Skip to main content

Do mortality improvement assumptions reflect reality

A new framework for future mortality – part 1 of a 7-part series

Insurance Consulting and Technology
Insurer Solutions

By Richard Marshall | September 9, 2019

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. This first article questions our industry’s reliance on commonly used mortality benchmarks and the role of group-think.

As one of the most material assumptions for writers of annuities and for pension schemes alike, accurate mortality improvement assumptions are vital for pricing and maintaining adequate funding. However, in a day when each country or region has developed its own favoured approach to improvement modelling, is there a danger of over-reliance on benchmarking and casual acceptance of the status quo at the expense of understanding what is driving mortality improvements? Do organisations really know how improvement assumptions relate to the real world and their own books of business or pension beneficiaries?

Current assumption setting processes vary in complexity. More sophisticated approaches involve analysis of recent and longer-term trends (e.g. causes of death) to set parameters within the best-estimate improvements model. More common is the use of benchmarks and reinsurance rates when selecting model parameters.

However, with an "a priori" choice of improvement model, many insurers effectively reduce their basis setting to choosing a model variant and/or two or three parameters in the model.

Whilst this is great for communicating improvements, the practice has encouraged the use of benchmarking by both insurers and auditors, at the expense of interpretation of those improvements.

Benchmarking of assumptions: pros and cons

Benchmarking can be a blessing or a curse. At best, it can show actuaries where they need to justify their assumptions more carefully if out of line with the industry; at worst, it can be used as a stick to force insurers to run with the pack (for example, due to auditors’ use of benchmarking data). In recent years, surveys of improvement assumptions carried out by Willis Towers Watson have shown growing convergence of assumptions between participants.

If the benchmark consists of multiple independent well-thought-out views, it can provide validation of the reasonableness of those views. Convergence might be due to insurers evaluating the same set of systemic drivers of mortality, relying upon the same datasets, having similar groups of policyholders and being supported by a small number of reinsurers (potentially with similar views on improvements).

In territories in which the Solvency II regime applies, the Delegated Acts (Article 10) could even apply if these views represented a deep liquid market underlying a market price of mortality improvements.

On the other hand, if many companies’ contributions to a benchmark reflect their own reliance upon an earlier benchmark, it represents ‘group-think’ or ‘herding’ of assumptions and can have negative consequences, including:

  1. Internal models (for Solvency II) or other regulatory or economic capital models unrealistically predicated on how an insurer ’would change’ their best-estimate improvements in various situations.
  2. Outsourcing of responsibility for assumption setting for reporting in the local regulatory framework.
  3. Potential ‘tick-box’ audit exercises based on benchmarking, not the insurer’s approach in arriving at their improvement assumptions.

Is group-think actually occurring?

In our 2018 survey of UK improvement assumptions, long-term rates became slightly more tightly clustered and about half of participants had chosen to use the default period smoothing parameter. Very few firms were making any other adjustments to the core CMI model parameters. All of this suggests herding, even if it’s not definitive proof that it is happening.

And whilst we do not have benchmarking information for non-UK improvement assumptions, it is likely that there is a degree of convergence within other territories as well, as ‘best practice’ is shared through industry conferences and academic publications.

How to avoid group-think

Asking some key questions can reduce reliance on benchmarks:

  1. Ignoring the choice of improvements model, what do we expect short-, medium- and long-term improvements to look like and why?
  2. Is there any model which allows these views to be reflected (and if so, with what parameters)?

The aim is to make sure that the best estimate improvements correspond to views around actual drivers of mortality over each time-period.”

Richard Marshall

The aim is to make sure that the best estimate improvements correspond to views around actual drivers of mortality over each time-period. If we are able to clearly articulate the rationale behind our improvement assumptions in terms of those drivers, then our views will be less susceptible to the inappropriate influence of benchmarks (or pressure from users thereof).

Appropriate governance of the assumptions process can also reduce the risk of herding or, worse, choosing assumptions to control their impact on the regulatory balance sheet contrary to the realism required in regimes such as Solvency II.

Next time

In the next instalment, we will introduce an overall framework for mortality improvement assumption setting which: supports rigorous independent assumption setting; uses benchmarking for validation, not assumption setting; and (where relevant) aligns best-estimate assumptions with a company’s internal model.


Upcoming articles

  • Developing a short-term view of mortality improvements
  • 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 In the UK, the CMI series of projection models is the lingua franca for improvement modelling and the a priori choice of projection tool. Elsewhere in the world, different approaches are preferred, ranging from stochastic models such as variants of the Lee-Carter model or the Cairns-Blake-Dowd model to simple period-independent improvement factors.


Richard Marshall

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