Hacker News Reader

Comment by encoderer | Hacker News Reader

> Just to give you an idea of how bad that is, a three-sigma or a one-in-200-year catastrophe would be 10% increase over pre-pandemic,” he said. “So 40% is just unheard of.”

Do we have an actuarial crisis on our hands? We’ve seen the same thing in finance, disaster preparedness and other unrelated industries. We have “once in 200 years” events happening it seems far more often than once in 200 years.

And how does it even pass muster that a 10% increase in deaths would be a “once in 200 years” event. When has there ever been a 200 year period without major war, disease or disaster? What exactly are they smoking?

View on HN

I commented on this elsewhere, but I think your point is fair. Many real-world processes follow power laws, but they are modeled as normal distributions. The issue with that is that a power law can "look like" a normal distribution, except it has fat tails, where you get these "once until the heat death of the universe" type events much more frequently.

Black Swan by Nassim Taleb is an entire book about this very common mistake.

It's a case where highly-educated people often make errors that laypersons would not.

I think Taleb’s point was more that humans don’t have a good intuition for extreme events, layperson or expert. See the housing crisis during which experts and laypeople were equally fooled. We’re good at understanding eg height of people (small variation) but have no sense of wild scale, eg stock market movements or a huge bank imploding within days.

> See the housing crisis during which experts and laypeople were equally fooled.

Not discounting your main point, but there were plenty of experts and laypeople that were completely unsurprised by the housing crisis. If anything, I think the housing crisis was more of a case that so many people had a vested interest in thinking/pretending the music would never stop.

Totally, I’ll claim to have been very pessimistic before it happened. Referenced the likely downturn in an economics assignment that didn’t impress my very optimistic lecturer. I missed the bottom though, was convinced we were going down a lot further. Don’t fight the Fed, I guess.

Been a while since I read it, but I seem to remember a false sense of confidence (caused by belief in a normal distribution) being a major theme.

Regardless, we should be skeptical of the assumption that we're dealing with a normal distribution.

Yeah I think he used the word intuition. Exactly same insight though. I think “recency” is the bias that causes our view that tomorrow will be pretty similar to today. I’ve noticed it in myself, for sure. Market at X? Yup that seems rational. Hard to imagine the market at 2X or 1/2X. Bitcoin blew my mind during the ultra-growth phases.

Anecdotally, I see a lot of pseudo-rigor in Business oriented statistics. Leadership selects the statistics and statisticians which make the most rigourous plausible model that achieves the desired outcome.

You might enjoy this video:

Nonlinearity Breeds Contempt

https://www.youtube.com/watch?v=C6eX6KaSBjc&list=LL&index=32...

> And how does it even pass muster that a 10% increase in deaths would be a “once in 200 years” event. When has there ever been a 200 year period without major war, disease or disaster? What exactly are they smoking?

I don't know if it's a once in 200 years event or not, but this is a US insurance company, presumably talking about numbers in the US. Aside from ww2, in which the US had around 400,000 deaths, the number of Americans who have died in wars or disasters has been fairly small relatively.

Even the 400,000 for WW2 was over four years, whereas the US has had around 400,000 covid deaths a year for the last two years, and even more if you just look at excess mortality (however the population is obviously larger now than during ww2).

I'm guessing your comment is about people who died since WW2 but an estimated 750k died in the civil war.

More than half of the covid deaths were the elderly.

>We have “once in 200 years” events happening it seems far more often than once in 200 years.

I just don't take it seriously anymore. Track 200 factors, 200 businesses, or 200 industries and you will see 1 event per year. Track all of the above and you will see 40,000 such events in a year.

More to the point, their actuarial model was probably wrong, not reality. Somewhere buried in their model were assumptions excluding black swan events

You can pretty much blame six sigma for these braindead once in 200 years predictions. When you beat your effecency numbers into the ground then have to explain a .000001 varianc will only happen in 200 years based on this totally detached from real world statistic model is hew you get here.

The person that sold that idea cashed out and left the company long ago.

I think the more thought-provoking question is what exactly is the Black Swan event? If only 37% of their ICU beds are used for COVID patients then why are so many more people dying right now?

Three years ago, 0% of ICU beds were used for COVID patients. Now it is 37%. If ICU admission correlates with risk of death, and those beds would have been empty otherwise, or used for lower risk patients, a 40% increase in deaths seems reasonable. Certainly not a rigorous analysis, but it passes the smell test.

It is the US so unlike most of the rest of the world even if there were a lot of surplus ICU beds most people can't afford the 100K USD with insurance or 1M USD with under-insurance or no insurance it would cost them out of pocket to receive treatment.

> most people can't afford the 100K USD with insurance or 1M USD with under-insurance or no insurance it would cost them out of pocket to receive treatment

Where are you getting these numbers? I know multiple people in the US who have needed a COVID ICU bed, and they certainly weren't paying $100k out of pocket to do so. In most areas of the US, $100k could buy you a small house.

Homicides are up, In Atlanta it's 65% from 2019, other cities are similar. I think car fatalities as well. Then the isolation, COVID fears and economic issues have made worse an already growing drug and suicide problem.

I've known of quite a few deaths of youngish people recently and only one was COVID related. Car crashes, ODs and other medical issues. Like I've known two girls both about 30 that died unexpectedly of non-covid medical issues last year.

