TOAMS 6 mortality trends: Highlights from WTW’s latest study

JUSTIN FOUNTAIN: So you mentioned the years that this edition of TOAMS covered. This is the first edition of TOAMS that covered the COVID era. How did COVID impact things this go around?

BOYANG MENG: Yeah, no it was definitely a key consideration even going into the study, before we even collected any data, had any results to look at, we had to be conscious of the impact COVID was going to have. We had to give thought into how we would study mortality, whether we would look at claims or causes of death to identify COVID claims and exclude them. That was a tricky proposition because of the tendency for that information to be incomplete or not fully accurate.

Ultimately, we did decide to look at COVID using an excess mortality method with predictive modeling. And the use of generalized linear models allows us to really isolate the effect of COVID. We did look at the relationship of a number of typical drivers of mortality with calendar year. So of those attained age was the most prominent.

So we do have a chart here of the pattern of some key age groups over time. And what we saw broadly was that the effect of COVID did track fairly closely to the general population in terms of the timing and patterns of it. So one thing that you can see is that for the younger ages, so below age 50, the peak mortality happened a little bit later in 2021.

So in this case peaking at about 35% to excess compared to our pre-COVID period. Compare that to our oldest age group. So like 90 plus that actually peaked the earliest. So it actually peaked in 2020 and it was just pretty much entirely wearing off after that. And it peaked at a much lower relative excess factor of 114%. So 14% higher than before COVID.

So yeah, helpful with the methodology, being able to figure out these levels of excess mortality, but also the patterns as well.

JUSTIN FOUNTAIN: OK, cool. So face amount is an interesting proxy for socioeconomic status or class. It's usually a pretty important factor in most of the studies that we do. What was the impact of COVID by face amount? And maybe you can also speak to how it differed from the non-COVID years.

BOYANG MENG: Yes, so one of the other major areas where we did see differences over time or across our study period with the effect of COVID was face amount. And I guess a little bit of background on how it works. We tend to build a primary effect for each of these variables and then we'll layer on a secondary effect by calendar year on top of it. So for now I'll focus on our primary face amount effect, which we build out as an attained age pattern.

And so this primary effect is representative of the experience in 2018 and '19. And then for later years, we would basically just add an additional multiplier on top to scale this to what we observed differently in 2020 through 2022. But the main effect that is present in all years is a pretty clean and nice linear wear off pattern for most face amounts above 100,000.

So 100,000 surfaces are our baseline are 100% baseline that we're comparing everything to and the relative differentials, especially at younger ages, as we increase in face amount, the socioeconomic effect lowers mortality. And then over time, as people age higher, those effects wear off and grade linearly towards 100% the main exception being at the very high end.

So at 5 million plus, where we don't really have a ton of claims, but where we do have very high face amount policies, they have a significantly lower mortality multiple compared to all the other face amounts. And that effect stays constant. They stay below everything else. There's no real grading or narrowing of that differential.

JUSTIN FOUNTAIN: Got it. So continuing on with the COVID theme. Risk class is obviously a very important predictor of mortality just in general. What was the impact that we saw in TOAMS stratified by risk class, inclusive of COVID and prior to COVID period?

BOYANG MENG: Yes. So risk class is obviously a major influence on the way we rate life insurance policies are underwritten. And within TOAMS the way we structure our data, the way we think about risk class, we see a policy smoker status, the underwriting risk class structure, the number of levels available for that product when it was originally issued, as well as the actual class assigned to any individual policy. We see those three pieces of information as a single package, a single holistic group or cohort, if you will.

And because life insurance underwriting design has changed so much over time, those things tend to change together. And that's also one reason why we tend to focus on attained age. The use of attained age for our wear-off patterns rather than duration is because of the history element of how policies have changed. Today, we'll focus on just the simpler two class and aggregate, non-smoker and smoker policies.

So the first thing that we'll show is smokers, actually. So with smoker policies, we have a green line here for the residual standard, a blue line for the preferred, and then a purple for the aggregate smoker. So no preferred classes available, but still issued as identified as being a smoker policy. And what we see here is that the preferred class differentials do exhibit a nice symmetric and clean wear off pattern.

So the residual standard ends up being about 20% higher than the aggregate, preferred ends up being about 20% lower, and they both move pretty linearly and cleanly, so that by age 70 the differentials between those preferred classes and the aggregate smokers is almost entirely gone.

And now we'll switch over to the two class non-smokers. So with the two class non-smokers we do see a pretty different pattern than smokers, both in terms of being asymmetric. So our residual standard class has a bigger gap to aggregate. It's about 30% higher than aggregate compared to the preferred classes that are only about 22% lower than aggregate. So they start off in a different place and then the patterns end up being quite different.

So as you can see, our residual standard policies, they start wearing down almost immediately, even at the youngest attained ages and they come down pretty cleanly to about 100% as our data starts to get thin in the 90s. But our preferred risk classes, they stayed level for a while up until about 65. And it's only at that point where we start to see a difference, where we start to see their mortality start to come up.

It's a much shallower slope such that by 95 there is still 17% differential to aggregate nonsmokers. So that preferred discount does not completely wear off even at high attained ages. So it's still noticeably better than residual standard two class non-smoker as well as aggregate non-smoker.

JUSTIN FOUNTAIN: Very cool. Yeah, utilizing predictive analytics really does allow us to parse out some interesting findings from this data. Obviously, there's so much more that we could discuss on mortality and experience studies.

BOYANG MENG: Yeah, and we covered a lot more in TOAMS, things on product, underwriting, a lot of other different topics were covered as part of the mortality analysis and TOAMS. But we do have a lot of interesting findings on policyholder behavior that we'd like to share. And we'll have an article and video on that coming shortly as well. Thanks for joining.