Macro ReviewHow to read
Outlook · Macro expectations · Inflation nowcast

The site's models put November CPI at +0.09% a month, near the Cleveland Fed

What will the next CPI and core PCE prints say on the site's own , and where is the pressure coming from?

Data to 10 Nov 2026 · updated 17 Nov 2026Next CPI: Nov 2026
CPI, Nov 2026 (site)+0.09%m/m SA · Cleveland Fed +0.14% · data to 10 Nov 2026 · energy −0.15pp
Core CPI, Nov 2026 (site)+0.25%m/m SA · 2.9% y/y · record ±0.11pp
Core PCE, Oct 2026 (site)+0.30%m/m · 3.2% y/y · bias-corrected line, raw +0.26% · mapped from the published CPI · record ±0.08pp; raw bridge runs 0.04pp low
Core PCE, Nov 2026 (site)+0.25%m/m · 3.3% y/y · bias-corrected line, raw +0.20% · CPI still nowcast · record ±0.10pp
Pace implied by the 2026 core PCE median+0.33%a month, Oct–Dec · median 3.4% Q4/Q4 (16 Sep 2026)
Pump price, first weeks of the month−5.6%2 weeks to 9 Nov 2026 vs last month's average (EIA)

In brief

The site's numbers for the next prints beside the Cleveland Fed's, the energy leg, what moved them and how good they have been
  • The site's models put Nov 2026 CPI at +0.09% on the month, seasonally adjusted, on data released to 10 Nov 2026.→ Exhibit 2
  • Core CPI for Nov 2026 comes out at +0.25% on the month, 2.9% on the year.→ Exhibit 2
  • The Cleveland Fed's model puts Nov 2026 CPI at +0.14% (10 Nov 2026), against the site's +0.09%.→ Exhibit 2
  • Core PCE for Oct 2026 maps to +0.30% on the month (3.2% on the year) from the CPI components already published.→ Exhibit 2
  • The Committee's 2026 core PCE median of 3.4% implies +0.33% a month from October to December.→ Exhibit 2
What would change this view
What would change this viewTriggerTodayDecided by
Core PCE for Oct 2026 missespublished change outside 0.21 to 0.38% ( ± record )nowcast +0.30% (10 Nov 2026)pendingPersonal Income and Outlays for Oct 2026, 25 Nov 2026 (BEA)
Core CPI for Nov 2026 missespublished change outside 0.14 to 0.36% (nowcast ± record RMSE)nowcast +0.25% (10 Nov 2026)pendingCPI for Nov 2026, 10 Dec 2026 (BLS)
The pump price does not reach the CPICPI energy commodities for Nov 2026 moving against the pump price's −5.6% by more than the 's residual (2.0pp)nowcast −3.5% for energy commoditiespendingCPI for Nov 2026, 10 Dec 2026 (BLS)
Core PCE for Nov 2026 missespublished change outside 0.14 to 0.35%nowcast +0.25% (10 Nov 2026)pendingPersonal Income and Outlays for Nov 2026, 23 Dec 2026 (BEA)
The reasoning behind each point
  • CPI
    The site's models put Nov 2026 CPI at +0.09% on the month, seasonally adjusted, on data released to 10 Nov 2026.

    That would be 3.9% on the year. Energy contributes −0.15pp: the pump price's first 2 weekly readings of the month average −5.6% against last month's, and energy commodities follow it about one for one (β = 1.02). Food adds +0.03pp and core +0.20pp.

  • Core CPI
    Core CPI for Nov 2026 comes out at +0.25% on the month, 2.9% on the year.

    Half the bottom-up (core goods +0.09%, core services +0.31%, weighted +0.26%) and half the direct autoregression (+0.25%).

  • Cleveland Fed
    The Cleveland Fed's model puts Nov 2026 CPI at +0.14% (10 Nov 2026), against the site's +0.09%.

    Its core CPI reads +0.21% against the site's +0.25%; its core PCE for Oct 2026 +0.27% against +0.30%. The Cleveland Fed's model reads daily oil and weekly gasoline prices and the past prints; the site's adds each CPI part's own autoregression and maps the published CPI into core PCE.

  • Core PCE
    Core PCE for Oct 2026 maps to +0.30% on the month (3.2% on the year) from the CPI components already published.

    The bias-corrected line: the bridge reads +0.26% and the record's mean error to date is −0.037pp a month, so the published line is +0.30% — the correction declared on 3 Oct 2026; both lines are scored from the next Personal Income and Outlays release on. Nov 2026 follows at +0.25% (3.3% on the year; the raw bridge +0.20%), with the CPI components still nowcast. Core PCE runs above core CPI on the year (3.2% against 2.9%) because the two weight prices differently: PCE gives shelter half CPI's weight and health care far more. The largest weights are health care (19.4% of core spending), shelter (17.6%) and other goods (14.5%); health care's CPI stand-in explains little of it, because BEA prices hospitals and physicians off PPI indexes whose history this site does not hold.

  • Projection
    The Committee's 2026 core PCE median of 3.4% implies +0.33% a month from October to December.

    That is 4.0% annualised, against the site's +0.30% for Oct 2026 and +0.25% for Nov 2026. The median is Q4 to Q4 and rounded to a tenth. The raw core PCE record's mean error is −0.037pp a month (Newey–West t −4.1); the published line removes it, so the gap of −0.03pp between this nowcast and the median's pace is read on the bias-corrected line (the raw bridge's gap is −0.07pp).

  • News
    Since 13 Oct 2026 the Nov 2026 CPI nowcast moved −0.06pp, most of it from the November pump prices (−0.11pp).

    Core PCE for Oct 2026 moved +0.12pp, most of it from the October CPI (+0.06pp).

  • Record
    Since 2015 the bias-corrected core PCE nowcast has missed by 0.08pp a month (root mean square), against 0.11pp for last month's print.

