Macro ReviewHow to read
Outlook · Macro expectations · Labour nowcast

The site's equation puts November payrolls at +163k, above the recent three-month pace

What will the next payrolls print say on the site's own , and where does that pace take unemployment?

Data to 12 Nov 2026 · updated 17 Nov 2026Next jobs report: Nov 2026
Payrolls, Nov 2026 (site)+163kfour-week average line · September equation +149k · data to 12 Nov 2026 · typical miss ±111k
Payrolls, Oct 2026+165kfirst estimate (6 Nov 2026) · three-month average +137k
Three-month pace with the nowcast+129kto Nov 2026 · +82k vs breakeven ≈47k (Oct 2026)above breakeven
Unemployment path, 12 months−0.60ppfrom 4.1% (Oct 2026) if the pace holds · ±0.71pp record · Dec 2026 4.0% vs median 4.1%
Initial claims, latest week202thousands, SA · week ending 7 Nov 2026 · reference week 14 Nov 2026
Record since 2012111thousands, 161 months vs first print · three-month average 115knarrowly beats the naive benchmarks

In brief

The site's payroll number, the pace it implies against breakeven, what moved it and how good it has been
  • The site's four-week average equation puts Nov 2026 payrolls at +163k, on data released to 12 Nov 2026.→ Exhibit 2
  • With it, payrolls would average +129k a month over three months, against about 47k that holds unemployment steady.→ Exhibit 3
  • If that pace held, unemployment would move −0.6pp over twelve months, from 4.1% in Oct 2026.→ Exhibit 3
  • Since 15 Oct 2026 the moved +43k; the largest single move came from the October jobs report (+33k).→ Exhibit 4
  • Since 2012 the four-week average equation has missed the first payroll print by 111k (root mean square), better than the three-month average's 115k.→ Exhibit 5
What would change this view
What would change this viewTriggerTodayDecided by
Nov 2026 payrolls land outside the nowcast's rangefirst estimate outside +52k to +274k (nowcast ± record )nowcast +163k (12 Nov 2026)pendingEmployment Situation for Nov 2026, 4 Dec 2026 (BLS)
The pace falls below breakeventhree-month average < 47k a month+129k with the nowcast (Nov 2026)pendingEmployment Situation for Nov 2026, 4 Dec 2026 (BLS)
The equation loses to the three-month average|nowcast − first estimate| > |+137k (the three-month average) − first estimate|record RMSE 111k vs 115kpendingEmployment Situation for Nov 2026, 4 Dec 2026 (BLS)
The four-week average line earns its placeon the jobs reports from 4 Dec 2026 on, |four-week average − first estimate| ≤ |September equation − first estimate| in most months+163k against +149k for Nov 2026; pseudo-real-time records 111k and 107kpendingEmployment Situation reports from 4 Dec 2026 (BLS)
The reasoning behind each point
  • Payrolls
    The site's four-week average equation puts Nov 2026 payrolls at +163k, on data released to 12 Nov 2026.

    Three lags of the payroll change as BLS last published them (Aug 2026 +186k, Sep 2026 +60k, Oct 2026 +165k) and the four-week average of initial claims to the week ending 7 Nov 2026 (196k), standing in for the reference week ending 14 Nov 2026, which is not published yet. One record RMSE either side spans +52k to +274k: the record's typical miss, not a probability. The September equation line, published beside it since the respecification was declared on 3 Oct 2026, reads +149k (its own record ±107k).

  • Pace
    With it, payrolls would average +129k a month over three months, against about 47k that holds unemployment steady.

    The breakeven is the Payrolls page's central estimate for Oct 2026, copied with its inputs: payroll jobs per employed worker 0.974 and a labour force of 170.7 million.

  • Unemployment
    If that pace held, unemployment would move −0.6pp over twelve months, from 4.1% in Oct 2026.

    By Dec 2026 that is 4.00%, against the Committee's 2026 median of 4.1% for the fourth quarter. The identity's twelve-month call has missed by 0.71pp out of sample (root mean square) on the Payrolls page.

