The Black Swan - Nassim Nicholas Taleb

Given how arrogant this guy comes across in the press, I was avoiding buying a copy of this book. Fortunately, a copy appeared in Oxfam, so I could buy it without him receiving money, and helping charity to boot. Woo. I have now finished it.

It felt like a long read. It's a slow-going 300 pages, and it doesn't feel like it covers a lot of ground. Apart from Taleb. It does cover him in excruciating detail. The only author I've seen who's more self-absorbed is Will Self!

So, does this book actually have any justification in existing? Well, yes. The author does have a point - we underestimate outliers, both intuitively and by modelling phenomena with normal distributions when we shouldn't. The point's drilled home until you can't really be bothered to argue, and it all seems pretty obvious anyway. I suppose the main contribution is a terminology which, once you're used to it, seems a non-contribution. Wittgenstein's ladder, I suppose.

The book therefore does contribute something, but I found the writing style particularly poor. As well as the slow, self-absorbed approach, Taleb introduces the occasional flimsy, 2D fictional characters to illustrate his ideas. He argues throughout against the habit of creating narratives to bypass real scientific thinking, but it turns out he has a big weakness in this area. I can imagine him resisting such stories, being strong-armed by his editor, and writing this stilted stuff under duress.

One of these narrative interludes is the story of Fat Tony versus Dr. John. The details of this dull and stereotyping story are irrelevant, but the interpretation is interesting. The previous chapter complains about people who look at history (i.e. a single sample), rather than taking a more scientific approach to see if history was a fluke (rather important when deciding if someone's good, or just lucky). In Fat Tony vs Dr. John, apparently the answer is that, given a model and a set of observations from the model, we should trust the probabilities implied by the set of observations, rather than the given model parameters.

Of course, there's more to this than him making a complete u-turn in his approach without noticing, but he doesn't provide any justification, or even note this change. Throughout, he doesn't really go into detail, so despite the length of the book you don't really feel you've received much more than an outline.

Another oddity is in his approach to skepticism. He does take a rather sensible, empirical and skeptical approach in the face of the unknown, and is a strong believer in falsification. He emphasises the difference between "evidence of no X' and 'no evidence of X'. Fair enough. He also laughs at people who make long-term plans in the face of Black Swans, since Black Swans will break the plans. However, evidence that in some (indeed, most) situations planning cannot stop disaster, is not the same as saying that not planning will not increase the risk of disaster. Our economists may not stop credit crunches or recessions, but at least we don't have the hyper-inflation of Zimbabwe, as we try to avoid the policies of the hyper-stupid.

Finally, his views are obviously influenced by his time as a trader, but his arguments lack depth in a book for general readership. While he complains about the models used currently, he doesn't offer any alternatives, and by not talking about these models in detail you don't really see his specific objections. Local and stochastic volatility models can fatten the tails of a normal distribution, while staying in a Gaussian framework. Is this acceptable to him? Who knows.

While Taleb complains about the shape of a normal distribution, compared to something more, er, fractal, he seems to ignore the fact that the Brownian motions which generate the normal distribution are fractal. I'm sure there's a good distinction, but he doesn't make it.

If you want to move away from models based around normal distributions, Lévy models introduce discontinuities, and I'm sure you can tweak them to get the kind of distribution he's after. All in all, it's not clear what he would consider a satisfactory model, and whether it would require a reinvention of financial maths from scratch using statistical physics to make him happy.

On top of that, anyone with the slightest clue realises that the models are not reality. Quants make models. They make assumptions. They make very silly assumptions, but it's either that or you have no mathematical models, and you're not going to manage a book of severaal thousand exotic options on a gut feeling and a hope that you're playing the outliers to your advantage. The models have limitations, which is why the traders should understand the models, the reality, and the gap between them, and be able to deal with this. Otherwise, we wouldn't need traders.

I believe this goes against his vehemently held views. Anyone who reads his work, but goes back to a market model with a sniff of a normal distribution, is apparently missing his point, even if such models are employed while bearing in mind their limitations. He claims to be a pragmatist, but since we lack other models, I can't see a pragmatic alternative.

This book is a big diatribe on What Not To Do. While he offers vague advice on what can be done instead, this is next to useless. Moreover, it's written in the clever-person-explaining-to-the-dumb style, so important subtleties are glossed over. I'm now very tempted to read his Dynamic Hedging, assuming that this volume might actually contain concrete advice on what to do, targetted at the mathematically literate, and therefore provide more than sound bites.

Posted 2008-06-05.