Tagging trades by session reveals leaks
Posted: Sat Oct 31, 2026 9:04 pm
Aggregate statistics hide a lot. A strategy might be profitable overall but lose money in one particular hour, and the loss gets absorbed by gains elsewhere.
Tagging each trade by session, and sometimes by hour, exposes the pockets. Typical findings: the first fifteen minutes of the London open work well, the late London hours chop, the NY open is volatile, and the end of the NY session drags.
Once visible, the pockets can be avoided. Cutting the worst hour often improves overall results more than any parameter tweak.
The tags should use UTC, to avoid the server time confusion mentioned in other threads. The sample per bucket needs to be large enough to mean something, ideally fifty trades or more.
Be cautious about overfitting. A bucket that looks terrible after twenty trades may just be unlucky. Confirm with more data before making permanent rules.
Which sessions or hours turned out to be the worst, and did cutting them help?
Tagging each trade by session, and sometimes by hour, exposes the pockets. Typical findings: the first fifteen minutes of the London open work well, the late London hours chop, the NY open is volatile, and the end of the NY session drags.
Once visible, the pockets can be avoided. Cutting the worst hour often improves overall results more than any parameter tweak.
The tags should use UTC, to avoid the server time confusion mentioned in other threads. The sample per bucket needs to be large enough to mean something, ideally fifty trades or more.
Be cautious about overfitting. A bucket that looks terrible after twenty trades may just be unlucky. Confirm with more data before making permanent rules.
Which sessions or hours turned out to be the worst, and did cutting them help?