A trader sorted trades by hour and found the favorite session was the problem
Posted: Sat Oct 31, 2026 8:06 pm
A trader assumed the London open was the best part of the day. It felt that way, since it was the busiest and the most exciting. After a couple of months of trades, the entries were sorted by hour of day, using broker server time converted to local time, and the picture was different.
The London open hour was roughly break-even. The profit was coming from the hour or so after it, once the first burst had settled and moves became more orderly. The hours losing money were not the quiet ones either. They were the ones right around the New York open, where a lot of trades were taken in a hurry.
The first twist: the strategy was fine, but the hour it was applied in was not. The second twist was subtler. When the trades from the losing hours were removed, the average winning trade barely changed. What changed was the number of losers. Fewer low-quality trades, not better ones.
Some cautions apply when doing this. Sample size per hour needs to be reasonable, because with ten trades in an hour a result is mostly noise. Grouping into blocks of two or three hours gives steadier numbers. Daylight saving shifts between the US and UK for a few weeks each year, so the same server hour can mean different market moments, and the data needs to be aligned to the actual session, not the clock.
Day of week deserves a look too. Fridays and Mondays often behave differently from midweek, and so do the days around major data releases.
It also helps to split by outcome type: win rate, average win, average loss and trade count per block. A block with a high win rate but one outsized loss tells a different story from a block that bleeds slowly.
When time-of-day stats get compared to gut feeling, the gut often picks the most memorable session, not the most profitable one.
Which hour or session turned out to be a surprise when trades were sorted by time, and did it hold up in the following months?
The London open hour was roughly break-even. The profit was coming from the hour or so after it, once the first burst had settled and moves became more orderly. The hours losing money were not the quiet ones either. They were the ones right around the New York open, where a lot of trades were taken in a hurry.
The first twist: the strategy was fine, but the hour it was applied in was not. The second twist was subtler. When the trades from the losing hours were removed, the average winning trade barely changed. What changed was the number of losers. Fewer low-quality trades, not better ones.
Some cautions apply when doing this. Sample size per hour needs to be reasonable, because with ten trades in an hour a result is mostly noise. Grouping into blocks of two or three hours gives steadier numbers. Daylight saving shifts between the US and UK for a few weeks each year, so the same server hour can mean different market moments, and the data needs to be aligned to the actual session, not the clock.
Day of week deserves a look too. Fridays and Mondays often behave differently from midweek, and so do the days around major data releases.
It also helps to split by outcome type: win rate, average win, average loss and trade count per block. A block with a high win rate but one outsized loss tells a different story from a block that bleeds slowly.
When time-of-day stats get compared to gut feeling, the gut often picks the most memorable session, not the most profitable one.
Which hour or session turned out to be a surprise when trades were sorted by time, and did it hold up in the following months?