Mandelbrot's fat tails, Taleb's black swans. Why haven't we learned anything from them?
I'm putting this in trading but it might as well go in any other section.
Taleb/Mandelbrot's criticism: the overwhelming majority of economic and/or financial models or research papers are based on normal distributions, represented by the Gaussian bell curve. Thus the focus is on the ''typical'' observation, the outlier is ignored by models. However in reality outliers exist and some of them have gigantic impacts, a game-changing effect on reality. However unlikely they are, they happen and you can't predict them. Thus the ''empirical'' or ''scientific'' validity of financial/economic models using the Gaussian is void.
By post-2007-8 basically everyone has read those books and accepted the explanation, it took a couple of Noble prizes to be wiped out of the market (twice!) and countless public humiliations. Yet when I look at methodology..... nothing has changed. Economists still (ab)use normal distributions. When I was in university studying models (finished in 2012), intuition would tell me what I was being given as ''truth'' did not reflect reality, but I couldn't put it into words and it wasn't until Taleb that I realized what was wrong. Same goes for any other discipline that originates from social studies yet wants to give itself a lick of scientific credibility and thus adds some statistical ''evidence''. Even worse, evidence of the contrary, or outliers are dismissed as ''anecdotal''. Finance from my (relatively small) competence hasn't learned much either.
So, uhm what the fuck? To put it into Taleb's words as he frenquently chastises intellectuals for being ''frauds'', is much of modern education, based on the abuse of the Gaussian, a fraud?
We haven't learned shit. We're still worried about quarter to quarter returns. doing any out of the money option and essentially losing premi costs is too dangerous for any risk team to approve.
I'm with you though.
Stop listening to Taleb would be my advice.,..
Why?
Because this idea that NNT is so fond of pushing - namely, that he's the only person in the universe and particularly in academia, who appreciates and understands the various "distribution model issues" that arise in economics and finance - is utter bollox. There's a lot of work that people have done and continue to do in all sorts of areas of mathematics, economics and cross-discipline, but you just don't get to hear about most of it. Taleb happens to be loud and obnoxious, which drowns out a lot of the more interesting reasonable voices.
Taleb might be a dick (he is, I don't think he even denies that), however the abuse of the Gaussian in the academia is true, especially in non-pure maths degrees the way statistics is taught is ''ignore the outliers''.
It's not even his argument, but it's Mandelbrot's, Taleb just embellished it with non-financial examples. I'm not saying there aren't studies,I'm questioning how it is applied. Normalization of data is presented as necessary and ''more accurate'' representation of reality, when it's actually the opposite. Many papers also stop there, they do the normal, forget the outliers, present an argument as ''truth''. If you don't analyze the outliers and weigh their impact, then it's bullshit and Taleb/Mandelbrot are right.
It's also more about the widespreadness of non-Gaussian analysis and quite frankly I haven't really seen that many papers that go outside the ''normal'' methodology. Granted I didn't develop a keen eye on the topic until recently so it might be cognitive bias.
I think there are a few relevant issues here...
Firstly, everything that happens in academia is going to be based on simplified models of the world, by construction. I am not sure if anyone, other than the really die-hard nutcases, would be naive enough to believe that these simplified models are flawless. Most academic papers that I've read are also quite careful about pointing out areas where model flaws and dependencies might affect conclusions. Neither do I find the lack of focus on outliers surprising or all that unreasonable, tbh. Aren't we all primarily interested in what happens most of the time, rather than all the possible things that can happen? NNT contends that this blinds everyone (other than himself) to the possibility of outliers, but I just don't see any evidence of this.
Secondly, a lot of the academia is about education, rather than research. Isn't it perfectly natural to expect that, when teaching, you'd want to focus on simplified models of relevant phenomena?
Needless to say, I have no beef with any critique such as Mandelbrot's, which has been phrased in a pretty rigorous and scientific fashion. I just intensely dislike NNT's rants on the subject, since I find them misdirected, disingenuous and highly counter-productive.
