Use the data vintage that existed at the decision time
An economic observation can refer to an earlier month while being published later and revised again afterward.
Separate observation time from availability time
An economic observation can refer to an earlier month while being published later and revised again afterward. A historical simulation needs to know which version was available when the simulated decision occurred. Today’s downloaded history may contain revisions that a past observer could not have seen. ALFRED archives versions of economic data to support this kind of distinction.
In this invented publication timeline, the January value is 100 from February 5 until its March 5 revision to 103. A February 10 simulation cannot use the later revision. No intraday release timing is assumed.
Open full-size diagram ↗Build an invented revision timeline
Suppose a January observation is first published on February 5 as 100, then revised on March 5 to 103. A hypothetical decision on February 10 can use the published value 100, but not the later value 103. Both numbers refer to January. Selecting rows only by reference month would fail to separate the information available on those two dates.
Worked example: First published January value: 100; Revised January value: 103. Invented values; units follow the row labels and article.
Open full-size diagram ↗Treat exact release timing as another input
A date alone may be insufficient for intraday work. If a release becomes available after the simulated decision time, using it earlier introduces unavailable information even on the same calendar day. Check what the source’s vintage dates actually represent and whether exact publication timestamps are supplied. Do not invent time precision absent from the archive.
Calculation checklist: February 5 publication: January value = 100; February 10 decision: Only 100 available; March 5 revision: January value = 103. Preserve these definitions when reproducing the example.
Open full-size diagram ↗Save an audit trail with the experiment
Keep series ID, reference period, vintage selection, download timestamp and the rule linking publication to decision time. Preserve the original extract so later revisions do not silently change a rerun. Comparing latest-vintage and historical-vintage results can reveal sensitivity to revisions, but this check does not by itself validate a model or prove that its historical performance can be repeated.
Sources and conventions
Examples are hypothetical. Worked examples state their formulas and assumptions in the article; our methodology explains the related calculators. Broker contract specifications determine actual lot, margin and execution conditions. These sources document definitions and data; a source link does not imply endorsement of SanForex or a trading service.
These materials are prepared with AI assistance and checked against worked examples and deterministic calculation tests where applicable. They are not an expert suitability assessment or investment advice.
Open trading journal tool ↗