Conventional metrics (mean and variance) used in Gaussian distributions are uninformative for fat-tailed phenomena. In fat-tailed epidemiology, 95-99.9% of observations fall below the mean, so being a good predictor of the mean doesn't capture the distribution's true properties; sampling error is monstrous, the law of large numbers barely applies, and prediction error dominates.
factualpending
Speaker
Nassim TalebEvidence Quote
“the gaussian distribution is based on mean and variance and these are uninformative...you have 95% or to 99.9% of the observation falling below the mean”
Source
The Merge: Ep. 03 | The past is not gone; it lives on in the future— Maroun Semaan Faculty of Engineering and ArchitectCreated: 8/11/2026, 7:38:57 AM
Your notes
Loading notes...