Data Revisions and Vintages

Government statistics agencies almost never get the final number right on the first try. GDP, employment, and inflation reports are released as preliminary estimates built from partial, incoming data, then revised — often more than once — as fuller surveys and tax records arrive weeks or months later. Each release date's version of history is called a vintage, and the same quarter's GDP can look meaningfully different depending on which vintage you're reading.

Why it matters for trading and policy

Markets react instantly to a first print, but that print can later be revised away entirely. A "surprisingly strong" jobs report can be quietly revised down the following month with far less market attention than the original headline got. Backtests that use only the latest, fully revised data (rather than the vintage that was actually available on the day) can look far more accurate than any real-time strategy could have been, because they are cheating with hindsight.

Worked example

The first estimate of a quarter's GDP growth is released at +2.8% annualized, and equities and bond yields both move on the headline. Two months later, the second estimate (built on more complete trade and inventory data) revises it down to +1.9%, and a further revision a year later, once annual benchmark data arrives, settles at +2.3%. A researcher testing a rates strategy against "actual" GDP growth of 2.3% would have back-tested on information that didn't exist on the trading day, overstating how predictable that reaction should have been.