New Home Sales Fall 10.5% to 607K — a Drop Census Calls Statistically Insignificant

Fundamentals · 2026-08-25

New single-family home sales 607,000 SAAR in July, 10.5 percent (±14.0) below June's 678,000 and 6.3 percent (±19.6) below July 2025's 648,000 — both confidence intervals include zero, so neither decline is statistically significant; homes for sale 488,000, up 1.9 percent (±1.2) — the only statistically significant change in the entire release; months' supply 9.6, up from 8.5 and the highest since January; median price $393,800, down 2.3 percent on the month and 0.9 percent on the year, while the average price rose to $508,800, up 5.4 percent on the year; Midwest sales halved from 75,000 to 43,000; next release September 24.

What Is This?

• What it is: The Census Bureau and HUD's joint monthly survey of new single-family home sales, released around the 24th at 10am ET. A "sale" is recorded when a deposit is taken or a sales agreement is signed, not at closing, which makes it the most forward-looking of the housing demand series. The survey is drawn primarily from a sample of building permits. Release CB26-128, covering July 2026.
• Why it matters: New home sales capture the marginal buyer at current mortgage rates, and they land today alongside FHFA data showing house prices flat in June and a Conference Board survey where homebuying intentions softened. Housing is the sector where restrictive policy shows up first, and three separate readings this morning are describing the same stall.
• How to read it: This is a sample survey with wide error bars, and Census publishes them for a reason. Every percent change comes with a 90 percent confidence interval; in the Bureau's own words, "if a range does not contain zero, the change is statistically significant. If it does contain zero, the change is not statistically significant; that is, it is uncertain whether there was an increase or decrease." Two further cautions from the release: "it takes 4 months to establish a trend for new houses sold," and the preliminary seasonally adjusted sales estimate is revised by about 5.0 percent on average.

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

Summary

New single-family home sales ran at a seasonally adjusted annual rate of 607,000 in July, down 10.5 percent from June's 678,000 and 6.3 percent below the 648,000 rate of July 2025. Those are the numbers that will lead every write-up of this release, and by Census's own standard neither of them means anything.

The confidence interval on the monthly change is ±14.0 percent. That puts the true change somewhere between a 24.5 percent decline and a 3.5 percent increase. The annual change carries ±19.6 percent, spanning a 25.9 percent fall to a 13.3 percent rise. Both ranges contain zero. The Bureau flags each with an asterisk defined at the foot of the release as: "there is insufficient statistical evidence to conclude that the actual change is different from zero." A 10.5 percent plunge in new home sales is, statistically, a month in which we cannot say whether sales rose or fell.

Run the same test across the rest of the release and almost everything fails it. Months' supply at 9.6 months is up 12.9 percent, ±21.3 — not significant. The median price at $393,800 is down 2.3 percent, ±7.4 — not significant. The average price at $508,800 is up 4.1 percent, ±11.8 — not significant. The annual comparisons on for-sale stock, months' supply and both price measures all carry intervals containing zero too.

Exactly one figure in the entire release clears the bar. The seasonally adjusted count of new houses for sale at the end of July was 488,000, up 1.9 percent from June's 479,000, with a confidence interval of ±1.2 percent. That range runs from +0.7 to +3.1 percent and does not contain zero. The only thing this release establishes with statistical confidence is that the number of new homes sitting unsold went up. It is the least dramatic number on the page and the only one that survives scrutiny.

That matters more than it sounds, because it is the same direction the rest of the housing data is pointing. At 488,000 the for-sale count is the highest of 2026, though still below July 2025's 496,000. Months' supply of 9.6 is the highest since January and roughly half again the six months conventionally treated as a balanced market. Builders are accumulating unsold homes.

The price split is the other detail worth having. The median sale price fell to $393,800, down 0.9 percent on the year, while the average rose to $508,800, up 5.4 percent. Median down and average up is a mix shift: the sales that are happening skew more expensive. Census warns explicitly that "changes in sales price data reflect changes in the distribution of houses by region, size, etc., as well as changes in the prices of houses with identical characteristics" — so this is not a like-for-like price signal in either direction. Read as a mix story, it says the entry-level buyer is the one stepping back, which is what a 9.6-month supply at current mortgage rates would predict.

