Sunday 16 January 2011

"The Black Swan" by Nassim Nicholas Taleb

The author of this book got up my nose from the start when in the Prologue he describes the probability estimates of a risk manager as having "no better predictive value for assessing the total risks than astrology". Which is clearly wrong. Maybe the predictions of the risk manager are imperfect but they are almost certainly better than astrology.

Taleb tends to over-argue his case and he is rhetorically rude about people who have a different point of view (which prompts my in-built bullshit detectors to suggest that he hasn't got a case). He describes people who live normal lives as "dull" and says they live in "Mediocristan"; "their minds are closed". Teachers and academics have him spitting dust.  "Only military people deal with randomness with genuine, introspective intellectual honesty."

The insults become far worse and far more frequent towards the end of the book. He really doesn't like statisticians, claiming that their entire disciplines are rubbish and that they are closed minded idiots. Most other academics are similarly rubbished. His ad hominem remarks include their dress sense. When someone gets so angry and so personal I always suspect that their arguments are inadequate.

So as a teacher and academic I tried very hard to keep an open mind but as a human I found it very difficult. What are his arguments?

Firstly, their is the problem of induction. He clearly quite likes David Hume. He describes it thus (borrowing an example from Bertrand Russell; Taleb uses intellectuals when he wants to): if a turkey has been fed every day for a thousand days it will probably assume that it will always be fed when in fact the next day it is to be killed. I would suggest that the problem of induction is not so much that we are not aware that recurring examples do not lead to proof but that it would be ridiculous for the turkey to be surprised at being fed on the 998th and 999th days. So there has to be trust in induction whilst at the same time being sceptical. If you were a turkey and you had to bet what what would happen on day 1000 it would make sense to bet on being fed.

The second problem is that of hindsight; he quotes, amongst other examples, the Anthropic Principle. He also quotes Cicero who gives an example of an irreligious man being told of sailors who had prayed and not drowned, and asking about the sailors who prayed and still drowned. Thus the Earth seems a planet perfectly tailored for the evolution of mankind because had it been any different we would not have evolved and therefore would not be here to admire its tailoring (or we would be somehow different and thus be admiring the tailoring that led to the new different us). Another lovely example is that of gamblers most of whom believe in beginner's luck because of their own personal experiences; because, claims Taleb, the gamblers who lost on their first gambles got disenchanted and never became gamblers.

So I admit that these are fallacies that mean that we can never absolutely trust what is to happen but I still cling to the idea that both these fallacies give a pretty good approximation of what is likely to happen.

In the latter part of the book I think his main point is that Gaussian bell curves do not work in financial markets. The evidence for this is the existence of Black Swan events on a frequency that is significantly greater than the Gaussian predictions. I do not disagree with him. Gaussian bell curves work brilliantly on probability distributions characterised by true randomness, that is where one throw of the die does not affect the next. However, in situations where there is feedback, such as with earthquakes, fashions and money markets, Mandelbrotian fractal power distributions seem to work much better. This is still not perfect, as NNT points out you still base your models on historical data so there is always the possibility of something utterly novel happening in the future which you can't predict, but this is surely the best we can do.

But this is all explained much better in Ubiquity by Mark Buchanan and Critical Mass by Philip Ball. And these authors do not spend page after page rudely rubbishing their opponents with the most angry and arrogant invective. NNT does not seem to realise that Gaussian statistics are utterly valid and really useful in their sphere. He does not seem to realise why his particular context is different. But he does seem to believe that he is indubitably correct, there is no room for self-doubt whatsoever, which in a book about the limits of our ability to know is really rather ironic.

There was one concept that gave me pause for thought. He is talking about scalability, which for him is the idea that sometimes the winner takes all. Thus a best selling author doesn't need to put any extra work in once his book has hit the shops but he reaps the reward; another author will have to keep working on the day job. Scalability applies to authors, move stars, bankers etc. Some scientists work all their lives in hop of a big breakthrough, sustaining endless disappointment with hope that one day they will find a cure for cancer. Dentists, on the other hand, work every day for reasonable reward. NNT clearly thinks the former lifestyle is better (if you strike lucky) and maybe it is what I should have done. Next life perhaps!

There are some good ideas in this book though none are original. Nevertheless, in essence he is right. But he does not have the monopoly on truth and the people he is so rude about are also developing truths in different contexts. I prefer an author who presents his ideas more concisely; in the Black Swan the ideas are padded out with an awful lot of rudeness.

Jan 2011; 300 pages

No comments:

Post a Comment