Look-ahead bias

Look-ahead bias is using information in a backtest that the strategy could not actually have had at the moment it made a decision. It’s the most common way backtests overstate results, and it’s dangerous precisely because it makes performance look better than it was.

It creeps in through same-bar peeking (acting on a bar’s close while assuming you traded earlier), computing indicators or signals across the full dataset at once, or joining data without respecting when it was really available. The result is a strategy that performed well on data it secretly knew in advance, and that falls apart live, where the future is genuinely unknown.

Event-driven backtesting reduces look-ahead bias by construction: the strategy is handed events in sequence and simply isn’t given data from the future to peek at.