Why would session timing plausibly matter?
Session timing has a structural mechanism behind it, which distinguishes it from most technical filters. Liquidity, participation and volatility genuinely differ by time of day — the London open and the London/New York overlap concentrate the great majority of activity, while Asian hours are typically quieter and more range-bound for the major pairs.
That mechanism generates a testable prediction rather than a vague expectation. If a setup depends on a level being defended, it should work better when more participants are present to defend it. If a setup depends on a range holding, it should work better when fewer participants are pushing through it.
A filter with a mechanism is a fundamentally better prior than one discovered by search. It is the difference between testing a hypothesis and mining for one, and it is why session was worth examining carefully rather than dismissing alongside the rest.
What did the data show?
The same setup performed materially differently depending on when it fired. In our testing, mean-reversion and fade-style setups fared better in the higher-participation windows, which is consistent with the mechanism — a level is more meaningful when more participants are transacting around it.
The effect was large enough to matter for selection and sizing rather than being a statistical curiosity that survives only in aggregate. It also held with reasonable consistency across the instruments we checked, which is the property that separates a finding from an artefact.
We are deliberately not publishing per-session expectancy tables. The measurement is real, it informs how the platform grades and sizes setups, and it is one of the things a subscription buys. What is published here is the shape of the result and the reasoning behind it, which is the part that is useful to a trader deciding whether to test session effects in their own work.
How do you use a session filter without overfitting?
Use it by deciding which session you expect to work, and why, before you look at the results — because session is a small search and small searches still manufacture findings.
There are roughly four plausible session buckets. Testing all four and keeping the best is a four-way search, which is modest but not free, and it needs out-of-sample confirmation exactly like anything else. A session filter chosen by mechanism and then confirmed by data is far more trustworthy than one selected because it happened to produce the best number.
Watch the sample carefully too. Splitting a 400-trade backtest four ways leaves 100 trades per bucket, and each per-session conclusion is correspondingly weaker than the headline that preceded it. If the overall result was already marginal, the per-session results are noise.
Be careful with time zones as well, which is a mundane trap that invalidates more session research than anything statistical. Broker server time is frequently not the time zone you assume, it shifts with daylight saving on a different schedule from the sessions themselves, and a one-hour offset can move trades between buckets wholesale.
Which session is best for forex trading?
There is no universally best session, because the answer depends on what the strategy needs from the market rather than on the clock.
In our testing, higher-participation windows — the London session and the London/New York overlap — produced better results for fade and mean-reversion setups. That is not a general endorsement of trading those hours; it is a statement about a specific family of setups whose logic depends on participation.
A breakout strategy plausibly has the opposite requirement, wanting the conditions in which a move continues rather than reverts. A range strategy may well prefer the quieter Asian hours where ranges actually hold. The mechanism tells you which to expect, and the test tells you whether you were right.
The practical advice is to test your own setup across sessions rather than adopting somebody else's answer, and to have a reason for the answer you expect before you look.