How GDP revisions work
The Bureau of Economic Analysis (BEA) publishes GDP on a schedule:Advance estimate (~30 days)
First look at the quarter. Based on incomplete survey data. This is the number that moves markets.
Second estimate (~60 days)
More survey responses incorporated. Typically moves the number by a few billion.
Querying revisions
Pass a specific observation date to see every published value:| Published | Value | What changed |
|---|---|---|
| 2023-10-26 | 22,491.567 | Advance estimate |
| 2023-11-29 | 22,506.365 | +$14.8B |
| 2023-12-21 | 22,490.692 | -$15.7B |
| 2024-09-26 | 22,780.933 | +$290.2B (annual revision) |
| 2025-09-25 | 22,840.989 | +$60.1B (annual revision) |
Which series get revised?
| Series | Revised? | Pattern |
|---|---|---|
| GDPC1 (GDP) | Yes | 3 estimates + annual revisions |
| CPIAUCSL (CPI) | Rarely | Seasonal factors updated each February |
| UNRATE (Unemployment) | Yes | Monthly revisions, annual benchmark |
| PAYEMS (Payrolls) | Yes | Two monthly revisions + annual benchmark |
| FEDFUNDS (Fed Funds) | No | Set by the Fed, published once |
| Treasury yields | No | Market rates, final on the day |
Revisions + point-in-time
Revisions tell you how a data point changed. Point-in-time queries tell you what was known when. Use both:as_of for backtesting, revisions for understanding how much a number moved.