Macro ReviewHow to read
Macro · Housing · Construction pipeline

Completions have fallen to 1.19 million a year, the fewest since 2026

Is the housing pipeline growing or shrinking?

Census/HUD New Residential Construction · 20 Oct 2026Data to Sep 2026 · updated 26 Oct 2026
Completionsⓘ1.19mannual rate, Sep 2026 · lowest since Aug 2026significant fall
Housing startsⓘ1.22mannual rate, Sep 2026 · −7.4% on the yearchange not significant
Building permitsⓘ1.40mannual rate, Sep 2026 · −3.0% on the year
Under constructionⓘ1.28munits, Sep 2026 · −3.3% on the year
pipelineⓘ24.6 monthsmultifamily, Sep 2026 · 21.3above norm
Single-family starts859kannual rate, Sep 2026 · permits 873k (3-month average)

In brief

Is the building pipeline growing or shrinking?
  • Housing starts averaged 1.29 million a year over the last three months, down 4.1% on a year earlier.→ Exhibit 1
  • Completions averaged 1.21 million a year over the last three months, down 17.1% on a year earlier.→ Exhibit 1
  • The pipeline holds 24.6 months of completions at the current pace, against 21.3 in 2015–19.→ Exhibit 4
  • Single-family permits (873k) are running below single-family starts (878k) on three-month averages.→ Exhibit 1
What would change this view
What would change this viewTriggerTodayDecided by
Multifamily starts fall back below completions5-or-more starts < 5-or-more completions, three-month averages, three months runningstarts 403k vs completions 330k (Sep 2026)not crossedthe Oct 2026 new residential construction release (Census)
The 5-or-more pipeline normalises5-or-more pipeline < 22.9 months of completions (halfway to its )24.6 months (Sep 2026)not crossedthe Oct 2026 new residential construction release (Census)
The reasoning behind each point
  • Starts
    Housing starts averaged 1.29 million a year over the last three months, down 4.1% on a year earlier.

    In Sep 2026 alone starts ran at 1.22 million a year, down 7.4% on the year; the month's dip is inside the survey's error band. Single-family starts −4.6% to 859k; 5-or-more −7.2% to 349k. Over the last twelve months (unadjusted) starts are −1.9% on the prior twelve: 1 unit −3.5pp, 5 or more +1.9pp; by region the West moved most (−1.5pp).

  • Completions
    Completions averaged 1.21 million a year over the last three months, down 17.1% on a year earlier.

    In Sep 2026 alone completions ran at 1.19 million a year, down 14% on the year, the lowest rate since Aug 2026. Single-family 840k (−11.0% y/y), 5 or more 339k (−20.4% y/y). Units under construction 1,280k, −3.3% y/y (90% CI ±2.7%). Starts less completions (three-month averages) is +79k at an annual rate: starts are running above completions, so the stock under construction has stopped shrinking.

  • Multifamily
    The multifamily pipeline holds 24.6 months of completions at the current pace, against 21.3 in 2015–19.

    Multifamily means buildings of five or more units (5+). 677k units under construction ÷ completions of 330k a year (three-month average). Multifamily starts are running at 403k, above completions by 73k a year, so the stock under construction is refilling. Permits authorized but not started: 123k (+13.9% y/y).

  • Permits
    Single-family permits (873k) are running below single-family starts (878k) on three-month averages.

    Permits lead starts by about a month for single-family homes (Census: 1.4 months from authorization to start in 2025). Total permits 1,400k, −0.2% m/m and −3.0% y/y; authorized but not started 279k (+9.8% y/y), 2.6 months of starts.

Next: New-home supply — the stock it feeds

U.S. Census Bureau and HUD, New Residential Construction (release and historical workbooks); Macro Review calculations.housing.construction-read

Permits, starts and completions against the survey's own error

Permits, starts and completions at an — and is the change bigger than the survey's own error?

