Spread Widening at Rollover: The 10pm Trap for EAs

Around 5pm New York time - 10pm or 11pm in the UK depending on the season - the forex trading day rolls over. For a few minutes either side of that moment, liquidity providers step back, swap charges are applied, and the spread on many pairs widens dramatically. A pair that trades at a 1-pip spread all day can briefly quote 5, 10, sometimes 20 pips or more. For a human trader it's a non-event: don't trade at 10pm. For an EA running around the clock, it's a nightly trap.

Why rollover spreads widen

The daily rollover sits in the quietest part of the 24-hour cycle: New York has closed, Tokyo hasn't properly opened, and many liquidity providers briefly withdraw quotes while the trading day resets and swaps are booked. Fewer quotes means wider spreads - mechanically, not maliciously. The pattern repeats every trading day, which makes it unusual among trading costs: it's predictable. You know exactly when it will happen; the only question is whether your EA is exposed when it does.

How it bites an EA

• Entries cost multiples of normal. A signal that fires at 10:01pm might pay 8 pips of spread instead of 1. An illustrative example: a system with a 15-pip average profit per trade gives up over half of it on entry cost alone if it routinely fills at rollover.
• Stops trigger on the spread, not the market. Stops on long positions are triggered by the bid. When the spread widens, the bid drops even if the mid-price hasn't moved. A stop 6 pips away can be taken out purely by a 10-pip spread spike - the market then snaps back, and the EA has banked a loss on a move that never really happened.
• Backtests don't see any of it. Most bar-based backtests apply one average spread to every trade. Every rollover fill in the test is costed at daytime prices - a specific case of the gap described in spread modelling in backtests, and a contributor to why most EAs fail live.

Who is most exposed?

Tight-stop, short-timeframe systems suffer most, because a temporary 10-pip spread is enormous relative to a 10-pip stop but trivial relative to a 100-pip stop. Systems that hold through the night with wide stops mostly experience rollover as a swap charge rather than a spread problem. Session-aware strategies - built around the rhythm described in trading session times - often avoid the window naturally.

Common mistakes

1. Blaming the broker for “stop hunting”. Rollover spread spikes explain many mysterious overnight stop-outs. Check the bid and ask history at the time of the fill before assuming foul play.
2. Placing pending orders just before rollover. A stop entry resting near the market at 9:58pm can fill on a spread spike and then instantly show a large open loss.
3. Testing with fixed spread and trading a 24-hour strategy. If signals cluster in quiet hours, the backtest systematically undercharges the very trades that are most expensive live.

Practical defences

• Block new entries in a window around rollover - even 15 minutes either side removes most of the exposure.
• Give overnight positions stops sized for overnight conditions, not daytime ones - and size positions accordingly, per how to calculate position size.
• Compare your live fills around 10pm with your backtest assumptions. The difference is your rollover tax.

This is one of the quiet reasons Nebula models spread inside the simulation rather than bolting a flat cost on afterwards: strategies that only looked profitable by trading the cheap hours in a fantasy-cost world tend to get found out in testing, not in your account.

Trading carries a high level of risk. Nothing here is financial advice; past performance does not predict future results.