On October 29, 2026, the Bureau of Economic Analysis reported that real gross domestic product rose at an annual rate of 3.4 percent in the third quarter. The number arrived with a label attached: advance estimate. BEA’s own release language says the figure rests on source data that are incomplete or subject to further revision by the source agency. The second estimate is scheduled for November 25, 2026.

That label is not boilerplate. It is a quantified warning, and BEA publishes the numbers behind it. The useful question for anyone reading a GDP headline is not whether the advance print will change. It will, sometimes. The useful question is how much movement is ordinary, and which comparison actually measures the draft’s noise.

Three prints, one quarter

BEA produces three current quarterly vintages of GDP. The advance estimate lands near the end of the first month after the quarter closes. The second estimate follows near the end of the second month, and the third near the end of the third month. BEA describes the advance as based on incomplete source data, with the second and third built on more detailed and more comprehensive data as they become available.

The Q3 2026 calendar shows the rhythm plainly. The advance release came October 29. The second estimate and preliminary corporate profits are set for November 25. The third estimate, along with GDP by industry and revised corporate profits, is scheduled for December 23.

Each print is a repeated estimate of the same quarter, built from a different mix of survey returns, trade data, and projections where data are missing. BEA’s 2018 Survey of Current Business article on revisions puts the underlying problem directly: the earliest GDP and GDI estimates are based on partial and preliminary data and on trend projections when data are not available. A revision, in this framework, is not a correction of an error. It is a better-informed estimate of a quantity no one can observe directly.

What BEA’s own revision table says

BEA publishes average revisions to quarterly percent changes in real GDP, without regard to sign, across vintages. Two recent vintages of that table are worth setting side by side.

Vintage comparison 1993–2017 (pp, annual rate) 1996–2022 (pp, annual rate)
Advance to second 0.5 0.5
Advance to third 0.6 0.6
Second to third 0.2 0.3
Advance to latest 1.2 —

The stability across the two windows is the first thing to notice. The advance-to-second average revision sits at half a percentage point in both. The advance-to-third sits at six-tenths. The second-to-third step is smaller, a fifth to three-tenths of a point, which fits the pattern: most of the missing information gets filled in by the second print, and the third print refines rather than rebuilds.

These are average magnitudes, not signed averages. BEA’s mean revisions, which allow positive and negative values to offset, are much smaller. In the 2018 paper, mean revisions for 1993–2015 ranged between roughly −0.15 and +0.14 percentage point. Revisions run in both directions and partly cancel. A reader looking for a systematic bias in the advance estimate will not find one in these tables.

The 1.2-point figure is a different animal

The advance-to-latest average revision of 1.2 percentage points gets quoted often, and it is routinely misread as the advance estimate’s typical error. It is not. BEA explains why in the same table’s note: the larger average revision from advance to latest reflects the fact that periodic comprehensive updates include major statistical and methodological improvements.

That distinction matters for how a reader interprets a moving number. The advance-to-second and advance-to-third comparisons measure how much the estimate of a given quarter shifts as source data arrive. The advance-to-latest comparison folds in comprehensive revisions, which occur roughly every five years and can change definitions, classifications, and methods. The 2018 paper gives a concrete example: the 2013 comprehensive revision recognized research and development and artistic originals as investment, and the dollar level of GDP was revised up an average of 3.1 percent over 1993–2012.

So when a quarterly figure is revised by a few tenths between prints, that is the draft being finished. When a long historical series shifts after a comprehensive update, that is the measuring instrument being recalibrated. Conflating the two turns a routine data update into a story about error, and it turns a methodological revision into a story about a quarter.

A band around the headline, labeled honestly

The advance-to-second average revision of 0.5 percentage point gives a rough sense of scale. If the advance print is 3.4 percent, a reader should not be surprised by a second print a few tenths away in either direction. That is the normal working range of the draft.

But the average is not a confidence interval, and it is not a forecast for this quarter. BEA’s 2018 paper shows how wide the underlying distribution is. Using the standard deviation of the advance-to-second revision, 0.79 percentage point, the paper constructs a 90 percent interval for the revision spanning roughly −1.24 to +1.36 percentage points. The paper applies the same logic to the advance estimate for the first quarter of 2016, 0.54 percent, and notes that a 90 percent interval for the second estimate ran from about −0.7 percent to 1.90 percent. The actual second estimate was 0.84 percent.

