What Is Data Snooping

Data snooping in five lines

Data snooping is trying many rules on the same bars and keeping the one that worked. It fools because with enough tries something always works; that is arithmetic, not skill. The tell is that the winner was found after looking rather than written down before. The cure is to write the rule first, count every try, and fence off bars you have not looked at. In the example, 276 buy-and-sell hour-pairs were tried on the euro on each of 27 weeks of 2026, and every single week had a winner: the best pair of the week made between 21.9 and 220.0 pips, where a pip is the fourth decimal place of the price, and it was a different pair almost every week.

The target painted afterwards

Picture a wall of darts with a target painted on afterwards. The darts were thrown first, all of them, at nothing in particular; then the rings were painted round whichever dart happened to land furthest from the rest, and there it is, a bullseye. The bullseye is real. It was never aimed at, and it will not be there next time, because the next throw scatters differently and the rings are still where they were. A rule found by trying 276 rules is the dart that got the rings painted round it.

This paper is locked