Macro ReviewHow to read
Outlook · Macro expectations · This quarter's growth

The site's model tracks Q4 growth at 2.7%, close to the New York Fed

How fast is the economy growing this quarter on the site's own , and how does it compare with the published ones?

Data to 17 Nov 2026 · updated 17 Nov 2026BEA's first estimate: 28 Jan 2027
Site nowcast, Q4 2026+2.7%% · data to 17 Nov 2026 · typical miss ±1.9pp
New York Fed Staff Nowcast, Q4 2026+2.9%13 Nov 2026 · close to the site's · 80% band 0.2–5.5%
Pace implied by the 2026 median+1.1%rest of 2026 · median 2.3% Q4/Q4 (16 Sep 2026)
Nowcast change since 29 Oct 2026+0.03ppto 17 Nov 2026 · 3 releases · since this quarter's nowcast began
Record since 20021.86pp, 93 quarters · last quarter's growth 2.52ppbeats the naive benchmark
Q3 2026 miss−0.88ppsite 2.49% vs BEA 3.37% (4 Nov 2026)within one RMSE

In brief

The site's against the New York Fed's and the Committee's, what moved it, and how good it has been
  • The site's put Q4 2026 growth at 2.7% , on data released to 17 Nov 2026.→ Exhibit 1
  • The New York Fed's model reads 2.9% (13 Nov 2026), close to the site's number.→ Exhibit 2
  • The Committee's 2026 projection implies about 1.1% annualised growth for the rest of the year.→ Exhibit 2
  • Since this quarter's nowcast began on 29 Oct 2026 it has moved +0.03pp; the largest single move came from industrial production (+0.12pp).→ Exhibit 3
  • Since 2002 the nowcast has missed BEA by 1.9pp (root mean square), against 2.5pp for last quarter's growth.→ Exhibit 4
What would change this view
What would change this viewTriggerTodayDecided by
Advance GDP lands outside the nowcast's rangeBEA's first Q4 2026 estimate outside 0.8–4.6% (nowcast ± one record )nowcast 2.71% (17 Nov 2026)pendingBEA advance GDP, 28 Jan 2027
The bridges lose to the naive benchmark|nowcast − advance| > |Q3 2026 growth − advance| (3.4%)record RMSE 1.86 vs 2.52pppendingBEA advance GDP, 28 Jan 2027
The site's nowcast leaves the New York Fed's bandnowcast outside 0.2–5.5% (its 80% band, 13 Nov 2026)2.71% (17 Nov 2026)not crossedSeptember retail sales (Census, 16 Dec 2026)
The reasoning behind each point
  • Now
    The site's bridge equations put Q4 2026 growth at 2.7% annualised, on data released to 17 Nov 2026.

    The five equations range from 2.1% (housing starts) to 3.4% (real retail sales, control group). One record RMSE either side spans 0.8–4.6%; the band is the record's typical miss, not a probability. 13 of the quarter's 15 indicator-months are still filled by the indicators' own autoregressions.

  • Published
    The New York Fed's model reads 2.9% (13 Nov 2026), close to the site's number.

    Its 80% band runs 0.2–5.5%. The two models share no inputs beyond the published releases, and they differ in structure. The New York Fed's reads a large set of indicators through a dynamic factor model, including trade and inventories, which the site's bridges leave out.

  • Projection
    The Committee's 2026 projection implies about 1.1% annualised growth for the rest of the year.

    Its median is 2.3% for 2026, fourth quarter to fourth quarter, over the last 1 quarters of the year. Real GDP has grown 2.0% since the fourth quarter of last year (BEA, to Q3 2026). The median is rounded to a tenth, so the implied pace is exact only to about ±0.2pp.

  • News
    Since this quarter's nowcast began on 29 Oct 2026 it has moved +0.03pp; the largest single move came from industrial production (+0.12pp).

    Aggregate weekly hours, production and nonsupervisory employees for October (6 Nov 2026) −0.10pp; real retail sales, control group for October (17 Nov 2026) +0.01pp; industrial production for September (17 Nov 2026) +0.12pp.

  • Record
    Since 2002 the nowcast has missed BEA by 1.9pp (root mean square), against 2.5pp for last quarter's growth.

    The average miss is the root mean square error, 93 quarters from 2002 to 2026, pandemic quarters excluded; the long-run average missed by 2.2pp. This is pseudo-real-time on revised data: each quarter is estimated only with data before its equivalent data date, but today's revised data. Over the 15 quarters since 2023Q1, the site's RMSE is 1.28pp and the New York Fed's final nowcasts' 1.49pp — its errors are real time, the site's are not. Against last quarter's growth the Diebold–Mariano t is 3.45 (p < 0.01, 3 Newey–West lags): the lower RMSE is significant at 5%; the mean error is +0.33pp (Newey–West t 1.5). Made at day 0 of the quarter, before any of its months is out, the record's RMSE is 2.22pp; at day 90, 1.64pp.