There is one possibility which this community will flag & downvote anyone for suggesting, which is shameful; the idea should be entertained and then dismissed, if appropriate.

You may be interested to see Kaiser Permanente's recent study, suggesting that more than one in 2,000 young men are experiencing untimely heart disease with a known cause: https://www.medrxiv.org/content/10.1101/2021.12.21.21268209v...

1) 37% of ICU beds for Covid means about a 50% increase in ICU patients.

2) If you go into the ICU for Covid you're probably leaving via the morgue. The survival chance for a Covid patient in the ICU is a lot lower than the typical ICU patient.

Ok, but the important ones are resource limited and possibly highly correlated. So if you see 1/200 for all sorts of (possibly correlated) events you might be misreading the implication.

Keep in mind that "once in two hundred years" is (an estimate of) the probability of a given event. The more events you track, the more "once in two hundred years" events you'll see in any given year. If you're tracking (or reporting on) many events...

Think of the number of hundred year floods you see reported on each year. Then look at the number of floodplains that your news source would report on...

I doubt there's an actuarial crisis brewing, for the simple reason that changes in life expectancy are a known risk which can be hedged. In the case of life insurance companies, the usual solution is to also sell life annuities; if insurance payouts go up, annuity payouts go down.

I'd guess war is usually excluded from insurance policies.

I wonder if the 1 in 200 year example wasn't communicated very well; it could be the actuaries view a long-lasting 10% increase to be the 1 in 200 year event; it does sound low for a single year stress.

I'm not sure about the case at hand here, but one reason for hearing about these events more often than once every 200 years might be because they are uncorrelated.

E.g. a "once in 200 years drought" in California might happen at the same time as a "once in 200 years over-mortality event" in Indiana.

News would tend to report on these, so we see many of these even though they are rare.

Yup. There are a *lot* of events to have a 1 in 200 chance of. Each is a separate chance, the odds add up.

> We have “once in 200 years” events happening it seems far more often than once in 200 years.

"Experts" making such a statement usually assume the variable in question to be standard distributed – which is close enough to the truth for non-extreme outcomes. But a lot of variables do not behave like the perfect standard distribution at all: due to non-linear effects and self-enforcing feedback loops, their distribution deviates more and more from the standard distribution the more extreme their values gets (they are "fat-tailed"). Referring to the n-sigma of such extreme events is plain and simple stupid, because there's no such thing as a standard deviation (or variance) for a fat-tailed distribution.

Many probability distributions have a variance[1], not only the standard distribution. Why do you think there is no variance for your "fat-tailed distribution"?

[1] https://en.wikipedia.org/wiki/Variance

We might. Could be a Black Swan event. Just the same, all along we should have been asking about "collateral damage." We ignored it. That's sadly suspicious.

Obviously, we don't have other pandemics' data to lean on. But we did have the economic crisis circa 2007 - 2008. There was plenty of analysis about the socio-economic impact of that event. That is, for example, poverty rate goes up, so does X, Y and Z.

If we can model a pandemic, can we not also - at least try - to mobel the impact of "the cure" and possible collateral damage?

Maybe it's similar to the well known "more likely to be hit by a meteorite than to win the lottery" style statements. Obviously that's completely wrong: i've never heard of anyone being hit by a meteorite, but people win the lottery every month.

I'm not saying that they're actively lying, but it seems some assumptions in these frequentist's numbers are just not correct.

They're not lying. It's just meteorite kills are very clumpy.

I'm not a lottery player so I can't give a good estimate on the number of lottery winners, but for the major jackpots I'll guess many per year.

AFIAK we have no documented meteorite deaths--but it looks like the destruction of Sodom (of biblical fame) was a meteorite. While obviously we have no death toll it's obviously a *lot* of years of major lottery wins. The atmosphere stops most of the stuff but when something's big enough to get through it makes quite a boom. (Chelyabinsk was half a megaton but was high enough up the blast only caused harm by throwing broken windows at people. Tunguska, however was a few megatons and got low enough to be a city-killer. All it blew up was forest for probably zero deaths, but had it fallen 7 hours earlier the world would be a different place because it would have wiped out Leningrad.)

AFAIK the Sodom story was pretty much debunked

https://retractionwatch.com/2021/10/01/criticism-engulfs-pap...

Interesting! Pretty obviously some sloppy science but I'm not sure that's a debunk. For something in the Bible to turn out to be based on real events with religious overtones added. Religions are always poaching from history, to find some truth to a story in a religious work is not surprising, nor is it a confirmation of the religious aspects of it.

https://astronomy.com/news/2020/05/death-from-above-7-unluck...

Just last year a woman had just gotten out of bed when a meteorite came through her ceiling and hit her pillow where her head had been.

I was actually aware of that story. But she wasn't hit, so it doesn't invalidate my statement.

There’s a confounding effect arising from the difficulty in comparing false negative rates between person-meteor-strike and person-lottery-jackpot. Whereas lottery-jackpot events are rarely misinterpreted as some prat dropping rubbish onto person, person-meteor-strike too often is.

A lot more people try to win the lottery than jump in front of meteorites.

> We have “once in 200 years” events happening it seems far more often than once in 200 years.

Past performance is not indicative of future results.

Reply on news.ycombinator.com