    The average miss is the root mean square error over 99 months, made after the month's CPI. On those months the raw bridge misses by 0.09pp with a mean error of −0.047pp; the bias-corrected line misses by 0.08pp with a mean error of −0.031pp, and its ± RMSE band held the print in 66% of months (the raw band 73%). Diebold–Mariano t of corrected against raw 3.6 (p < 0.01); the correction was declared on 3 Oct 2026, not chosen by this score. The raw record's mean error (bias) is −0.037pp over 123 months, and the intercepts are re-estimated every month. Core CPI: 0.11pp against 0.12pp for last month's print (123 months); headline CPI: 0.15pp against 0.28pp. Core PCE before the CPI: 0.10pp against 0.11pp. Pseudo-real-time on revised data: each month's models see only data before its equivalent data date, but today's seasonally adjusted figures, which BLS and BEA revise. Core PCE after the CPI: mean error −0.037pp (Newey–West t −4.1, p < 0.01); against the random walk, Diebold–Mariano t 2.3 (p 0.02). Core CPI: mean error −0.029pp (Newey–West t −2.4, p 0.02); against the random walk, Diebold–Mariano t 0.7 (p 0.47). The raw core PCE bridge runs persistently below the print — about 0.44pp a year; the intercept correction declared on 3 Oct 2026 is the published line, and the raw line stays beside it. Newey–West lags: floor(4(n/100)^(2/9)), 4 here. Made 28 days before the CPI release, headline CPI's record RMSE is 0.150pp; by the release day, with the month's pump prices in, 0.124pp.

Housing supply 2027–28 · The thesis fails if CPI shelter runs above 3.5% annualised over three months (now 3.9%, Oct 2026) or Apartment List rents rise above +2% y/y by December (subscription, not tracked; −0.8% in August per the September report), or if NAR months' supply falls below 4.0 (4.9 in August, quoted); the housing objection to a sequence falls if new-home months' supply drops below 8.0 by Q1 2027 (now 8.4).

Construction pipeline · The read that the multifamily pipeline is refilling fails if 5-or-more starts (three-month average) fall back below 5-or-more completions for three consecutive months or the 5-or-more pipeline falls below 22.9 months of completions (halfway to its 2015–19 norm).

Next: Labour nowcast — this month's hiring, nowcast

BLS; EIA; BEA; Federal Reserve Board (projection median); Federal Reserve Bank of Cleveland; Macro Review calculationsinflation.nowcast-read

November CPI is tracking +0.09% on the month, give or take 0.15pp

The next CPI print through the last four weeks with the news that moved it and the record's typical miss, beside the Cleveland Fed's ; the next core CPI and core PCE prints and the pace the projection median implies

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

Month-on-month change, % seasonally adjusted, at each release, 13 Oct 2026 to 10 Nov 2026; band: the record's typical miss; bars: each release's news, pp
Range: out-of-sample record
About this exhibit
Line chart of the site's nowcast of the next CPI print (m/m %, seasonally adjusted) 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; buttons switch to core CPI and core PCE
CPI, Nov 2026: the nowcast through four weeks and the news that moved it
Month-on-month change, % seasonally adjusted, re-run at each release date on today's data; news in percentage points by release.
The shaded band is the nowcast ± the record's RMSE at the nearest horizon (28, 21, 14, 7, 0 days before the release) at or beyond the point's days to the release; 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 10 Dec 2026.
Source: BLS; EIA; BEA; Federal Reserve Board (projection median); Federal Reserve Bank of Cleveland; Macro Review calculations · data to 10 Nov 2026inflation.nowcast-13

Exhibit 2 · Site and Cleveland Fed nowcasts of the next CPI and core PCE

Month-on-month change, % seasonally adjusted, Nov 2026; range = each model's typical miss, not a probability; Cleveland Fed, 10 Nov 2026; then each CPI part, pp
About
PrintSite, m/m %Typical missCleveland Fed, m/m %Site, y/y %Latest published
CPI, Nov 2026+0.09%±0.15 out-of-sample record+0.14%3.9%+0.35% (Oct 2026)
Core CPI, Nov 2026+0.25%±0.11 out-of-sample record+0.21%2.9%+0.30% (Oct 2026)
Core PCE, Oct 2026mapped from the published CPI · bias-corrected+0.30%±0.08 out-of-sample record+0.27%3.2%+0.32% (Sep 2026)
Core PCE, Oct 2026, raw bridgebefore the record's mean error (−0.037pp) is removed+0.26%±0.09 out-of-sample record–3.2%–
Core PCE, Nov 2026the CPI components still nowcast · bias-corrected+0.25%±0.10 out-of-sample record+0.26%3.3%+0.32% (Sep 2026)
Core PCE, Nov 2026, raw bridgebefore the record's mean error (−0.044pp) is removed+0.20%±0.10 out-of-sample record–3.3%–

The Committee's 2026 core PCE median of 3.4% (Q4/Q4, 16 Sept 2026) implies +0.33% a month to December.

Declared on 3 Oct 2026, the author's decision of that day: the published core PCE line is the bridge less the record's mean error to date (−0.037pp a month after the CPI, −0.044pp before it); the raw bridge stays beside it; both are scored from the next Personal Income and Outlays release.

