TradeHouse Fidelity Lab · Guides
Trading strategy robustness tests
A robust strategy keeps most of its performance when the conditions of the test change a little. A fragile one only works on the exact settings, period and cost assumptions it was tuned on. Robustness tests deliberately change those conditions and watch what happens.
The core tests
| Test | What you change | Warning sign |
|---|---|---|
| Parameter sensitivity | Each parameter moved one or two steps either side of the chosen value | Performance collapses away from one exact value |
| Cost stress | Fees and slippage raised, for example doubled | The edge disappears with a modest cost increase |
| Sub-period split | Results by year, quarter or regime | Most of the profit comes from one short period |
| Cross-market | Same rule on related instruments or venues | Works on one ticker only, with no structural reason |
| Entry and exit delay | Fills shifted by one bar or a few seconds | A small delay removes the edge, which suggests timing leakage |
| Resampling | Trade order reshuffled or trades bootstrapped | Wide spread of drawdowns; see Monte Carlo backtests |
| Out-of-sample | A frozen rule scored on unseen data | In-sample edge does not reproduce; see out-of-sample testing |
Robustness is not the same as validity
A strategy can be robust to every perturbation above and still be built on contaminated data. If every trade was filled at a stale quote, doubling the fee will not reveal it. Data checks come first; robustness tests come second.
Where Fidelity Lab fits
Fidelity Lab covers the parts that are hardest to do by hand on your own data: 11 contamination detectors (look-ahead, oracle lag, crossed books, sign errors, hindsight selection, mark tautology and more), cost re-netting from the venue's fee schedule, a frozen-threshold walk-forward and multiplicity control. Parameter-sensitivity sweeps still belong in your backtester, because only it can re-run the rule.
Try it on your own trades. The free Fidelity Lab tier runs one report a month on up to 5,000 legs: all 11 contamination detectors, cost re-netting and the walk-forward verdict. Run a free report or see pricing.
Frequently asked questions
How much performance loss is acceptable under stress?
There is no universal number. What matters is a gradual decline rather than a cliff, and that the strategy stays positive after realistic costs.
Is a robustness test the same as a walk-forward test?
Walk-forward is one robustness test, the one focused on time. A full set also varies parameters, costs and markets.
Related guides
- Backtest overfitting: a practical guide
- Walk-forward validation vs cross-validation
- Backtest validation tool: what it should check
- Probability of backtest overfitting (PBO)
- Trading strategy audit service
- Overfitting in backtrader
- Monte Carlo backtests
- Look-ahead bias in backtests
- Survivorship bias in backtests
- Data snooping in trading
- Out-of-sample testing in trading
- What is backtest overfitting?
- Deflated Sharpe ratio calculator
- Fidelity Lab
Educational content, not investment advice. Updated 2026-10-28.