Look-ahead Bias
Look-ahead bias is using information that was not available at the time the forecast was made. It can dramatically inflate measured performance.
Definition
Look-ahead bias happens when evaluation uses information from the future, even accidentally. In forecasting, it occurs when you score or construct forecasts using data that was not available at the time of prediction.
Why it matters
Look-ahead bias makes results look far better than reality. It is one of the most common ways backtests and scorecards become misleading.
Common examples
• Using a closing price as the “forecast” when the goal is to evaluate earlier forecasts.
• Backfilling a probability after seeing late breaking news.
• Using final settlement sources when building features for a model.
How to prevent it
• Store strict timestamps for forecasts and data pulls.
• Use fixed evaluation checkpoints.
• Run out of sample tests based on time splits.
Related
Look-ahead bias is closely related to data leakage and to evaluation integrity.