Backtesting mistakes to avoid before automating a trading strategy

Look-ahead bias occurs when a decision uses information that would not have been known at the time. Examples include using a completed higher-timeframe bar before it closed, entering at the same bar's ideal price after calculating with its close, or referencing a revised data series.

Signals should be evaluated using only information available at the decision timestamp. Intrabar assumptions need special care when the stop and target are both touched inside one historical bar.

Backtests often assume immediate fills at the signal price. Live orders face spread, slippage, queue position, partial fills, rejected orders, and commissions. Futures strategies also encounter contract rollover and changing liquidity.

Use conservative costs and test worse scenarios. A strategy that remains viable after higher costs and delayed entries is more robust than one that depends on perfect fills.

Trying many parameters and selecting the best result can fit random noise. The more combinations tested, the more likely one will look successful by chance.

Keep the rule set understandable, reserve data for out-of-sample testing, examine neighbouring parameter values, and prefer stable performance over one narrow optimum. Document every change so later results are not mistaken for independent confirmation.