Taleb’s data dump

By Felix Salmon
April 16, 2009

Back in March, Rick Bookstaber published a blog entry entitled “The Fat-Tailed Straw Man”, which seems to imply that Nassim Taleb attacks Wall Street practices on the grounds that its practitioners believe in normal distributions.

The result of that blog entry, and of a few other criticisms of Taleb in various places, is that Nassim has now published an extremely useful technical appendix to The Black Swan. It includes substantially all his relevant scientific and technical papers, and essentially comprises a gauntlet being thrown down to those who would criticize him. If you want to attack my ideas, he’s saying, that’s fine, but please first do me the favor of looking at where they’re laid out in detail, as opposed to where they’re laid out in newspaper quotes or in a literary book.

There is certainly a lot of confusion in the public mind when it comes to philosophical arguments about fat tails and risk management, and it’s easy for journalists to distill things into easy soundbites about normal distributions and the like. I, for one, would like at some point to write about Taleb’s important paper on Errors, Robustness, and the Fourth Quadrant, which looks at the history of thousands of data series going back decades — but I haven’t found a very clear way of trying to explain its details in English.

The Black Swan has proved to be a very popular book largely because it doesn’t even attempt to do that: instead it talks philosophically about the conclusions one draws after looking at the data. But if you want to attack Taleb for drawing the wrong conclusions, I think he’s right: you shouldn’t attack the book, but rather the empirical research which underlies it.

10 comments

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see http://blogs.reuters.com/fulldisclosure/2010/09/27/toward-a-more-thoughtful-conversation-on-stories/

Why don’t I attempt to read everything ever written before I criticize a generalization or conclusion made by an author? Because I do not have time. And because the author is also creating a short-cut when they sum-up very complicated ideas – or large data sets – into easy bite-sized summaries. It is perfectly valid to then say that summary is a generalization or is in fact too narrow.

Posted by MrBill, Eurasia | Report as abusive

Sell side and buy side still to reluctantly accept normal distributions. It’s because there is nothing in most systems — from Excel to in house — that handle other distributions well. When you need an answer that works “most of the time” lognormal is accepted as fine, w/ the caveat that “well, we’re assuming the returns are lognormal.”

Taleb, like Roubini, is a pontificating pedant. But that doesn’t mean they’re wrong. They’re just selling books/themselves too.

“There are three kinds of lies: lies, damned lies and statistics”

-Attributed to Benjamin Disraeli (1804-81), British statesman and Prime Minister (1868, 1874-80), in:

Mark Twain (Samuel Langhorne Clemens), U.S. writer and humorist (1835-1910), Autobiography, “Notes on Innocents Abroad”

Posted by S. Hellinger | Report as abusive

Felix, There is an “English” version that Taleb wrote for the EDGE:

THE FOURTH QUADRANT: A MAP OF THE LIMITS OF STATISTICS
http://www.edge.org/3rd_culture/taleb08/ taleb08_index.html

Posted by Matthew | Report as abusive

But Taleb DOES attack Wall Street practices on the grounds that its practitioners believe in normal distributions. And that’s stupid, because everyone I’ve ever talked to in Wall Street or academia (who he attacks for the same oversight) knows this. Fat tails in finance were discussed in my intro econometrics textbook – it’s not a secret. Normal distributions are often assumed for a variety of reasons, but everyone knows that it’s not accurate. For example, many option markets quote (normal) volatility numbers when making trades because it’s a more stable convention than quoting the exact price (twenty minutes later when the underlying asset moves in price the implied volatility on the option won’t have changed much). But all of the firms’ internal models use much more complicated measures of volatility that take into account fat tails.

And the other general point, which is that unexpected things will happen that the models haven’t taken into account, is equally useless as Bookstaber points out. As he says, that’s exactly as helpful as saying “Be Careful!” It’s nice to remind people that you need to be careful, but that doesn’t make you some kind of prophet.

But yeah, why should people respond to his academic work instead of responding to his several hundred page book? I don’t think the problem is that his academic work is bad (although I haven’t read it). The problem is that his advice is useless. And if he can’t give any useful advice in his books, than he needs to think a bit harder.

Posted by Kyle | Report as abusive

[i]And the other general point, which is that unexpected things will happen that the models haven’t taken into account, is equally useless as Bookstaber points out. As he says, that’s exactly as helpful as saying “Be Careful!” It’s nice to remind people that you need to be careful, but that doesn’t make you some kind of prophet.[/i]

This is a ludicrous generalization- a kind of reductio ad absurdum. Have you actually read the book ‘The Black Swan’? ‘Be careful’- is that all you gleaned from, say, the the ‘Inductive Turkey’ example he gave? (Something which he borrows from Bertrand Rusell.)

What about difference between Mediocristan and Extremistan?

Surely, you could disagree with his arguments, but to suggest that book is nothing but a bunch of ‘blindingly obvious, something-which-everyone-knows-anyways advice’ is a a bit unfair, and not to mention, a bit erroneous.

[i]The problem is that his advice is useless. And if he can’t give any useful advice in his books, than he needs to think a bit harder.[/i]

It’s YOU who needs to try harder. Try critiquing something more specific rather than trashing the whole book as useless.

Posted by Jaideep Dave | Report as abusive

His whole point is ‘good enough’ isn’t, fat tails can’t be accurately modeled, and avoidance is most important. That is more perceptive than what ‘everyone knows’ and willfully ignores.

Posted by Lord | Report as abusive

I did read The Black Swan when it first came out. I do agree with many of his observations with the way things are – unlikely events are much more common than we often appreciate, and I did like the Mediocristan bit that you mentioned. However, I didn’t find his advice on what to do with this state of affairs very useful and his claims that people in finance and academia have never heard of, or just ignore, his ideas is simply false. The claim that Bookstaber is making, which is absolutely true, is that risk managers appreciate this problem of “fat tails.” I’ll admit, though, that I find both Taleb’s writing and speaking to be incredibly pretentious and annoying, which certainly colors the way I view him.

Lord – There is a tradeoff when building models. You can increase the complexity of the model, which will enhance it’s precision, but it will also be harder to understand why the model is giving the outputs that it is. It is often better to use simple ‘good enough’ models that the trader or risk manager understands very well. That frees them up to think about all of the other issues that could affect their position (including black swans).

Both Taleb’s and Bookstaber’s book are concerned about how to deal with complexity that is difficult to model. Personally, I found the later to be much more insightful and useful.

That really worked out well now, didn’t it?

Posted by Lord | Report as abusive

I would say the problem is that traders and risk managers focused far too much on complicated models and statistics. Those models will never be perfect and if they had spent more time thinking about the ways they could get in trouble instead of trying to improve the models it would have worked out a lot better.