Recent mortality experience increasingly supports a return to pre-COVID-19 pandemic expectations in the UK and the U.S., but the pattern is not universal. Analysis using three philosophical views calibrated to the CMI_2025 model shows why assumption-setters must align model parameters with the specific population being modeled.
This paper revisits three common philosophical views of how a pandemic would be expected to impact future mortality rates. By applying these views to the latest data in the UK, U.S., and the Netherlands, we investigate how the pandemic has affected mortality.
Considering data up to 2026, in the UK and the U.S. at a high-level, mortality rates appear to be returning to pre-pandemic expectations at older ages. This is an important observation that should guide how we set best-estimate mortality assumptions.
However, the emerging pattern is not the same everywhere. In the Netherlands, mortality rates have remained persistently high. This paper demonstrates the challenge of setting mortality assumptions in that country and highlights the risk of applying approaches used elsewhere to additional countries without sufficient care.
Modeling note: For this analysis we will fit the UK industry model, the Continuous Mortality Investigation (CMI) model,1 to each of the three countries. We consider only ages 65-95, as these are the most material ages for longevity insurance products. Mortality rates at younger ages often exhibit a different experience and would require a separate analysis.
Each view has been fitted with a calibration of the CMI model. The calibrations have been selected to meet the following criteria:
This paper uses “underlying mortality” in later descriptions, referring to the mortality rates with temporary effects due to COVID-19 being stripped out. This is the most material element when projecting future mortality because it is what drives expected outcomes beyond the pandemic period.
Once the projected 2026 datapoint is included, Figures 1A and 1B show the “return to pre-pandemic” curve seems the most likely scenario. If full-year 2026 data is consistent with the projection, it is reasonable to expect best-estimate scenarios set in 2027 to be more consistent with the “return to pre-pandemic” view.
This result should not be particularly surprising. After all, it was the most popular baseline view in the early days in 2020 and, indeed, before the COVID-19 pandemic. It has also been seen in historical pandemics across the 20th century, where more reliable data is available.3
With the benefit of hindsight, it is possible to question whether it was premature to move away from this baseline and toward other philosophical views.
Figures 2A and 2B show the same lines as 1A and 1B and the addition of two new lines. One is the default CMI_2025 model, with a long-term rate of 1.5% per year. The other is the default CMI_2025 model but with the half-life parameter set to 1.75 and a long-term rate of 1.5% per year.
The default parameterization of CMI_2025 is closest to the “change in direction” view. This is at odds with the emerging 2026 experience, which suggests a return to pre-pandemic expectations.
A move from a “change in direction” to a “return to pre-pandemic” would see life expectancies increase by around 1%. This shows how material decisions about the underlying philosophical view can be, even when the model is a good fit to observed data to date.
Figure 3A shows that male mortality rates returned to pre-pandemic expectations in 2023 and have been consistent with those expectations since then. It is also clear that the curves for a “change in direction” and “return to pre-pandemic” are very similar. This is because 2024 and 2025 mortality rates were very close to the pre-pandemic curve, so the “change in direction” philosophical view is not reflected in the data. Therefore, the paper’s calibration for this view does not produce the intended shape. A bespoke “change in direction” calibration for the U.S. has not been developed because the data clearly contradicts this view.
Figure 3B shows that female mortality rates have been broadly parallel to pre-pandemic expectations while being slightly heavier in 2023-2025. However, these could soon drop below pre-pandemic expectations if full-year 2026 data is aligned with this paper’s current estimate.
Figures 4A and 4B show mortality rates in the Netherlands have been decreasing steadily since 2020, but they are still significantly higher than pre-pandemic expectations. Unlike the other examples in this paper, it is not yet clear which philosophical view will ultimately prove correct. However, it does seem clear that none of the three views presented in this paper is unfolding consistently across males and females.
In the Netherlands, the “change in direction” curve looks aggressive and leads to life-expectancies about 3% lower than pre-pandemic expectations; however, the same method applied to the U.S. leads to an answer very similar to the “return to pre-pandemic” view. This highlights how the same model parameterization can react very differently to input data from different countries.
When using a mortality model, such as the CMI, across different countries, it is crucial to appropriately fit it to the specific population being modeled. The best approach is to first determine an appropriate philosophical view and then to choose a model parameterization to represent this. Assuming the best-estimate parameterization from one country will apply to another is risky and could lead to unintended outcomes.
Other non-COVID-19 effects have impacted mortality since 2019. It may be reasonable to expect underlying improvements to have changed because of these. This analysis does not explore other mortality drivers, but they may contribute to the choice of philosophical view.
Knock-on effects from the pandemic may still be affecting emerging population mortality data, and observed recent mortality rates alone should not therefore be used to justify a particular choice of mortality driver. The preferred approach would be first to take a view on the trajectory of mortality in terms of temporary COVID-19 effects and then overlay any additional views on top of that path.
During the acute phase of the pandemic, it was essential to fit mortality models using philosophical views. There was simply too much uncertainty and too little useful data for setting assumptions. This analysis has shown that even now, six years after the start of the pandemic, the choice of philosophical view is still a material judgment.
In the UK and the U.S., mortality rates would appear to be returning to pre-pandemic expectations. The next step is to ensure models adequately reflect this recent experience.
In the Netherlands, the picture is still not clear. Special care should be exercised when setting assumptions in this country. Simply carrying over a model or parameterization from another country could lead to material errors. It is worth remembering that mortality projection models, even those as sophisticated as the CMI, are only as good as the parameterizations used.
The pattern of future mortality is uncertain. Rates may not remain on the pre-pandemic trend for a prolonged period. However, perhaps an answer to the question of how the pandemic affected short-term mortality rates is closer at hand.
RGA’s biometric risk expertise drives meaningful business results. See how we apply in-depth insights to help our partners grow.