  • Read
    The New York Fed's mean error, −0.39pp, is inside two standard errors (±0.74pp): treat its nowcast as unbiased but imprecise.

    One of its RMSEs either side of the New York Fed's nowcast today spans 1.4–4.4% (the New York Fed's range, not this site's). Of the 3 recent quarters with the New York Fed's own bands on file, 3 landed inside the final update's 80% band. Errors here are against BEA's latest estimates, which include revisions the model could not have seen; against the advance estimates in the New York Fed's metadata (4 quarters) the average error is +0.58pp.

The New York Fed's model · The read that the nowcast is an unbiased but imprecise guide fails if BEA's advance estimate for 2026 Q4 lands outside one RMSE of the final nowcast (today 1.4–4.4%) and outside its 80% band (today 0.2–5.5%), or if the next two completed quarters miss in the same direction by more than 1.5pp each.

Next: Inflation nowcast — this month's prices, nowcast

BEA; Census Bureau; Federal Reserve Board; BLS; Federal Reserve Bank of New York (New York Fed Staff Nowcast, not an official forecast of the New York Fed, the Federal Reserve System or the FOMC); Macro Review calculationsgrowth.nowcast_gdp-read

Q4 is tracking 2.7%, give or take 1.9pp

The site's through the quarter, the news that moved it and its record's typical miss, beside the New York Fed's nowcast and the pace the Committee's projection implies

Exhibit 1 · Site and New York Fed GDP nowcasts for the quarter, by release

GDP growth, Q4 2026, % q/q annualised, at each release to 17 Nov 2026; band: the record's typical miss; bars: news, pp; New York Fed dashed
Range: out-of-sample record
About this exhibit
Line chart of the site's real GDP nowcast for the quarter (% q/q annualised) re-run at each release date, with a band of the record's typical miss at that point, bars splitting each step by the release that caused it (pp), and the New York Fed's weekly nowcast dashed
Q4 2026: the nowcast through the quarter and the news that moved it
Real GDP growth, % q/q annualised, re-run at each release date on today's data; news in percentage points by indicator; the New York Fed's weekly nowcast dashed.
The shaded band is the nowcast ± the record's RMSE at that day of the quarter (the accuracy-by-horizon table, days 0, 30, 60, 90; the latest on or before the point); at today's point, the record at today's equivalent data date — 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).
Source: BEA; Census Bureau; Federal Reserve Board; BLS; Federal Reserve Bank of New York (New York Fed Staff Nowcast, not an official forecast of the New York Fed, the Federal Reserve System or the FOMC); Macro Review calculations · data to 17 Nov 2026growth.nowcast_gdp-11

Exhibit 2 · Nowcast GDP growth by source, and by each of the site's indicators

GDP growth, Q4 2026, % q/q annualised, to 17 Nov 2026; each range is its source's own; then each indicator's growth, months in hand and prediction
About
EstimateGrowth, %RangeDated
The site's bridge equations2.7%0.8 to 4.6% out-of-sample record17 Nov 2026
New York Fed Staff Nowcast2.9%0.2 to 5.5% (its 80% band)13 Nov 2026
Pace the 2026 projection median implies for the rest of the year1.1%median 2.3% Q4/Q416 Sept 2026
BEA, Q3 2026 (the quarter before)3.4%–4 Nov 2026
EquationQuarter's growth, %Months in handPrediction, %
Real consumer spending (PCE)+2.8%0 of 32.6%
Real retail sales, control group+4.7%1 of 33.4%
Industrial production+3.3%0 of 33.1%
Aggregate weekly hours, production and nonsupervisory employees+0.9%1 of 32.3%
Housing starts−14.8%0 of 32.1%
About this exhibit
Two tables. Upper: the quarter's growth estimates by source (the site with its typical miss, the New York Fed with its band, the pace the Committee's median implies, and BEA's last quarter). Lower: each bridge equation's indicator growth, months in hand and prediction
Q4 2026: the site's nowcast and the published ones
Real GDP growth, % q/q annualised. Each range is its source's own: the site's is its record's typical miss, the New York Fed's its 80% band.
Source: BEA; Census Bureau; Federal Reserve Board; BLS; Federal Reserve Bank of New York (New York Fed Staff Nowcast, not an official forecast of the New York Fed, the Federal Reserve System or the FOMC); Macro Review calculations · data to 17 Nov 2026growth.nowcast_gdp-1

