Backtest quality: what a ninety-nine percent label really means
Posted: Thu Oct 29, 2026 6:59 am
Platforms often report a modelling quality figure for a backtest, and a high percentage can create false confidence. The number describes how much of the test used tick data that the platform considered real, not whether the simulated trading resembles live conditions.
The fine print is revealing. In some cases, a figure of ninety-nine percent only means that the data was complete from the tick source, which may itself be generated from minute bars through interpolation.
What matters more is whether the test accounts for variable spreads, commissions, slippage, rejected orders, and realistic latency. A strategy that earns two pips per trade in a perfect simulation can lose money once those factors are applied.
A more honest approach is to run the test on real tick data with recorded spreads, then add pessimistic assumptions for slippage and delay, and see whether the edge survives.
If it does, there is a decent basis for going forward. If it does not, the strategy has no edge to speak of, whatever the quality label says.
Running a walk-forward test, where the strategy is optimised on one period and tested on the next without changes, adds another layer of honesty. Results that hold up out of sample deserve far more trust than those fitted to a single block of history.
What checks do you run beyond the quality percentage before trusting a backtest?
The fine print is revealing. In some cases, a figure of ninety-nine percent only means that the data was complete from the tick source, which may itself be generated from minute bars through interpolation.
What matters more is whether the test accounts for variable spreads, commissions, slippage, rejected orders, and realistic latency. A strategy that earns two pips per trade in a perfect simulation can lose money once those factors are applied.
A more honest approach is to run the test on real tick data with recorded spreads, then add pessimistic assumptions for slippage and delay, and see whether the edge survives.
If it does, there is a decent basis for going forward. If it does not, the strategy has no edge to speak of, whatever the quality label says.
Running a walk-forward test, where the strategy is optimised on one period and tested on the next without changes, adds another layer of honesty. Results that hold up out of sample deserve far more trust than those fitted to a single block of history.
What checks do you run beyond the quality percentage before trusting a backtest?