Housing Starts 1.275M vs 1.32M Est — but July Was Revised Up 70K and Single-Family Rose 7.6%

Fundamentals · 2026-09-17

Housing starts fell 2.6% to a 1,275,000 annual rate in August against a 1.320M consensus — but the miss is mostly a revision artifact: July was revised up 70,000, from 1,239,000 to 1,309,000, so August's level sits *above* what the market believed July was. Single-family starts rose 7.6% to 918,000; the decline was entirely multifamily, where 5+ unit starts fell 22.5% to 344,000. Building permits fell 2.7% to 1,394,000 against 1.400M expected, with single-family permits -1.8% to 878,000. Completions collapsed 11.9% to 1,128,000 and are 27.1% below a year ago. Of every change in this release, only three clear the Census Bureau's own significance test: Northeast starts -44.5%, and completions on both horizons. The headline decline does not. Next release 20 October.

What It Changes

• The headline is not statistically distinguishable from zero, and Census says so. The 2.6% decline carries a ±12.0% confidence interval. So does the 7.6% single-family rise, at ±14.0%. Read either as a signal and you are reading noise the agency has explicitly labeled as such.
• The revision is bigger than the print. July went up 70,000 — a 5.6% upward revision — so the "decline" is measured from a base that did not exist at 08:29. Against the previously published July figure of 1,239,000, August is higher.
• The composition inverts the story. Single-family starts added 65,000 while multifamily removed 100,000. The part of housing that responds to mortgage rates and builder sentiment went up; the part that responds to apartment financing and is notoriously lumpy went down.
• The real number is completions, and nobody will lead with it. Down 27.1% year over year with a ±8.9% interval — comfortably significant — to 1,128,000. That is 420,000 fewer homes a year being finished, into a market that has spent three years short of supply.
• The forward-looking series is the soft one. Permits fell 2.7%, single-family permits fell 1.8%, and single-family authorizations are down 2.5% year to date. Builders are working through an existing backlog while authorizing less new work — which is what NAHB's buyer traffic index at 23 predicts.

Impact

• USD — Neutral — housing starts has no reliable dollar transmission, and this print lands inside a five-release 08:30 block where claims and the Philadelphia Fed have more claim on attention.
• The dollar arrives strong: DXY tagged 100.00 during Wednesday's decision and closed up 0.35% at 99.97, having broken the 99.80 confluence.
• US Indices (ES / NQ / YM / RTY) — Mixed — a headline miss that is not significant, over a single-family rise that is also not significant, is not a reason to reprice equities.
• RTY holds the specific exposure and it is genuinely two-sided. Homebuilders get a single-family starts number up 7.6% and an upward revision, which is good for volumes. Building-products names get a completions figure down 27.1% year over year, which is where their revenue actually lands.
• Futures came in bid overnight — ES +0.85%, NQ +1.10%, YM +0.72%, RTY +0.79% — on 13% to 18% of average volume, so these are thin-tape levels ahead of Friday's triple witching.
• Gold (GC) — Neutral — no mechanism from a construction print to a rates instrument worth writing down.

Inside The Number

Privately-owned housing starts ran at a seasonally adjusted annual rate of 1,275,000 in August, 2.6% below the revised July rate of 1,309,000 and 1.2% below August 2025's 1,291,000. Consensus was 1.320M on this morning's calendar, with Trading Economics carrying 1.31M — so a miss of roughly 35,000 to 45,000 depending on which panel you read.

Before any of that, read the confidence intervals, because this release publishes them and almost nobody quotes them. Census reports the monthly change as -2.6% (±12.0%). That interval runs from -14.6% to +9.4% and contains zero. In the agency's own words, when the range contains zero "it is uncertain whether there was an increase or decrease." The single-family rise of 7.6% (±14.0%) fails the same test, as does the multifamily fall of 22.5% (±24.9%).

Run the test across the whole release and exactly three changes clear it: Northeast starts at -44.5% (±18.5%), total completions at -11.9% (±9.7%) on the month and -27.1% (±8.9%) on the year, and single-family completions at -10.4% (±9.3%). Everything else in the starts table is noise by the publisher's own standard.