Industrial production moved it most since 29 Oct 2026: +0.12pp

Each recent release, back four weeks at most, and how much it moved the

Exhibit 3 · Each recent release's effect on the GDP nowcast

Q4 2026 nowcast change by release, pp, 29 Oct 2026 to 17 Nov 2026, with the nowcast after each; rows add to the total
About
ReleaseReleasedMonth addedMoved the nowcast, ppNowcast after, %
Nowcast on 29 Oct 20262.68%
Aggregate weekly hours, production and nonsupervisory employeesEmployment Situation (BLS)6 Nov 2026Oct 2026−0.10pp2.58%
Real retail sales, control groupAdvance Monthly Retail Sales (Census)17 Nov 2026Oct 2026+0.01pp2.59%
Industrial productionG.17 Industrial Production (Federal Reserve Board)17 Nov 2026Sept 2026+0.12pp2.71%
Total+0.03pp2.71%
About this exhibit
Table of the releases since the previous data date, each with release date, month added, the pp it moved the nowcast and the nowcast after
What moved the nowcast between 29 Oct 2026 and 17 Nov 2026
Each release added back in date order to the data as they stood on 29 Oct 2026; each row is its equation's slope ÷ 5 × the change in its quarterly growth, and the rows add to the total.
Source: BEA; Census Bureau; Federal Reserve Board; BLS; Macro Review calculations · data to 17 Nov 2026growth.nowcast_gdp-3

It has missed by 1.9pp (root mean square), against 2.5pp for last quarter's growth

Both models scored against BEA, the site's re-run at every past quarter's equivalent data date on today's data; how good it is at each point in the quarter; and the as the site published them

Exhibit 4 · Site and New York Fed nowcasts against BEA's estimate, quarterly since 2002

GDP growth, % q/q annualised, Q1 2002 to Q3 2026: site re-run at each quarter's data date, BEA's latest, the New York Fed; pandemic quarters dropped
About
About this exhibit
Line chart, quarterly: the site's nowcast re-run at each quarter's equivalent data date on today's revised data, BEA's latest estimate, and the New York Fed's final nowcasts as published since 2023; pandemic quarters left out
Both models against BEA, quarter by quarter
% q/q annualised, Q1 2002–Q3 2026. The site's nowcast is re-run at each quarter's equivalent data date on today's revised data; the New York Fed's final nowcasts are as it published them, since 2023. Pandemic quarters left out.
Source: BEA; Federal Reserve Bank of New York; Macro Review calculations · data to Q3 2026growth.nowcast_gdp-4

Exhibit 5 · Nowcast error by day of the quarter, against repeating last quarter's growth

Root mean square error and bias, pp of annualised growth, made at each day shown; Q1 2002 to Q3 2026, 93 quarters, pandemic quarters out
Made onRMSE, ppLast quarter's growth, RMSE, ppBias, ppMonths in hand, of 3Last quarter's GDP in handQuarters
Day 0 (the quarter's first day)2.222.83+0.470.0no93
Day 302.042.52+0.420.0yes93
Day 601.782.52+0.320.8yes93
Day 901.642.52+0.261.8yes93
About this exhibit
Table of the record's root mean square error, the benchmark's (last quarter's growth), bias and months in hand at days 0, 30, 60 and 90 of the quarter
The miss narrows from 2.2pp to 1.6pp through the quarter
The record re-run at each day of the quarter, Q1 2002–Q3 2026; pp of annualised growth, pandemic quarters left out. Today is day 47 of Q4 2026.
Each indicator is cut to the months its publisher had released by that day, on a stylised calendar: pce 29 days after the month's end (25–30); ctrl 16 days after the month's end (14–17); ip 48 days after the month's end (46–49); hours 5 days after the month's end (2–7); starts 18 days after the month's end (16–21). BEA's advance estimate comes about 29 days after the quarter ends.
Source: BEA; publishers' release calendars; Macro Review calculations · data to Q3 2026growth.nowcast_gdp-9

Exhibit 6 · Site and New York Fed nowcast errors against simple benchmarks, by sample

