Many testers let the user set a fixed spread, say 0.5 pip, and apply it to every trade. Real spreads move constantly: tight in London, wide at rollover, enormous around news.
Run a scalping system with a fixed spread and the results look smooth. Run the same system with the real recorded spread for each tick and the picture changes, often drastically. Trades during wide spread moments, which a fixed model treats as cheap, are the ones that bleed.
A fairer test applies the actual historical spread, or at least a time-of-day model with realistic widening around events. Add slippage as well, with a bigger value for stop orders than for limit orders.
The honest approach is to stress the system: double the spread, add a pip of slippage, and see if anything survives. A strategy that dies under mild stress will likely die live.
What happens to results when costs are doubled?
Advertisement
Fixed spread backtests are fiction
Recommended broker for automated trading & scalping
