How to read a backtest without fooling yourself

Before looking at the equity curve, write down the instrument, timeframe, date range, spread model, commission, swap, execution assumptions, and whether the data includes real ticks. A result without its assumptions cannot be reproduced or fairly compared.

The test should use information that would have been available at the time of each trade. Entries at a future close, accidental look-ahead, or an unrealistically perfect fill can make a strategy look better than it is.

Win rate answers how often trades were profitable. It does not tell you whether the average win is large enough to pay for the average loss and trading costs. Profit factor compares gross profit with gross loss, but it can be distorted by a small number of large trades.

The period used to create or tune a strategy is in-sample data. A separate out-of-sample period is a more honest check because the rules were not selected against it. Walk-forward testing repeats that idea across several historical windows and shows whether the logic can adapt without being rewritten for every period.

Parameter stability matters too. If only one exact combination works and nearby settings fail, the result may be curve-fit. A robust strategy usually has a sensible region of acceptable settings rather than one magical number.

Monte Carlo tests can reshuffle trade order, vary execution, or perturb outcomes to show how much the result depends on luck. They do not predict the future, but they can expose a system that only survives in its most favourable sequence.

Even a careful backtest cannot model every future market regime, broker outage, spread event, or implementation mistake. Treat it as evidence for a measured experiment, not a promise of returns.