  • News
    Since 15 Oct 2026 the nowcast moved +43k; the largest single move came from the October jobs report (+33k).

    Initial claims, week ending 17 Oct 2026 (DOL) (22 Oct 2026) −3.7k; Initial claims, week ending 24 Oct 2026 (DOL) (29 Oct 2026) −1.0k; Initial claims, week ending 31 Oct 2026 (DOL) (5 Nov 2026) +5.7k; Employment Situation for Oct 2026 (BLS) (6 Nov 2026) +32.5k; Initial claims, week ending 7 Nov 2026 (DOL) (12 Nov 2026) +9.6k.

  • Record
    Since 2012 the four-week average equation has missed the first payroll print by 111k (root mean square), better than the three-month average's 115k.

    The average miss is the root mean square error over 161 months, so read the ±111k range, not the point. Last month's print missed by 126k. The September equation (reference-week claims): RMSE 107k, inside its band 76% of months, mean error +24k. Four-week average of claims to the reference week: RMSE 111k, inside its band 78% of months, mean error +26k. Both are scored from the next Employment Situation release on; neither was chosen by this record (the respecification was declared on 3 Oct 2026). Without claims the equation's RMSE is 103k. Against the latest estimates the four-week average equation's RMSE is 120k. Pseudo-real-time on revised data: the equation is estimated on today's benchmark-revised series, while its lags are the estimates as published at the time. Made 28 days before the jobs report its RMSE is 115k; on the day, with the reference week's claims in, 108k.

Next: Range and scenarios — the named paths these readings feed

BLS; Department of Labor; Federal Reserve Board (projection median); Macro Review calculationslabour.nowcast-read

November payrolls are tracking +163k, give or take 111k

The next payroll change through the last four weeks with the news that moved it and the record's typical miss, beside the recent prints and the unemployment rate the pace implies

Exhibit 1 · The site's payroll nowcast through the month, and what moved it

Total nonfarm payroll change, Nov 2026, thousands, at each release, 15 Oct 2026 to 12 Nov 2026; band: the record's typical miss; bars: news, thousands
Range: out-of-sample record
About this exhibit
Line chart of the site's payroll nowcast for the next month (thousands) re-run at each release date since the previous data date, with a band of the record's typical miss by horizon and bars splitting each step by release
Nov 2026 payrolls: the nowcast through four weeks and the news that moved it
Total nonfarm payroll change, thousands, re-run at each release date on today's data; news in thousands by release.
The shaded band is the nowcast ± the record's RMSE at the nearest horizon (28, 21, 14, 7, 0 days before the jobs report) at or beyond the point's days to the report; at today's point, the record the tiles and the falsifier print — the one range the page prints: the record's typical miss, not a probability. The bars below split each step of the line by the input released that day; they add to the step (checked in the pipeline). The next release of the target is due 4 Dec 2026.
Source: BLS; Department of Labor; Federal Reserve Board (projection median); Macro Review calculations · data to 12 Nov 2026labour.nowcast-12

Exhibit 2 · The payroll nowcast and recent prints, and the unemployment path they imply

Nonfarm payrolls, monthly change, thousands: the site's equation lines for Nov 2026, recent prints, the three-month pace against breakeven, and the unemployment rate it implies
About
Site nowcast, Nov 2026Four-week average of claims to the reference week+163k+52k to +274k out-of-sample record
The September equation (reference-week claims)published beside it since 3 Oct 2026+149k+42k to +257k out-of-sample record
Aug 2026, third estimate+186k–
Sep 2026, second estimate+60k–
Oct 2026, first estimate+165k–
Three-month pace with the nowcast+129k+82k against breakeven
Breakeven: the pace that holds unemployment steady (Oct 2026)47kthe Payrolls page's central estimate
Unemployment rate, Oct 20264.1%published
… if the pace held, Dec 20264.00%−0.10pp over 2 months
… if the pace held, Oct 20273.50%−0.60pp over 12 months · record ±0.71pp
The Committee's 2026 median, fourth-quarter average4.1%16 Sept 2026

Declared on 3 Oct 2026, the author's decision of that day: initial claims enter the published equation as the four-week average to the reference week; the September equation stays beside it; both are scored from the next Employment Situation release.

