Sornette vs. Taleb Diametrically Opposite Approaches to Risk & Predictability
ฝัง
- เผยแพร่เมื่อ 23 พ.ค. 2014
- An ETH sponsored meeting and debate between Nassim Taleb and Didier Sornette. Edited to eliminate the parts of the conversation not involving either but otherwise comprehensive.
A "diplomatic debate" is defined by Taleb as a conversation in which one looks for synthesis (as opposed to one in which one is trying, as in war, to win the argument, as with political debates). Clearly this type of debate only works when the two parties have mutual respect for each other.
![[4K] TREASURE(트레저) “KING KONG” Band LIVE Concert 킹콩은 라이브를 찢어🦍 [it’s KPOP LIVE 잇츠라이브]](http://i.ytimg.com/vi/p8bLLOxPDD8/mqdefault.jpg)
When there are no cars on the street, Sornette still waits for the green light.
Taleb looks both ways, and jaywalks if there are no cars.
Once the light turns green, Sornette crosses the street.
Taleb looks both ways and crosses if it's safe.
Sornette gets hit by a speeding self-help guru in a Lamborghini.
Taleb lets the guru wheel past him.
This comment is the best summary of the debate I have read here.
Nice
That is not correct. Taleb is assuming the market is a wild beast, and on that basis his approach is 100% correct, to survive long term one has to be anti-fragile. But the financial system is not a market macroscopically, it is a market only in the micro, it is a simple public monopoly, the currency-issuers (banks and governments) control the traffic light switch. The problem is politicians and policy makers do not know this (hence make poor policy choices that create fragile markets), that's mostly because they've been infected with fraudulent neoclassical economics crap their whole life, including Taleb, (which sees no role for government or banks) and the result is they believe the government has no power to reduce private debt and thus stabilize, e.g., housing mortgage, markets in favour of the debtors rather than the rentier creditors, nor to regulate banks to prevent fraudulent loans. If you've got the traffic light switch and fail to use it you can easily create financial chaos.
Never mind about the relative merits of the two participants. This is a fantastic video to watch.
Taleb's sense of humour is infectious.
For those having difficulty - Sornette is an academic interested in making correct statements of facts, or perhaps correct (probabilistic) predictions. Nassim is interested in the optimal betting strategy. They are mostly talking past each other at cross purposes. For a scientist knowledge is the goal in itself. For Nassim knowledge is only a tool for making bets, and the only goal is to be ahead in your bets (in the long run). Note how at 25:40 Sornette completely conflates the concept of a prediction with that of a trading strategy.
You're missing the point of the entire talk...
Thank you for the insight :)
Both are practicioners and academics
No, that's not true. For Nassim knowledge is the goal in itself as well. Nassim is concerned with developing a philosophy of uncertainty, of how to deal with it and that's why betting systems are relevant to him. You got it backwards.
18.28 "It's not your opinion" "It's How you express your opinion" "The financial instrument or the strategy is vastly more important than if you are right or wrong"
Mathematician vs. Philosopher. This is hard to watch because no matter how hard the philosopher tries to inform the mathematician, the latter refuses to listen because he is convinced he can't be wrong with his maths.
Brilliant to watch post COVID-19, especially the part after 32:35. The urge to optimize everything (run no ICU "buffer", have JIT supply chains for your whole economy, outsource production of critical items etc.) leads to a fragile positioning even for "white swans". Modelling dynamics of systems is fine if you have skin in the game and want to play it but Taleb's approach feels more appropriate for policy and systemic risk-management.
have to side with Taleb on the financial markets. Once it is WIDELY known that such and such dynamics is a predictor, the phenomena is then dead. But for books, disease, and other more "real" items Sornette has great point.
yes. I see Sornette approach more like timing the market whereas Nassim's approach is identifying fragilities in the system and act in consequence to become "convex" and not suffer the consequences of the possibility to precisely mis-timing. Both approaches are similar because they are trying to identify fragilities in a system. Event though Sornette's approach is interesting and Nassim says it in the debate, Sornette makes the mistake of becoming blind with data not realizing that his approach could one day be wrong in the real life because of unexpected little changes like Nassim says and because once people know about Sornette's approach, it will no longer work. I prefer to apply Nassim's approach in real life and especially financial markets. You need to have something that works all the time
@@rickp6628 yes. Hedging techniques in financial markets for example often work as long as not everybody knows about it. It's the trick
Mises and Hayek show that mathematical models cannot be used to accurately predict human behavior or aggregates of human behavior. At most, a range prediction can be made.
I...agreed -- but only on trends that are near some kind of statistical mean -- not out on some kind of assumed skinny tail that may well be fat!
Were they familiar with machine learning?
