A study of insurer death-claims during covid first-wave
Data suggests that 9MFY21 death-claim growth is in line with or even below trend of prior six years. LIC and private-insurers show divergent trends, though.
“It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts.” – Sherlock Holmes
Summary
I analyze death-claim data across nine life-insurers, accounting for 95% of India’s insurance death-claims. My aim is to compare trends between covid wave-1 period (9-months of FY21) and prior years. Data suggests that 9MFY21 death-claim growth (~4%) is in line with or even slightly below trend of prior six years (6-7%). LIC and private-insurers show divergent trends, though.
Long-form version
Anecdotes make for shoddy analysis and pointless debates. As do small, biased samples. Reliable extrapolation from a story to a country is impossible. In my day-job and otherwise, I try to find relatively broad-based, representative data where possible. Since India’s all-cause mortality data is patchy, I attempted to find a surrogate that, though imperfect, beats anecdotes.
Methodology
10 million Indians die each year. 2 million death-claims are paid out by Indian life-insurers each year. At 20% of relevant universe, this is as large as a sample can get. It’s possible claims and deaths don’t match 1:1 due to a person having multiple policies. Still, no other sample comes close in scale. IRDAI requires and collates quarterly and annual filings from every life insurer with data on claims paid. I analyze this data across FY14-FY20 and 9-months of FY21 (9MFY21) to see how death-claims trended during India’s covid first wave. Note that I only use number of claims and not claim amount (value of life uncorrelated with value of insurance purchased). Also, wave-1 peaked in September-2020 and ~80% of wave-1 deaths occurred by end-October. So, data till December-2020 can shed some light on wave-1. I’d have ideally liked to wait another quarter to work with full FY21 numbers. However, regulatory filings have been delayed due to ongoing covid/lockdown and it’s unclear how long I’ll have to wait. Since I had to manually peruse quarterly filings, I limited myself to top-9 life insurers (8 private + LIC) who account for 95% of all death-claims over these 7.75 years.
Base-line (with one adjustment)
Covid-first-wave data (9MFY21) needs to be framed against a prior trend to make meaningful inferences. My baseline is as-reported FY14-20 death-claim data, with one adjustment.
But, FY20 data was distorted by year-end lockdown that messed with claim filing, approval and payment. This resulted in a highly depressed Q4 compared to trend till then.
Depressed Q4 reduced FY20 growth rate to 8%, down from 14% over 9MFY20. We’ve seen year-end lockdown distortion across nearly all companies. In this case, deaths weren’t impacted (covid wasn’t a factor in last week of March-2020) and life insurance had been purchased beforehand. Only claim filing, processing and payments were impacted. It’s reasonable to assume that nearly all of these claim payments simply got deferred from end-FY20 to FY21. Counting FY20 deaths/claims under FY21 runs counter to our ultimate goal of reliably comparing deaths across periods. So, I explicitly adjust for this.
Had trend of first 3-quarters continued into 4th quarter, extra 95,000 claims would’ve been paid out in FY20. In subsequent data that I present, I shift 95,000 claims from 9MFY21 to FY20. My adjustment restores FY20 growth to pre-lockdown 14% and reduces growth during 9MFY21 from as-reported 11% to 4%. If any reader wants to play around with scenarios, you’re welcome to shift varying fractions of this amount or better still, view my numbers are directional not precise. Like-to-like FY20 growth was in low-double-digits and like-to-like 9MFY21 growth was in mid-single-digits. The crucial methodological point is that not making any adjustment is more erroneous than making an imprecise, but reasonable, one.
So, how did death-claims from covid first-wave trend?
Here’s the picture of how FY21 compared to baseline:
Death-claim growth of ~4% in April-December 2020 is slightly below trend growth of 6-7% over previous 6 years and comfortably within normal range of fluctuations over this period. For folks who are cautious or inclined to make a smaller lockdown-adjustment, a safe conclusion is “mid-single-digit growth, consistent with prior trend, in 9MFY21”. Note that life insurance industry (and death-claims) have grown well above India’s ~1% death cagr due to increasing penetration.
What do we make of this?
Countries where covid was highly lethal showed massive deviation in 2020 from historic trend in all-cause-mortality. In Western countries, historic growth in all-cause-mortality has been near-zero or low-single-digit percentage. In 2020, this rose to between 8% and 18% (except in a few Nordic countries). Delta was dramatic. While I have used an imperfect surrogate, it doesn’t show anything close to a massive deviation from trend. Even if I had gone with an unadjusted 11% growth in 9MFY21 (which is plain wrong) or a smaller adjustment, it’s within the band of historic variation.
This is consistent with 2020 all-cause-mortality data of a few states. Media reports indicated lower deaths in 2020 than 2019 in Kerala and Gujarat (separately, Mumbai showed higher deaths in 2020). Death-claim trend is consistent with our understanding of lower impact of covid on all-cause mortality in India compared to many other countries.
What are the problems with using death-claims?
While 20 million claims/year is significant, sample doesn’t perfectly mirror population. Sample could under-represent the poor, although there may be some offset due to large contribution of micro-insurance in private insurer books. Group life insurance policies which are typically purchased by an employer may overweight organized sector in death-claim data. That said, I haven’t directly compared death-claims to mortality. I only compare death-claims to its own history. Biases are common to covid-period as well as baseline. Trend shown is based on a like-to-like comparison, although there’s no guarantee it mirrors trend in mortality. I wouldn’t overstate precision or implications of above analysis. I merely tried to pick a reasonable surrogate and see if trend fell under “Whoa” or “Meh” (sorry to use technical language).
