Point-in-time data and vintages
Which vintage of an economic release may a historical strategy actually use?
Use the value that had been released by the simulated decision date. A present-day FRED download may contain revisions published later; ALFRED and FRED real-time parameters expose earlier vintages for the historical check.
A point-in-time series records each value as it was published on a given date. A backtest that reads a later, revised value uses information that was unavailable when the trade was placed.
Release, revision and vintage
An economic observation has a reference period, such as a quarter, and one or more release dates. Statistical agencies revise many first releases as more complete source data arrive. A vintage is the set of values a series showed on one publication date.
A point-in-time dataset retains past vintages together with their release dates. A decision dated d may use a value from a vintage released on or before d. Croushore and Stark assembled a real-time data set for macroeconomists on this principle, and Orphanides compared monetary policy rules evaluated on real-time data with the same rules evaluated on revised data.
Value available at decision date d = value in the latest vintage released on or before dRetrieving past vintages from ALFRED
The Federal Reserve Bank of St. Louis publishes FRED, which returns the current values of its series, and ALFRED, an archive of vintage versions. In the FRED API, the optional realtime_start and realtime_end parameters take YYYY-MM-DD dates and default to today's date, so a default request returns history as it stands today. Setting both parameters to a past date returns the observations as they had been published on that date. The documentation defines this real-time period as closed at both boundaries.
“ALFRED makes it possible to gather data as reported by a source on past dates in history.”
A revised growth signal
Suppose a rule buys an equity index when reported quarterly GDP growth exceeds 2.0%. The first release reports 2.4%, so the rule enters. A later vintage revises the figure to 1.6%, a revision of -0.8 percentage points. Over the holding period the index returns -3.0%.
A point-in-time backtest records the trade and its -3.0% return. A backtest on the revised series reads 1.6%, skips the trade and records 0%. On this decision the two tests differ by 3.0 percentage points, and the revised-data result looks better because it avoided a trade that a live trader acting on the first release would have taken.
Revision = value in a later vintage − value in the first releaseIn Stochastly
The alternative-data lead-lag, momentum, mean-reversion and cointegration nodes align an external series on the date it became available. That date comes from a vintage column in the file or, when the column is absent, from the observation date plus a declared publication lag; a series with neither is rejected before alignment. Values are carried forward causally.
When a file holds several revisions of one observation, these nodes align the latest revision at its own vintage date and report how many earlier revisions were overwritten. They do not reconstruct a full history of vintages. A point-in-time validator is listed as roadmap in the node catalogue.
Frequently asked questions
What is a data vintage?
The set of values a series showed on one publication date. A later vintage can hold revised values for the same reference periods.
Which data need vintages?
Any series that its provider revises after first publication, such as macroeconomic releases and restated company fundamentals. Adjusted price histories also change after corporate actions.
Can a fixed publication lag replace vintages?
A lag moves the first availability date. It keeps the revised value, so the backtest can still read a number that differs from the first release.
Sources
Federal Reserve Bank of St. Louis (n.d.). ALFRED Help. ALFRED, ArchivaL Federal Reserve Economic Data.
Federal Reserve Bank of St. Louis (n.d.). Real-Time Periods. FRED API documentation.
Croushore and Stark (2001). A real-time data set for macroeconomists. Journal of Econometrics 105(1), 111-130.
Orphanides (2001). Monetary Policy Rules Based on Real-Time Data. American Economic Review 91(4), 964-985.
Croushore (2011). Frontiers of Real-Time Data Analysis. Journal of Economic Literature 49(1), 72-100.