Errors, pp of annualised growth, Q1 2002 to Q3 2026: root mean square, mean absolute and mean (+ = too high); pandemic quarters left out
Estimate and sampleQuartersRMSE, ppMAE, ppBias, pp
Site, Q1 2002–Q3 2026, against BEA's latest931.861.36+0.33
… previous quarter's growth932.521.99+0.02
… long-run average to the quarter before932.241.52+0.53
Site, since Q1 2010611.511.19+0.05
… previous quarter's growth612.441.89+0.09
Site, Q1 2023–Q3 2026 (the New York Fed's record)151.281.04−0.21
… New York Fed's final nowcasts, real time151.491.26−0.39
… previous quarter's growth151.981.62−0.04
Site against BEA's first estimate (Q4 2025, Q1 2026, Q2 2026, Q3 2026)40.880.83+0.39
… New York Fed against the same41.030.94+0.59
About this exhibit
Table of root mean square error, mean absolute error and bias for the site's nowcast over several samples (the whole record, since 2010, the New York Fed's years, against BEA's first estimate), the New York Fed's final nowcasts, and the benchmarks (previous quarter's growth, long-run average)
Both models' scores against the naive benchmarks
Errors in percentage points of annualised growth: RMSE (root mean square), MAE (mean absolute) and bias (mean, + = too high). Pandemic quarters left out.
Source: BEA; Federal Reserve Bank of New York; Macro Review calculations · data to Q3 2026growth.nowcast_gdp-5

Exhibit 7 · Every GDP nowcast the site has published, with its outcome

GDP growth, % q/q annualised: each nowcast published since 24 Sep 2026, with data date and range, unrevised; the last before BEA's first estimate is scored

5 nowcasts on the record, the first made on data to 17 Sept 2026. Real GDP: 2 quarters nowcast, 1 with an outcome so far. The real-time score is printed once 8 quarters of a nowcast have an outcome; until then the page counts them.

PublishedData toNowcast ofValueRangeOutcomeMiss
this page17 Nov 2026Real GDP, Q4 20262.71%0.85% to 4.56% out-of-sample record––
16 Nov 202617 Nov 2026Real GDP, Q4 20262.59%0.68% to 4.51% out-of-sample record––
26 Oct 202620 Oct 2026Real GDP, Q3 20262.60%1.00% to 4.20% out-of-sample record3.37%−0.77pp
18 Oct 202616 Oct 2026Real GDP, Q3 20262.60%0.99% to 4.21% out-of-sample record––
5 Oct 20262 Oct 2026Real GDP, Q3 20263.71%2.09% to 5.34% out-of-sample record––
24 Sept 2026first on the record17 Sept 2026Real GDP, Q3 20262.39%0.70% to 4.09% out-of-sample record––
About this exhibit
Table of each GDP 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 its data date and range, kept as published; the last before BEA's first estimate of its quarter is scored against it.
Source: Macro Review (its own published nowcasts); BEAgrowth.nowcast_gdp-10

What the New York Fed's model says

The New York Fed Staff 's path through the quarter with its own bands, its nowcasts across quarters, and the releases that moved it

Exhibit 8 · New York Fed's GDP nowcast by weekly update, with its bands

GDP growth, % q/q annualised, by weekly update through the quarter, to 13 Nov 2026; shaded: the New York Fed's bands, published only for recent quarters
About
About this exhibit
Line chart of the New York Fed's nowcast for a quarter by weekly update, with its 50% and 80% bands; buttons switch between the previous, current and next quarter
Nowcast path for 2026 Q4, with the New York Fed's bands
Real GDP growth, % q/q annualised, by weekly update. Bands are published only in the weekly files for recent quarters.
Source: Federal Reserve Bank of New York, New York Fed Staff Nowcast; BEA NIPA Table 1.1.6.growth.nowcasts-1

Exhibit 9 · New York Fed's GDP nowcasts for each quarter, by weekly update

% q/q annualised, by weekly update, 2 Dec 2022 to 13 Nov 2026; one line per horizon (previous-quarter backcast, current-quarter nowcast, next-quarter forecast)
About this exhibit
Line chart of the New York Fed's GDP nowcasts by weekly update since it resumed tracking, one line per horizon (previous-quarter backcast, current-quarter nowcast, next-quarter forecast)
The New York Fed's nowcasts across quarters since tracking resumed
% q/q annualised, by weekly update, 2 Dec 2022 to 13 Nov 2026
Source: Federal Reserve Bank of New York (chart-data workbook, by horizon)growth.nowcasts-4

Exhibit 10 · Weekly impacts on the New York Fed's GDP nowcast, by data category

Impact on the current-quarter nowcast, pp of annualised growth, by category and week, to 13 Nov 2026; line: each week's total, the nowcast's change
About
About this exhibit
Stacked bars of weekly impacts on the New York Fed's current-quarter nowcast by data category (construction, production, surveys, consumption, income, labor, trade, prices, other), with a line for each week's total; buttons for recent weeks or all
Weekly impacts on the New York Fed's 2026 Q4 nowcast, by data category
Percentage points of annualised GDP growth. The line is the week's total, equal to the change in the nowcast.
Source: Federal Reserve Bank of New York (weekly news decomposition)growth.nowcasts-5