...nor with AI. Unless someone discovers a probability distribution that is valid for the vagaries of human behavior, math models will be dependent on some form of randomness that applies only to purely physical, non-emotive processes -- which therefore have NO validity for human behavior, much less applied to financial risk, based allegedly on greed and fear.
I didn't watch most of this video but I've read the Black Swan and watched other lectures by him. He attacks a lot of issues with the bell curve/standard statistical procedures but not once did I come across anything related to the rise of AI/Machine Learning. I wish he would address this more thoroughly since it has and will continue to be used to predict human behavior in certain contexts.
@Miguel Bustamante ok. I guess there is no reason to point out the cognitive errors of an insane person.
The prediction becomes part of the predicted, yes. This dynamic is significant.
Dynamical systems and to some extent statical distributions stay still far behind signal processing in terms of early warning systems , just quick at power plots will you show a lot of things , so true question which needs to be answered for those who are in to profitable betting , how do you turn these early warning systems / features or properties of those dynamical systems or statistical distributions in to profitable trading decisions
Brilliant, brilliant debate ever!! I think doctor Sornette is probably someone who's able enough to debate Naseem! I really loved them both!
Fantastic discussion, helps clarify some things!
“So What are you now” 😂😂 Oh Nassim
My question is this: even if Sornette’s first hypothesis holds and dynamic systems can be used as an early detection system to warn that we are in a bubble, isn’t his second hypothesis (probability distributions for the end of bubbles) necessarily false? If Sornette goes on CNBC and tells us that there is a 75% chance of a market crashing in 6 months, I believe the entirety of the market’s reaction (and perhaps more reaction than what is supported by fundamentals) would be accelerated dramatically. Meaning the end of the bubble would necessarily come before Sornette’s “most likely” time band as a result of financial markets being driven by human agents with bounded rationality. Thoughts?
i think this is computational irreducible. the act of predicting provides information. The market is a function of information. The next step is to predict the effect of a prediction ad infinitum. As long as the market remains complex of course.
"you can use my graphs if you want" :)...
I need subtitles!
Those who do not speak French will not appreciate the irony of the situation, but "Sornette", is used in a derogatory way to call out bullshitters :-) Honestly, over the past two decades we have seen and heard about so many so called "dynamic" system predictors. But I am sorry, they just don't work in practice. I stand with Taleb on this one !
Of course 2008 was not a black swan - but for most people it was 99.9% turkeys
So much agreement between the two.
An infamous example is the way Allianz Alpha fund blew up in 2020. They had hedge puts coming to play only with a 50% drawdown on SnP500. And a 25% drop in SnP500 blew up the portfolio
"What are you now" 😏
Problem with Sornette's approach is even if it is 90% accurate in predicting crashes, it will eventually miss one and lose virtually everything. Nothing less than 100% predictability suffices, and the signal must be clear and timely and impossible to misinterpret; and he does not and cannot claim that.
So, back to Taleb, which is to manage your affairs properly so you are not at great risk.
Sornette reminds me of the geniuses behind LTCM, and I would like to see him put together a hedge fund using his math and see what happens.
Let me ask you this straight, "Whose hedge fund would you bet on?" Nassim or Sornette?
yes. I see Sornette approach more like timing the market whereas Nassim's approach is identifying fragilities in the system and act in consequence to become "convex" and not suffer the consequences of the possibility to precisely mis-timing. Both approaches are similar because they are trying to identify fragilities in a system. Event though Sornette's approach is interesting and Nassim says it in the debate, Sornette makes the mistake of becoming blind with data not realizing that his approach could one day be wrong in the real life because of unexpected little changes like Nassim says and because once people know about Sornette's approach, it will no longer work. I prefer to apply Nassim's approach in real life and especially financial markets. You need to have something that works all the time
@@hasnainabbas3442 you don't need to put any money in NNT fund. He is very clear that one should be 90% risk free investment and only 10% in risky investments knowing very well that everything will blow up. Only that 10% has to be diversified into many pockets and from time to time, one pays off extremely well. During the dot com bust also, if one was invested in the top 30 stocks of NASDAQ with 10% of investment at the time, only MSFT, AAPL would have stayed today invested, everything would be paid off despite so many drawdowns in between.
@@12345ngb So barbell strategy:)
It's funny because "sornette" means nonsense in french
hahah thats too good
Amazing that nobody made a joke about ¨chasing the (purple) dragon¨. Choosing a fictional creature as a 'mascot' is just a great omen.
I hope the wine was good enough... :-)
why are they sitting so close?
What is the title of the book that Sornette handed Taleb?
Why Stock Markets Crash
It's interesting!👌🙏
NNT is correct in principle. No matter how accurate one is with prediction of x, it's accuracy of prediction of f(x) determines whether performance in financial markets will be good or not.