Also, there’s some risk that trend may change once Q4FY21 data is incorporated. While 9MFY21 covers the bulk of wave-1 fatalities, longer-than-expected lead-time in processing death-claims may imply that I haven’t captured its full effect.
Divergent trend between LIC and private insurers
While I showed (comforting) aggregate death-claim trend across 95% of industry, there’s a puzzling divergence in trend between LIC and private insurers.
By way of background, LIC dominates individual-life policy segment with 87% share. However, this segment and LIC show low historical growth. Private insurers dominate fast-growing group-life policy segment with ~80% share. So, LIC is more of a ‘retail’ business with a diverse customer base. Private insurers have more of an ‘institutional’ business. However, since micro-insurance seems to be the largest segment within group policies, there’s an underlying retail element to it. LIC and private-insurer client-base could have different demographic or age profiles.
If I create two separate charts for LIC and 8-private-insurers, using same methodology/adjustment, here’s what it looks like. FY21 is above-trend for LIC and below-trend for private-insurers.
Before commenting on LIC’s above-trend FY21, it’s worth looking at its absolute numbers (for graphic convenience, I project FY21 using 9MFY21 growth rate).
First, I don’t know why this divergence. My theorizing will only confirm that theory is a fancy euphemism for guess. That said, a few comments on LIC data. At 11 million (adjusted) claims in FY21, LIC’s 8% above its own FY14-18 average and 5% above its prior FY17 peak. Further, FY19 and FY20 are below-trend years and deaths/claims data should be mean-reverting. Typically, excess-mortality data is presented after adjusting for population growth (1% a year), while death-claim data shown here is unadjusted. Framed against this context, LIC’s deviation from trend is less stark that its 16% rise in 9MFY21 indicates. But, it’s certainly above trend.
Also, individual company data (within my 8 private-insurer sample) is all over the place. Widely disparate growth rates, year-wise swings and magnitude of deviation from trend in FY21, although 7 of 8 showed below-trend growth in FY21. At a single-company level, idiosyncratic factors prevail. While I separated out LIC due to its relative size and differing segment-focus, it’s still a single company where company-specific factors play a larger role than in aggregate multi-company data.
A philosophical digression, before closing
My goal isn’t to provide an answer. It is to attempt a better approach to an important question (how did aggregate deaths trend in a pandemic year) that cannot yet be directly answered. I’d like you to focus on process rather than on inference. I’d like to highlight a few process aspects.
· Sizeable, pan-India dataset directly linked to deaths
· Robust, audited, regulatorily-filed, exhaustive data
· Decadal data, facilitating time-series comparison and contrast of year of interest to trend
· Careful data extraction, going quarter-wise when required
· Transparent adjustments, where required for meaningful comparison
· Show entire distribution, not averages
· Show inconvenient sub-trends, not merely comforting aggregate trend
· Strike balance between drawing cautious inferences and not stretching it too far
· Ability to repeat the same exercise at future points (refine wave-1 post FY21 and attempt wave-2 post FY22)
I have zero domain expertise in either life insurance or data science. Apart from curiosity and comfort with company filings, I have no relevant skills. If I can make a reasonable attempt despite shortcomings, I hope better qualified folks will do more. Do refine or correct what I’ve done. Find better data sources. Overlay qualitative insights if you have industry expertise. And please publish your work, so that we can discuss off a more informed starting point.
Conclusion
I analyze death-claim data across nine life-insurers, accounting for 95% of India’s insurance death-claims. My aim is to compare trends between covid wave-1 period (9-months of FY21) and prior years. Data suggests that 9MFY21 death-claim growth (~4%) is in line with or even slightly below trend of prior six years (6-7%). LIC and private-insurers show divergent trends, though.
Sources
Start from IRDAI website
https://www.irdai.gov.in/ADMINCMS/cms/NormalData_Layout.aspx?page=PageNo764&mid=31.1
Click links to respective life insurer websites and go through quarterly/annual statutory filings. Apart from LIC, I downloaded company-wise data for Aditya Birla, Bajaj, HDFC, ICICI, Kotak, Max, Pramerica & SBI.
Annual data is more easily downloaded from IRDAI’s statistical handbook available at
https://www.irdai.gov.in/ADMINCMS/cms/frmGeneral_List.aspx?DF=Creport&mid=11.2
Other links:
https://health.kerala.gov.in/pdf/Technical-paper-All-Cause-Mortality-Kerala.pdf
https://www.cdc.gov/mmwr/volumes/70/wr/mm7014e1.htm
I’ll send you XL on request. Not going to format it for you, though!
A study of insurer death-claims during covid first-wave
Great analysis Anand. This must have been a lot of work and your point about evaluating data before coming up with theories is an excellent one.
One caveat - it is important to wait for the full year (and perhaps even QI FY22) data as life insurance claims tend to come with a lag of 6m. Our data suggests that there was a tick-up in overall mortality due to Covid in FY21. It was not as much as one might have initially expected. Regretfully, the current wave is more serious and will lead to much higher mortality. The key question now is whether this should be treated as a one-off epidemic-related event or will those who had Covid have lower life expectancy in general? Of course, the actual answer will take time to emerge.
This is really good analysis! very often noise and narrative take front cover while truth gets hidden..