Most trading rules have numbers in them: a 14-bar RSI, a 50-bar moving average, a 20-bar breakout. A robust strategy keeps working when those numbers move a little. A fragile one only works at one exact setting, which usually means it was fitted to noise.
Plateaus and peaks
Imagine plotting a strategy’s return for every possible lookback length.
- A plateau is a broad region where many nearby settings give similar results. That suggests the rule captures something real about the market.
- A sharp peak is a single setting that stands out while its neighbors do poorly. That is what optimization finds when it fits randomness.
If you picked your parameters by trying many values and keeping the best, you are more likely to be sitting on a peak. See overfitting in backtesting.
How the nearby-parameters check works
Backte.st’s Nearby parameters check shifts every indicator period in your rules by ±10% and ±20% at the same time and re-runs the backtest for each. It reports the worst of those results, and you can see the return for each scaling in the report.
For example, a 20-bar and 50-bar moving-average rule is also tested at roughly 16/40, 18/45, 22/55 and 24/60 bars.
Reading the result
- All nearby settings profitable: the rule is not hanging on a single number.
- One direction fails: the strategy may depend on a particular speed. Faster or slower settings might be worth testing on their own.
- Most nearby settings fail: treat the original result with heavy suspicion.
Other ways to build robustness
- Use round, conventional numbers unless you have a reason not to. They are less likely to be the product of a search.
- Prefer fewer parameters. Every extra one is another dimension to overfit.
- Test across markets and timeframes. Robust ideas tend to transfer, at least partly.
Try a moving-average trend strategy, such as the 50/200 golden cross, in the lab and open the robustness checks to see the return at each scaling.