secondly, the CMI have made a technical change to the ‘core’ parameterisation of the model, which has had the effect of making the projections more sensitive to recent (heavy) mortality experience.  This change was intended to provide a better ‘fit’ to the historical England & Wales population data used to calibrate the model.

These two changes have had similar impacts (i.e. have each contributed around half of the change from CMI_2017).  We discuss below the relevance of both changes to pension schemes.

In addition, the CMI have added an extra parameter to the model, which allows users to make an addition to the initial rates of mortality improvement.  This is intended to enable users to reflect the different experience of specific populations, for example the higher levels of improvement seen in more affluent pensioners over recent years.

What is driving recent mortality experience?

The slowdown in the rate of improvement in life expectancy had been cause for much debate among industry professionals.  A number of factors have been proposed as potential drivers:

  • Having seen significant improvements in death rates from cardio-vascular disease in recent decades (and major contributor to overall death rates), we have simply run out of room for improvement.
  • Austerity has led to marked reductions on health and social care budgets – has this had an impact on life expectancies, particularly among the old and vulnerable?
  • Death rates from Dementia and Alzheimer’s have increased markedly in recent years (faster than simply as a result of aging populations). The reasons behind this rise are not yet fully understood.
  • We have had a series of heavy winters and virulent flu seasons in recent years. Could these be having a combined impact of wearing down frail people causing premature mortality?
  • Were the strong improvements in life expectancy seen over the 2000s simply an anomaly, and the lower levels seen since a more ‘normal’ level?

The reality is that it’s likely to be some combination of the above factors (and possibly others).

How does pension scheme experience differ?

The focus of the CMI’s analysis is England & Wales population data. However, this is not the most relevant data set from a pension scheme context (it includes, for example, those too ill to have ever been in employment).

Club Vita analysis has shown that pension scheme members have seen higher improvements in life expectancy than the general population.  In particular, the most affluent pension scheme members have been largely resilient to the heavy experience over recent years (while less affluent members have seen life expectancy plateau). This analysis of differences in experience has also been replicated by studies carried out by the likes of Public Health England and the CMI itself.

Why the change in core smoothing parameter?

The CMI have reduced the smoothing parameter in the ‘core’ parameterisation of CMI_2018. This change was intended to give a better fit to population data experience.  However, when considering pension scheme data, there is less justification for the amendment made, given how pension scheme experience has varied.

The CMI themselves recognised this point, as shown in the following quote from Working Paper 116 (which covered the results of the CMI’s consultation on changing the core parameterisation).

“We stress that the Core value of 𝑆𝜅 and the Core value of the addition to initial improvements are intended to be suitable for the general population of England & Wales.... A number of respondents referred to evidence that members of pension schemes and higher socio-economic groups have had higher mortality improvements than the general population in recent years. We agree, and we urge users of the Model to ensure that they use values of 𝑆𝜅 and the addition to initial improvements that are appropriate for their specific populations.”

CMI Working Paper 116

As a result, many pension schemes are expected to want to assume higher initial rates of improvement than under the ‘core’ CMI_2018 model (for example by using the new ‘initial addition to mortality improvements’ parameter). As such, the impact of moving to CMI_2018 will be much reduced.

Will rates of improvement continue to fall?

The future is clearly uncertain. The steady rates of improvement in life expectancy in the 2000s were expected by many to continue in the short term. However, we have now seen several years of falling improvement rates. How will this trend evolve?

One source of information is the details of weekly deaths, as published by the ONS. While 2018 saw another year of heavy deaths (so slowing improvements in life expectancy), this was largely driven by a very heavy first quarter, offset by much lower death rates in the rest of the year. 

Early evidence from 2019 suggests that this low death rate has persisted into the new year. It may be that a combination of a relatively mild winter, and potentially a successful flu vaccination programme, could result in a noticeable drop in deaths over the year (although it’s clearly early days). So, it’s possible that we could see a pickup in mortality improvement rates, potentially leading to an increase in life expectancies in the near future.

Clearly this is something that we, and many others, will be keeping a close eye on in the months ahead.

Share this article:

Interpreting the new CMI_2018 model

Download a print friendly version of this article

Join our mailing list

Stay up to date with the latest news, insights, events and webinars

Join the discussion

Our Friends of Club Vita LinkedIn group brings together like-minded individuals with a shared interest in longevity. Share insights on longevity and connect with industry peers.