Gross domestic product in the United States is estimated by the Bureau of Economic Analysis in three monthly vintages per quarter — advance, second, and third estimates — and then re-benchmarked for decades afterward. The advance estimate lands about 30 days after a quarter ends, built while most source data is incomplete, and the BEA's own evaluations show the advance estimate captures the direction of growth in the vast majority of quarters while missing turning points at the margins. The 2024 example: third-quarter 2024 growth was first reported at an annualized 2.8 percent in the advance release of October 2024. The number readers quote is always a snapshot of what was knowable at a date, not a final truth.
Where do the numbers come from?
GDP is measured three ways — production, income, and expenditure — and the expenditure approach publishes monthly: consumption, investment, government spending, net exports. The source data is a patchwork: monthly retail sales and durable-goods orders from the Census Bureau, monthly payroll and hours from the Bureau of Labor Statistics, corporate profits from tax records that arrive a quarter late, trade balances that take months to settle. The advance estimate uses whatever fraction of a quarter's data exists — often under half of the survey coverage the third estimate will have — plus BEA staff judgment and trend extrapolation for the rest. Health-care services, the largest consumption component, is a standing extrapolation problem: its source data is annual, interpolated, and revised hard when benchmark numbers land.
Why three estimates per quarter?
Because data arrives on its own schedule. The second estimate, about 30 days after the advance, folds in the first complete month of late survey responses and early trade figures; the third, 30 days later, adds corporate profits and fuller source data. Typical revision sizes are modest: the BEA's mean absolute revision between advance and third estimates has run around 0.5 percentage point on the quarterly growth rate over recent decades, with over half of quarters revised within 0.3 point — but the distribution has fat tails around recessions, when new data arrives fastest and extrapolations fail worst, as 2008's sequence showed.
What happens after the quarter closes?
Annual updates each summer incorporate newly complete source data — tax returns, census benchmarks — and methodological improvements, routinely rewriting several years of history. Comprehensive benchmark revisions, roughly every five years, rebase everything; the 2023-2024 comprehensive updates revised levels of GDP upward noticeably for 2017-2022 by capturing intangible investment better. And in 2025 the BEA changed its release calendar to align with new-source timing, the latest in a series of transparency measures that publish revision histories alongside each release so anyone can score the record.
Why does nominal and real matter?
Nominal GDP is dollars spent; real GDP strips price change to measure volume, and the deflator used to do it is itself an estimated chain price index subject to the same revision cycle. The distinction drove the pandemic era's headline confusion: nominal growth of about 10 percent in 2021 was mostly inflation, with real growth near 6 percent. Quarterly rates are also annualized — a 2 percent quarterly growth rate prints as about 8 percent at an annual rate — the convention that most exaggerates small quarterly wiggles, and the one readers most often mistake for a year's growth.
Should anyone trust a first print?
As a first approximation with a published error band, yes; as a decimal-precise fact, no. The honest use, which the BEA itself encourages, treats the vintage structure as information: the advance estimate is a fast reading designed to be superseded, and the revision record — public, scored, and studied — is the statistical system's quality control. The analysis: revisions are not scandal but method, the cost of publishing 30 days after quarter-end instead of waiting years for complete data; the genuine costs concentrate at turning points, where first prints mislead most and policy, markets, and commentary react anyway. What would change the reading is real-time data sources — card transactions, scanner data — shrinking advance-estimate error, an experiment the BEA has begun in limited components.
Frequently asked questions
How often is GDP revised?
Three estimates per quarter — advance, second, third, a month apart — then annual updates each summer and comprehensive benchmark revisions about every five years. A quarter's number is never technically final; it is stabilized by the third estimate in most cases.
How accurate is the advance GDP estimate?
The mean absolute revision from advance to third estimate is roughly half a percentage point on quarterly growth, with most quarters within 0.3 point. Direction is right in most quarters; magnitude at turning points is where errors concentrate.
What is the difference between real and nominal GDP?
Nominal GDP measures spending in current dollars; real GDP adjusts for inflation using chain price indexes to capture volume. Real GDP is the growth measure; nominal matters for debt ratios and dollar comparisons.
Why do quarterly GDP numbers get annualized?
U.S. convention reports quarterly growth as if that rate continued for a year, making quarters comparable with annual rates. A 0.5 percent quarterly growth prints as about 2 percent annualized — which exaggerates one quarter's noise.
For more context, read How the IMF Assembles Its World Economic Outlook.
For more context, read nber recession dating.
For more context, read What Purchasing Managers Indexes Tell Us About the Economy.