How many nonfarm payroll jobs will the U.S. add in the September 2026 Employment Situation report?

TL;DR

My forecast is +125,000 jobs for the initial September 2026 nonfarm-payroll print. The distribution has a median of +130,000, an 80% interval of −7,000 to +251,000, and an 11% chance of a negative print. September's survey data beat a weak payroll trend, while August's school and restaurant surge should not repeat in full.

Context

The target is the first published, seasonally adjusted monthly change in total nonfarm payroll employment, not the later revised estimate. The release was scheduled for October 2, 2026 at 8:30 a.m. ET (BLS schedule); the evidence cutoff is September 24 at 10:15 UTC. The CES sample covers about 119,000 businesses and government agencies, 622,000 worksites, and 26% of payroll jobs (BLS technical note).

August's initial gain was +162,000, against a prior-12-month average of +31,000. But food services added +59,000 and local-government education added +42,000, so those two volatile sectors supplied 101,000, or 62%, of the headline (BLS August report). September starts from a stronger level than July, but not from a clean +162,000 trend.

Evidence

The historical backbone is the real-time first print, because that is what resolves the question. The full 2026 history available at the cutoff is below; units are thousands of jobs, seasonally adjusted (BLS revision table).

The eight initial prints average +87,000, with a sample standard deviation of about 99,000; the latest six average +110,000. The sequence is noisy and has little stable momentum. BLS's longer record since the 2003 sample redesign shows a mean absolute first-to-third revision of 51,000, while its approximate 90% sampling interval for one monthly total-nonfarm change is ±122,000 (revision history, sampling error). The preliminary March 2026 benchmark was only −79,000, or −0.1%, for total nonfarm employment, so it supports a mild downward trend adjustment rather than a collapse story; it will not enter official monthly levels until February 2027 (BLS benchmark release).

I reconstructed same-vintage first prints from ALFRED PAYEMS and ran expanding-window forecasts from January 2011 through August 2026, excluding March 2020 through June 2021. The common evaluation sample had 137 monthly forecasts. A curated ensemble of six- and twelve-month means, a two-lag autoregression, and an autoregression with initial and continuing claims produced +108,000 for September, with out-of-sample RMSE of 94,000 and MAE of 72,000. Simple averages performed about as well as richer models. Claims-only models had RMSE near 145,000 and overpredicted when layoffs were low, so I did not translate low claims into a huge payroll number.

The current-period data move the +108,000 model baseline upward. Initial claims were 196,000 in the week ended September 12, their four-week average was 203,250, and insured unemployment was 1.730 million in the week ended September 5; the release was published September 17 (Department of Labor). S&P Global's September 23 flash composite PMI rose from 56.0 to 58.4, and its employment measure showed the fastest job growth since June 2022 (S&P Global). ADP's September 22 pulse, based on administrative payroll data, rose to an average 20,000 private jobs per week for the four weeks ended September 5 (ADP). Indeed's seasonally adjusted, seven-day postings index rose from 101.92 on September 1 to 103.16 on the reference date and 103.45 on September 18; FRED updated the series on September 23 (FRED/Indeed).

Regional surveys support positive hiring, but not a uniform boom. Philadelphia nonmanufacturing full-time employment rose to 19.0, while Philadelphia manufacturing employment was 11.8 (Philadelphia services, Philadelphia manufacturing). Empire manufacturing employment was 10.6 and its workweek index was 17.0 (New York Fed). Richmond manufacturing employment improved to 7, but Richmond nonmanufacturing employment eased to 2 (Richmond manufacturing, Richmond nonmanufacturing). New York's service-sector employment index was −4.9 (New York business leaders). I read this as broad improvement in manufacturing and selected services, not evidence for a +200,000 central forecast.

The counterweight is hiring flow. July JOLTS showed 5.054 million hires, a 3.2% rate, while layoffs and discharges stayed low at 1.0%; professional and business services hires fell by 188,000 (BLS JOLTS, released September 1). ADP's full August report showed only +38,000 private jobs in a sample covering more than 26 million workers (ADP August report). NFIB's August survey found hiring plans still positive at a net 17%, but down 3 points from July (NFIB, released September 3). Temporary staffing was 0.5% above the August monthly report, though the Labor Day week fell 1.4% (ASA, released September 22). These data describe a low-fire, uneven-hire market.

I set the statistical baseline at +110,000, added a rounded +20,000 for the exceptional national PMI, rising ADP pulse, postings, and regional breadth, then subtracted 5,000 for weak gross hiring and August sector mean reversion. That gives a mean forecast of +125,000. The final distribution is a three-state Student-t mixture: 55% baseline at +110,000, 35% survey-upside at +180,000, and 10% downside/reversal at +15,000. Its standard deviation is 109,000. The code treats the published integer as the latent result rounded to the nearest thousand, then uses half-thousand cutoffs to assign exact mass to the 202 buckets.

What's non-obvious

Low layoffs are more useful as a floor than as a point forecast. In my backtest, claims-only models were among the worst monthly predictors. July JOLTS tells the same story: layoffs were low, yet hiring was weak (BLS JOLTS). A high survey diffusion index can also come from many firms adding one or two workers. That produces strong survey language without a proportionate jump in the national job count.

A separate check gave a mixed signal. Treasury's Daily Treasury Statement showed withheld individual and FICA receipts of $207.994 billion through September 22, versus $220.223 billion through the same date in 2025, a 5.6% nominal decline; fiscal-year-to-date receipts were $3.496 trillion versus $3.355 trillion, a 4.2% increase (Treasury Daily Statement). The conflict points to pay-date and tax-timing noise, not a clean employment signal, so it did not move the forecast.

Uncertainties

• The next weekly-claims observation was not available by the 10:15 UTC cutoff. The latest included release covers initial claims through September 12 and continuing claims through September 5 (Department of Labor).
• The official September strike report was scheduled for September 25, after the cutoff (BLS schedule). I found no verified national strike adjustment before the cutoff, which is a data gap rather than evidence of zero impact.
• August JOLTS was scheduled for September 29, and the full September ADP report for September 30, so both arrive too late for this forecast (BLS JOLTS schedule, ADP release notice).
• S&P Global did not publish the numerical employment subindex or the flash survey's response count in its public note (S&P Global).
• September is exposed to school-calendar residuals. BLS notes that education payrolls normally fall about 20% after the spring term and rise again in fall, so small calendar shifts can create large seasonally adjusted surprises (BLS technical note). BLS also changed the birth-death model for first estimates starting in January 2026 to use current sample information, which limits how cleanly older errors transfer to the current regime (BLS methodology).

The unresolved risks are large enough to keep the distribution wide. The 90% interval is −57,000 to +289,000, with fat tails on both sides. That width reflects first-print noise, school and food-service seasonality, and disagreement between count-based hiring data and unusually strong diffusion surveys.