If you're going to see a 6 sigma day you're going to be fucked no matter the distribution you're choosing from. The cost of hedging that 6 sigma risk is obscene, to the point of giving up a huge portion of day-to-day activities. I find that better risk control and returns comes from limiting maximum loss, not trying to hedge unlikely (but devastating) events.
I lost you when you said economists.
economists are not PMs, don't confuse the two.
I take everyone's opinion with a grain of salt, and I rarely put much stock in people who have no capital riding on their predictions. economists desperately want financial markets to operate in a controlled environment like a physics lab, but when you have free thinking actors in it, it doesn't work. since you can't accurately model human behavior, their job is a crapshoot, so they rely on normal distributions to justify their worth.
Models are models are models are models. Literally any continuous probability distribution has a tail. It's unavoidable. It's much more dangerous to overfit your model than it is to accept that there is always tail risk
All models are wrong, but some are useful-- Too many people seem to forget this.
Love the people shitting on NNT as if anyone is saying his word is gospel.
The core of the problem is humans are dumb, we do not learn from history and we do not plan for unseen, black swans.
NNT is the main voice in this message. That is what matters. Debating on the merits of other people's analysis is worthless. They aren't sounding the alarm.
This is a very poor and misleading interpretation of the various bits of criticism that people, including myself, have directed at NNT.
Furthermore, if you think that NNT is the main voice in the "message" that you have described, you clearly haven't bothered to listen to many other people (who, IMHO, have much more useful things to say). That illustrates precisely one of the issues I have with NNT.
Also, Taleb mixes statistics and analysis with a historical outlook. His reading and writing transcends normal statistics writing.
I think all you have to do is look at the sub-par performance of his (former) hedge fund... I don't think theta was very friendly to him...
I think you answer your own question when you say "however unlikely they are, they happen and you can't predict them."
If you can create a model that predicts some type of behavior better than any other model out there and with 99.99% accuracy would you not use it? Of course you would and since you can't predict when something is going to happen to make the model irrelevant you have no choice but to use it and hope for the best. If you didn't use it because of the .01% chance it might not be the day it works then there would be nothing to ever go off of and decisions could never be made because of that lurking chance something different happens.
And obviously I cannot speak for everyone, but at all levels of my education any model regarding economics behavior was paired with the sentiment that their are events we do not know how to predict and these models do not account for that, BUT these models have been shown to be true the vast majority of the time and are the best thing we have created.
It is the same idea as driving your car to work every morning. You know that getting in, putting your seat belt on, and driving on the right side of the road to work will get you there safe and alive pretty much every time. Say something happens though and a drunk driver comes barreling down the street the wrong way and hits you head on. In that case, people don't sit around and ask how you could be so dumb to drive to work everyday when you knew that something like that might happen to you.
The possibility of Long Term Capital Managemment model blowing up was 1 to 10^52. A lot less than 0.01% and it happened, twice, within a decade.
The point isn't not using a model that is unlikely to blow up, it's to prepare for that eventuality regardless of whether it'll happen or not.
I have a good friend who's a great poker player. However he often makes the mistake of confusing improbable with impossible. He goes all in at the river with a 93-95% of winning, that relatively unlikely card arrives and he's out of the tournament.
I think you're using the wrong model to assess LTCM's probability of failure. It was inevitable. The Black & Scholes Models failed because they lacked historical depth to back-test under extreme conditions. The main underlying assumption was that implied volatility on the date of the analysis will remain unchanged until expiration. That has never been the case.
Tournament Poker is different than cash games. Sometimes it is correct to take large risks. Your chance of winning increases when your roll (big blinds) is relatively greater than the average opponent. Phil Ivey is notorious for pressing small edges early in tournaments (the utility of getting back his buyin is trivial, while utility of final tabling is high).
Ipsam quam et iusto rem aut accusamus voluptas. A nihil aperiam aut aut. In est ut sed.
Doloremque nihil ut et ut culpa nesciunt. Repudiandae ratione aperiam labore dolores atque. Ut minima porro perspiciatis amet ab. Quos aut ea enim ad pariatur accusamus aliquam.
Voluptatem officiis quae quia omnis ut. Eum expedita dignissimos ea.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...