CPI part, Nov 2026Nowcast, %Contribution, pp
Food+0.24%+0.03pp
Energy commodities−3.53%−0.15pp
Energy services+0.18%+0.01pp
Core goods+0.09%+0.02pp
Core services+0.31%+0.19pp
About this exhibit
Two tables. Upper: the next prints (CPI, core CPI, core PCE), each with the site's m/m nowcast, typical miss, the Cleveland Fed's nowcast, the site's y/y and the latest published print. Lower: each CPI part (food, energy commodities, energy services, core goods, core services) with its nowcast and contribution in pp
The next prints on the site's models and the Cleveland Fed's
Month-on-month change, % seasonally adjusted; the range is each model's typical miss, not a probability. The Cleveland Fed's is its nowcast of 10 Nov 2026.
Source: BLS; EIA; BEA; Federal Reserve Board (projection median); Federal Reserve Bank of Cleveland; Macro Review calculations · data to 10 Nov 2026inflation.nowcast-1

Exhibit 3 · Core CPI and core PCE monthly changes, with the site's nowcasts

% month on month, seasonally adjusted, Nov 2024 to Nov 2026; lines: published prints; diamonds: the site's nowcasts; no CPI for Oct 2025: the line breaks
About this exhibit
Line chart of published core CPI and core PCE (m/m %, seasonally adjusted) over two years, with diamonds for the site's nowcasts of the months not yet published; the CPI line breaks at the month BLS did not publish
The site's next core prints: CPI +0.25%, core PCE +0.25% a month
Core CPI and core PCE, % month on month, seasonally adjusted; the diamonds are the site's nowcasts for the months not yet published. BLS published no CPI for October 2025, so the line breaks there.
Source: BLS and BEA; nowcasts computed here.inflation.nowcast-2

The November pump prices moved the CPI most: −0.11pp

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

Exhibit 4 · Each release's effect on the CPI, core CPI and core PCE nowcasts

Change in each Nov 2026 nowcast by release, pp, 13 Oct 2026 to 10 Nov 2026; pump prices summed; rows add to the total
About
ReleaseReleasedCPI, Nov 2026Core CPI, Nov 2026Core PCE, Oct 2026Core PCE, Nov 2026
Nowcast on 13 Oct 2026+0.15%+0.20%+0.17%+0.17%
CPI for Sep 2026 (BLS)14 Oct 2026+0.02+0.03+0.02+0.02
Pump prices, Oct 2026 weeks (EIA) · 3 weeks27 Oct 2026±0.00±0.00±0.00±0.00
PCE for Sep 2026 (BEA)29 Oct 2026±0.00±0.00±0.00±0.00
CPI for Oct 2026 (BLS)10 Nov 2026+0.03+0.02+0.06+0.01
Pump prices, Nov 2026 weeks (EIA) · 2 weeks10 Nov 2026−0.11±0.00±0.00±0.00
Total change−0.06+0.05+0.08+0.03
About this exhibit
Table of the releases since the previous data date (the CPI, the weekly pump prices, PCE), each with how much it moved the CPI, core CPI and core PCE nowcasts, and the totals
What moved the nowcasts between 13 Oct 2026 and 10 Nov 2026
Each release added back in date order to the data as they stood four weeks earlier; percentage points of the month-on-month change. The weekly pump prices are summed by calendar month. The rows add to the total.
Source: Computed here on the publishers' data and release calendars.inflation.nowcast-3

Bias-corrected core PCE has missed by 0.08pp a month (root mean square), against 0.11pp for last month's print

The models re-run at every past month's equivalent data date on today's data, against the published print and a random walk; how good they are at each point before the release; and the as the site published them

Exhibit 5 · The site's core PCE nowcast against the published print, monthly since 2015

Core PCE, % month on month, since 2015: the site's nowcast with that month's CPI in hand, against the published change; March 2020–June 2021 left out
About
About this exhibit
Chart of the core PCE record: the site's nowcast re-run each month with that month's CPI in hand, on today's revised data, against the published core PCE change, % m/m
The record: core PCE mapped from the CPI, re-run each month
% month on month. Each month's bridges are estimated on data before that month's PCE, with that month's CPI in hand, on today's revised data; March 2020 to June 2021 left out.
Source: Scores computed here on BEA and BLS data.inflation.nowcast-4

Exhibit 6 · Each inflation nowcast's error by days before its release

RMSE and bias, pp a month, by days before release, since Jan 2015: CPI 123 months, core PCE 125; random walk = the latest published month
Days before the CPI releaseCPI, RMSEBiasCore CPI, RMSEBiasRandom walk, coreWeeks of pump prices in hand
28 days0.150−0.0290.110−0.0280.1181.9
21 days0.135−0.0280.110−0.0280.1182.9
14 days0.125−0.0270.110−0.0280.1183.9
7 days0.124−0.0320.110−0.0280.1184.3
release day0.124−0.0320.110−0.0280.1184.3
Days before the PCE releaseCore PCE, RMSEBiasBias-corrected, RMSERandom walkThe month's CPI in hand
28 days0.105−0.0440.1030.105no
21 days0.105−0.0440.1030.105no
14 days0.084−0.0360.0830.10598% of months
7 days0.084−0.0360.0830.10598% of months
release day0.084−0.0360.0830.10598% of months

Publication lags, each the median of its latest releases on the publishers' calendars: the CPI 12 days after the month's end, PCE 29, the pump price 1 day after its Monday. Record months no horizon row scores (a horizon could not be run there): Mar 2015 (CPI).

About this exhibit
Two tables: root mean square error and bias of headline CPI, core CPI and the random walk at each horizon before the CPI release, with weeks of pump prices in hand; and core PCE (raw and bias-corrected) and the random walk at the same horizons before the PCE release
Core CPI's typical miss is 0.11pp at every point from 28 days before the release
The record re-run 28, 21, 14, 7, 0 days before each release, each input cut to what its publisher had released by then: the CPI (headline and core) before the CPI release, since Jan 2015 (123 months); core PCE before the PCE release, since Jan 2015 (125 months). Percentage points of the month-on-month change; months where every horizon can be run, so the rows compare like with like. The random walk is the latest published month at each point.
Source: Scores computed here; release lags from the publishers' calendars.inflation.nowcast-11

