Hacker News new | past | comments | ask | show | jobs | submit login
Statistical Consequences of Fat Tails by Nassim Nicholas Taleb (arxiv.org)
89 points by ArtWomb 27 days ago | hide | past | web | favorite | 31 comments

While Taleb himself is kind of an asshole an endearing thing about him (to me at least) is that he pisses off bigger assholes and doesn't pull his punches when calling them out on their quackery. See e.g. the whole IQ debacle. Regardless of what you may think about him, his problematic views and his personality, call him abrasive, inflammatory, full of himself, whatever, but he's not ignorant, he's not a charlatan, he's not pretending to be someone that he's not.

In a way NNT reflects the original hacker ethos in all of its pristine and glorious ugliness.

A few things:

-It's basically a HBD blog. HBD proponents love to pretend they're doing "science" by posting huge walls of texts that masquerade as papers but would be thrown out of any respectable peer-reviewed popgen journal. (popgen, not psychology.). Here's an example of titles on this blog: "The Catholic Church and Western Genetics", "Racial Ancestry and IQ", "Population Differences in IQ-Related Genes", "Expert Surveys on Race and IQ". Note that at no point in any of those posts, any actual geneticist is mentioned, only "intelligence researchers". Many of these, like Lynn, Rushton and Kanazawa were found to be guilty of conducting absolute trash science or outright fraud (https://medium.com/@evopsychgoogle/a-critique-of-rushton-and..., https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3332228/, http://risk-resilience.berkeley.edu/sites/default/files/jour...). Many of these researchers also get funding from very shady sources like the Pioneer Fund. This alone should send all kinds of warnings about the motivated agenda of the people putting forward arguments here.

-It does nothing to address any of the points raised by Taleb, such as nonlinearity and convexity. This sentence right here shows how mathematically illiterate the author is: >I’ll say more about this below, but here note in passing that Taleb never explains why a non-linear trend would invalidate IQ in the first place. The author also doesn't seem to understand that covariance between a fat-tailed distribution and a gaussian one isn't well-defined in the first place and as such measuring samples' covariance makes no sense. You don't need "a study" to prove it, it's just math.

-It basically parrots the same points over and over again, distorting the word "predict" to the point of meaninglessness: IQ predicts this, IQ predicts that. Well of course when we live in a society where many of your life outcomes are somehow tied to your ability to sit down and take a pen and paper test, IQ is going to correlate well with these! "An artifact number that drives a bunch of correlates" is not what normal people (read, people outside the Pioneer Fund buddies) mean by intelligence! In fact, nobody really agrees what is meant by intelligence, which is why AI isn't a thing and likely won't be within our lifetimes.

-You're basically a one-reply account, who posted just so that challenges to your debunked notion of IQ wouldn't get the last word. Again, let me doubt good faith is involved here.

The problem with Taleb is that he's incapable of recognizing any error, and he's not smart enough to do novel intellectual work. I recall pointing out some trivial mistake in the stat reasoning on some Facebook post of his, in a comment section full of fawning adulation by sycophants, and was immediately banned.

What really draws people to him is that controversial overconfidence, mostly from folks who don't understand the math anyway but are drawn to the personality. It's the standard confidence con frankly not much more sophisticated than Alex Jones. He mostly got lucky publishing a book about market crashes right as the market crashed, that's pretty much the extent of what his fans understand about the matter.

When evaluating Taleb’s work you have to force yourself into a mindset of ignoring the messenger and focusing on the message. Load the content into your right-brain analytical mind and then firewall it off from your other reactions. Don’t let his personality induce bias into your analysis and judgement of his work.

Maybe you’ll ultimately find value in it, or maybe not. But you won’t know for sure if you’re spending most of your CPU cycles on hating the messenger.

It’s annoying to have to do, but a useful skill to work out and keep fit from time to time, not just for dealing with Taleb but plenty of other irksome sources in life.

Have to do this with Wolfram too. There’s a good mind and a good man behind all the puffery.

Is there? What is it, exactly? Could you show us? All I see is an Etonian, a work-stealer and plagiarist, a slapper of his own name onto things built by groups of people. His lone interesting contribution, the idea of cellular automata for quantum theories, failed Bell's requirements and had to be rescued by others.

This weird guy despises university professors while being one himself. Nevertheless, he has something to say and, personally, I have learned a few things from his books. Recommended.

That doesn’t seem tremendously uncommon from my college experience. There’s definite inter-department rivalry and even within the department you get one or two savants who kind of does their own thing.

Thanks for the link. Just ‘flipped’ through the first few pages and my blood pressure is already way up.. the guy just can’t help himself antagonize his readers. It’s going to be a tough read..

