Calibration

How closely the likelihoods you state match how often those things actually happen, measured across many stated likelihoods rather than any one of them.

Example

Of the things called about 70% likely, roughly 70% should occur. Five of nine is 55.56%, which is a gap worth watching on nine observations and not yet a finding.

Important

Calibration is a property of a set of statements, never of a single one. Six predictions produce a share that one different outcome would move by more than sixteen percentage points, so a thin band describes the sample rather than the judgement behind it.

Related terms

• Sample size — The number of independent observations a conclusion rests on.
• Base rate — How often an outcome occurs across the whole population of similar cases, before you know anything specific about the one in front of you.
• Overconfidence bias — Systematically overestimating the accuracy of your own judgement - in particular, being far more certain of a forecast than its track record justifies.

Lessons that explain it

• The memo is worth more at the two-year review than on the day you wrote it — A memo's real value is not in getting a deal approved. It is in being specific enough, on the day it was written, that reading it again later tells you something you could not otherwise have known. (10 min)
• Reading an old entry without rewriting it — The entry is only half the practice. The other half is going back to it later and reading what you actually wrote, rather than what you now wish you had. (11 min)
• A short review you actually do beats a thorough one you abandon — Decision journals fail through ambition. A fixed, brief monthly pass over recent entries and a longer annual review are enough to produce most of the learning. (9 min)

All 578 terms in the Investing glossary