About this exhibit
Table: the site's equation lines for the next month with their ranges, the last prints, the three-month pace against breakeven, the unemployment rate and its path if the pace held, and the Committee's median
The next print on the site's equation, and where its pace leads
Total nonfarm payrolls, monthly change in thousands, seasonally adjusted, the next month and the three-month pace; the range is the record's typical miss, not a probability.
Source: BLS; Department of Labor; Federal Reserve Board (projection median); Macro Review calculationslabour.nowcast-1

Exhibit 3 · Monthly payroll changes and their three-month average, against breakeven

Nonfarm payroll change, thousands a month, as last published; Nov 2026: the nowcast, whisker its typical miss (+52k to +274k); lines: three-month average and breakeven
About this exhibit
Chart of monthly payroll changes as BLS last published them, the next month's nowcast with its typical-miss whisker, the three-month average and the breakeven line
Payrolls, the nowcast and the pace against breakeven
Thousands a month, as BLS last published them; the nowcast in blue with its typical miss (+52k to +274k) as the whisker; the three-month average; the breakeven.
Source: BLS; Department of Labor; Federal Reserve Board (projection median); Macro Review calculations · data to 12 Nov 2026labour.nowcast-2

The October jobs report moved it most: +33k

Each release since the previous data date four weeks earlier, and how much it moved the

Exhibit 4 · Each recent release's effect on the payroll nowcast

Change in the nowcast by release since 15 Oct 2026, thousands; a jobs report adds its month and revises the earlier ones; rows add up
About
ReleaseReleasedMoved the nowcastNowcast after
Nowcast on 15 Oct 2026+120k
Initial claims, week ending 17 Oct 2026 (DOL)22 Oct 2026−3.7k+116k
Initial claims, week ending 24 Oct 2026 (DOL)29 Oct 2026−1.0k+115k
Initial claims, week ending 31 Oct 2026 (DOL)5 Nov 2026+5.7k+121k
Employment Situation for Oct 2026 (BLS)6 Nov 2026+32.5k+153k
Initial claims, week ending 7 Nov 2026 (DOL)12 Nov 2026+9.6k+163k
Total+43.1k+163k
About this exhibit
Table of the releases since the previous data date (the weekly claims releases and any jobs report), each with how much it moved the nowcast and the nowcast after
What moved the nowcast between 15 Oct 2026 and 12 Nov 2026
Each release added back in date order to the data as they stood four weeks earlier, when the latest jobs report and the latest claims were a month older; thousands. A jobs report brings its month and the revisions to the two before. The rows add to the total.
Source: Computed here on BLS and DOL data and release calendars.labour.nowcast-3

It has missed the first print by 111k (root mean square), better than the three-month average's 115k

The equation re-run at every past month's equivalent data date, against the first print, the latest estimate and two naive benchmarks; how good it is at each point before the jobs report; and the as the site published them

Exhibit 5 · Site payroll nowcasts against BLS's first estimate, monthly since 2012

Nonfarm payroll change, thousands a month, since Jan 2012: the site's nowcast at each month's data date against BLS's first estimate; March 2020–June 2021 out
About
About this exhibit
Chart of the record, thousands a month: the site's nowcast re-run at each month's equivalent data date against BLS's first estimate; March 2020 to June 2021 left out
The record: the equation re-run each month, against BLS's first estimate
Thousands a month. Each month's equation is estimated on data before it and fed the claims week the same distance before its reference week; March 2020 to June 2021 left out.
Source: Scores computed here on BLS and DOL data.labour.nowcast-4

Exhibit 6 · Payroll nowcast error by days before the jobs report, against benchmarks

RMSE, MAE and bias, thousands, by days before each jobs report since Jan 2012 (161 months); benchmarks: last month's print and the three-month average
Days before the jobs reportRMSEMAEBiasLast month's printThree-month averageWeeks between the claims in hand and the reference weekLast month's report in hand
28 days11582+261261152.2yes
21 days10877+251261151.2yes
14 days10878+271261150.2yes
7 days10878+271261150.0yes
release day10878+271261150.0yes

Publication lags, each the median of its latest releases on the publishers' calendars: the jobs report 5 days after the month's end, claims 5 days after the week's end. Record months no horizon row scores (a horizon could not be run there): Oct 2025.

