What missing the best days costs — and what missing the worst days saves

Over last 30 years, missing the ten best sessions takes the compound return from 8.50% to 5.62% a year. That is the number in circulation, and it is correct. What is almost never printed beside it: missing the ten *worst* sessions over the same window raises it to 11.72%. The second effect is larger than the first.

Taken alone, each half is an argument for the opposite behaviour, which is a good sign that neither is an argument at all. A statistic that supports staying invested when you show one column and supports timing the market when you show the other is not evidence about either; it is evidence that removing the extreme observations from a fat-tailed series moves the result a lot.

The column that does carry an argument is the clustering. Of the fifty best sessions in the entire record, 33 — 66% of them — happened within ten trading days of one of the fifty worst. They are not separate events you could sort into a good pile and a bad pile. They are the same handful of crises, and the enormous up days are what crises look like from the inside. Sitting out the fortnight to avoid the crash means sitting out the rebound, because it is the same fortnight.

One more thing worth checking whenever you meet this statistic: which window it used. The full record since 1928 and the last thirty years give visibly different figures on this page, because the extremes of the 1930s dominate the long window. Any version quoted without its period cannot be reproduced, and most of the versions in circulation do not name one.