Now the revision, which is larger than the change it sits under. July's starts were originally published at 1,239,000. This release revises them to 1,309,000 — up 70,000, or 5.6%. That matters twice over. It means the 2.6% "decline" is measured from a base the market learned about at 08:30. And it means August's 1,275,000, while below the revised July, is 36,000 above the July figure everyone was working from yesterday. A market that had priced 1.239M as the starting point received a number higher than that starting point and will report it as a fall.

Census notes that preliminary seasonally adjusted estimates of starts, permits and completions are revised by 3.8% or less on average. July's 5.6% revision exceeds that, which is a reminder that the monthly print is a provisional read on a non-probability sample with a 75.8% response rate.

The composition is where the actual information is. Single-family starts rose to 918,000 from 853,000, adding 65,000. Starts in buildings with five units or more fell to 344,000 from 444,000, removing 100,000. So multifamily accounted for more than the entire net decline, and the single-family sector — the one tied to mortgage rates, builder sentiment and household formation — accelerated.

That single-family figure deserves care rather than celebration. At 918,000 it is the highest since March's 1,017,000 and up 5.2% on the year, but the interval swallows both comparisons. What can be said without a caveat is that it did not fall, in a month when the 30-year mortgage rate hit 6.97%, its highest since May 2025, and builder confidence printed 32, its lowest in a year.

Permits are the forward-looking series and they went the other way. Total authorizations fell 2.7% to 1,394,000 against a 1.400M consensus, essentially in line. Single-family permits fell 1.8% to 878,000. Because permits come from an administrative count rather than a sample, Census publishes them without confidence intervals — they are not subject to sampling error at all, which makes them the more reliable half of this release even though they get a fraction of the coverage.

And the year-to-date permit split is the cleanest structural fact in the report. Through August, single-family authorizations are down 2.5% on 2025 while authorizations for buildings with five or more units are up 5.9%. Total permits are flat at -0.1%. The mix of what America is being authorized to build is shifting toward apartments, even as this particular month's *starts* went the opposite way. Multifamily is lumpy month to month; the year-to-date figure is the trend.

Completions are the number that should lead, and they are alarming. Housing completions fell 11.9% to 1,128,000, and are 27.1% below August 2025's 1,548,000. Both changes are statistically significant. Single-family completions fell 10.4% to 816,000 against 1,059,000 a year ago.

Set that against units under construction, and the pipeline arithmetic gets interesting. Units under construction stand at 1,271,000, essentially unchanged from July's 1,267,000 and only 3.2% below the 1,313,000 of a year ago. So the backlog has barely moved while the rate of finishing homes has fallen by more than a quarter. Units are being started, they are sitting in construction, and they are coming out the other end far more slowly. Whether that is labor shortage, materials, financing or lot availability, this release cannot say — but NAHB's survey yesterday had 42% of builders rating lot availability poor and the chairman citing labor shortages and immigration enforcement.

Against the week's other housing data, the picture is consistent and it is not about demand for new construction. Mortgage applications fell 4.1% with purchases down 1%. Builder confidence fell three points to 32, with buyer traffic stuck at 23 and 38% of builders cutting prices. Building materials were the only one of thirteen retail categories to decline in August, and building-material inventories carry the highest stock-to-sales ratio in retail at 2.15, and rising. Now completions down 27%. Materials are piling up at the distributor while fewer homes get finished — those two facts are the same fact.

And for the Fed, this is the ambiguous one. Chair Warsh said on Wednesday he is "hard-pressed to call financial conditions restrictive." Housing is the sector where that claim is most testable, and today's answer is genuinely mixed: single-family construction accelerated, July was revised up, and the headline weakness fails a significance test — but authorizations are falling, completions have collapsed, and the forward-looking half of the report agrees with a builder survey at a one-year low.

The Internals

The three headline series, seasonally adjusted annual rates in thousands:

Series · August · July revised · Change · August 2025 · vs year ago

Building permits · 1,394 · 1,433 · -2.7% · 1,347 · +3.5%

Housing starts · 1,275 · 1,309 · -2.6% · 1,291 · -1.2%

Housing completions · 1,128 · 1,280 · -11.9% · 1,548 · -27.1%

Units under construction · 1,271 · 1,267 · +0.3% · 1,313 · -3.2%

Single-family against multifamily, where the month actually happened:

