In a Gaussian distribution, the probability of a remote event (6, 7 or 8 sigma) is extremely nonlinear in the standard deviation. Increasing sigma by 10% multiplies its probability by thousands, millions or billions. So any uncertainty in estimating the standard deviation raises the small probability and magnifies the error about it; most parameter uncertainty in a model flows to the tails. Fukushima's 'once-in-a-million-years' event, slightly perturbed, becomes roughly one in thirty.

causalpending

Speaker

Nassim Nicholas Taleb

Evidence Quote

“most of the uncertainty in parameterizing a model goes to the tails”

Source

Nassim Taleb on Antifragility 01/16/2012— EconTalk
Created: 10/29/2026, 8:31:14 PM

Your notes

Loading notes...