Exhibit 7 · Each inflation nowcast's errors against repeating last month's print

Errors, pp of the monthly change, since Jan 2015: RMSE, MAE, bias, bias t; DM t: Diebold–Mariano test against the random walk, + = the site's smaller
EstimateMonthsRMSEMAEBiasBias tDM t vs random walk
CPI, site1230.1470.104−0.029−1.7 (p 0.09)+3.0 (p < 0.01)
… random walk1230.2780.198+0.001
Core CPI, site1230.1100.081−0.029−2.4 (p 0.02)+0.7 (p 0.47)
… random walk1230.1180.088+0.003
Core PCE after the CPI, site1230.0850.066−0.037−4.1 (p < 0.01)+2.3 (p 0.02)
… random walk1230.1060.082±0.000
Core PCE before the CPI, site1230.1050.078−0.044−3.5 (p < 0.01)+0.8 (p 0.42)
… random walk (two months back)1230.1130.089+0.002
Core PCE line, against the published changeMonthsRMSEInside ± RMSEMean errorDM t, corrected vs raw
Bias-corrected (the published line), after the CPI990.08466%−0.0313.6
Raw bridge, same months, after the CPI990.09373%−0.047–
Bias-corrected (the published line), before the CPI990.10476%−0.0293.1
Raw bridge, same months, before the CPI990.11476%−0.055–
About this exhibit
Two tables: root mean square error, mean absolute error, bias, the bias's t and the Diebold–Mariano t against the random walk for CPI, core CPI and core PCE (after and before the CPI), each beside its random walk; and the bias-corrected and raw core PCE lines against each other
Core PCE beats the random walk (DM p 0.02); core CPI does not
Errors in percentage points of the month-on-month change, since Jan 2015: RMSE, MAE and bias (+ = too high); the bias's t with Newey–West errors (4 lags), and Diebold–Mariano on squared errors against the random walk (+ = the site's smaller), two-sided. The random walk is the latest published month as it stood at the time.
Source: Scores computed here.inflation.nowcast-5

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

% month on month: each inflation nowcast published since 24 Sep 2026, with data date and range, unrevised; outcome and miss once the print is out

6 nowcasts on the record, the first made on data to 15 Sept 2026. CPI: 3 months nowcast, 2 with an outcome so far; Core CPI: 3 months nowcast, 2 with an outcome so far; Core PCE: 4 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.

PublishedData toNowcast ofValueRangeOutcomeMiss
13 Nov 202610 Nov 2026CPI, Nov 2026+0.09%−0.06% to +0.23% out-of-sample record––
13 Nov 202610 Nov 2026Core CPI, Nov 2026+0.25%+0.14% to +0.36% out-of-sample record––
13 Nov 202610 Nov 2026Core PCE, Oct 2026+0.30%+0.21% to +0.38% out-of-sample record––
13 Nov 202610 Nov 2026Core PCE, Nov 2026+0.25%+0.14% to +0.35% out-of-sample record––
8 Nov 20263 Nov 2026CPI, Oct 2026+0.26%+0.13% to +0.38% out-of-sample record+0.35%−0.09pp
8 Nov 20263 Nov 2026Core CPI, Oct 2026+0.26%+0.15% to +0.37% out-of-sample record+0.30%−0.04pp
8 Nov 20263 Nov 2026Core PCE, Oct 2026+0.24%+0.14% to +0.35% out-of-sample record––
31 Oct 202629 Oct 2026CPI, Oct 2026+0.26%+0.13% to +0.38% out-of-sample record––
31 Oct 202629 Oct 2026Core CPI, Oct 2026+0.26%+0.15% to +0.37% out-of-sample record––
31 Oct 202629 Oct 2026Core PCE, Oct 2026+0.24%+0.14% to +0.35% out-of-sample record––
26 Oct 202620 Oct 2026CPI, Oct 2026+0.25%+0.12% to +0.38% out-of-sample record––
26 Oct 202620 Oct 2026Core CPI, Oct 2026+0.26%+0.15% to +0.37% out-of-sample record––
26 Oct 202620 Oct 2026Core PCE, Sep 2026+0.25%+0.17% to +0.33% out-of-sample record+0.32%−0.07pp
26 Oct 202620 Oct 2026Core PCE, Oct 2026+0.24%+0.13% to +0.34% out-of-sample record––
5 Oct 202630 Sept 2026CPI, Sep 2026+0.50%+0.37% to +0.62% out-of-sample record+0.65%−0.15pp
5 Oct 202630 Sept 2026Core CPI, Sep 2026+0.21%+0.10% to +0.32% out-of-sample record+0.33%−0.13pp
5 Oct 202630 Sept 2026Core PCE, Sep 2026+0.22%+0.11% to +0.32% out-of-sample record––
24 Sept 2026first on the record15 Sept 2026CPI, Sep 2026+0.40%+0.25% to +0.54% out-of-sample record––
24 Sept 2026first on the record15 Sept 2026Core CPI, Sep 2026+0.21%+0.10% to +0.32% out-of-sample record––
24 Sept 2026first on the record15 Sept 2026Core PCE, Aug 2026+0.19%+0.09% to +0.28% out-of-sample record+0.26%−0.07pp
24 Sept 2026first on the record15 Sept 2026Core PCE, Sep 2026+0.18%+0.07% to +0.29% out-of-sample record––
About this exhibit
Table of each inflation nowcast the site has published (CPI, core CPI, core PCE), with publication date, data date, value, range, outcome and miss
Every nowcast as published, kept unrevised and scored once its month is out
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 print is scored against it.
Source: The site's own published nowcasts; outcomes from BLS and BEA.inflation.nowcast-12

Shelter beyond the next print — the rent rule, tested and applied

CPI rent projected quarter by quarter from new-tenant rents four quarters earlier, on the rule's two fits with its band, against the quarters already published — carried forward from the last edition BLS published

Exhibit 9 · CPI rent growth projected from new-tenant rents, against the actual

CPI rent, % on a year, Q4 2025 to Q3 2026: projected from new-tenant rents 4 quarters earlier on pre-pandemic and full fits; actual where published
About
CPI rent inNTR y/y (quarter)Pre-2020 fitFull fit95% bandActual
Q4 2025+1.2% (Q4 2024)2.4%2.5%1.1 to 3.9%3.0%
Q1 2026+1.5% (Q1 2025)2.5%2.7%1.3 to 4.1%2.7%
Q2 2026−1.9% (Q2 2025)0.7%0.9%-0.5 to 2.3%2.8%
Q3 2026+1.2% (Q3 2025)2.4%2.5%1.1 to 3.9%2.8%

Model-conditional, not a forecast: the rule carried forward from the last edition BLS published. It is a correlation of changes, not a model of lease renewal, and the band ignores parameter uncertainty and the revisions the newest new-tenant quarters still carry. How far ahead new-tenant rents lead, and the view the rule tests, are on the Housing supply theme page.

