Fat Tails, explained.
Fat tails describe distributions in which extreme outcomes are more probable than under a thin-tailed comparison, such as a normal distribution.
Why it happens
When a few large observations account for much of a total, the average can change sharply after one more event. This makes the shape of the tail important, even if most everyday observations look unremarkable.
A fat-tailed distribution assigns more probability to very large deviations than a thin-tailed benchmark such as a normal distribution. Rare observations can dominate totals and make estimates unstable.
Look for this pattern
Compare typical outcomes with the contribution of the largest observations. A stable-looking center says little about whether the model gives enough weight to rare, consequential events.
A worked example
A service with occasional huge jobs
Most requests take little time, but a few require hours of processing.
An average based on a quiet week may underestimate the resources needed when one large job arrives.
Inspect extremes and capacity constraints alongside the average; do not assume every workload follows a bell curve.
Where this idea is useful
A model that understates extreme outcomes can make a system look safer than it is.
A common misconception
“Fat-tailed means that disasters happen all the time.”
An extreme event can still be rare. The claim is about its probability relative to a specified alternative model.
What this explanation leaves out
- Fat-tailed does not mean every extreme event is likely. Tail definitions and the relevant comparison distribution need to be made explicit.
Are every skewed distribution and every power law fat-tailed?
Skewness describes asymmetry, while tail behavior describes how probabilities decline far from the center. These properties are related in some distributions but are not interchangeable.
Which single unusually large outcome could dominate a total you normally describe using an average?
Associated thinkers
Associations marked provisional are awaiting source review.