Measure · August · July revised · Change · Read

Single-family starts · 918 · 853 · +7.6% · Added 65,000 to the annual rate

Starts, 5 units or more · 344 · 444 · -22.5% · Removed 100,000

Single-family permits · 878 · 894 · -1.8% · The forward-looking softness

Permits, 5 units or more · 467 · 482 · -3.1% · Apartments authorized more slowly too

Single-family completions · 816 · 911 · -10.4% · Significant, and the real weakness

Completions, 5 units or more · 302 · 359 · -15.9% · Apartment deliveries falling

What clears the Census Bureau's own significance test, and what does not:

Change · Value · Confidence interval · Verdict

Northeast starts, m/m · -44.5% · ±18.5% · Significant

Total completions, y/y · -27.1% · ±8.9% · Significant

Total completions, m/m · -11.9% · ±9.7% · Significant

Single-family completions, m/m · -10.4% · ±9.3% · Significant

Total starts, m/m · -2.6% · ±12.0% · Not significant

Single-family starts, m/m · +7.6% · ±14.0% · Not significant

Starts, 5 units or more, m/m · -22.5% · ±24.9% · Not significant

Midwest starts, m/m · -12.0% · ±31.8% · Not significant

South starts, m/m · -1.3% · ±16.5% · Not significant

Total starts, y/y · -1.2% · ±10.8% · Not significant

Single-family starts, y/y · +5.2% · ±10.9% · Not significant

Starts by region, seasonally adjusted annual rate in thousands:

Region · August · July revised · Change · Single-family August · Single-family July

South · 658 · 667 · -1.3% · 531 · 536

West · 323 · 244 · +32.4% · 198 · 154

Midwest · 198 · 225 · -12.0% · 143 · 116

Northeast · 96 · 173 · -44.5% · 46 · 47

Permits by region, seasonally adjusted annual rate in thousands:

Region · August · July revised · Change · Single-family August · vs year ago total

South · 745 · 748 · -0.4% · 527 · +4.9%

West · 313 · 308 · +1.6% · 169 · -0.6%

Midwest · 208 · 225 · -7.6% · 130 · +3.5%

Northeast · 128 · 152 · -15.8% · 52 · +5.8%

Year-to-date permits, not seasonally adjusted, in thousands:

Category · 2026 to date · 2025 to date · Change · Read

Total permits · 969.1 · 970.1 · -0.1% · Flat year on year

Single-family · 620.9 · 636.5 · -2.5% · Authorizations shrinking

Buildings with 5 units or more · 313.8 · 296.4 · +5.9% · The mix is shifting to apartments

Buildings with 2 to 4 units · 34.4 · 37.2 · -7.5% · The smallest and weakest category

Regional And Structure Detail

The Northeast is the only region that moved beyond doubt, and it moved hard. Starts fell 44.5% to 96,000 from 173,000, a change whose ±18.5% interval excludes zero. Single-family starts in the region barely budged, at 46,000 against 47,000 — so the entire collapse was multifamily, which in a region dominated by New York and Boston apartment financing is one or two large projects failing to break ground. Northeast permits fell 15.8% to 128,000, the largest regional permit decline, which suggests the pipeline behind it thinned too.

The West did the opposite and nobody will mention it. Starts rose 32.4% to 323,000 from 244,000, with single-family up to 198,000 from 154,000. The confidence interval on that regional change is ±41.6%, so it is statistically meaningless in isolation — but it is the reason the national number fell only 2.6% while the Northeast lost nearly half its starts.

The South is the market that matters for scale and it was flat. At 658,000 starts the South is 52% of all US housing starts, and it fell 1.3% — inside any plausible margin. Single-family starts there were 531,000 against 536,000. Southern permits were similarly flat at 745,000, down 0.4%, and up 4.9% on the year. The largest homebuilding region in the country is doing neither of the things the national headline implies.

The Midwest fell 12.0% to 198,000, but its single-family starts rose to 143,000 from 116,000 — another region where the headline and the single-family line point opposite ways.

On structure type, the year-to-date permits table is the one to keep. Single-family authorizations are running 2.5% below last year while buildings with five or more units are running 5.9% above, leaving the total flat at -0.1%. The two-to-four unit category — the "missing middle" that housing policy keeps trying to revive — is down 7.5% and is now just 34,400 units year to date.

