Fat Tails

Fat tails describe probability distributions in which extreme events are significantly more likely than in a normal (Gaussian) distribution. In a normal distribution, the probability of an event five standard deviations from the mean is effectively zero. In fat-tailed distributions — characteristic of financial returns, earthquake magnitudes, and cascading system failures — such events are rare but non-negligible, and the tail probability decays polynomially rather than exponentially.

The concept is central to risk management because most standard financial models, including the Black-Scholes model, assume returns are log-normal. This assumption makes extreme risks invisible until they materialize. The financial crisis of 2008, the COVID crash of 2020, and numerous smaller crises all involved events that standard models treated as astronomically improbable.

Fat tails are not merely a statistical nuisance. They are a signature of systems with feedback loops, network effects, and reflexive dynamics — systems where small perturbations can propagate and amplify. A market with fat-tailed returns is not a random walk with the wrong distribution; it is a system that occasionally undergoes regime changes that the normal regime cannot predict.

The normal distribution is not wrong because it fails occasionally. It is wrong because it is taught as the default, the unmarked case, the distribution you assume until you have evidence otherwise. But in any system with interconnected agents, feedback, and history-dependence, the default should be fat tails. Normality is the special case — and it is a special case that almost never applies to the systems we actually care about.