Except that Taleb doesn't realy care about predictions i.e. probabilistic inference. In fact, he encourages doing away with predictive models and use heuristics instead. He domonstrated this in most of his books.
“To the wine!”
Sornette is out to lunch. He thinks systems reach maturation, but then illogically assumes one can know when this is. That would assume perfect knowledge.
Sornette is certainly more optimistic about seeing dynamical processes as they tend towards instability. But then he says in reality we might only find 'pockets of predictability' for these regimes. Sounds a bit like a married bachelor.
+gmshadowtraders exactly! His theory leaves a very sour taste; figures, he is French. What do married bachelors do? They get into affairs with the rather handy excuse- c'est realité. A lot of CEOs use this excuse when asked for a comment on the economic crash.
Scholars for dollars except Nassim
Oh boy this one is going to be a WAR lol. Watching it now.
sort of like JK Rowling vs Suzanne Collins LOL
which author can promote their own book better.
It's like a Monty Python sketch
39:25 ⭐️💫🌟✨⚡️
@23:30 "anti-fragility... interesting but too static"??? That made no sense. I think perhaps the way Taleb describes it dumbed-down for the audience seems static, but there's nothing inherently static about convexity. You want convexity over time, what's so hard about that to grok?
My friend, they are in ETH, where Einstein studied. The audience there is not dumb. There are better mathematicians in ETH than Nassim himself.
@23:00 phase transitions are controllable, quasi-static. So I did not get Sornette here. He is dead right about dynamics, but unpredictability arises from chaotic dynamics which is unpredictable, even without Thom catastrophes or phase transitions --- Taleb I think would be right here, even on the dynamics: to predict behaviour of a complex system (chaos + many bodies) you need to simply gather enormous amounts of data, truly enormous, even for a short-run predictive accuracy[^]. Sornette should have emphasised QUALITATIVE dynamics, not quantitative. Chaotic systems "at the edge of chaos" have highly regular patterns. But are not predictable. Many biological evolutionary processes are like this, easily discoverable patterns, total micro- and meso-scale unpredictability. Another example is the Minksy macroeconomics model (Steve Keen) whihc qualitatively predicted the GFC (absent government intervention) but such a macroeconomics models have no accuracy, they can only reveal patterns, not exact times for events, and only if monopolists do not intervene (eg., widespread debt cancellation for poor black mortgage holders around 2006 woudl've done the trick).
[^] At least on biosphere time scales. The solar system is also chaotic, but the characteristic time scales are so vast we have high predictive accuracy on our Earthly time scales. We live (by thousands of generations) well within the solar system's Lyapunov time.
Beautiful.
I think you're both right. One of you is saying, "don't go outside in a Hurricane" (approx) and the other, "when everything's calm in the middle of a Hurricane-- Get Inside quick!"
Grandma still knows best.
I'm remembering Hegel-- Perhaps we don't know we've slayed a Dragon King until after he's dead.
Just like, "maybe that wasn't the eye of the Hurricane. Ok."
"Stay inside" is the less risky proposition.
Non-reductionism vs reductionism
Taleb - everybody is stupid
Same as daniel kahneman
Perhaps The only way to make it is to be less stupid that others around us
@@Regular.Biceps Yes. Or even more precisely according to Taleb is to "be aware that we don`t know." That gives you an edge
volatility asmr
Polyglot of size/ lol 27:53
Chucking at 2:17 A.M
Sornette destroys his own position at 44:15.
Peter Schiff predicted the 2008 crisis and it may have value for these speakers (except Nassim) to analyze how he was able to do that.
I think there were many people who predicted the crisis including Taleb, Schiff and Michael Burry (I'm aware only of these)
Problem is, he's predicted dozens of 2008's before and after 2008.
Sornette is a ludic who has never really practically used his ideas (he hasn't taken any risk), whereas Taleb has (he was a trader). Sornette has no skin in the game (but he has many degrees). Its nice that the red dragon (or whatever) works in purely theoretical, after the event, type simulations. But, in reality (where it doesn't work) its useless, time is the ultimate detector of fragility. Models and predictions fail over time. Try to predict any black swan event with any fancy shmancy analysis you can throw together and you will see that, in practicality, randomness is not predictable.
