Fixed spreads in backtests: the assumption that quietly flatters results

[Amanhasnoname]

Default settings in many testers use a constant spread. A scalper then sees stable results that ignore a basic reality: the spread moves all day.

At rollover it widens. At the news it jumps. In the Asian session it's wider than in London. A strategy that trades at all hours effectively pays a higher average cost than the tester assumes.

Variable spread data fixes the problem, and it's available with good tick sources. Testing with real bid and ask streams shows the actual cost of each entry and exit.

Where that isn't possible, a rough alternative is to build a spread model by hour and apply it. Even a crude one beats a constant.

Compare the two versions of a result. It's common to see a smooth, profitable curve on fixed spread turn flat or negative with realistic spread behaviour.

The difference is a good indicator of how dependent the strategy is on tight spreads, which is an important thing to know.

How different do your results look with real spreads versus fixed?