The Larger World ArchiveClaims · Sources · Evidence
States, secrets and documentsLive debateRisk and probability · 2001 – present

Black Swans

Nobody predicted 2008, the pandemic or the smartphone. The interesting question is not that surprises happen but whether our models are built to be surprised by them.

A black swan. The bird that falsified a thousand years of European certainty is a common A
A black swan. The bird that falsified a thousand years of European certainty is a common Australian species. Image: Francis C. Franklin · CC BY-SA 3.0 — Wikimedia Commons.

The definition and the three properties

Taleb's 2007 book defines a black swan as an event with three properties: it is an outlier relative to past expectations, it carries extreme impact, and human beings produce explanations for it after the fact as though it had been predictable.

The name comes from the European assumption that all swans are white, falsified by the first sighting in Australia. The point is epistemic: the observation that had never occurred is the one that changes the theory.

The practical programme that follows is not prediction but robustness: build systems that survive being wrong, prefer insurance to forecasts, and be suspicious of any model whose tail is thin because nobody has seen a fat one yet.

Where the concept is strongest and weakest

It is strongest as a description of financial and historical data, where returns and event sizes are genuinely fat-tailed and normal-distribution risk models systematically understate extremes. The 2008 crisis and the failures of Gaussian copula models are the canonical case.

It is weakest when used as a universal solvent. Some domains have thin tails and predictable distributions; a bridge has a load rating because the physics of bridges is well behaved. Calling every wide confidence interval a black swan turns a statistical statement into a rhetorical one.

There is also a falsifiability problem. If black swans are by definition unpredictable, then the claim that forecasting is futile cannot be tested and cannot fail — which is precisely the structure the book warns about elsewhere.

What the research actually says

Expert forecasting: Philip Tetlock's long study found experts barely beat chance on long-range political forecasts, but a subset — his superforecasters — did substantially better, with small teams, base rates and frequent updating. That is evidence against fatalism, not for it.

Tail risk is real: the empirical case that financial returns are fat-tailed is not seriously disputed, and risk models that assume normality are wrong in a measurable direction.

Hindsight: the robust finding is the narrative fallacy — the ease with which we construct stories that make past surprises seem inevitable. That is a fact about cognition, and it survives regardless of how the forecasting argument is settled.

The archive labels this live debate because the phenomenon is real, the response is contested, and 'nobody could have known' is often true and often a convenient excuse.

Why it is in the archive

Because it is used both as a serious critique of statistical modelling and as a way to avoid accountability, and the two uses are hard to tell apart from the outside.

Related: the Lindy effect, market efficiency.

Sources for this page

  • Wikipedia, “Black swan theory” — the definition and its criticsen.wikipedia.org
  • Taleb (2007), The Black Swan — the book that defined the termen.wikipedia.org
  • Tetlock & Gardner (2015), Superforecasting — the evidence against fatalismen.wikipedia.org
  • Wikipedia, “Fat-tailed distribution” — why the tails matter more than the meanen.wikipedia.org