Parameter sensitivity is the test that asks whether a strategy still works when the settings move a little. If the backtest only survives one exact combination, the edge may be a memory of the sample, not a property of the market.
For a prop trader, that matters before the first live trade. The account does not fail because a parameter looked elegant in the optimiser. It fails when a small change in spread, session, volatility, fill, or broker data pushes a fragile system through the loss rules.
What parameter sensitivity means
Parameter sensitivity measures how much a backtest result changes when the strategy settings are nudged around the chosen values. A robust system should have a neighbourhood of acceptable settings, not one isolated dot that collapses when touched.
The setting can be anything the strategy depends on: stop distance, take-profit distance, moving-average length, volatility filter, session window, breakout lookback, risk scale, or entry threshold. The exact type does not matter. The question is whether the result depends on an unrealistically precise setting.
| Pattern | What it suggests | Prop-trader read |
|---|---|---|
| Broad plateau | Nearby settings behave similarly | The idea may be robust enough to test further |
| Needle peak | One setting works and neighbours fail | Possible curve fit or data-snooping result |
| Smooth degradation | Performance weakens gradually | The edge has a tolerable margin of error |
| Chaotic surface | Small changes flip the result | The rules are probably too brittle |
| Cost-sensitive edge | Results break when costs rise | Broker portability is weak |
Why the best optimiser result is usually suspect
An optimiser is built to find the highest score in the search space. That is useful, but it is not proof. The more combinations you test, the easier it is to find one that was lucky on that exact historical sample.
This is the same family of problem as backtest overfitting. Overfitting is not only too many indicators or a beautiful equity curve. It can be one parameter value that happened to line up with the past.
The clean question is not "which setting won?" The clean question is "what happens around the winner?"
If the winning setting is surrounded by similar settings that also survive, the result has a stronger case. If the winner stands alone, it deserves suspicion. A live account will not experience the past again with identical spread, volatility, trade sequence, and fills.
The plateau test
The plateau test is a simple robustness check: map the area around the chosen settings and look for a stable region. You are not trying to make the equity curve prettier. You are trying to find out whether the strategy has breathing room.
For one parameter, the check is straightforward. Move the setting lower and higher, then compare the key outputs. For two parameters, map them together. The table should not be judged only by net profit. A prop backtest needs the path metrics too.
| Metric to inspect | Why it matters |
|---|---|
| Max equity drawdown | Shows whether the rule path still survives |
| Losing streaks | Reveals whether a slight change clusters losses |
| Average trade duration | Shows whether risk stays open longer |
| Trade frequency | Shows whether the setting creates cost pressure |
| Profit factor | Checks whether gross wins still cover gross losses |
| Out-of-sample behaviour | Tests whether the setting survived unseen data |
The best-looking return can be the wrong choice if it adds drawdown, clustering, or holding time that the account cannot tolerate. For prop trading, a calmer plateau often matters more than the tallest point on the chart.
Parameter sensitivity is a cost test
Parameter sensitivity is also a hidden cost test. A system that only works at one exact setting may also only work under one exact cost model.
That is why spread, commission, slippage, and swap belong in the robustness check. If a small cost change destroys the result, the parameter was probably tuned to an idealised path. That is especially dangerous for strategies with tight stops, short targets, high frequency, or overnight holds.
The same logic sits behind why the same cTrader backtest changes across brokers. Broker data and costs do not need to be dramatic to matter. A fragile setting gives small differences too much power.
How to test settings without fooling yourself
A useful parameter test separates discovery from validation. The discovery phase can explore. The validation phase has to be stricter.
Use this order:
- Define the parameter range before looking for the winner.
- Optimise on the in-sample window only.
- Inspect the nearby settings, not just the top result.
- Prefer stable regions over isolated peaks.
- Test the chosen region on out-of-sample data.
- Re-check the result with realistic costs and floating equity drawdown.
This is where out-of-sample testing and walk-forward testing become practical rather than academic. A parameter that keeps working as the window rolls forward has a stronger claim than one selected once and protected from further inspection.
The prop-firm layer
Prop accounts add a second filter: the setting must survive the firm's rules, not just the backtest report. A parameter can improve profit while making the account less suitable for a daily loss limit, max loss floor, or weekend-hold constraint.
That is why settings should be read through account behaviour:
| Question | What a weak setting hides |
|---|---|
| Does it raise open risk? | Floating drawdown can break the account before trade close |
| Does it increase overlap? | Several trades may become one correlated loss |
| Does it hold longer? | Swap, weekend gaps, and news exposure become larger |
| Does it trade faster? | Spread, commission, and slippage get more chances to bite |
| Does it depend on one market regime? | The rule may fail when volatility changes |
The broader rule context belongs on the funding page. The point is simple: a setting is not robust if it only looks good in a closed-trade summary while the equity path carries too much rule pressure.
Where realbacktesting draws the line
realbacktesting is a trading-software studio for cTrader: prop-firm cBots, indicators, and plugins built around verifiable backtests rather than screenshots.
The public methodology uses intrabar M1 execution, cTrader broker M1 bars + tick-measured spread from 2021-2026, real per-symbol spread, real commission, swap, 1 bps slippage, and an 80,000 EUR model base. The drawdown ceiling is the worst floating-equity low and is checked against a 30% out-of-sample hold-out. The method is laid out on the methodology page.
Those details do not make parameter sensitivity disappear. They make it harder to hide. If a setting only works after costs are softened, out-of-sample is ignored, or floating drawdown is replaced by a closed-balance summary, the backtest is answering the wrong question.
Frequently asked
What is parameter sensitivity in backtesting?
Parameter sensitivity in backtesting is how much the result changes when strategy settings are adjusted. A robust strategy should remain usable across nearby settings instead of depending on one exact value.
Is the best optimiser setting the best trading setting?
Not necessarily. The best optimiser setting may be the luckiest setting in the tested sample. For a prop account, a stable setting inside a broad plateau is usually more informative than the highest point in the optimiser table.
How does parameter sensitivity relate to overfitting?
Parameter sensitivity is one way overfitting reveals itself. If tiny setting changes destroy the result, the strategy may have fitted historical noise rather than a repeatable market behaviour.
Should prop traders optimise for return or drawdown?
A prop backtest should treat drawdown and rule survival as first-class outputs. Return matters, but a setting that improves return while increasing fragile floating drawdown can be worse for a prop account.
The stubborn takeaway
The setting that survives is rarely the prettiest number in the optimiser. It is the one with room around it, because live trading is all room around the backtest.