Fixed-spread backtests are flattering the system

[Amanhasnoname]

A test with a fixed 0.8 pip spread assumes the cost is identical at 03:00 and at the 17:00 New York rollover, and identical a minute after a major release. Live, it is not. A pair quoted at 0.6 pips most of the day can print 2.0 to 3.0 pips for a few minutes around rollover, and anyone trading those windows pays it.

The bigger issue is that cost and opportunity are correlated. The fast moves scalping setups love tend to arrive exactly when spreads are widest. A fixed spread severs that link, so the test hands the system the moves without the toll.

Check it properly. Log the live spread every minute for a few weeks, or export the broker's own history if that is available, then replay the system with spread looked up by time of day. Even a crude hour-by-hour average table beats a single number.

Expect the biggest hit on systems aiming at under 4 pips, where an extra 1 pip of cost eats a quarter of the target. Spread behaviour differs by broker and account type, so the table has to come from the account actually being traded.

Has anyone built an hourly spread table from live logs, and how far did it move the test result?