BLS Annual Benchmark Revision Expected to Revise Up Nonfarm Payrolls, First Positive Revision Since 2022
nashnova research
The Bureau of Labor Statistics will release its preliminary benchmark revision on Friday. Goldman Sachs expects an upward revision of 50,000 to 450,000 jobs — the first positive revision since 2022 — but monthly job gains remain historically low.
What exactly is being revised?
Every year the BLS recalibrates nonfarm payroll figures using more complete data. This round covers April 2025 through March 2026.
Goldman Sachs estimates the revision at 50,000 to 450,000 jobs, which would lift average monthly job gains from roughly 25,000 to 30,000–65,000.
This means → the past year's job numbers were somewhat better than first reported, but monthly gains in absolute terms remain subdued.
Why is "first positive revision since 2022" significant?
For the past two years, the BLS benchmark revision was a large downward adjustment — the initially published payroll figures had overstated actual employment.
This time the direction is expected to flip to an upward revision, meaning the published data actually understated true employment.
In plain terms = after two straight years of "overcounting," the data finally switched to "undercounting" — a directional inflection point.
How reliable is the preliminary estimate?
Goldman notes that over the past six years, the preliminary estimate has fallen short of the final revision every time, by an average of roughly 100,000 jobs.
The reason: the QCEW data — the Quarterly Census of Employment and Wages, drawn from state unemployment-insurance records and considered more accurate than BLS surveys — keeps getting revised upward in subsequent quarters.
This means → Friday's number is likely not the final word. The ultimate revision could be larger; the final figure won't arrive until February 2027, bundled with that month's jobs report.
Why did the data overcount for two years straight?
Goldman's core explanation: QCEW systematically undercounted employment among undocumented immigrants.
Undocumented workers typically lack unemployment-insurance eligibility, and employers may not pay UI taxes on their behalf — so those jobs never entered the count.
The BLS's own error decomposition supports this: only 14% of the past two years' downward revision came from the birth-death model. The rest fell into the residual category. This reflects a structural blind spot, not normal business-turnover noise.
Will this problem persist?
Goldman points out that recent immigration has slowed sharply, meaning QCEW's undercount problem weighs much less on this revision.
Future benchmark revisions are also expected to be more accurate, avoiding the outsized negative swings of recent years.
In plain terms = the variable that distorted the statistics is fading, so the data "noise" should shrink — but monthly job gains still sit in a historically low range, and the true strength of the labor market needs further data to confirm.
市场有风险，内容仅供研究参考，不构成投资建议。