Exhibit 1 · Three-month averages of new-home permits, starts and completions

Thousands of units, seasonally adjusted annual rate, three-month averages, monthly, last ten years to Sep 2026; dashed: completions' 2015–19 average (1,125k)
About
About this exhibit
Permits, starts and completions of new homes, three-month averages, thousands at a seasonally adjusted annual rate, monthly, over the last ten years, with completions' 2015–19 average dashed.
Permits, starts and completions, three-month averages
Thousands of units, seasonally adjusted annual rate, three-month averages; dashed: completions' 2015–19 average
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculations · data to Sep 2026housing.construction-1

Exhibit 2 · Homes under construction and authorized but not started, by structure type

Thousands of units, seasonally adjusted, month-end, last ten years to Sep 2026; lines: under construction (total, 5+ units, one unit) and authorized not started (dashed)
About
About this exhibit
Units under construction (total, multifamily 5+ and single-family) and units authorized but not started, thousands, seasonally adjusted, end of month, over the last ten years; completions are not plotted.
The stock in the pipeline
Thousands of units, seasonally adjusted, end of month
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Constructionhousing.construction-2

Exhibit 3 · The selected stage by structure and region, with survey error bands

Thousands at a seasonally adjusted annual rate, Sep 2026; % changes with Census's 90% confidence interval (±), 'sig.' where it excludes zero; year to date unadjusted
About
SeriesLevelFrom Aug 2026From Sep 2025Unadjusted, monthYTD 2026YTD 2025YTD change
Total1,221−5.2%±11.8−7.4%±11.5103.81,036.51,048.0−1.1%±3.6
1 unit859−4.6%±10.0+2.6%±9.572.6699.0728.0−4.0%±3.2sig.
5 units or more349−7.2%±27.7−25.9%±21.5sig.30.1327.0305.9+6.9%±10.9
Northeast114±0.0%±34.2−30.9%±32.19.9109.2100.9+8.3%±11.1
Midwest158−19.0%±20.7−19.4%±21.014.4150.4153.5−2.0%±8.6
South714+6.1%±15.5+11.0%±20.459.5554.8556.7−0.3%±4.4
West235−23.2%±22.9sig.−25.4%±14.3sig.19.9222.2237.0−6.3%±7.0
Northeast, 1 unit49+4.3%±32.3−7.5%±26.64.340.547.4−14.7%±5.2sig.
Midwest, 1 unit120−13.7%±18.7+2.6%±22.011.197.3101.2−3.9%±6.9
South, 1 unit506−3.3%±12.4+8.8%±13.841.6411.6415.3−0.9%±4.3
West, 1 unit184−3.7%±21.3−8.9%±20.015.5149.6164.0−8.7%±9.3
About this exhibit
A table of housing starts (switchable to other stages) by structure type and region: level in thousands at an annual rate, % changes on the month and the year with Census's 90% confidence intervals and 'sig.' flags, and unadjusted year-to-date totals, in the latest month.
Housing starts, Sep 2026 (preliminary)
Published levels (thousands; seasonally adjusted annual rate) and percent changes with Census's 90% confidence interval (±). “sig.” marks changes whose interval excludes zero. Unadjusted year-to-date totals from Table b.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction, release tables 3a–bhousing.construction-10

Pipeline

Are units entering the pipeline faster than they leave it?

Exhibit 4 · Construction pipeline in months: multifamily, single-family and authorized not started

Months; under construction ÷ monthly completions (three-month average), authorized not started ÷ monthly starts; monthly, since 2000; dashed: multifamily 2015–19 average (21.3) · loading history
About
About this exhibit
Months of completions under construction for multifamily (5+ units) and single-family homes, and months of starts authorized but not started, monthly, since 2000, with the multifamily 2015–19 average dashed.
Months of completions under construction
Units under construction ÷ monthly completions (three-month average); authorized-not-started ÷ monthly starts
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculations · data to Sep 2026housing.construction-3

Exhibit 5 · Housing starts less completions: total and by structure type

Thousands of units at a seasonally adjusted annual rate, three-month averages, monthly, last ten years to Sep 2026; above zero the stock under construction grows
About this exhibit
Housing starts less completions for total, single-family and multifamily (5+ units), thousands at an annual rate, three-month averages, monthly, over the last ten years.
Starts less completions
Thousands of units, annual rate, three-month averages: above zero the stock under construction grows
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculations · data to Sep 2026housing.construction-4

Stages, structures and regions

Which stage, which structure type and which region is moving?