Exhibit 11 · Each release's impact on the New York Fed's latest nowcast update

Each release in the 13 Nov 2026 update, largest first: forecast, actual and impact, pp; then impact by category, recent and since tracking began
ReleaseForecastActualImpact
Data revisionsOther––−0.001
CPI-U: All itemsPrices · Oct · MoM % chg.0.330.35−0.001
Parameter revisionsOther––±0.000
CPI-U: All items less food and energyPrices · Oct · MoM % chg.0.260.30±0.000
CategoryLast 4 weeksSince tracking began
Construction−0.02−0.07
Production±0.00+0.02
Surveys+0.01−0.04
Consumption−0.03+0.10
Income+0.12+0.11
Labor+0.02−0.12
Trade±0.00±0.00
Prices±0.00−0.08
Other±0.00+0.38
About this exhibit
Table of each release in the New York Fed's latest update with forecast, actual and impact on the nowcast, largest first; a second table of impact by data category over the last four weeks and since tracking began
The New York Fed's latest update, 13 Nov 2026
Each release's impact on the nowcast, largest first; units as published.
Source: Federal Reserve Bank of New Yorkgrowth.nowcasts-6

How robust is each record?

The site's record for each equation alone, for one joint regression, with the pandemic quarters and one quarter earlier; the New York Fed's record in full

Exhibit 12 · The site's nowcast errors by equation, and under alternative setups

RMSE, MAE and bias, pp, Q1 2002 to Q3 2026: each equation alone, one regression on five indicators, pandemic quarters scored, and one quarter earlier
VariantQuartersRMSE, ppMAE, ppBias, pp
Pooled bridge equations (the page's nowcast)931.861.36+0.33
Real consumer spending (PCE) alone932.031.41+0.36
Real retail sales, control group alone931.901.41+0.25
Industrial production alone931.881.39+0.23
Aggregate weekly hours, production and nonsupervisory employees alone931.941.43+0.41
Housing starts alone932.201.59+0.42
One joint regression on the five indicators (not the model chosen before the record was run)931.631.30+0.16
Pandemic quarters scored too993.361.79+0.32
… previous quarter's growth, same quarters997.813.14−0.02
One quarter earlier (the next quarter's extrapolation)932.241.51+0.49
… growth two quarters before932.832.12±0.00

One joint regression on the five indicators scores better on this record (RMSE 1.63 against 1.86pp); the page keeps the pooled equations it set out with rather than choose the model by its score.

About this exhibit
Table of root mean square error, mean absolute error and bias for the pooled equations, each equation alone, one joint regression, the pandemic quarters scored and one quarter earlier, against benchmarks
How robust is the site's record?
The same pseudo-real-time record for each equation alone, for one regression on all five indicators together, with the pandemic quarters scored, and one quarter earlier.
Source: Macro Review calculationsgrowth.nowcast_gdp-7

Exhibit 13 · The New York Fed's nowcast error each quarter

15 quarters from Q1 2023 · RMSE 1.49pp · MAE 1.26pp · bias −0.39pp; bars: final nowcast less BEA's current estimate, percentage points
About
About this exhibit
Bar chart of the New York Fed's final-nowcast error (final nowcast less BEA's current estimate), one bar per quarter of its current model, with its final nowcasts and BEA's estimates; a button switches to the 2016–2021 real-time record
Track record: final nowcasts against BEA
15 quarters · RMSE 1.49pp · MAE 1.26pp · bias −0.39pp. Bars: error (final nowcast − BEA current estimate), % points.
Source: Federal Reserve Bank of New York (nowcasts; advance estimates from its history metadata); BEA NIPA Table 1.1.6, current estimates; scores computed here.growth.nowcasts-2

Exhibit 14 · The New York Fed's error by horizon, 2023Q1–2026Q3

Nowcast k weeks before the final update against BEA's current estimate.
About
Weeks before final updateRMSE, ppBias, ppQuarters
Final (week of advance release)1.49−0.3915
4 weeks1.57−0.5015
8 weeks1.48−0.5515
12 weeks1.62−0.4215
Source: Federal Reserve Bank of New York; BEA; Macro Review calculationsgrowth.nowcasts-7

Exhibit 15 · The New York Fed's nowcast errors, by sample period

Final nowcast errors, pp: RMSE, MAE and bias by sample, against BEA's current and advance estimates; 2026 quarters against its band
SampleQuartersRMSEMAEBias
2023–present vs BEA current151.491.26−0.39
2023–present vs BEA advance41.030.94+0.59
2016–2021 real time, ex 2020 Q2–2021 Q1181.911.43−0.46
2016–2021 real time, all225.742.90−0.77
2026 Q1: 80% band 0.12–4.50%, BEA 2.49% — inside
2026 Q2: 80% band 0.69–5.01%, BEA 2.22% — inside
2026 Q3: 80% band 0.07–5.42%, BEA 3.37% — inside
About this exhibit
Table of the New York Fed's final-nowcast root mean square error, mean absolute error and bias by sample (its current model against BEA's current and advance estimates, the 2016–2021 real-time record with and without the pandemic), plus this year's quarters against its 80% band
The New York Fed's scores side by side
Final nowcast errors, percentage points.
Source: Federal Reserve Bank of New York; BEA; Macro Review calculationsgrowth.nowcasts-8

