Cobus Daneel, Chair of the CMI Mortality Projections Committee, summarises plans for the next version of the CMI Model.
The CMI publishes annual updates to the CMI Mortality Projections Model. It has recently proposed significant structural changes to the CMI Model, ahead of the release of the next version, CMI_2024. This blog describes the changes and the reasons for them.
The COVID-19 pandemic was challenging for all mortality projection models. It was uncertain how mortality rates would react beyond the initial shock and the CMI took a pragmatic approach, avoiding material revisions to projected life expectancies until a clearer picture emerged. We excluded data from the peak of the pandemic, as these rates were unlikely to be a good guide to the future, and smoothed through data for other years, focusing on a plausible level of mortality rates in the medium term.
That approach worked well initially, but required subjective judgements regarding how much emphasis should be placed on recent data. Now that we’ve experienced three years of more typical mortality, we have taken the opportunity to improve the CMI Model, providing a better fit to recent data across the age range during and after the pandemic.
The first proposed change is to explicitly model the impact of the pandemic on mortality.
We’ll give full weight to every year of data, and model mortality as a combination of ‘underlying’ mortality, reflecting long-term trends, and an ‘overlay’ that reflects the impact of the pandemic. For the default version of the CMI Model the overlay will start in 2020 and halve every year.
As for earlier versions, the CMI Model provides a framework that allows users to adjust parameters to reflect their own portfolios and views on the impact of the pandemic, and to communicate those to others. The ‘half-life’ – the time it takes for the overlay to halve – will be a key decision for users of the CMI Model and a way to communicate views. We expect this to be a more intuitive parameter than weights, with a real-world interpretation.
The second major change will allow the model to cope better with situations where different trends in mortality emerge at different ages.
The structure of the current version of the model, CMI_2023, leads to similar changes in mortality trends over time at all ages. However, this has not been seen in practice in recent years. Our end-2024 mortality monitor showed that mortality in 2024 for ages 75 to 100 was lower than in any other year, but for ages 20 to 44 mortality was materially higher than before the pandemic.
To address this, we’ll update the structure of the CMI Model to allow for different changes in trends at young, middle, and old ages to better reflect the recent divergence in experience.
The two other planned changes are to the method used to calibrate the CMI Model, making it more robust in light of the additional complexity of the overlay and trends at different ages; and the way in which we constrain cohort components at young and very old ages, so these do not have unintended consequences at other ages.
We have also set out a policy for when we would consider further changes to the CMI Model in light of unusual mortality experience, including triggers based on the level of mortality and changes in cohort life expectancy.
Taken together, the proposed changes and the default parameters for the model would tend to increase cohort life expectancy at older ages relative to CMI_2023 and reduce it at younger ages.
Proposed life expectancies at age 65 would be about five weeks higher for males and about one week lower for females than in CMI_2023.
There will be a webinar on our proposals for CMI Subscribers on 5 March at 10am – details on how to register are available in Working Paper 197.
We are actively seeking views from authorised users of the model. Our consultation is open until 25 March and we intend to provide an update on our plans for CMI_2024 in light of consultation responses in April.
We aim to publish CMI_2024 in the second quarter of 2025 with the exact timing depending on the nature of consultation responses.
