Backtesting applies a strategy’s rules to historical data and records the trades it would have made. The output is a simulated equity curve and the statistics that describe it.
The result is only as honest as the assumptions behind it, and most backtests are dishonest in ways their authors don’t notice.
The steps
Define the rules precisely enough that a machine can execute them without judgement. Source point-in-time data — prices and any other inputs as they were known at each moment, not as revised later. Simulate execution with realistic fees, slippage and fill assumptions. Run, record, evaluate.
The errors that inflate results
Look-ahead bias — using information not yet available at decision time. Survivorship bias — testing only on assets that still exist. Overfitting — tuning parameters until the strategy fits the historical noise perfectly and predicts nothing. Ignoring costs — the most common one, and enough on its own to turn a profitable backtest into a losing strategy.
What a backtest can tell you
That a strategy is not obviously broken. That its return profile — win rate, drawdown shape, trade frequency — is roughly what you expected. That it survives a reasonable range of costs.
What it can’t
That the strategy will work. Every strategy that has ever failed live had a backtest that said it wouldn’t. Walk-forward testing narrows the gap; only a live, dated track record closes it.