Seven knobs on a one-minute chart is a noise-fitting machine
Posted: Sat Oct 31, 2026 8:20 am
Opinion: most over-optimised scalpers share the same recipe. An RSI period, two moving average lengths, a session filter, an ATR multiple for the stop, another for the target and a spread cap. Seven free parameters on one-minute data, where a month offers tens of thousands of bars. That is plenty of room to fit noise, and the optimiser will happily find the combination that looks perfect.
One quick sanity test is to look at the neighbours. If period 14 makes +300 pips while 13 and 15 lose, the peak is a fluke. A real edge shows a plateau, where nearby values behave similarly even if none is the best.
Other ways to cut the risk:
- fix parameters that have a structural reason, such as sizing the stop from volatility rather than leaving it free
- run the final settings on a different pair without re-tuning
- drop each rule in turn and see which ones actually earn their place
Expect the transfer to be weaker on the second pair, but a total collapse is a bad sign. The goal is fewer knobs, not a prettier curve.
How many free parameters is too many for you, and what do you do to cut them back?
One quick sanity test is to look at the neighbours. If period 14 makes +300 pips while 13 and 15 lose, the peak is a fluke. A real edge shows a plateau, where nearby values behave similarly even if none is the best.
Other ways to cut the risk:
- fix parameters that have a structural reason, such as sizing the stop from volatility rather than leaving it free
- run the final settings on a different pair without re-tuning
- drop each rule in turn and see which ones actually earn their place
Expect the transfer to be weaker on the second pair, but a total collapse is a bad sign. The goal is fewer knobs, not a prettier curve.
How many free parameters is too many for you, and what do you do to cut them back?