Night-Session Mean Reversion
Fade Bollinger-band extremes during the quiet overnight session, expecting reversion to the mean.
A disciplined, well-built clean-room EA with no edge — price reverts to the band-center only ~21% of the time while stopping out ~66%.
- Why it fails
- The mean-reversion thesis does not hold inside the night window: the few trades it takes lose, and no parameter neighbourhood is profitable.
- When / how it stopped
- Backtested on real 2018–2026 Dukascopy M5 data, the EA posts PF 0.655, Sharpe −0.48, and is net-negative in every out-of-sample segment.
Night-session mean reversion is a popular retail idea: in the quiet overnight hours, fade the Bollinger-band extremes and bet that price snaps back to the band center. The appeal is that low volatility “should” make ranges stick.
We built a clean-room, 985-line MQL5 implementation — Bollinger-band rejection gated to low-volatility, low-trend regimes, with genuinely well-built risk controls — and ran it on real 2018–2026 Dukascopy M5 data. It lost:
- Profit factor 0.655, Sharpe −0.48, win rate 29.6% across 71 trades — net −$181 (−3.62%).
- It is net-negative in every out-of-sample segment: validation PF 0.42, locked OOS PF 0.47.
- The core thesis fails directly: price reverts to the band-center target only 21% of the time and stops out 66% of the time.
This is not a tuning problem. A one-at-a-time parameter sweep finds no profitable neighbourhood — the least-bad perturbation reaches PF 1.001 only by trading nothing at all, and loosening filters to generate trades makes results worse. The “tunings” that nudge PF toward 1.0 do so by deleting losing trades until nothing remains. That is the signature of no edge, not overfitting.
It is also too selective to validate: only ~8 trades per year across four pairs, missing a 200-trade prerequisite by ~8×. The engineering is sound; the signal those controls gate is absent.
The full report, gate scorecard and equity charts are in our validation write-up.
→ Read our full validation report: /strategy/nightmeanrevert
Sources
- Our validation report — NightMeanRevert EA, PF 0.655, no edge
- Lo, Mamaysky & Wang (2000), "Foundations of Technical Analysis", Journal of Finance
Frequently asked
Does mean reversion work in the overnight forex session?
Not in this clean-room test. A Bollinger-band night-session mean-reversion EA backtested on real 2018–2026 M5 data returned a profit factor of 0.655 and was net-negative in every out-of-sample segment. Price reverted to the band-center target only 21% of the time while stopping out 66% of the time, and a 29.6% win rate needs to be at least 39.1% just to break even.
Is night scalping profitable in 2026?
Not on this evidence. The selective night-window approach produced only about 8 trades per year across four pairs, and the few it took lost. Every one-at-a-time parameter perturbation was break-even at best, and loosening filters to get more trades made it worse — there is no profitable parameter island. That is the signature of no edge, and a backtest that loses before live frictions (night spreads, swap, slippage) only loses more with them.
Not investment advice — your mileage may vary, but the burden of proof is on the person claiming an edge. This entry describes general research and published evidence (or its absence), not a recommendation. See the full disclaimer.