Taleb writes thus: “Many have the illusion that ‘because Kolmogorov-Smirnov is nonparametric’, it is therefore immune to the nature specific distribution under the test….”
What does that mean? The Kolmogorov-Smirnov test is a standard measure of how a certain data sample compares with a reference probability distribution. It focuses on the largest observed distance between the actual set and the corresponding figure in the referenced distribution. If the largest observed distance is acceptably low, then presumably the one is acceptably conformable to the other. This test is widely said to be “non-parametric,” which in this context means that it is free of question-begging assumptions.
So what Taleb contends here is that the nonparametric nature of that test is an illusion. This is part of his effort to discredit patch-up jobs. In his view, Kolmogorov-Smirnov is one of many ways in which quants persuade themselves that they at last have a fix on how fat those fat tails can really get. If true, this would allow the quants to return to their pre-crisis mechanical ideas about risk management. But it isn’t, and they shouldn’t.
Or, as Taleb puts it at his most eloquent, “Shkmolgorov-Smirnoff.” Though I’m not sure I could manage the pronunciation of that.
via Would-be Error Detectors: Suit Up | AllAboutAlpha: Hedge Fund Trends & Alternative Investment Analysis.