About this exhibit
A table of CPI rent, percent on a year, by quarter: new-tenant rent growth some quarters earlier, the rule's pre-pandemic and full-sample projections, the band, and the actual where published
The rule, tested and applied
CPI rent y/y = 1.72 + 0.54 × NTR y/y 4 quarters earlier, fitted Q1 2006–Q4 2019 (R² 0.82). Out of sample since 2020: RMSE 1.11pp against 2.28pp for rent's own change 4 quarters earlier.
Source: U.S. Bureau of Labor Statistics research indexes R-CPI-NTR and R-CPI-ATR; rent rule calculated on this site. From Shelter and rents (data to Jul 2026)themes.housing_supply-20

The supply behind rents — completions implied by starts already made

Completions fitted and projected from past starts by a distributed lag, and the share of a month's starts completed each month after — , not a forecast

Exhibit 10 · Housing completions: actual and fitted from earlier starts

Thousands, seasonally adjusted annual rate, monthly: completions, the model's fit from starts made, and starts; single-family by default, multifamily on the switch; weights fitted 1990–2019
About
About this exhibit
Housing completions, the model's fit from starts already made, and starts, thousands at a seasonally adjusted annual rate, monthly; single-family by default, multifamily (5+) on a switch; five years, ten years or all.
Completions, 1 unit: actual and fitted from past starts
Thousands, seasonally adjusted annual rate: completions, the model's fit from starts already made, and starts; the next twelve months' level is in the sensitivity table, and the Housing supply theme draws the projection against the multifamily pipeline.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculations (weights fitted 1990-01–2019-12)housing.construction-13

Exhibit 11 · Share of single-family starts completed after each delay, model and Census

% of a month's starts completed within each band of months: model weights (fitted 1990–2019) against Census's Survey of Construction, buildings with 1 unit, 2025
Model weights fitted 1990–2019; Census: units completed in 2025, buildings with 1 unit.
Start to completionModelCensus
3 months or less26%17%
4 to 6 months29%42%
7 to 9 months25%20%
10 to 12 months15%9%
13 months or more5%12%
About this exhibit
Bars of the share of a month's single-family starts completed each month after the start, and a table of the share completed within each band of months, the model's weights against Census's survey.
Lag weights, 1 unit
Share of a month's starts completed k months later (β_k ÷ Σβ). K = 16 by BIC; Σβ 0.987; mean lag 6.3 months against Census's average 7.4 for buildings with 1 unit in 2025.
Source: U.S. Census Bureau, Survey of Construction (months from start to completion); Macro Review calculations.housing.construction-16

How robust is it?

Each CPI part's own record, the bottom-up and direct core separately, and the core PCE bridge before and after the CPI

Exhibit 12 · Nowcast errors by CPI part, core CPI method and core PCE timing

RMSE, MAE and bias, pp a month, since Jan 2015: each half of core CPI's blend, each CPI part, core PCE before and after CPI
VariantMonthsRMSEMAEBias
Core CPI, the blend (the page's nowcast)1230.1100.081−0.029
Core CPI, bottom-up only1230.1110.080−0.027
Core CPI, direct autoregression only1230.1140.083−0.031
Food, its own change1230.1580.124−0.006
Energy commodities, its own change1232.3211.666−0.204
Energy services, its own change1230.9080.647−0.063
Core goods, its own change1230.2520.176−0.029
Core services, its own change1230.1090.082−0.022
Core PCE after the CPI1230.0850.066−0.037
Core PCE before the CPI1230.1050.078−0.044
About this exhibit
Table of root mean square error, mean absolute error and bias for core CPI's blend and each half, each CPI part on its own change, and core PCE after and before the CPI
How robust is the record?
Core CPI's two halves of the blend on their own, each CPI part's own record, and core PCE before and after the CPI is published.
Source: Scores computed here.inflation.nowcast-8

The projection's record and the weights' sensitivity

Root-mean-square error of the projection at each horizon against carrying the latest pace forward, and the same model with a different polynomial, sample or lag length

Exhibit 13 · Single-family completions projection error by horizon, against carrying the latest pace forward

Root-mean-square error of h-month-ahead projections (thousands, SAAR) against carrying the latest three-month average forward, samples since 1990; bias = mean of actual − projection
About
Window3 months6 months12 months18 months
1990–2019 (in sample)model / naive · bias49 / 74
+2k · n 348
60 / 103
+3k · n 348
107 / 164
+4k · n 348
166 / 219
+5k · n 348
2020–2023model / naive · bias103 / 50
−21k · n 48
118 / 59
−11k · n 48
127 / 66
−4k · n 48
136 / 68
±0k · n 48
2024 onwardsmodel / naive · bias50 / 66
+14k · n 30
47 / 69
+2k · n 27
61 / 90
−24k · n 21
85 / 103
−66k · n 15
About this exhibit
A table of the root-mean-square error of single-family completions projections at each horizon against carrying the latest three-month average forward, with bias, for the in-sample window and the later test windows.
Projection record, 1 unit
Root-mean-square error of h-month-ahead projections (thousands, SAAR) against carrying the latest three-month average forward; bias = mean of actual − projection.
Source: Macro Review calculations.housing.construction-14