Exhibit 6 · Housing permits to completions by structure and region: levels and changes

Thousands, seasonally adjusted (flows at annual rates, stocks month-end), Sep 2026; % changes shaded against 2015–19; percentile since 2000; select a row to chart it
About
SeriesLevel3m avg1m3m1y3m avg y/yLevel pct since 20003m avg y/y, 36 monthsAs of
Building permits · SAAR
Total1,400k1,412k−0.2%+1.9%−3.0%+1.1%49Sep 2026
1 unit845k873k−4.0%−3.1%−4.8%−0.8%43Sep 2026
2–4 units52k52k+8.3%−1.9%−10.3%−5.4%57Sep 2026
5 units or more503k487k+5.9%+12.0%+1.0%+5.3%87Sep 2026
Northeast145k142k+13.3%−6.5%−2.0%+6.5%57Sep 2026
Midwest213k217k+0.5%+4.9%+4.4%+3.7%60Sep 2026
South759k752k+1.5%+7.2%−1.8%+0.5%59Sep 2026
West283k302k−10.2%−8.1%−11.3%−1.5%31Sep 2026
Authorized but not started · SA, end of month
Total279k278k+0.4%+3.7%+9.8%+10.6%93Sep 2026
1 unit151k150k±0.0%+5.6%+6.3%+9.5%94Sep 2026
5 units or more123k124k±0.0%+0.8%+13.9%+11.4%89Sep 2026
Northeast23k23k+4.5%−14.8%+21.1%+25.5%72Sep 2026
About this exhibit
A table of permits, authorized but not started, starts, under construction and completions by structure type and region: level, three-month average, changes over one and three months and a year, the percentile in the series' history and a sparkline, in the latest month.
Every stage by structure type and region, Sep 2026
Seasonally adjusted: flows at annual rates, stocks end of month. Changes in %; shading against each series' own 2015–19 distribution. Select a row to chart it.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculationshousing.construction-5

Exhibit 7 · Construction pipeline in months and starts less completions, by structure type

Months (ratios) and thousands at an annual rate (starts less completions), three-month averages, Sep 2026; changes over three months and a year; percentile since 2000
About
SeriesLevel3m change1y changePct since 2000Level, 10 yearsAs of
Pipeline ratios
Under construction ÷ completions, total12.7+1.7+1.891Sep 2026
Authorized not started ÷ starts, total2.6+0.2+0.399Sep 2026
Starts − completions, total (3m avg)79k+109.7k+195.0k59Sep 2026
Under construction ÷ completions, 1 unit8.1+0.3+0.989Sep 2026
Authorized not started ÷ starts, 1 unit2.1+0.2+0.1100Sep 2026
Starts − completions, 1 unit (3m avg)6k+13.7k+119.3k42Sep 2026
Under construction ÷ completions, 5 units or more24.6+7.2+5.889Sep 2026
Authorized not started ÷ starts, 5 units or more3.7+0.3+0.787Sep 2026
Starts − completions, 5 units or more (3m avg)73k+91.7k+75.0k75Sep 2026
About this exhibit
A table of months of completions under construction, months of starts authorized but not started, and starts less completions, for total, single-family and 5+ units: level, three-month and one-year changes, the percentile in the series' history and a sparkline, in the latest month.
Pipeline ratios
Months of completions under construction and of starts authorized but not started; starts less completions in thousands at an annual rate (three-month averages). Changes in the ratio's own unit.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculationshousing.construction-6

Exhibit 8 · Total housing starts: monthly and three-month average

Thousands, seasonally adjusted annual rate, last five years to Sep 2026; bars: monthly; line: three-month average; select a row above to chart another series
About this exhibit
The series selected in the tables above (total starts by default), thousands at a seasonally adjusted annual rate: monthly (bars) and three-month average (line), over the last five years.
Starts — Total: 1,221k in Sep 2026
Thousands of units, seasonally adjusted annual rate: monthly (bars) and three-month average (line). Select a row above to chart it.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Constructionhousing.construction-7

Where the change comes from

Single-family or ? Which region?

