US nonfarm payroll figures have been revised lower in 21 of the past 30 months, a cumulative downward revision of roughly 1.05 million jobs, according to data highlighted by The Kobeissi Letter. That works out to an average of about 35,000 jobs erased from previously reported totals each month over the period, a pattern that has repeatedly made the labor market look stronger in real time than it turns out to have actually been.
The most recent revisions, drawn from Bureau of Labor Statistics data, cut May's payroll gain from an initially reported 129,000 to just 63,000, and June's from 57,000 down to 20,000, a combined 103,000 jobs lower than first reported. The three-month average payroll gain has fallen to roughly 20,000 as a result, down sharply from 77,000 as recently as June.
Why the Pattern Matters for Markets
July's initial jobs report itself showed the economy losing 23,000 jobs even as unemployment ticked down to 4.1% from 4.2%, a weak print that sharply reduced market expectations for a Federal Reserve rate hike in September; odds of a hike fell toward levels some analysts read as pricing in cuts as early as October instead. Treasury yields fell and stocks rallied toward their strongest week since April on the initial report, before the pattern of subsequent downward revisions raised a separate question: whether even the softer initial prints have been overstating labor market resilience.
For a market that has spent much of 2026 trying to read the Fed's next move off each month's jobs number, a track record of the data being revised weaker after the fact 21 out of 30 times complicates that exercise considerably. Investors reacting to the initial headline number have repeatedly been reacting to a figure that later proved too optimistic.
Related: Big Tech's AI Spending Set to Hit a Record 2.4% of US GDP in 2026
Crypto and risk assets broadly have tended to benefit from rate-cut expectations, so a labor market that keeps turning out weaker than first reported adds another data point supporting looser policy, even as the unreliability of the initial prints themselves makes each individual report harder to trade on with confidence.