The reference week's claims. Forced to the reference week's own claims, the September equation's record RMSE is 117.3k, against 107.4k a week earlier and 107.4k two weeks earlier (Diebold–Mariano t 2.34 against two weeks earlier; 2.77 over 2013–2019), and 42% of the extra squared error falls in July, 21% in June, 17% in August, 17% in March — months when claims' seasonal adjustment is hardest — while the equation without claims scores 103.5k.

About this exhibit
Table of root mean square error, mean absolute error and bias at each horizon before each jobs report, with the two benchmarks (last month's print, three-month average), claims timing and report in hand
The miss falls from 115k 28 days out to 108k on the day
The record re-run 28, 21, 14, 7, 0 days before each jobs report since Jan 2012 (161 months), payrolls and claims cut to what BLS and DOL had released by then, scored against the first estimate; thousands. The benchmarks are last month's print and the three-month average as published at each point.
Source: Scores computed here; release lags from the publishers' calendars.labour.nowcast-10

Exhibit 7 · Payroll nowcast errors against first and latest estimates, with benchmarks

Errors in thousands since Jan 2012: RMSE, MAE and bias (+ = too high); benchmarks: last month's print and the three-month average as published; then claims lines
Estimate, againstMonthsRMSEMAEBias
Site, against the first estimate16111179+26
… last month's print16112689+2
… three-month average16111580+10
Site, against the latest estimate16212085+28
… last month's print162140108+5
… three-month average16212296+11
Line, against the first estimateMonthsRMSEInside ± RMSEMean error
The September equation (reference-week claims)16110776%+24
Four-week average of claims to the reference week (the published line)16111178%+26
About this exhibit
Two tables: root mean square error, mean absolute error and bias of the site's equation and two benchmarks (last month's print, three-month average) against BLS's first and latest estimates; and the two claims lines' errors, hit rate and mean error
The equation misses by 111k, the three-month average by 115k
Errors in thousands since Jan 2012: RMSE, MAE and bias (+ = too high). The benchmarks are last month's print and the three-month average as published at the time.
Source: Scores computed here.labour.nowcast-5

Exhibit 8 · Every payroll nowcast the site has published, with its outcome

Payroll change, thousands: each nowcast published since 24 Sep 2026, with its data date and range, unrevised; scored against BLS's first estimate

7 nowcasts on the record, the first made on data to 10 Sept 2026. Payrolls: 3 months nowcast, 2 with an outcome so far. The real-time score is printed once 8 months of a nowcast have an outcome; until then the page counts them.

Correction. The nowcast published on 13 Nov 2026, +353k for Nov 2026, counted the claims week ending 7 Nov 2026 when 3 of 53 jurisdictions had reported it. That number was wrong and is withdrawn. A week now enters only once every jurisdiction has reported; the row stays below as it was published.

