Publication lag is the interval between the date or period that a dataset describes and the time when that information becomes publicly available. In official statistics this is commonly treated as a dimension of timeliness: a monthly observation may describe activity during one month but not be disseminated until days or weeks after that month has ended. The important distinction is that an observation’s reference date is not necessarily its information-availability date.
The timestamp that matters depends on the question
For descriptive analysis, organizing data by reference period may be entirely appropriate. For historical trading research, event studies or signal construction, however, the relevant timestamp is usually the first time the information was publicly available. Suppose a value describes March but is first released in April. A backtest that allows the March value to influence a March decision has used information that a real participant could not yet have observed. Publication lag is therefore not itself look-ahead bias, but ignoring it can create look-ahead bias.
The lag is not necessarily a fixed number of calendar days. It can vary with the dataset, reporting frequency, compilation process, business-day calendar and release schedule. Researchers should therefore prefer actual historical release metadata where available rather than mechanically shifting every observation by an assumed constant delay. When a release has a specified publication time, date-only alignment can also be too coarse for intraday or same-day analysis.
Publication lag is different from revision
Publication lag concerns when an observation first becomes available; data revision concerns how a value changes after it has already been published. Real-time datasets can preserve both dimensions by recording successive vintages: the same reference period may first appear on one date and then receive different values on later dates. A database containing today’s best historical estimates therefore does not necessarily reproduce the information set that existed at an earlier decision date.
There is also a practical distinction between official publication and downstream data delivery. A statistical agency may release a figure at one time while a particular database or vendor makes it available later. For research concerned only with public information, the official release timestamp may be the relevant boundary; for a simulation of a specific executable data pipeline, the later usable timestamp may matter as well.
For point-in-time research, the safest question is not simply “What value belongs to this date?” but “What value had actually been released and was available by this date?” Preserving that distinction is essential when comparing indicators with different release delays, reconstructing historical decisions or testing whether an apparent predictive relationship could genuinely have been observed in real time.