Any discussion about Taleb always leads to questions about why he picks such public fights. I've said this before:

The most salient anecdote Taleb tells about himself is the one where he's going onstage to debate an opponent. He asks his publisher if punching the other guy in the face would be against his contract, and his publisher notes that it would be very good for book sales.

All Taleb does is punch people in the face nowadays. It's probably very good for book sales. He notes elsewhere, rightly I think, that the goal of anybody seeking PR should be to get the attention of somebody more famous then them: since it's much easier to pick fights than make friendships, and either will do, he picks fights with anybody he thinks has prominence. I believe it's a persona, and he's a very good method actor.

He is a method actor. I’ve seen him remarkably civil in small groups when that’s what he’s incented or inclined to be.

I can confirm that as well. I’ve been at the same small table with him for lunch and he is very civil and thoughtful.

Reading Antifragile always felt like he's got some interesting insights, some of which feel quite quackery ... and the constant pot shots are unnecessary to the point of distraction.

We get it, you don't like academics and The Establishment. But do you have to mention it every other paragraph like a teenage boy?

Guess nobody gave him the Kill Your Darlings writing advice.

I just bought Antifragile and can't believe this guy gets away with the constant petty swipes. There are enough interesting ideas to keep you going but there is so much petty noise in these books. If he has an editor they are doing a terrible job.

Yes, his prose is terrible. Initially I thought he never took any english composition class. But then I realized that perhaps he thinks he is too smart for such things. That is, he writes the way he think, the way he speaks.

I'm pretty sure he discusses editors at some point and he doesn't really like them or have one.

He takes the swipes, but he'll show up with a truckload of evidence supporting them if you push back.

I think Black Swan is much, much better.

Thanks for the warning. I'll go with my normal approach to non-entertaining non-fiction and just read a couple of online summaries.

E.g. https://www.nateliason.com/notes/antifragile

You can't cheat with reading. Whatever you think of NNT he carefully picks metaphors to simplify underlying fabric. When I read a book it feels like a conversation with the writer, it takes me through a new door and unlocks another paradigm. Try to read the whole Skin in the Game book. It's worth the investment. Summaries are like a hand job it doesn't compare with the real deal.

I'd give you multiple points for the handjob metaphor of I could.

> and the constant pot shots are unnecessary to the point of distraction.

For each their own I suppose. I found it unnecessary and also entertaining.

Oh they’re entertaining alright. They’re just also unnecessary and get old after the first 50 :D

I very much love and hate NNT. I think my biggest gripe about him is most of his books and podcats are kind of the same story over and over again: fat tail/Poisson distribution realizations. I do find him interesting most of the time but I wish he would also cover more things. And sorry this is a generalization, I know not everything he puts out is this.

He has some legitimately very interesting ideas, but yeah there's more than a whiff of man-with-a-hammer syndrome. Understanding the tails of distributions made him rich, so obviously they're the most important mathematical construct ever!

I've not regretted reading some of his books, but I strongly advise anyone who wants to maintain some respect for him to avoid his Twitter feed at all costs. And I'm one of those highly analytical types that isn't overly burdened by emotional sensitivity. The man really is just that grating in real time. I get the impression that in his books he does some rewriting to limit his more egregious impulses.

The book "Dynamic Hedging" stands apart from the others with none of the typical NNT personality. Instead, it reads more like a math book with amazing theorems such as predicting the exact order that various kinds of hedgers will get wiped-out in a volatile market.

My guess the reason he is not moving on to different topics is that there thousands of people with official credentials who completely misunderstand statistics, and fat tails in particular.

The real reason that fat-tails aren't widely used isn't ignorance, it's because they are tricky to model and work with and have there own unexpected results.

While assuming normality will very often lead you to underestimate risk, fat tails can easily cause you to over estimate your risk.

Or think about the case of "robust" regression in which errors in a linear model are assumed to be from a fat tailed distribution like Student's T. This means that your model is not suprised by outliers. The consequence of this is that your model assumes that mode data is near the center and ignores extreme observations.

As a philosophical device fat tails are very helpful and interesting, but as a practical tool for modeling they are far from a panacea

So, you got to Section 2.2.15?

He is so unnecessarily pedantic it comes out as great insecurity. He is so difficult to read. I would love it if a great copy editor just rewrote his work because the ideas are important. You could probably cut it half and make it accessible to more people. A shame.

It's not supposed to be accessible to everyone. I think he says as much in one of the books. He's not interested in being liked or broadly read, so much as saying what he wants to say.

Applications are open for YC Summer 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact