Skip to main content
← All research papers

Explainer · Kresmion Research

How Economic Data Gets Revised: Jobs, CPI and GDP

November 8, 2026 · 12 min read

Published by Kresmion Research. Read our editorial approach and data methodology.

ShareXLinkedInReddit
Macroinflationdata-literacy

See it liveLive central bank policy rates →Free, no account needed.

An economic data revision is a later change to a published figure, made when the agency adds late survey replies, new seasonal factors or a fuller benchmark.

The first estimate of a month's jobs or a quarter's growth is built fast, from incomplete reports, so that it is useful while it is still news. The agencies then replace it on a published schedule as fuller data arrive. This page covers why first prints change, how payrolls, the Consumer Price Index and GDP are each revised, a worked example from the jobs report of 6 November 2026, how to read the revision lines in a release, and why revisions matter to markets as well as the headline. It is descriptive throughout.

Why first prints change

Three mechanisms drive the scheduled revisions to US headline data:

  • Late responses. Surveys keep collecting after the first deadline. A first estimate uses the replies in hand; the next one adds the businesses and agencies that reported later.
  • Seasonal factors. Seasonally adjusted figures remove the usual calendar pattern, and the size of that pattern is estimated from the data itself. New data change that estimate, so adjusted figures for past months can move even when the raw numbers do not.
  • Benchmarks and annual updates. Once a year the sample-based estimates are lined up with far more complete sources, such as tax records or annual business surveys, and recent history is rewritten to match.

These revisions are built into the calendar: the agencies publish their timing in advance and report the size of each one.

Payrolls: two monthly revisions and an annual benchmark

The Bureau of Labor Statistics publishes the first estimate of nonfarm payrolls early in the following month, usually on a Friday, about three weeks after the survey's reference period (the pay period that includes the 12th). The estimate is then "revised twice, before being held constant until the annual benchmarking process": a second estimate a month later and a final sample-based estimate two months after the first. The Employment Situation release gives the reason in one sentence: monthly revisions come from "additional reports received from businesses and government agencies since the last published estimates and from the recalculation of seasonal factors."

The annual benchmark is the larger event. Each year the March level of payroll employment is set against the Quarterly Census of Employment and Wages, a count drawn primarily from the unemployment insurance tax records that nearly all employers file. BLS publishes a preliminary estimate of the benchmark revision in late summer and folds the final one into the data with the January jobs report the following February, revising the months before and after it. Because the monthly sample cannot see businesses opening and closing in time, BLS estimates that part with a model; the benchmark replaces the March estimate, model included, with the full count, and BLS then updates the model's forecasts.

BenchmarkPreliminary estimateFinal revision
March 2025-911,000 not seasonally adjusted (-0.6 percent), published 9 September 2025-862,000 not seasonally adjusted (-0.5 percent), -898,000 seasonally adjusted (-0.6 percent), incorporated 11 February 2026
March 2026-79,000 not seasonally adjusted (-0.1 percent), published 28 August 2026Due with the January 2027 jobs report in February 2027

In its August 2026 release BLS put the annual benchmark revisions of the previous ten years at an absolute average of 0.2 percent of total nonfarm employment, so the March 2025 revision was more than twice the usual size.

Worked example: the jobs report of 6 November 2026

The October 2026 report showed payrolls up 165,000 and carried two revisions. In the release's words, August "was revised up by 53,000, from +133,000 to +186,000", September "was revised up by 31,000, from +29,000 to +60,000", and "employment in August and September combined is 84,000 higher than previously reported."

Each month's path through its three estimates, read from the archived vintages of the payroll series in thousands, seasonally adjusted:

MonthFirst estimateSecond estimateThird estimate
July 2026-23 (7 Aug)+21 (4 Sep)-10 (2 Oct)
August 2026+162 (4 Sep)+133 (2 Oct)+186 (6 Nov)
September 2026+29 (2 Oct)+60 (6 Nov)due 4 Dec
October 2026+165 (6 Nov)due 4 Dec

August shows that revisions do not run in one direction: its first estimate was cut by 29,000 a month later and then raised by 53,000. July's third estimate is now fixed until the next benchmark.

The useful sum on release day is the headline plus the net revision. The level of payroll employment the 6 November report gives for October, 159,293,000, is 249,000 above the level the 2 October report had given for September, 159,044,000: 165,000 of new jobs plus 84,000 of revisions to the two months before.

CPI: final when published, except seasonal adjustment and the chained index

The Consumer Price Index works differently. BLS says revisions to the unadjusted CPI-U and CPI-W "can be costly for the users of these indexes, so these series are final when issued", and it recommends unadjusted indexes for escalation clauses. Seasonally adjusted indexes are another matter: "Each year with the release of the January CPI, seasonal adjustment factors are recalculated", and the new factors "are used to revise the previous 5 years of seasonally adjusted data." The chained C-CPI-U is the one series published as preliminary, revised quarterly and final 10 to 12 months later. The CPI explainer covers the index itself.

The seasonal update released on 13 February 2026, with the January 2026 index, shows what that means in practice. Three monthly changes from 2025, computed from the published seasonally adjusted index levels:

MonthFirst publishedAfter the February 2026 update
January 2025+0.5% (12 Feb 2025)+0.4%
March 2025-0.1% (10 Apr 2025)0.0%
August 2025+0.4% (11 Sep 2025)+0.3%

The unadjusted index levels for those months did not change. The same rules apply to the September 2026 index, released on 14 October 2026: BLS reported prices up 0.6 percent on the month, seasonally adjusted, and 3.7 percent over the year before seasonal adjustment, with the index excluding food and energy up 0.3 percent on the month and 2.6 percent over the year. The 12-month figures are computed from unadjusted indexes and stay as published; the 0.6 percent monthly change is a seasonally adjusted figure, so it is open to revision with each February update for the next five years, the first coming with the January 2027 index.

GDP: advance, second, third, then annual updates

The Bureau of Economic Analysis publishes three estimates of each quarter's GDP. The advance estimate comes "about a month after the quarter's end", and the second and third estimates "each incorporate additional source data that weren't available the month before." Each year an annual update revises several years of history with more complete annual source data, and a more extensive comprehensive update takes place approximately every five years (as of November 2026, the most recent was in 2023). The 2026 annual update, released on 30 September 2026, covered the first quarter of 2021 through the first quarter of 2026.

Real GDP growth at an annual rate, as each estimate reported it:

QuarterAdvanceSecondThirdAfter the 2026 annual update
Q1 20262.0% (30 Apr)1.6% (28 May)2.1% (25 Jun)2.5% (30 Sep)
Q2 20261.5% (30 Jul)1.5% (26 Aug)2.2% (30 Sep)released with the update

The first quarter's growth went down, then up, then up again, and ended half a point above its advance estimate. The second quarter's third estimate, released on the same day as the annual update, was 0.7 percentage point above the second. The advance estimate for the third quarter of 2026, released on 29 October 2026, was 3.4 percent; BEA's schedule, read on 7 November 2026, listed its second estimate for 25 November 2026. The recession explainer shows how GDP revisions can change which quarters count as negative.

How to read a release's revision lines

  • Jobs report: the release text includes a revision paragraph giving each of the two prior months' old and new change and the combined revision. Add the headline and the net revision to see how far the employment level moved from the previous report. In February, check the benchmark note before comparing any level with an older one.
  • CPI: in most months nothing earlier is revised. Around the release of the January index each February, BLS publishes recalculated seasonal factors; compare old and new seasonally adjusted monthly changes, and remember that the unadjusted 12-month rate is not affected.
  • GDP: each later estimate (second and third) states how far growth was revised from the previous one ("revised up 0.7 percentage point from the second estimate"). In the annual update release, read the revised window and the revised growth of the earlier quarters as well.
  • Any series, after the fact: ALFRED, the St. Louis Fed's archive of past FRED data, keeps every vintage of a series as it stood on each date. Asking for the payroll series as of 4 September 2026 returns August's first print; as of 6 November 2026, its third. Every vintage figure on this page was read that way.

Why revisions matter to markets

On release day the new month and the revisions arrive together, and what changes the picture of the economy is the level, not the headline alone. A 165,000 headline with 84,000 of upward revisions describes a labor market about 249,000 jobs larger than the previous report did; the same headline with revisions of the opposite sign would describe a smaller one. Thomas Gilbert (2011) studied payrolls, GDP and industrial production and found that S&P 500 returns on announcement days were related to the later revisions of the number just released (positively in expansions, negatively in recessions), which suggests investors "care about the final revised value of a macroeconomic series" and read through the preliminary number toward it. The event study explainer covers how a market's response to a release is measured.

Honest limitations

The worked examples are specific releases and do not measure the typical size of revisions; the only average quoted here, the 0.2 percent benchmark figure, is BLS's own. The vintage figures are as ALFRED recorded them, which follows each agency's release schedule. Payroll changes are computed from levels in thousands, so a monthly change can differ by a thousand from a figure the agency computed from unrounded levels. The CPI monthly changes are computed from the published three-decimal index levels and rounded to one decimal, the way BLS states them. Release dates after 7 November 2026 are the agencies' schedules, which can move.