Exhibit 10 · The selected stage's twelve-month change, split into its parts

Percentage points of the % change in the unadjusted twelve-month sum on a year earlier, monthly, Oct 2021 to Sep 2026; line: total % change
About
About this exhibit
Contributions of one-unit, two-to-four-unit and 5+ unit buildings, in percentage points, to the % change in unadjusted twelve-month housing starts on a year earlier, with the total's change as a line, monthly; another stage, or the split by region, can be selected.
Housing starts: contributions to the 12-month change by structure type
Percentage points of the change in the unadjusted 12-month sum on a year earlier; the line is the total's % change. Sep 2026: 1 unit −3.5pp, 2–4 units −0.2pp, 5 units or more +1.9pp.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction, unadjusted; Macro Review calculationshousing.construction-11

Exhibit 11 · The selected stage's growth by structure and region, last twelve months

% change of the three-month average on a year earlier, seasonally adjusted, Oct 2025 to Sep 2026; shading against each series' 2015–19 distribution; regions indented
Three-month average growth heat table
SeriesOct ’25Nov ’25Dec ’25Jan ’26Feb ’26Mar ’26Apr ’26May ’26Jun ’26Jul ’26Aug ’26Sep ’26
Total−5.8−2.9−4.7−1.6−5.3+1.5+1.1+2.1−0.8−2.7−0.2−4.1
1 unit−14.0−12.0−8.5−9.8−12.3−6.6−4.4−0.6−4.2−5.8−2.9−1.3
5 units or more+19.5+24.7+5.6+24.0+17.0+25.1+16.6+9.7+9.5+6.5+6.9−9.2
Northeast+2.8+15.3+14.7+28.6+21.1+29.4−1.9−1.2−16.3+16.5+20.1+18.5
Midwest−3.0−6.1−11.1−7.5+5.5+3.8+0.9−9.3+8.1+11.6+6.6−11.8
South−9.0−2.5−5.0−2.3−7.6+3.6+1.3+2.4−1.7−8.5−5.1−3.4
West−3.8−9.6−7.4−7.9−14.5−13.7+2.0+11.7+3.5−6.1−1.1−9.4
About this exhibit
A heat table of the year-on-year % change in three-month-average housing starts (another stage can be selected) for the total, one unit, 5+ units and the four regions, each of the last twelve months, shaded against 2015–19.
Last 12 months — three-month average, year on year
%, seasonally adjusted; each cell shaded against the series' own 2015–19 distribution. Regions indented.
Source: U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction; Macro Review calculationshousing.construction-12

Appendix

Surveys, adjustment, intervals, revisions and the model52 checks, all pass
About this page

Every new home passes through the same stages: a permit, a start, months under construction, a completion. Census counts each stage every month by structure type and region; this page follows the flow through the pipeline, separates signal from sampling noise with the published confidence intervals, and estimates when today's starts become tomorrow's supply.

Completions from past starts (Almon distributed lag)

Ct=∑k=0Kβk St−k+εtC_t = \sum_{k=0}^{K} \beta_k \, S_{t-k} + \varepsilon_t
βk=∑j=0daj kj,β−1=βK+1=0\beta_k = \sum_{j=0}^{d} a_j \, k^j, \quad \beta_{-1} = \beta_{K+1} = 0
kˉ=∑kk βk∑kβk\bar{k} = \frac{\sum_k k\,\beta_k}{\sum_k \beta_k}
SymbolMeaningFromAssumed or estimated
CtC_tcompletions, seasonally adjusted annual rateCensus, New Residential Construction–
St−kS_{t-k}starts k months earlier, seasonally adjusted annual rateCensus, New Residential Construction–
KKlag length, chosen by the Bayesian information criterion––
aja_jpolynomial coefficients (degree 3) with Almon's endpoint restrictions––
kˉ\bar{k}mean lag in months––

Exhibit 12 · Definitions and caveats

  • Surveys
    Permits are counted; starts are sampled.

    Permits come from the Building Permits Survey of about 20,000 permit-issuing places (imputation for non-respondents was revised from the February 2025 release). Starts, units under construction and completions come from the Survey of Construction, a sample of permits followed through construction plus an area sample for places that issue no permits; they carry sampling error, which is why Census publishes 90% confidence intervals for their changes and none for permits.

  • Definitions
    Start, completion, authorized.

    A start is the beginning of excavation for the footings or foundation; all units in a multi-unit building are counted as started when the building is. A unit is completed when all finished flooring is installed, or at occupancy if that comes first. Authorized but not started counts units permitted in this or any earlier month on which construction had not begun at the end of the month.