Appendix

The model in notation, its parameters, its inputs and what it leaves out61 checks, all pass
About this page

A model-conditional nowcast: five bridge equations read the month's releases into this quarter's real GDP growth, and the page sets the result beside the New York Fed's model and the Committee's projection, shows which release moved it, and scores both models against BEA.

The bridge equations: GDP growth on one indicator at a time

yq=100[(GDPqGDPq−1)4−1],xi,q=100[(Xˉi,qXˉi,q−1)4−1]y_q = 100\left[\left(\frac{GDP_q}{GDP_{q-1}}\right)^{4}-1\right], \qquad x_{i,q} = 100\left[\left(\frac{\bar X_{i,q}}{\bar X_{i,q-1}}\right)^{4}-1\right]
yq=ai+bi xi,q+ei,q,ei,q∼(0, σi2),i=1,…,Ky_q = a_i + b_i\,x_{i,q} + e_{i,q}, \qquad e_{i,q} \sim (0,\ \sigma_i^2), \qquad i = 1,\dots,K
y^q=1K∑i=1K(a^i+b^i x^i,q),K=5\hat y_q = \frac{1}{K}\sum_{i=1}^{K}\left(\hat a_i + \hat b_i\,\hat x_{i,q}\right), \qquad K = 5
SymbolMeaningFromAssumed or estimated
yqy_qReal GDP growth in quarter q, % annualisedBEA, NIPA Table 1.1.6 (chained dollars)–
Xˉi,q\bar X_{i,q}The quarter's mean monthly level of indicator iReal consumer spending (PCE); Real retail sales, control group; Industrial production; Aggregate weekly hours, production and nonsupervisory employees; Housing starts–
xi,qx_{i,q}Indicator i's quarterly growth, % annualised; the target quarter's months not yet released are filled (below)computed here–
ai, bia_i,\ b_iIntercept and slope of bridge iOLS, the parameter tableestimated
σi\sigma_iStandard deviation of bridge i's residualOLS, the parameter tableestimated
KKNumber of equations, pooled with equal weights 1/Kchosen before any record was computedassumed

Ordinary least squares, one equation per indicator, on the quarters from 1993Q1 to 2026Q3 with 2020Q1–2021Q2 left out (129 quarters an equation); classical (homoskedastic) OLS standard errors in the parameter table. The weights are equal by design, not estimated.

The ragged edge: each indicator's missing months

gi,t=100 Δln⁡Xi,tg_{i,t} = 100\,\Delta \ln X_{i,t}
g^i,t+h=c^i+ϕ^i g^i,t+h−1,X^i,t+h=X^i,t+h−1 eg^i,t+h/100,h=1,…,3−mi\hat g_{i,t+h} = \hat c_i + \hat\phi_i\,\hat g_{i,t+h-1}, \qquad \hat X_{i,t+h} = \hat X_{i,t+h-1}\,e^{\hat g_{i,t+h}/100}, \qquad h = 1,\dots,3-m_i
SymbolMeaningFromAssumed or estimated
Xi,tX_{i,t}Indicator i's monthly levelthe publisher's file (Data tab)–
mim_iMonths of the target quarter released at the data datethe publishers' release calendars–
ci, ϕic_i,\ \phi_iIntercept and autoregressive coefficient of indicator i's monthly growthOLS, the parameter tableestimated

One lag (an assumption), estimated by OLS on the months from 1993 to the last month released, Mar 2020 to Jun 2021 left out (389 months for real consumer spending (pce)). Forecasts feed the next month's lag. The filled months and their growth rates are printed in the parameter table.

The news: what one release does to the nowcast

Δy^q(r)=b^iK Δx^i,q(r)\Delta \hat y_q^{(r)} = \frac{\hat b_i}{K}\,\Delta \hat x_{i,q}^{(r)}
y^q(t)−y^q(t′)=∑r ∈ (t′, t]Δy^q(r)\hat y_q(t) - \hat y_q(t') = \sum_{r\,\in\,(t',\,t]} \Delta \hat y_q^{(r)}
SymbolMeaningFromAssumed or estimated
rrA release of indicator i between two data dates t′ and tthe publishers' release calendars–
Δx^i,q(r)\Delta \hat x_{i,q}^{(r)}The change release r makes to indicator i's (filled) quarterly growthcomputed here–

An indicator enters only its own equation, so a release's news does not depend on the order the releases are added. The identity is checked at every step of the path and over the news window (tolerance 1e−9).

