Night Scalper EAs: Why the Backtests Look Too Good

Night scalper backtests look too good because they assume quiet hours spreads and swap costs that are worse live, and their settings are tuned to years of calm nights. Judge one on real session spreads, swap and out-of-sample results.

Night scalpers - EAs that trade the quiet Asian session, fading small moves back to the mean - are a perennial bestseller. Their backtests are frequently the most beautiful you will ever see. That is not a coincidence, and it is not good news.

Why the strategy exists

The idea is sound on paper: in the hours after New York closes, major pairs often drift in tight ranges. Mean-reversion logic with small targets can harvest those ranges. Real desks have traded versions of this. The problem isn't the concept - it's that the conditions that make it work on historical data are precisely the conditions that break down when you trade it for real.

Where backtest and reality diverge

• Night spreads are the widest of the day. The session with the least liquidity is the session where a backtest's fixed "typical" spread is most wrong. At the time of writing, many brokers run Asian-session spreads at multiples of their London-session numbers - on a strategy targeting a handful of pips. Spread modelling isn't a detail here; it's the whole verdict.
• Rollover happens in the middle of it. Swap is charged, spreads spike at 5pm New York, and some feeds print odd ticks around rollover. Backtests glide through this; live accounts don't.
• Thin liquidity means real slippage. Small orders usually fill fine - until a news wire hits an empty book and your "tight stop" fills far away.
• Broker sensitivity. Night scalpers are notoriously feed-dependent. The same EA can look profitable on one broker's historical data and be unviable on another's live spreads.

The overfitting bonus round

Because night ranges are small and regular, mean-reversion parameters tune beautifully to history - which is exactly the trap. A system fit tightly to a decade of quiet nights has memorised those nights, not learned a law of markets. When volatility regimes shift, the range disappears and the losses arrive. The mechanics are in our overfitting and walk-forward guide, and the aftermath is documented in why most EAs fail live.

How to judge a night scalper honestly

1. Demand backtests using real recorded spreads for the Asian session, not a daytime average.
2. Check swap costs are included - the strategy holds through rollover by design.
3. Ask for out-of-sample results and a walk-forward pass, not one glorious optimisation window.
4. Look at average profit per trade after costs. If it's under a pip or two, tiny execution differences decide everything.
5. Verify enough trades to matter - sample size - across different volatility years, not just calm ones.

Build instead of buy: our biased take

Full disclosure: we make Nebula, a no-code builder that breeds strategies with a genetic algorithm and pushes every candidate through walk-forward, out-of-sample and blind forward validation. We won't tell you Nebula's strategies beat night scalpers - we don't make performance claims for anything. The relevant point is narrower: when you build rather than buy, you control the spread and swap assumptions, and you can watch a too-good-to-be-true candidate fall apart under honest costs before it costs you anything. With a purchased night scalper, you usually find out live.

The bottom line

A night scalper with a flawless backtest is a claim about the quietest, most cost-hostile hours of the market - treat it accordingly. Judge it on real session spreads, swap and out-of-sample survival, or not at all. More in evaluating an EA before buying.

Trading carries a high level of risk and automated systems do not eliminate it. Nothing here is financial advice; past performance does not predict future results.