forecasting//black swan

A black swan is an event outside the range a model or a forecaster considered possible, with a large effect, that looks predictable only in hindsight, and the term is used to mark what no extrapolation of a trend can contain. The name is Nassim Taleb's, after the European belief that all swans were white until black ones were found in Australia: no number of white swans proves the next one is white.

A black swan is an event outside the range a model or a forecaster considered possible, with a large effect, that looks predictable only in hindsight, and the term is used to mark what no extrapolation of a trend can contain. The name is Nassim Taleb's, after the European belief that all swans were white until black ones were found in Australia: no number of white swans proves the next one is white.

Every growth rate is a slope fitted to the past, so a forecast built on slopes is silent about the next discontinuity by construction. In machine learning such discontinuities are part of the field. An ablation study of a decade of language-model efficiency (Gundlach and colleagues, 2025) found that one change, from recurrent networks to the transformer, accounts for most of the gain it could explain: a decade of progress was mostly one event plus a slope.

A forecast states its black swans as a premise.

The honest form is conditional: the conclusion holds if no event of a named kind (a new architecture, a method that makes models legible, a measurement that stops working) arrives before the date, and it has to be redone from the event if one does.

A black swan differs from a risk with a known mechanism. A credit crunch, a fall in revenue or a change of accounting rule can be priced and weighed as a scenario of a scenario mixture; an event with no defensible distribution cannot, since it is the world nobody listed and sits outside the mixture by definition, and pretending to weigh it gives a number with no meaning.

The term is easy to abuse. Many events called unforeseeable had a base rate that someone ignored, and calling them black swans excuses the forecast instead of correcting it.