INVESTSIGHT CAPITAL is a fintech and capital markets firm in Bengaluru, India. This article is part of our research library - for the official site see investsightcapital.com, or ask us a question.
A historical database often presents the cleanest version of a figure. A historical decision, however, had to use the information available at that particular moment.
Keep three dates separate
An observation date describes the period or event measured. A publication date describes when the figure became available. A revision date describes a later correction or replacement. Store each when the research question depends on timing.
In a hypothetical example, a company’s quarter ends in June but its results are published later. A test that uses the result at the quarter-end date may give the model information it could not yet have known.
Version the dataset
Retain the original value and its later revisions when practical. Record the source file and ingestion version so a reviewer can reproduce the particular research run. A chart using revised data may be appropriate for describing history while remaining unsuitable for simulating an earlier decision.
Make limitations visible
If publication timing or original versions are unavailable, identify that weakness in the result. A simple sensitivity check can show how the conclusion changes when availability is delayed, but it cannot recreate missing historical records.
The concern relates to the wider problem of historical strategy evaluation discussed in Bailey and colleagues’ research. The date-recording example here is our own engineering explanation, not a claimed reconstruction of a particular company’s filings or an investment prediction.