Try to sound more brainwashed lol
Trying to impress people copy-pasting Taleb’s words and re-organizing them
yes. I see Sornette approach more like timing the market whereas Nassim's approach is identifying fragilities in the system and act in consequence to become "convex" and not suffer the consequences of the possibility to precisely mis-timing. Both approaches are similar because they are trying to identify fragilities in a system. Event though Sornette's approach is interesting and Nassim says it in the debate, Sornette makes the mistake of becoming blind with data not realizing that his approach could one day be wrong in the real life because of unexpected little changes like Nassim says and because once people know about Sornette's approach, it will no longer work. I prefer to apply Nassim's approach in real life and especially financial markets. You need to have something that works all the time
Sornette: this one thing is kinda predictable
Nassim be like “nah sHIt JuSt HaPpEnS bRo don’t even try”
If Sornette's "Early warning" would have been a valid approach, it would have been signalling ever since 1972. Buit it didn't and doesn't.
Heuristics vs mathematical models
Math models are "lecturing the bird how to fly" instead of just letting the bird to learn to be able to fly on its own after a number of trials i.e. heuristics.
One is a student of Mandelbrot, another of de Gennes. I think Didier won the debate.
Mandelbrots ideas can be found in both their ideas
Mises wins.
It seems Sornette has a nice dynamical model. He can see from the behaviour of the system in a small window of time, that some big event is due. It seems Taleb has some gut feelings that there might be even bigger big events, which are not that rare, that cannot be predicted at all.
The pro of Sornette's approach is that it is done scientifically. Certain mechanics is defined, hypothesis can be postulated, and it can be tested. The con of Taleb's approach is that he doesn't do anything like that. I've also read his book antifragility. It's void of anything of use to an engineer. There is no way to measure antifragility, even less how to construct an antifragile system.
From an optimization perspective the difference between convexity/concavity is nonsense. Minimization of a convex function is the same as maximization of a concave function.
All that being said, it's still the case of course that Taleb has interesting ideas. To position yourself in such way that you would be able to benefit from unpredictable events, the bigger, the better, is definitely interesting. I just wished he used a more normal academic approach to convey his ideas. His slides are typical...
Did you just say that there's no way to measure antifragility? Cuz there's a really easy way to do it--just find dose-response curves. Taleb even talks about it too. The proofs for what he's saying is easy too given that people have a sufficient mathematical background. Just do Monte Carlo simulations using dose-response curves by sampling from the initial random variable. He talks about this in his Antifragile book, although he deals with the more technical parts in the index of his book.
Kaysuv Boyina An MCMC simulation is not a proof. I know my share about regression. If antifragile is just the same as hormesis, it is nothing new. If antifragile is really the opposite of fragile: the more the box get shaken, the better it gets. Then it is something new, but I don't want that dose. :-)
Anyway, back to mathematics. Please, postulate a model of a system (might be an electronic circuit, might be a cybernetic system, might be something else) and tell me how to calculate its antifragility.
It's really simple. Take some variable in the dynamic system and simply measure the dose-response curve where the dose is something like the measurement error of the variable. With a financial firm, you could take a look at the shifts in a certain asset (or exposure to other firms) and calculate the losses with regard to the shifts of the asset price. That's what fragility is about. Antifragility is simply the exact opposite. In a physical system, like a bridge, you could have the damage to the bridge as a response to the wind speed in the area.
The problem with Sornette's view (which has many benefits BTW) is that as soon as the market finds out about it, it adapts. Another problem about the application of Sornette's approach has to do with how small measurement errors of the parameters can create large deviations because, as you said, these are dynamic system that are usually very chaotic. Small perturbations or shifts in the initial conditions can throw the results off dramatically.
I kindly object. Please, I would like to see a mathematical model. Pick a dynamical system, say a double pendulum, or a Henon map, and show me fragile and an antifragile variant.
I've yet to see a bridge that benefits from the wind. But apart from that, words give you the opportunity to be vague. Mathematics gives you the benefit of being crisp and clear. I would really appreciate a simple model that just shows the concept.
For now I see much more in the concept of "empowerment" by Klyubin et al. See www.prokopenko.net/empowerment.html. It has a thorough mathematical definition and its possible to design a system which maximizes it. It maximizes "optionality" and I think this is similar to what Taleb tries to achieve, but doesn't want to write down mathematically. If someone wants me to engineer a system that maximizes antifragility, they have to come up with the math, so I can write the code. No excuses. :-)
Anne van Rossum Nassim lays out antifragility with mathematical rigor here: papers.ssrn.com/sol3/papers.cfm?abstract_id=2392310
He makes the point that in most cases where probabilistic risk models are used, they underestimate tail event probabilities because they will rarely have enough data/examples of tail events and the systems are too open/dynamic/sensitive to assumptions. Instead of trying to assess 'risk'/probability of events, he recommends you test for fragility/angifragility - essentially whether your payout is convex(antifragile) or concave(fragile) to events in different directions. The point is not to predict when or if an event will happen, but to find systems where you benefit disproportionately vs your losses as random events accumulate.
Sornette WIN !
I love the contrast between the Sornette and Taleb commentators. LOL