And the completions detail is where the supply story sits. Total completions at 1,128,000 are 27.1% below a year ago. Single-family completions at 816,000 are down from 1,059,000, a fall of 243,000. Completions in buildings with five or more units are 302,000 against 470,000. Both halves of the market are finishing far fewer homes than they were twelve months ago, while units under construction have fallen only 3.2% — which means the constraint is in finishing, not in starting.

A methodology note that matters for how much weight to put on any of this. Census states that it takes three months to establish a trend in permits, six months in starts and six months in completions. The permit numbers come from an administrative count of permit-issuing places and carry no sampling error at all. The starts and completions numbers come from sample surveys with a 75.8% response rate and the intervals printed above. When those two halves disagree, as they do this month, the permits are the more reliable signal and the starts are the more quoted one.

Against This Morning's Open

• The Open asked the right question and the answer is no, not this month — The Open wrote that "a second soft housing month would be the first hard data arguing the stance is already doing more than the Chair credits."
• It did not deliver that. Single-family starts rose 7.6%, July was revised up 70,000, and the headline decline fails the Census Bureau's own significance test. This is not the hard evidence that Warsh's "hard-pressed to call financial conditions restrictive" is wrong.
• Where the Open's framing does get support is the forward-looking half. Permits fell 2.7%, single-family permits fell 1.8%, and single-family authorizations are down 2.5% year to date — which is consistent with the builder sentiment at 32 and the 6.97% mortgage rate it cited in the same paragraph.
• The Open's prior figure was correct as published and is now obsolete. It listed starts prior at 1.239M, which is what Census had published. The revision to 1.309M arrived with this release, and it is the single largest reason the print reads as a miss.
• The calendar was complete again today — nine rows including both Thursday fixed-cadence items, against three rows and five omissions yesterday.

What This Sets Up

• Next release — Tuesday 20 October, covering September, the first month fully after the Fed's hike and the first to capture whether 6.97% mortgage rates deter authorizations further.
• Whether single-family permits keep falling is the number to watch, not starts. Permits carry no sampling error, they lead starts by roughly three months, and single-family authorizations are already down 2.5% year to date.
• Whether completions stabilize. Down 27.1% on the year with under-construction down only 3.2% is a pipeline that has stopped clearing. Another month like it and the finishing bottleneck becomes the story of the housing market rather than demand.
• Whether the Northeast rebounds. A 44.5% single-month collapse concentrated in multifamily usually reverses, because it typically reflects the timing of one or two large projects rather than a change in conditions.
• Census says give it six months. The agency states explicitly that establishing a trend takes three months for permits and six for starts and completions. On that standard, no single month in this release — including this one — should change anybody's view.

What Is This?

• What it is: New Residential Construction, published jointly by the Census Bureau and the Department of Housing and Urban Development around the middle of each month. It reports three stages of the homebuilding pipeline at a seasonally adjusted annual rate: building permits (units authorized in permit-issuing places), housing starts (ground broken), and housing completions (finished and ready to occupy), plus units authorized-but-not-started and units under construction. Each is broken out by single-family, two-to-four unit, and five-or-more unit buildings, and by the four Census regions. The permit figures come from an administrative count and carry no sampling error; the starts and completions figures come from sample surveys with published confidence intervals and, this month, a 75.8% response rate.
• Why it matters: Residential construction is the most interest-rate-sensitive major component of GDP, so this is where monetary policy shows up first and most violently. Permits lead starts, starts lead completions, and completions lead the supply of homes available to buy — so the three series together read as a forward calendar of housing supply. Right now it is also the test case for the Fed's claim that policy is not yet restrictive: if a 3.75%-4.00% funds rate and a 6.97% mortgage are doing real damage, this is the release where it appears.
• How to read it: Four things, and the first governs everything. Check the confidence interval before believing any change — Census prints one beside every figure, monthly starts changes routinely carry intervals of ±10% or more, and a great many published housing-starts headlines describe movements the agency itself calls statistically indistinguishable from zero. Revisions are large, averaging up to 3.8% and sometimes more, so this month's "prior" is frequently not the number the market was working from. Split single-family from multifamily always, because apartment projects are enormous, lumpy and financed differently, and a single development can move the national figure. And permits are more reliable than starts — no sampling error, three months to a trend rather than six — which makes them the better series and the less quoted one.

Want the full explainer? Economic News Events, Explained breaks down this release and every other one we cover, in plain English.

_For informational purposes only. Not investment advice._