  • Adjustment
    Annual rates, regional adjustment, suppressed 2–4 units.

    Seasonally adjusted flows are annual rates (unadjusted monthly × 12 ÷ seasonal factor); stocks are adjusted end-of-month levels. The adjusted U.S. total equals the sum of the four adjusted regions in every month (checked), so regions add exactly. Adjusted 2–4 unit starts, units under construction and completions are suppressed (Census: tests for identifiable and stable seasonality do not meet reliability standards); 1 unit and 5 or more therefore do not add to the adjusted total. Published factors reproduce the adjusted total, 1-unit and regional series of starts, under construction and completions within rounding (checked); permits and the 5-or-more series do not satisfy that identity in every month, consistent with being aggregated from separately adjusted components, and are left out of the check.

  • Intervals
    90% confidence intervals.

    A change is statistically significant when its 90% confidence interval excludes zero. In the Sep 2026 release the interval on the monthly change is ±11.8% for total starts and ±10.8% for total completions, so most monthly moves are not significant. Three-month averages and year-on-year changes of three-month averages are the steadier readings; the tables show both.

  • Revisions
    Preliminary, revised, annual.

    The latest month is preliminary (p) and earlier months in the table may be revised (r); seasonal factors and recent history are also revised annually. Each update compares the release table with the historical workbook cell by cell (checked). Census: year-to-date permits estimates reflect revisions not reflected in the monthly data (release Table 1b note), so permit year-to-date totals differ from the sum of the monthly estimates; the check applies to starts and completions.

  • Contributions
    Unadjusted parts that add.

    Contribution of a structure type or region to a stage's 12-month change: (x_i,t − x_i,t−12) ÷ X_t−12 × 100 on unadjusted data — 12-month sums for flows, end-of-month levels for stocks — so the parts add to the published total's change up to rounding (checked). 2–4 units are included unadjusted.

  • Model
    Distributed lag of completions on starts.

    Almon distributed lag of completions on current and past starts, without intercept: weights follow a degree-3 polynomial with Almon's endpoint restrictions (zero weight one month before the first lag and one month after the last), so they rise and fall smoothly. The lag length is chosen by the Bayesian information criterion — 1 unit K = 16 months (grid 6–24), 5 or more K = 35 (grid 18–42) — fitted 1990–2019 on seasonally adjusted annual rates. Σβ is completions per start in the long run (below 1 where units are abandoned or reclassified); the mean lag is Σkβ_k ÷ Σβ_k. Projections hold starts after Sep 2026 at their three-month average; the 90% band is ±1.645 × the root-mean-square error of the same h-step procedure at every origin from 2020 (it includes the 2021–23 jump in construction times and ignores parameter uncertainty). Sensitivity to the polynomial degree, sample and lag length is shown beside the model.

  • Record
    How the model has done.

    Mean lags: single-family 6.3 months (Census's average start-to-completion time in 2025: 7.4); multifamily (5+ units) 13.9 months (Census, buildings with 2+ units: 16.5). Out of sample (2024 onwards) the single-family projection six months ahead has missed by 47k (root mean square) against 69k for carrying the latest completions forward; the 5-or-more projection twelve months ahead by 119k against 133k, with completions running 97k above the model on average — large buildings are taking longer than the 1990–2019 weights assume (fitted through Sep 2026 the mean lag is 18.4 months). The projection itself is drawn on the Housing supply theme page, which copies it from this page.

  • Stock-flow
    Why the stock does not equal starts less completions.

    Units under construction should change by starts less completions. On unadjusted data the 12-month residual (change in the stock − starts + completions) has averaged -12k since 2000 (σ 16k; latest -36k): sampling error in three separately estimated series, abandoned projects and reclassified units. It is shown for context, not used as a check.

  • Norms
    Own history, 2015–19.

    Shading compares each change with its own 2015–19 distribution (full colour at two standard deviations); percentiles run since 2000.

  • Licence
    Federal data only.

    Private construction trackers (Dodge, CoStar, RealPage multifamily pipelines) and builder surveys (NAHB/Wells Fargo HMI) are paid data and are not used.

Source: U.S. Census Bureau, New Residential Construction methodology and explanatory notes; Macro Review calculations.housing.construction-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.