PublishedData toNowcast ofValueRangeOutcomeMiss
16 Nov 202612 Nov 2026Payrolls, Nov 2026+163k+52k to +274k out-of-sample record––
13 Nov 2026withdrawn12 Nov 2026Payrolls, Nov 2026+353k+242k to +464k out-of-sample record––
31 Oct 202629 Oct 2026Payrolls, Oct 2026+102k−6k to +210k out-of-sample record+165k−63k
26 Oct 202622 Oct 2026Payrolls, Oct 2026+102k−6k to +210k out-of-sample record––
18 Oct 202615 Oct 2026Payrolls, Oct 2026+107k−4k to +218k out-of-sample record––
5 Oct 20262 Oct 2026Payrolls, Oct 2026+81k−36k to +198k out-of-sample record––
24 Sept 2026first on the record10 Sept 2026Payrolls, Sep 2026+118k+10k to +226k out-of-sample record+29k+89k
About this exhibit
Table of each payroll nowcast the site has published, with publication date, data date, value, range, outcome and miss
The nowcasts as published
Each nowcast this page has published, with the data it was made on and its range — kept as it was published, so later revisions to the data cannot rewrite it. The last one before a month's jobs report is scored against BLS's first estimate.
Source: The site's own published nowcasts; outcomes from BLS.labour.nowcast-11

How robust is it?

The equation without claims, with the household survey added, and scored with the pandemic months

Exhibit 9 · Payroll nowcast errors without claims, with household data and with pandemic months

RMSE, MAE and bias, thousands, against BLS's first estimate since Jan 2012, for each variant of the equation
VariantMonthsRMSEMAEBias
The equation (the page's nowcast)16111179+26
Without the claims terms16110372−10
With the household survey's employment change a month before15911482+24
Pandemic months scored too (the equation still estimated without them)177334126+41

Two lines, each scored against the first estimate on the same 161 months: the September equation line RMSE 107k, hit rate 76%, mean error +24k; the four-week average line RMSE 111k, hit rate 78%, mean error +26k. Without the claims terms the equation scores RMSE 103k. The headline is the four-week average line by the declaration of 3 Oct 2026, not by this record; both are scored from the next Employment Situation release on.

About this exhibit
Table of root mean square error, mean absolute error and bias for the equation, the equation without claims, with the household survey added, and with the pandemic months scored
How robust is the record?
The same record, scored against the first estimate, for the equation without the claims terms, with the household survey's employment change added, and with the pandemic months scored.
Source: Scores computed here.labour.nowcast-7

Appendix

The equation in notation, its coefficients, its inputs and what it leaves out53 checks, all pass
About this page

A model-conditional nowcast of the next payroll change from the recent prints and the week's unemployment claims, set against the pace that holds unemployment steady. The page shows what moved it and how often an equation like this has beaten simply repeating the recent average.

The payroll equation — two declared claims terms

ΔNt=c+∑L=13ρL ΔNt−L(v)+θ1 IC‾t(k)+θ2 (IC‾t(k)−IC‾t−1(k))+et,et∼(0, σ2)\Delta N_t = c + \sum_{L=1}^{3} \rho_L\,\Delta N^{(v)}_{t-L} + \theta_1\,\overline{IC}^{(k)}_t + \theta_2\,\left(\overline{IC}^{(k)}_t - \overline{IC}^{(k)}_{t-1}\right) + e_t, \qquad e_t \sim (0,\ \sigma^2)
IC‾t(k)=1k∑j=0k−1ICw(t)−j,k=4 (the published line), k=1 (the September equation)\overline{IC}^{(k)}_t = \tfrac{1}{k}\sum_{j=0}^{k-1} IC_{w(t)-j}, \qquad k = 4 \text{ (the published line)},\ k = 1 \text{ (the September equation)}
SymbolMeaningFromAssumed or estimated
ΔNt\Delta N_tPayroll change in month t, thousands (the target: BLS's first estimate)BLS, Current Employment Statistics–
ΔNt−L(v)\Delta N^{(v)}_{t-L}The payroll change L months before, as BLS last published it at the timeBLS, CES revisions triangle–
ICw(t)−jIC_{w(t)-j}Initial claims in the week j weeks before the CES reference week w(t) (the latest published week stands in for the reference week when it is not yet out), thousandsDOL, weekly claims–
kkThe claims term's width in weeks: 1 (September equation) and 4 (four-week average) — declared 3 Oct 2026, the author's decision of 3 Oct 2026 (D-2026-10-03-09)the page's declared settings (the respecification of 3 Oct 2026)assumed
c, ρL, θ1, θ2c,\ \rho_L,\ \theta_1,\ \theta_2Coefficients, estimated separately for each kOLS, the coefficient tablesestimated
σ\sigmaResidual standard deviationOLSestimated

OLS, 2010-01 to 2026-10 (186 months), March 2020 to June 2021 left out; R² 0.47, σ̂ 116k for the published line (the September equation line R² 0.44, σ̂ 120k). Each line is estimated with the same claims stand-in it is applied with (1 week before the reference week today).

