Fixed spread in a backtest is a quiet form of cheating, even when nobody intends it

[Shadow Trader]

Real spreads breathe. They are tight in the middle of the London and New York overlap, wider in the Asian lull, and they blow out around rollover and major releases. A scalper who targets 3 pips gives up a very different share of the move when the spread is 0.2 than when it is 1.5.

A fixed number in the tester, usually something near the typical daytime figure, treats every entry as if it happened at the best moment of the day. Trades taken near rollover or right after a release get filled much cheaper in the simulation than they would live. Those are often exactly the trades where the setup looks most appealing, because volatility is high.

Variable spread data fixes part of it, though it has its own trap. If the spread column was recorded on one broker and the strategy will run on another, the costs are still borrowed. A useful middle ground is to test with a fixed spread, then rerun with that spread doubled and see whether anything survives.

If the edge disappears at double the cost, it was thin to begin with.

How wide does the spread have to go in your testing before the strategy stops making sense?