Key takeaways

PointDetail
DefinitionA later change to a published figure, from late survey replies, new seasonal factors, or a benchmark or annual update
PayrollsFirst estimate early the following month, revised twice, then held until the annual March benchmark (March 2025: -862,000 not seasonally adjusted)
6 November 2026August revised from +133,000 to +186,000, September from +29,000 to +60,000: 84,000 more jobs than previously reported, on top of October's 165,000
CPIUnadjusted indexes are final when issued; seasonally adjusted changes are revised each February for five years
GDPAdvance, second and third estimates, then annual updates (Q1 2026: 2.0%, 1.6%, 2.1%, then 2.5%)
What to watchThe headline plus the net revision, and the vintage a figure comes from

Frequently asked questions

Why do jobs numbers get revised?

Because the first estimate uses the survey replies received by the deadline. BLS revises each month twice as more businesses and government agencies report and as seasonal factors are recalculated, then once a year aligns the March level with near-complete counts from unemployment insurance tax records. On 6 November 2026, August was revised from +133,000 to +186,000 and September from +29,000 to +60,000.

How many times is the jobs report revised?

Each month's payroll change is published three times in consecutive monthly reports (a first, second and final sample-based estimate) and then held until the annual benchmark, which is folded in with the January report each February. The benchmark can move the level of employment by hundreds of thousands: the March 2025 revision was -862,000 before seasonal adjustment.

Is CPI data revised?

The unadjusted CPI-U and CPI-W are final when published. Seasonally adjusted figures are revised once a year, around the release of the January index each February, for the previous five years. In the February 2026 update, March 2025's seasonally adjusted change moved from -0.1 percent to 0.0 percent while its unadjusted index stayed the same.

How many times is GDP revised?

Three estimates in the three months after a quarter ends (advance, second and third), then again in annual updates, which reach back several years, and in comprehensive updates about every five years. First-quarter 2026 growth was reported as 2.0 percent, then 1.6, then 2.1, and 2.5 percent after the September 2026 annual update.

How do you see what a number was when it was first released?

ALFRED, the St. Louis Fed's archival version of FRED, stores every vintage of a series, so a chart or download can be set to show the data as it stood on any past date. That is how the first, second and third estimates on this page were read.

This page is information, not investment advice.

---

Source: BLS, The Employment Situation, October 2026, released 6 November 2026, https://www.bls.gov/news.release/archives/empsit_11062026.htm ; BLS, CES revisions between over-the-month estimates, https://www.bls.gov/web/empsit/cesnaicsrev.htm ; BLS, CES national estimates revised to incorporate March 2025 benchmarks, 11 February 2026, https://www.bls.gov/web/empsit/cesbmart.htm ; BLS, CES birth-death model, https://www.bls.gov/web/empsit/cesbd.htm ; BLS, CES preliminary benchmark revision, 28 August 2026, https://www.bls.gov/news.release/archives/prebmk_08282026.htm, and 9 September 2025 ; BLS, CPI seasonal adjustment, https://www.bls.gov/cpi/seasonal-adjustment/ ; BLS, CPI questions and answers, https://www.bls.gov/cpi/questions-and-answers.htm ; BLS, Consumer Price Index, September 2026, released 14 October 2026 ; BEA, GDP (third estimate) and 2026 annual update, second quarter 2026, released 30 September 2026 ; BEA, GDP (advance estimate), third quarter 2026, released 29 October 2026 ; BEA, What to know about GDP, and BEA blog, 9 September 2024 ; Federal Reserve Bank of St. Louis, ALFRED vintages of PAYEMS, CPIAUCSL, CPIAUCNS and A191RL1Q225SBEA, read 7 November 2026, https://alfred.stlouisfed.org/ ; Gilbert, T. (2011), "Information Aggregation around Macroeconomic Announcements: Revisions Matter", Journal of Financial Economics 101(1), 114 to 131.

Kresmion Research.

Sources
See it live
Live central bank policy rates →

Every tracked central bank's policy rate, its latest change and the primary source behind it.

Free to view, no account needed.

FREE, NO ACCOUNT

Put this to work

Real filings, 13F flows, and positioning reads with their sources, in your inbox when there is something in the data, or live on Telegram. Free, no account.

Get the morning brief by email

No fixed schedule: it goes out when the data has something in it. Unsubscribe anytime.

Or get live alerts on Telegram
Join on Telegram

One tap. Live alerts, no email needed.

Kresmion publishes information, not investment advice. See our methodology and the latest research notes.