Exhibit 14 · Completions model variants and their twelve-month projections

Same model with a different polynomial, sample or lag length (baseline fitted 1990–2019); projection = average of the next twelve months (thousands, SAAR)
VariantKSampleΣβMean lagR²Next 12m
Baseline (degree 3)161990–20190.9876.30.981879k
degree 4161990–20190.9886.30.982879k
free polynomial, degree 2161990–20190.9886.20.982880k
fit 2000–2019162000–20190.9906.40.985881k
fit to latest161990–20260.9856.60.969879k
K −6101990–20190.9875.70.979876k
K +6221990–20190.9876.60.980880k

Census average months from start to completion (1 unit): 2018 6.7 · 2019 7.0 · 2020 6.8 · 2021 7.2 · 2022 8.3 · 2023 8.6 · 2024 7.7 · 2025 7.4.

About this exhibit
A table of completions-model variants (polynomial degree, sample, lag length) with the sum of the weights, mean lag, R² and the next-twelve-month projection.
Sensitivity of the weights
Same model with a different polynomial, sample or lag length; projection = average of the next 12 months (thousands, SAAR)
Source: Macro Review calculations.housing.construction-15

Appendix

The models in notation, their parameters, their inputs and what they leave out212 checks, all pass
About this page

A model-conditional nowcast of the next inflation prints: the CPI from its parts, with energy read off the week's pump prices, and core PCE mapped from the CPI components BEA prices it with. The page sets the result beside the Cleveland Fed's, shows what moved the numbers and how often models like these have been right.

Core PCE, bias-corrected: the published line

π^tPCE,bc=π^tPCE−eˉ<t,eˉ<t=1n<t∑s<t, s∈R(π^sPCE−πsPCE),n<t≥24\hat\pi^{PCE,bc}_t = \hat\pi^{PCE}_t - \bar e_{<t}, \qquad \bar e_{<t} = \frac{1}{n_{<t}}\sum_{s < t,\ s \in R} \left(\hat\pi^{PCE}_s - \pi^{PCE}_s\right), \qquad n_{<t} \ge 24
SymbolMeaningFromAssumed or estimated
π^tPCE\hat\pi^{PCE}_tThe raw bridge's nowcast of month t (the card above)this page–
eˉ<t\bar e_{<t}The mean error of the record's months before t whose outcome was published — today −0.037pp after the CPI, −0.044pp before itthe record (pseudo-real-time)estimated
RRThe record's months (the after-CPI record for a month whose CPI is in hand, the before-CPI record otherwise)this page–
2424The fewest months of errors before a correction is applied; until then the raw line standsdeclared 3 Oct 2026, the author's decision of 3 Oct 2026 (D-2026-10-03-13)assumed

An intercept correction declared before it was estimated (the author's decision of 3 Oct 2026 (D-2026-10-03-13)). Its record (99 months after the CPI, 99 before) is scored on the same months as the raw line's, beside it; both are scored on the releases that follow. The CPI lines are not corrected.

Headline CPI: five parts, weighted

πt=∑kwk,t πk,t,k∈{food, energy commodities, energy services, core goods, core services}\pi_t = \sum_{k} w_{k,t}\,\pi_{k,t}, \qquad k \in \{\text{food},\ \text{energy commodities},\ \text{energy services},\ \text{core goods},\ \text{core services}\}
πk,t=ck+∑L ∈ {1,2,3,12}ϕk,L πk,t−L+εk,t(k≠energy commodities)\pi_{k,t} = c_k + \sum_{L\,\in\,\{1,2,3,12\}} \phi_{k,L}\,\pi_{k,t-L} + \varepsilon_{k,t} \qquad (k \neq \text{energy commodities})
SymbolMeaningFromAssumed or estimated
πk,t\pi_{k,t}Part k's seasonally adjusted change on the month, %BLS, CPI-U flat files–
wk,tw_{k,t}Part k's relative importance, rolled forward with unadjusted pricesBLS relative importances–
ck, ϕk,Lc_k,\ \phi_{k,L}Intercept and lag coefficients of part kOLS, the CPI parts tableestimated

OLS on the months since 2012, March 2020 to June 2021 left out, estimated on the data before each nowcast; the lags (1, 2, 3, 12) are an assumption. The weights are BLS's, not estimated.

Energy commodities: the pump price passes through

πEC,t=α+β Δpt pump+∑m=212δmDm,t+εt\pi_{EC,t} = \alpha + \beta\,\Delta p^{\,pump}_{t} + \sum_{m=2}^{12} \delta_m D_{m,t} + \varepsilon_t
Δpt pump=100(pˉt (k)pˉt−1−1),k=2\Delta p^{\,pump}_{t} = 100\left(\frac{\bar p_{t}^{\,(k)}}{\bar p_{t-1}} - 1\right), \qquad k = 2
SymbolMeaningFromAssumed or estimated
pˉt (k)\bar p_{t}^{\,(k)}Mean of the month's first k weekly retail gasoline pricesEIA, Gasoline and Diesel Fuel Update–
pˉt−1\bar p_{t-1}Mean of the previous month's weekly pricesEIA–
β\betaPass-through of the pump price to CPI energy commoditiesOLS, the pass-through cardestimated
Dm,tD_{m,t}Calendar-month dummies–assumed

OLS with the same k weeks in every month of the sample (160 months to 2026-10); β̂ and its standard error are printed on the pass-through card.

