Point-in-time data

Point-in-time (PIT) data preserves what a dataset looked like at a specific moment in the past, including values that were later revised, restated, or corrected. This differs from an ordinary historical database, which typically stores only the current, most up-to-date figures and silently overwrites what an earlier vintage actually showed.

The distinction matters most for data that gets revised after first release, such as GDP, employment, or company fundamentals restated after earnings corrections. A backtest run on data that has already been revised implicitly assumes a trader on that historical date could see numbers that were not actually available yet, a form of look-ahead bias that produces inflated, unrealistic results. PIT data avoids this by tagging every value with both the period it describes and the date it was known.

Point-in-time data is essential infrastructure for rigorous backtesting of quantitative strategies and for research using alternative data, where fields are frequently revised or backfilled after initial delivery. It is closely related to guarding against survivorship bias, which similarly inflates historical results, in that case by excluding entities that no longer exist rather than by using revised figures. A common pitfall is assuming a vendor's "historical" dataset is automatically point-in-time; many are not unless explicitly built and documented as such.