Risk

Power Laws, Fat Tails, and the Loss That Has No Natural Ceiling

Recognise the exposures where there is no typical loss size, so you stop sizing limits and reserves from an average that describes nothing and start bounding the exposure instead of estimating it.

  • Advanced
  • 12 min total
  • 14 chapters

What decision this helps you make: Which of your exposures are fat-tailed, and for those, how to buy limit, cap the exposure contractually, and hold reserves against a loss whose largest observed value is simply the largest one so far.

What this topic is

A fat-tailed exposure is one where extreme outcomes are far more likely than a bell curve implies, and where there is no characteristic size at all. Human height is thin-tailed: nobody is twenty feet tall, the average describes almost everyone, and one more measurement barely moves it. Liability verdicts, cyber losses, catastrophe damage, and market crashes are not like that — the largest event in a sample can exceed the sum of everything else in it, and the average is a number nobody experiences.

Why it matters

Almost every default risk practice — average loss, historical maximum as the worst case, a limit sized from experience, a reserve set from a typical year — assumes a thin tail. Applied to a fat-tailed exposure they all understate it, in the same direction, at the same time. The failure is not that the estimate is imprecise; it is that the estimator is measuring something that does not exist.

Who should learn it

Anyone sizing insurance limits, capital reserves, or contractual liability caps; anyone whose exposure includes litigation, cyber, catastrophe, or a single dominant counterparty; and anyone who has ever been shown a worst case that turned out to be a historical maximum.

What you will understand

  • The test that separates fat-tailed exposures from thin-tailed ones
  • Why the sample average and the historical maximum both mislead here
  • Why bounding an exposure beats estimating it, and what bounding looks like
  • The honest counter-argument: most business exposures are not actually fat-tailed

Prerequisites

Common misconception

"Our worst-ever loss was $400,000, so we should be covered somewhere above that." In a thin-tailed world that reasoning is sound — the maximum of a large sample sits close to a natural ceiling and the next one will be similar. In a fat-tailed world the largest observation is simply the largest one so far, and it carries almost no information about the largest one available. The record is not a boundary. It is a sample from a distribution whose whole point is that it keeps producing new records, and treating it as a ceiling is how a limit gets set at exactly the level that has already been exceeded once.