The fixed spread assumption that flatters every backtest
Posted: Sat Oct 31, 2026 8:53 pm
Many backtests apply a constant spread, say 1 pip, to every trade. Real spreads move constantly. They are tight at the London open and wide at rollover, and they spike at news.
A strategy that trades through all hours with a fixed spread looks fine, because the expensive moments are hidden. In live trading, those moments hit the account.
For scalpers, the bias can be decisive. Imagine a system whose profit depends on trades at the NY close. In the backtest, each costs 1 pip. In reality, they cost 3 or 4.
The better approach is using historical spread data, either from a tick feed that includes bid and ask or from the broker's own records. The backtest should charge the actual spread at each entry and exit.
If that data is not available, a conservative alternative is applying a time-of-day spread table, with higher costs in the thin hours.
Comparing results with fixed and variable spreads shows how much of the edge was real.
How different did the results look after switching to variable spreads?
A strategy that trades through all hours with a fixed spread looks fine, because the expensive moments are hidden. In live trading, those moments hit the account.
For scalpers, the bias can be decisive. Imagine a system whose profit depends on trades at the NY close. In the backtest, each costs 1 pip. In reality, they cost 3 or 4.
The better approach is using historical spread data, either from a tick feed that includes bid and ask or from the broker's own records. The backtest should charge the actual spread at each entry and exit.
If that data is not available, a conservative alternative is applying a time-of-day spread table, with higher costs in the thin hours.
Comparing results with fixed and variable spreads shows how much of the edge was real.
How different did the results look after switching to variable spreads?