That example is the right mental model. The average revision tells you the typical size of the move. The distribution tells you how far an individual quarter can travel. A band built from the average is a scale marker, not a prediction, and it should be labeled that way in any chart that uses it.

Why the components move

The headline revision is the sum of revisions to components, and the components draw on different source data with different arrival schedules. The Q3 2026 advance release names several of them. Inventory investment leaned on Census Bureau inventory book value data. Equipment investment leaned on trade data and the Census Bureau’s Advance Economic Indicators Report for September. Consumer spending on health care leaned on Bureau of Labor Statistics Current Employment Statistics, and spending on goods leaned on the Census Bureau’s Monthly Retail Trade Survey.

Each of those inputs gets revised on its own timetable. A retail survey return that arrives after the advance cutoff shows up in the second print. A trade figure that is initially estimated gets replaced by a fuller report. The advance estimate is not a single guess but a bundle of estimates at different stages of completeness, and the second print is where most of the bundle gets upgraded.

This is also why the advance release includes a Key Source Data and Assumptions file. It lists what BEA assumed where data were missing, which is the part of the draft most likely to change.

A rule of thumb for the next headline

Read the advance estimate as a directional signal and a rough magnitude, not a verdict. Three practices follow from BEA’s own documentation.

First, treat a move smaller than roughly the advance-to-second average revision as within the draft’s normal range. At half a percentage point, a shift of a tenth or two between the advance and second prints is ordinary. A headline built on that shift is a headline built on the revision schedule, not on the economy.

Second, wait for the second and third prints before treating a decimal change as news. The second estimate is where most source data arrive. The third refines. A story that requires the advance number to be exactly right is a story that will need rewriting in four weeks.

Third, check the vintage before comparing across quarters. BEA publishes a Vintage History of Quarterly GDP and GDI Estimates file, and it notes that with each new release the prior data are superseded and the original figures move to the Data Archive. Comparing a current-quarter figure to a figure quoted from an older article can mean comparing two different vintages of the same series. The archive exists precisely so that comparison can be done correctly.

What the draft is for

The advance estimate exists because policymakers and businesses need a timely read on the quarter, and BEA’s reliability research concludes that the early estimates present the same general picture of economic activity as later vintages: long-term growth, the broad shape of the business cycle, and the trends in major components. The revisions do not substantively change that picture.

What the revisions do change is the decimal. The advance print is a draft with a documented revision pattern, and the pattern is the useful part. A reader who knows the advance-to-second average is half a point, and that the distribution is wider than the average, can read the next headline without mistaking a routine update for a reversal or a methodological recalibration for a quarterly miss.

Questions readers ask

Does a downward revision mean the advance estimate was wrong? Not in the sense of an error. BEA treats each vintage as a repeated estimate using more complete data. The mean revisions in BEA’s tables are small and offsetting, which is what you would expect if the revisions were filling in missing information rather than correcting a persistent bias.

Why is the advance-to-latest revision so much larger? Because it includes comprehensive updates that change methods and definitions, not just the arrival of source data for one quarter. BEA states this directly in the note to its revision table.

Can I build a confidence interval from the average revision? No. The average is a measure of typical magnitude. BEA’s 2018 paper shows the advance-to-second revision has a standard deviation of 0.79 percentage point, which produces a much wider 90 percent interval than the average alone suggests. Any band should be labeled as an average-based scale marker, not a statistical interval for a specific quarter.

Where do I find the original advance figure after the second estimate is released? In BEA’s Data Archive. The current release page will show the latest data, and the note on the release states that the original figures move to the archive with the next release.

When is the next print for Q3 2026? The second estimate and preliminary corporate profits are scheduled for November 25, 2026. The third estimate is scheduled for December 23, 2026.

Sources

The Advance Estimate Is a Draft: How Far GDP’s Three Prints Move Before the Number Settles
Roberta Meyer Blogging