Core CPI and core PCE

πtcore=12 wCG πCG,t+wCS πCS,twCG+wCS+12 π^tcore,AR\pi^{core}_t = \tfrac{1}{2}\,\frac{w_{CG}\,\pi_{CG,t} + w_{CS}\,\pi_{CS,t}}{w_{CG} + w_{CS}} + \tfrac{1}{2}\,\hat\pi^{core,AR}_t
πtPCE=∑jsj,t−1(γj+δj πm(j),tCPI)\pi^{PCE}_t = \sum_{j} s_{j,t-1}\left(\gamma_j + \delta_j\,\pi^{CPI}_{m(j),t}\right)
SymbolMeaningFromAssumed or estimated
π^tcore,AR\hat\pi^{core,AR}_tCore CPI's own autoregression (lags 1, 2, 3, 12)OLS, the core table–
sj,t−1s_{j,t-1}Source j's share of core PCE spending (Sep 2026)BEA, NIPA Tables 2.4.4U and 2.4.5U–
m(j)m(j)The BLS index that stands for PCE source jthe PCE sources table–
γj, δj\gamma_j,\ \delta_jBridge of source j's price change on its stand-in indexOLS, the PCE sources tableestimated
12\tfrac{1}{2}The weight of each half of core CPI, fixedset before the record was computedassumed

The half-and-half core is a fixed combination, not estimated. A PCE source no BLS index prices follows its own autoregression. Spending shares are the latest month's, so the weights move each month with spending.

The record and the band

RMSE(h)=1n∑t(π^t h−πt)2,bias=1n∑t(π^t h−πt),band=π^t±RMSE(h)\mathrm{RMSE}(h) = \sqrt{\frac{1}{n}\sum_{t}\left(\hat\pi_t^{\,h} - \pi_t\right)^2}, \qquad \text{bias} = \frac{1}{n}\sum_t\left(\hat\pi_t^{\,h} - \pi_t\right), \qquad \text{band} = \hat\pi_t \pm \mathrm{RMSE}(h)
SymbolMeaningFromAssumed or estimated
π^t h\hat\pi_t^{\,h}The nowcast of month t re-run h days before its releasecomputed here, on today's revised data–
hhDays before the release: 28, 21, 14, 7, 0the accuracy-by-horizon tableassumed

Scored each month since 2015 against the published change, the intercepts re-estimated every month. Pseudo-real-time on revised data: the seasonally adjusted figures are today's. The band is the record's typical miss, a backtest range — not a probability.

The record's tests: bias and the random walk

et=π^t−πt,teˉ=eˉV^NW(eˉ)e_t = \hat\pi_t - \pi_t,\qquad t_{\bar e} = \frac{\bar e}{\sqrt{\hat V_{NW}(\bar e)}}
dt=(πtRW−πt)2−et2,DM=dˉV^NW(dˉ),L=⌊4 (n/100)2/9⌋d_t = \big(\pi^{RW}_t - \pi_t\big)^2 - e_t^2,\qquad \mathrm{DM} = \frac{\bar d}{\sqrt{\hat V_{NW}(\bar d)}},\qquad L = \big\lfloor 4\,(n/100)^{2/9} \big\rfloor
SymbolMeaningFromAssumed or estimated
πtRW\pi^{RW}_tthe random walk: the latest published month as it stood at the timeBLS; BEA–
V^NW\hat V_{NW}Newey–West variance of a mean (Bartlett weights, L lags, no small-sample correction)computed here; statsmodels' HAC reproduces it (a check)–
LLNewey and West's (1994) plug-in lag for n monthsthe ruleassumed

Both two-sided, against the normal. The random walk is not nested in the nowcast models, so Diebold–Mariano, not Clark–West, is the comparison; a positive DM means the site's squared errors are the smaller.

Exhibit 15 · The CPI parts and their models

About
PartWeight, %ModelLatestNowcastContributionR²SampleRecord RMSE
Food13.48autoregression · σ 0.14+0.47%+0.24%+0.03pp0.692012-01 to 2026-09 (156)0.16
Energy commodities4.38pump price −5.6% (2 weeks) · β 1.02+0.81%−3.53%−0.15pp0.832012-01 to 2026-10 (160)2.32
Energy services3.22autoregression · σ 0.86+0.55%+0.18%+0.01pp0.102012-01 to 2026-09 (156)0.91
Core goods18.74autoregression · σ 0.21+0.16%+0.09%+0.02pp0.372012-01 to 2026-09 (156)0.25
Core services60.18autoregression · σ 0.09+0.34%+0.31%+0.19pp0.512012-01 to 2026-09 (156)0.11

Pass-through to CPI gasoline (all types), full-month pump-price change: 0.93 (se 0.02, 162 months).

About this exhibit
Weights: BLS's relative importances, rolled to the latest month with unadjusted prices. Energy commodities: the change in the average of the month's first 2 weekly EIA pump prices against last month's average, with calendar-month dummies; the others: their own autoregression (lags 1, 2, 3 and 12).
Source: BLS CPI-U and EIA weekly retail gasoline; estimated here.inflation.nowcast-6