Where the pace takes unemployment (an identity)

Δut+h=− h ΔNˉ−Br⋅LF×100\Delta u_{t+h} = -\,h\,\frac{\bar{\Delta N} - B}{r \cdot LF} \times 100
SymbolMeaningFromAssumed or estimated
ΔNˉ\bar{\Delta N}The three-month average payroll change with the nowcastthis pageassumed
BBBreakeven payroll growth that holds unemployment steadythe Payrolls page's model–
rrPayroll jobs per employed workerthe Payrolls page–
LFLFThe labour forceBLS, Current Population Survey–

An accounting identity applied to a pace assumed to hold for h months, not a forecast; the breakeven and its inputs are copied from the Payrolls page and checked.

The record and the band

RMSE(h)=1n∑t(ΔN^t h−ΔNtfirst)2,band=ΔN^t±RMSE(h)\mathrm{RMSE}(h) = \sqrt{\frac{1}{n}\sum_{t}\left(\widehat{\Delta N}_t^{\,h} - \Delta N_t^{first}\right)^2}, \qquad \text{band} = \widehat{\Delta N}_t \pm \mathrm{RMSE}(h)
SymbolMeaningFromAssumed or estimated
ΔN^t h\widehat{\Delta N}_t^{\,h}The nowcast of month t re-run h days before the jobs reportcomputed here–
ΔNtfirst\Delta N_t^{first}BLS's first estimateBLS, CES revisions triangle–

Scored each month since 2012 (161 months) against the first estimate. Pseudo-real-time on revised data: the equation is estimated on today's benchmark-revised series, its lags are the estimates as published at the time. The band is the record's typical miss — not a probability.

Exhibit 10 · The equation's coefficients

About
TermCoefficientStandard errorInput todayContribution
Constant49.39035.608–+49.4
Payroll change, a month before0.2180.065165+36.0
… two months before0.1420.06660+8.5
… three months before0.2620.064186+48.7
Initial claims, 4-week average to the reference week (stand-in)0.0440.110196+8.6
… change from the month before-3.8770.719-3+11.6

The September equation (reference-week claims): estimated 2010-01 to 2026-10 (186 months), R² 0.44, residual standard deviation 120k; its nowcast +149k.

TermCoefficientStandard errorInput todayContribution
Constant37.69936.994–+37.7
Payroll change, a month before0.2800.066165+46.2
… two months before0.1560.06860+9.3
… three months before0.2470.066186+45.9
Initial claims, reference week (stand-in)0.0660.114202+13.3
… change from the month before-1.6530.3962−3.3
About this exhibit
Estimated 2010-01 to 2026-10 (186 months), R² 0.47, residual standard deviation 116k. The reference week ends 14 Nov 2026; the week ending 7 Nov 2026 stands in for it.
Source: BLS; Department of Labor; Macro Review calculationslabour.nowcast-6

Exhibit 11 · Definitions and caveats

  • Model
    Model-conditional payroll equation.

    Model-conditional: the nowcast is what one equation, estimated on payroll and claims data since 2010, makes of the data released so far — not a forecast in any other sense, and no probability is attached to it. The payroll change (thousands) is regressed on its own three previous changes and on the level and month-on-month change of seasonally adjusted initial claims — in the published line the four-week average to the CES reference week, the respecification declared on 3 Oct 2026 (the author's decision of 3 Oct 2026 (D-2026-10-03-09)); in the September equation, published beside it, the reference week's own claims — with March 2020 to June 2021 left out. Both lines are estimated the same way and scored against the same first estimates, from the next Employment Situation release on; the record below is pseudo-real-time and chose neither.

