forex-scalping.com

How to backtest sweeps honestly with a small sample

Everyone has a few screenshots of perfect sweeps. That is not a backtest, it is a highlight reel. To test the idea honestly, write down the exact rules first: what level counts, what timeframe, what confirmation, where the stop and target go. Then scroll through the chart bar by bar, hiding the future, and record every occurrence, including the ugly ones. Aim for at least fifty samples before drawing any conclusion, and split them into two groups so you can check whether the result holds on data you did not use to shape the rules. Track win rate, average R, and how many trades you skipped because they were unclear. If the rules need judgment calls, note how often you made them. Be suspicious if your results look too good, because hindsight bias creeps into manual testing easily. A modest, repeatable edge that survives both halves of the sample is worth more than a spectacular result you cannot reproduce live.

Splitting the sample into two halves is the part most people skip, and it's the most useful. One extra step: shuffle which half you build on. If you shape the rules on the older half and test on the newer one, then do it the other way round, you get a sense of whether the result depends on one market period. Sweeps behave differently in trending and ranging months, so a result that only shows up in one half is a warning. Fifty samples is a starting point too. Split by session or pair, the groups get small fast, so read them loosely.

I'd also log the trades you would have been unsure about in real time as their own group. In a manual backtest it's easy to mark an ambiguous sweep as valid when you can see the result, even with the future hidden, because the shape still tips you off. Keeping a separate tag for borderline cases lets you check whether the edge comes from the clean examples or from the ones you'd struggle to call live. If the borderline group is what's carrying the results, that's a sign the rules need tightening before any of it goes near a real account.