The record and the band

RMSE(τ)=1n∑q(y^q τ−yq)2,bandt=y^q(t)±RMSE(τt)\mathrm{RMSE}(\tau) = \sqrt{\frac{1}{n}\sum_{q}\left(\hat y_q^{\,\tau} - y_q\right)^2}, \qquad \text{band}_t = \hat y_q(t) \pm \mathrm{RMSE}(\tau_t)
SymbolMeaningFromAssumed or estimated
y^q τ\hat y_q^{\,\tau}The nowcast of quarter q re-run with the information of day τ of the quartercomputed here, on today's revised data–
τt\tau_tThe latest horizon (days 0, 30, 60, 90) on or before the point's day of the quarterthe accuracy-by-horizon tableassumed
nnQuarters scored2002Q1–2026Q3, pandemic quarters left out–

Pseudo-real-time on revised data: each quarter is estimated only on data before its date, but on today's revised data, so the record flatters the model against a truly real-time one. The band is the record's typical miss, a backtest range — not a probability.

Exhibit 16 · The bridge equations' parameters

Indicatorâ (SE)b̂ (SE)σ̂R²QuartersAR ĉ (SE)AR φ̂ (SE)AR σ̂Filled months, % a monthPrediction
Real consumer spending (PCE)Q1 1993 to Q3 20260.49 (0.28)0.767 (0.085)1.750.391290.275 (0.021)-0.220 (0.050)0.337Oct 2026 +0.27; Nov 2026 +0.21; Dec 2026 +0.232.61%
Real retail sales, control groupQ1 1993 to Q3 20261.34 (0.22)0.439 (0.054)1.810.341290.306 (0.034)-0.294 (0.048)0.634Nov 2026 +0.23; Dec 2026 +0.243.41%
Industrial productionQ1 1993 to Q3 20262.10 (0.16)0.305 (0.034)1.750.391290.106 (0.032)0.170 (0.050)0.612Oct 2026 +0.22; Nov 2026 +0.14; Dec 2026 +0.133.11%
Aggregate weekly hours, production and nonsupervisory employeesQ1 1993 to Q3 20261.89 (0.18)0.518 (0.064)1.820.341290.126 (0.021)-0.127 (0.050)0.396Nov 2026 +0.15; Dec 2026 +0.112.34%
Housing startsQ1 1993 to Q3 20262.52 (0.19)0.031 (0.007)2.090.12129-0.015 (0.349)-0.409 (0.046)6.885Oct 2026 +2.17; Nov 2026 −0.90; Dec 2026 +0.352.07%
About this exhibit
Estimates with classical standard errors in brackets; the bridge in % annualised, the autoregression in % a month; the months each indicator lacks are filled by its autoregression.
Source: BEA; Census Bureau; Federal Reserve Board; BLS; Macro Review calculationsgrowth.nowcast_gdp-6

Exhibit 17 · Definitions and caveats

  • Model
    Model-conditional bridge equations.

    Model-conditional: the nowcast is what five bridge equations estimated on BEA's real GDP since 1993 make of the monthly data released so far, not a forecast in any other sense. Each equation regresses real GDP growth (% q/q annualised) on the quarterly growth of one monthly indicator — the quarter's average level against the previous quarter's, annualised — and the nowcast is the equal-weight average of the five predictions.

  • Ragged edge
    Months not yet published are filled.

    Ragged edge: an indicator's missing months up to the end of the quarter are filled by its own autoregression on monthly log growth (lag 1), estimated from 1993 with Mar 2020 to Jun 2021 left out; the bridge equations leave out the quarters that hold those months, Q1 2020 to Q2 2021.

  • News
    What moved it, release by release.

    News: the previous data date is 28 days before the latest release, or the day this quarter's nowcast began when that is later, as it is now (29 Oct 2026, BEA's first estimate of the quarter before). Each input is cut back to the months released by then, and the releases since are added back one at a time in date order; because each indicator enters only its own equation, each release moves the nowcast by its equation's slope ÷ 5 × the change in that indicator's quarterly growth, whatever the order.

  • Record
    Pseudo-real-time on revised data.

    Record: at each quarter since 2002Q1 the model is re-estimated with GDP to the quarter before and each indicator cut to the months of the quarter it carries today, then scored against BEA's latest estimate. That is pseudo-real-time on revised data: the data are today's, and both GDP and the indicators have been revised since, so the record flatters the model against a truly real-time one. Benchmarks: the previous quarter's growth and the average since 1993 to the quarter before.