  • Ragged edge
    The claims week that stands in.

    Ragged edge: when the reference week's claims are not yet published, the latest week before it stands in — for the four-week line, the four weeks ending on that week — and the equation is estimated with the same stand-in (claims 1 week before the reference week), so the coefficients fit what they are applied to. Lagged payroll changes are the first, second and third estimates as BLS published them, from its revisions table.

  • Path
    The breakeven identity.

    Unemployment path: Δu = −h × (pace − breakeven) ÷ (payroll jobs per employed worker × labour force) × 100, the Payrolls page's identity with its central breakeven and inputs copied; the pace is the three-month average including the nowcast, applied from the latest published rate. It is an accounting identity, not a behavioural model: it holds participation and population trends as the breakeven assumes them.

  • Record
    Pseudo-real-time on revised data.

    Record: each month since 2012 the equation is re-estimated on data before the month and fed the claims week the same distance before its reference week, and scored against BLS's first estimate (from the revisions table) and the latest. Benchmarks: last month's print and the three-month average, both as published at the time. Pseudo-real-time on revised data: the estimation sample is today's benchmark-revised series.

  • Path through the month
    Re-run at each release this month.

    Path through the month: the same equation re-run on the data released by each release date since the jobs report for the latest month — each Thursday's claims — on today's data, as the GDP nowcast's path is.

  • Horizon
    The record days before each jobs report.

    Accuracy by horizon: the record re-run 28, 21, 14, 7, 0 days before each jobs report since 2012, payrolls and claims cut to what BLS and DOL had released by then on a stylised calendar — the median lag of their latest releases on the publishers' calendars (the jobs report 5 days after the month's end, claims 5 days after the week's end) — with the claims week that stands in for the reference week wherever that week is not yet out. Pseudo-real-time on revised data, like the record.

  • Claims week
    The reference week's claims, forced.

    Claims week forced: the September equation's record re-run with the claims week set 0 to 4 weeks before the reference week, last month's report in hand, and scored against the first estimate. Forced to the reference week's own claims, the September equation's record RMSE is 117.3k, against 107.4k a week earlier and 107.4k two weeks earlier (Diebold–Mariano t 2.34 against two weeks earlier; 2.77 over 2013–2019), and 42% of the extra squared error falls in July, 21% in June, 17% in August, 17% in March — months when claims' seasonal adjustment is hardest — while the equation without claims scores 103.5k. This finding is why a respecification was declared on 3 Oct 2026: the four-week average of claims to the reference week is the published line, the September equation stays beside it, and the two are scored from the next Employment Situation release on. Neither was chosen by its score on this record: choosing a specification by its score here would be fitting the record.

  • As published
    Kept as published, never revised.

    As published: every nowcast this page publishes is kept as it was published — its data date, its value and its range — and never revised. Beside the path recomputed on today's data the page draws those published points; once eight months have a first estimate it scores them, a real-time record; until then it counts them.

  • Published
    What stands beside the site's number.

    No published payroll nowcast is read: consensus surveys and private payroll counts are not among the sources, so nothing stands beside the site's number but the recent prints.

  • Not used
    Inputs on disk left out, and why.

    Not used: JOLTS (to July on disk) arrives a month behind the payroll print it would inform; continuing claims were not in the September report's equation. The household survey's employment change is tested in the robustness table.

Source: BLS; Department of Labor; Federal Reserve Board (projection median); Macro Review calculationslabour.nowcast-8

Before you read

Macro Review is a personal research site, still being built.

Nothing on it is investment advice, a recommendation or a solicitation.

Check every figure for yourself. Each names its publisher — confirm it there before you rely on it.

Figures can be revised, delayed or wrong.

Cones, fans, scenarios, fair values and “what followed” tables are conditional on the model each page names: they describe, they do not predict.

No warranty is given; use is at your own risk.