When a strategy works only because of a single pair or day
Posted: Mon Nov 02, 2026 11:08 pm
When a strategy works only because of a single pair or day
Aggregate results can hide concentration. A strategy might show a profit factor of 1.5 overall, but removing one pair or one extraordinary day drops it to 1.0. The edge, in that case, depends on a single source.
Test for concentration by breaking down profits by pair, by day, by session, and by week. If a handful of items account for most of the result, the system lacks robustness.
Concentration is not always bad. A strategy might legitimately fit one pair, because that pair has unique characteristics. But the risk of overfitting increases, and it becomes more sensitive to changes in that pair's behavior.
Check by removing the top contributors and re-evaluating. Also test the strategy on similar pairs or different time periods.
If it works across a broad set of conditions, confidence grows. If not, size accordingly and monitor for signs of change.
Understanding where profits come from clarifies where the risk lies.
What happens to the results when the best pair or the best week is removed?
Aggregate results can hide concentration. A strategy might show a profit factor of 1.5 overall, but removing one pair or one extraordinary day drops it to 1.0. The edge, in that case, depends on a single source.
Test for concentration by breaking down profits by pair, by day, by session, and by week. If a handful of items account for most of the result, the system lacks robustness.
Concentration is not always bad. A strategy might legitimately fit one pair, because that pair has unique characteristics. But the risk of overfitting increases, and it becomes more sensitive to changes in that pair's behavior.
Check by removing the top contributors and re-evaluating. Also test the strategy on similar pairs or different time periods.
If it works across a broad set of conditions, confidence grows. If not, size accordingly and monitor for signs of change.
Understanding where profits come from clarifies where the risk lies.
What happens to the results when the best pair or the best week is removed?