Exhibit 16 · Core PCE by source, and the BLS index that stands for each

SourceWeight, %Stands inCoefficientsR²Oct 2026Nov 2026Contribution, Oct 2026Contribution, Nov 2026
Imputed bank and fund servicesBEA imputes it from bank balances and rates; no BLS index prices it3.2its own autoregressiona 0.24 · l1 0.500.24+0.34%+0.41%+0.011+0.013
Fees and commissions on assetsBEA uses the PPI for portfolio management, not among this page's inputs3.0its own autoregressiona 0.32 · l1 0.180.03+0.47%+0.40%+0.014+0.012
Insurance, net of claimsBEA uses premiums net of claims, not among this page's inputs2.8its own autoregressiona 0.24 · l1 0.030.00+0.25%+0.25%+0.007+0.007
Shelter17.6CPI sheltera 0.04 · SAH1 0.880.86+0.31%+0.32%+0.054+0.055
Health careBEA prices hospitals and physicians off PPI indexes, which are not among this page's inputs with history19.4CPI medical care servicesa 0.13 · SAM2 0.110.04+0.16%+0.15%+0.031+0.029
Air transport and package toursBEA uses the PPI for airline passenger services, not among this page's inputs1.3CPI airline faresa 0.30 · SETG01 0.340.17+0.44%+0.20%+0.006+0.003
Vehicle services and transit2.8CPI transportation servicesa 0.08 · SAS4 0.440.35+0.45%+0.28%+0.013+0.008
Education services1.9CPI tuition, other school fees and childcarea 0.01 · SEEB 0.860.61+0.28%+0.29%+0.005+0.005
Social services and religious1.9its own autoregressiona 0.18 · l1 0.310.11+0.26%+0.26%+0.005+0.005
Nonprofit final consumption3.3its own autoregressiona 0.34 · l1 -0.150.02+0.35%+0.29%+0.012+0.010
Recreation, restaurants, lodging12.2CPI food away from home and recreation servicesa 0.00 · SEFV 0.66 · SARS 0.260.53+0.57%+0.33%+0.070+0.041
Communication, personal, professional5.5CPI services less rent of sheltera 0.01 · SASL2RS 0.450.06+0.21%+0.16%+0.012+0.009
Water and sanitation0.7CPI water, sewer and trasha 0.00 · SEHG 1.001.00+0.61%+0.36%+0.005+0.003
Pharmaceutical and medical products3.8CPI medical care commoditiesa 0.00 · SAM1 1.060.91+0.55%+0.23%+0.021+0.009
Computers, software and phones2.0CPI information technologya -0.32 · SEEE 0.460.04−0.36%−0.39%−0.007−0.008
Motor vehicles and parts4.0CPI new and used motor vehiclesa 0.04 · SETA 0.800.84−0.37%−0.16%−0.015−0.007
Other goods (tradable)14.5CPI core goodsa 0.01 · SACL1E 0.630.33+0.11%+0.07%+0.016+0.010
About this exhibit
Each source's price change (its exact Fisher contribution ÷ last month's spending share) regressed on its BLS index, or on its own last change where no index on disk prices it; weights are the sources' shares of core spending in the latest PCE month.
Source: BEA NIPA Tables 2.4.4U and 2.4.5U, the Sources page's classification; BLS CPI-U; estimated here.inflation.nowcast-7

Exhibit 17 · Definitions and caveats

  • Model
    Model-conditional component models.

    Model-conditional: these are what the site's component models make of the data released so far, not forecasts in any other sense, and no probability is attached to them; the ranges printed are the models' own record (± one RMSE). CPI: food, energy services, core goods and core services each follow their own autoregression on the monthly change (lags 1, 2, 3 and 12); energy commodities follow the month's pump price; the headline adds the five with BLS's relative importances (Sep 2026, rolled forward with unadjusted prices); core is half the bottom-up and half a direct autoregression. October and November 2025 are each half of the published September-to-November change, because BLS did not publish October 2025. The forecasts read these two values, and no model is estimated on them (the author's decision of 24 Oct 2026, D-2026-10-24-NC1).

  • Pump price
    How energy reaches the headline.

    Pass-through: energy commodities' seasonally adjusted change regressed on the change in the average of the month's first 2 weekly EIA pump prices against the previous month's average, with calendar-month dummies, since 2012; slope 1.02 (se 0.04). The full-month pass-through to CPI gasoline is 0.93 (se 0.02).

  • Core PCE
    The CPI components mapped to PCE.

    Core PCE: the Sources & rate sensitivity page's components, classification and exact Fisher contributions give each core source's price change each month; each source is bridged to the BLS index that stands for it (or follows its own autoregression where no index on disk prices it), and the nowcast weights the bridged changes by the sources' shares of core spending in Sep 2026. BEA compiles PCE from many PPI indexes (hospitals, physicians, airline passenger services, portfolio management) whose history this site does not hold; the CPI stand-ins and the autoregressions are what the data on disk allow, and the record says how much that costs.

  • Record
    Pseudo-real-time on revised data.

    Record: each month since 2015 the models are re-estimated on data before that month's equivalent data date — for the CPI, the previous month's CPI and the month's first 2 weekly pump prices; for core PCE after the CPI, the month's CPI and the previous month's PCE; before the CPI, one month less of each — and scored against the published change. Pseudo-real-time on revised data: the seasonally adjusted figures are today's, not those published at the time. Months left out: March 2020 to June 2021, and 2025-10, 2025-11, 2025-12 (a lag falls in the October 2025 gap).

  • Path
    Re-run at each release this month.

    Path through the month: the same models re-run on the data released by each release date since the CPI for the latest month came out — each Tuesday's pump prices and any PCE release — on today's data, as the GDP nowcast's path is.

  • Horizon
    The record days before each release.

    Accuracy by horizon: the record re-run 28, 21, 14, 7, 0 days before each release since 2015, each input cut to what its publisher had released by then on a stylised calendar — the median publication lag of its six latest releases on the publishers' calendars (the CPI 12 days after the month's end, PCE 29, the pump price 1 day after its Monday). Core PCE is dated from the PCE release, so its nearer horizons have the month's CPI in hand. Pseudo-real-time on revised data, like 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 values and their ranges — and never revised. Beside the path recomputed on today's data the page draws those published points; once eight months have an outcome it scores them against the published prints, a real-time record; until then it counts them.

  • Published
    What stands beside the site's number.

    Published: the Federal Reserve Bank of Cleveland's inflation nowcast is read from the monthly chart file its Inflation Nowcasting page draws, on the latest day on or before this page's data date, and printed beside the site's number; its actual CPI for the latest month is checked against the change computed here from BLS's index. Its model uses daily oil prices, weekly gasoline prices and past inflation; the two models share no code.

  • Not modelled
    What the models leave out.

    Not modelled: headline PCE (food and energy in PCE); the PPI components BEA maps into core PCE, which are not among this page's inputs with history.

Source: BLS; EIA; BEA; Federal Reserve Board (projection median); Federal Reserve Bank of Cleveland; Macro Review calculationsinflation.nowcast-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.