  • Horizon
    The record at each day of the quarter.

    Accuracy by horizon: the record re-run at days 0, 30, 60, 90 of each quarter since 2002, each indicator cut to the months its publisher had released by that day on a stylised calendar — its median publication lag over its six latest releases on the publishers' calendars (real consumer spending (pce) 29 days after the month's end, real retail sales, control group 16 days after the month's end, industrial production 48 days after the month's end, aggregate weekly hours, production and nonsupervisory employees 5 days after the month's end, housing starts 18 days after the month's end) — and GDP to the quarters BEA had estimated, its advance estimate coming 29 days after the quarter's end on its schedule. Pseudo-real-time on revised data, like the record, and scored on the same quarters.

  • 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, so revisions to the data cannot rewrite it. Beside the path recomputed on today's data, the page draws those published points; once eight quarters have an outcome it scores them against BEA, a real-time record; until then it counts them.

  • Published
    Copied, not recomputed.

    The New York Fed Staff Nowcast and the projection median are reference copies of the site's Nowcasts and What is priced pages, re-read and compared on every update. The median is Q4-to-Q4 growth for the year; the pace it implies for the rest of the year is solved from BEA's growth so far. The Atlanta Fed's GDPNow is quoted on the Nowcasts page only and not used here.

  • Not modelled
    What the equations leave out.

    Not modelled: net exports — the site reads only the nominal goods and services balance, not the real (chained-dollar) goods trade exhibit; inventory investment — the site reads only nominal book values of business inventories and holds no inventory deflators; government spending — the site reads no monthly source for federal, state and local purchases; business equipment and structures — the site reads only nominal new orders of core capital goods, not shipments or construction spending.

  • Notice
    Source and non-affiliation.

    The New York Fed Staff Nowcast is not an official forecast of the Federal Reserve Bank of New York, its president, the Federal Reserve System, or the Federal Open Market Committee.

Source: BEA; Census Bureau; Federal Reserve Board; BLS; Federal Reserve Bank of New York (New York Fed Staff Nowcast, not an official forecast of the New York Fed, the Federal Reserve System or the FOMC); Macro Review calculationsgrowth.nowcast_gdp-8

Exhibit 18 · The New York Fed's model: definitions and caveats

  • Model
    A dynamic factor model that reads news.

    The New York Fed Staff Nowcast (version 2.0, reintroduced September 2023) is a dynamic factor model estimated with Bayesian methods and the Kalman filter on a broad set of monthly and quarterly indicators. Each Friday it reads the week's releases as news: the difference between each release and the model's own forecast of it, times the series' weight, is that release's impact on the GDP nowcast. Updates for a quarter begin after BEA's second estimate for the quarter two before and end with the advance estimate.

  • Bands
    The model's own uncertainty.

    The 50–80% bands are the New York Fed's, from the model's predictive density.

  • Scoring
    Against the latest GDP, and the advance where on file.

    Errors are final nowcast minus real GDP growth in BEA's latest estimates (4 Nov 2026), which include later revisions. A true real-time score would compare each nowcast with BEA's advance estimate for that quarter; the New York Fed's metadata carries advance estimates only for its most recent quarters (), and the page uses them where available. Quarters from 2016Q1 to 2021 come from the legacy file (real-time forecasts of the earlier model; 2002–2015 reconstructions are excluded). The legacy file ends in August 2021 and the current series begins with forecasts for 2023 Q1, so 2021 Q3 – 2022 Q4 are not scored.

  • Files
    Two files, one path.

    The New York Fed publishes this path twice: in a weekly file to three decimals, which the page draws, and in a chart-data workbook to two. They print different numbers for 2 weekly updates of the path for Q4 2026: 25 Sep 2026, 2.592 against 2.60; 2 Oct 2026, 2.548 against 2.54. Every other week agrees to two decimals (the author's decision of 16 Nov 2026 (D-2026-11-16-NYF1)).

  • Sensitivity
    The window matters.

    Scores shift with the window: including the 2020 Q2 – 2021 Q1 quarters in the legacy record raises its RMSE to 5.74pp.

  • Notice
    Source and non-affiliation.

    Source: Federal Reserve Bank of New York. The New York Fed Staff Nowcast is not an official forecast of the Federal Reserve Bank of New York, its president, the Federal Reserve System, or the Federal Open Market Committee. The New York Fed is not affiliated with this site and does not endorse it.

Source: Federal Reserve Bank of New York. The New York Fed Staff Nowcast is not an official forecast of the Federal Reserve Bank of New York, its president, the Federal Reserve System, or the Federal Open Market Committee. The New York Fed is not affiliated with this site and does not endorse it.growth.nowcasts-9

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.