Put slippage into the expectancy math before trusting a backtest
Posted: Sat Oct 31, 2026 7:44 am
Backtests with fixed spread and zero slippage are fiction for anything with a target under 5 pips. A system showing 0.9 pip net per trade in the tester can be at 0.2 live, and nobody changed the entry logic.
Do the arithmetic once, with real numbers. Gross edge 1.4 pips per trade. Average spread paid 0.7. Entry slippage averaging 0.15. Exit slippage on stops averaging 0.3 across the losing half, so 0.15 per trade overall. Commission 0.3 pip equivalent. Net is 1.4 - 0.7 - 0.15 - 0.15 - 0.3 = 0.1. That is a system that looks great in a tester and barely breaks even, and a single bad week of fills turns it negative.
Practical use: run the backtest with a stress cost, spread plus 0.3 pip as a flat penalty, and see whether the curve survives. If it dies, the edge is too thin for retail execution and no filter will rescue it. If it survives with a little left, the strategy has margin.
Then replace the guess with the live average once there are 100 or more trades, and revisit it every quarter, since execution quality drifts with account type, size and broker conditions.
How large a stress penalty can your own backtests take before the system stops working?
Do the arithmetic once, with real numbers. Gross edge 1.4 pips per trade. Average spread paid 0.7. Entry slippage averaging 0.15. Exit slippage on stops averaging 0.3 across the losing half, so 0.15 per trade overall. Commission 0.3 pip equivalent. Net is 1.4 - 0.7 - 0.15 - 0.15 - 0.3 = 0.1. That is a system that looks great in a tester and barely breaks even, and a single bad week of fills turns it negative.
Practical use: run the backtest with a stress cost, spread plus 0.3 pip as a flat penalty, and see whether the curve survives. If it dies, the edge is too thin for retail execution and no filter will rescue it. If it survives with a little left, the strategy has margin.
Then replace the guess with the live average once there are 100 or more trades, and revisit it every quarter, since execution quality drifts with account type, size and broker conditions.
How large a stress penalty can your own backtests take before the system stops working?