Regionally, the monthly fall was concentrated rather than broad, and the concentration is where the sampling noise lives. The Midwest fell from 75,000 to 43,000, a drop of 32,000 that on its own is nearly half the national decline, and against 87,000 in July 2025. The South fell from 440,000 to 383,000, a decline of 57,000 and the largest in absolute terms. Offsetting those, the Northeast rose from 33,000 to 43,000 and the West from 130,000 to 138,000. The four regional changes sum to the 71,000 national decline. But the Midwest and Northeast are the smallest samples in the survey and the most prone to exactly the kind of swing that reverses next month.

Set against the week's data, this belongs with housing and hiring rather than with the activity surveys. A Fed holding at 3.50-3.75 percent with three members dissenting for a hike has seen a five-year high in Philadelphia Fed manufacturing and a Leading Economic Index turning positive for the first time in over four years — but also flat house prices, private hiring near 51,000 a month, and now builders adding to unsold stock. What this release does not do is give either camp a clean number to argue with. The honest summary is that new home sales in July were statistically unchanged, and the supply of unsold homes rose.

The Internals

Headline figures and their confidence intervals — the asterisk marks changes whose 90 percent interval includes zero:

Measure · July 2026 · Comparison · Change · Interval · Significant

New homes sold, SAAR · 607,000 · June 678,000 · -10.5 percent · ±14.0 · No

New homes sold, SAAR · 607,000 · July 2025 648,000 · -6.3 percent · ±19.6 · No

New homes for sale · 488,000 · June 479,000 · +1.9 percent · ±1.2 · Yes

New homes for sale · 488,000 · July 2025 496,000 · -1.6 percent · ±4.0 · No

Months' supply · 9.6 months · June 8.5 months · +12.9 percent · ±21.3 · No

Months' supply · 9.6 months · July 2025 9.2 months · +4.3 percent · ±25.2 · No

Median sales price · 393,800 dollars · June 403,100 dollars · -2.3 percent · ±7.4 · No

Median sales price · 393,800 dollars · July 2025 397,300 dollars · -0.9 percent · ±6.9 · No

Average sales price · 508,800 dollars · June 488,900 dollars · +4.1 percent · ±11.8 · No

Average sales price · 508,800 dollars · July 2025 482,800 dollars · +5.4 percent · ±13.1 · No

One of ten published comparisons is statistically significant, and it is the rise in unsold homes.

Sales by region, seasonally adjusted annual rate, thousands:

Region · June 2026 · July 2026 · Change · Note

Northeast · 33 · 43 · +10 · Fourth straight rise, highest since December

Midwest · 75 · 43 · -32 · Nearly half the national decline

South · 440 · 383 · -57 · Largest absolute drop

West · 130 · 138 · +8 · First increase in three months

United States · 678 · 607 · -71 · Regional changes sum to the total

The monthly series through 2026, seasonally adjusted annual rate:

Month · Sales SAAR · For sale · Months' supply

January 2026 · 576,000 · 477,000 · 9.9

February 2026 · 630,000 · 486,000 · 9.3

March 2026 · 659,000 · 479,000 · 8.7

April 2026 · 641,000 · 483,000 · 9.0

May 2026 · 630,000 · 486,000 · 9.3

June 2026 · 678,000 · 479,000 · 8.5

July 2026 · 607,000 · 488,000 · 9.6

For context, sales averaged about 632,000 across the first seven months of 2026 against roughly 699,000 across the second half of 2025 — a run rate about 10 percent lower, which is the trend worth watching rather than any single month.

What Census Tells You About Its Own Numbers

The explanatory notes carry three warnings that change how this release should be read, and they are printed in every edition:

Caveat · What it says

Statistical significance · A range containing zero means it is uncertain whether there was an increase or decrease

Trend establishment · It takes four months to establish a trend for new houses sold

Revision size · The preliminary seasonally adjusted total sales estimate is revised about 5.0 percent on average

That third one deserves weight against a 10.5 percent headline. An average revision of about 5 percent is roughly half the size of the reported monthly change, so the June-to-July move could be substantially rewritten next month without anything unusual happening.

The fourth caveat is definitional and often missed: a sale is counted when a deposit is taken or an agreement is signed, which can occur before a permit is issued. That is what makes this series lead closings — and also what makes it jumpy.

Impact on USD

• Slight bearish — housing softness supports the easing case, but a statistically insignificant headline is weak evidence for anything.
• The one significant figure, unsold homes rising 1.9 percent, points the same way as flat house prices and softening homebuying intentions.
• Months' supply at 9.6 against a balanced-market convention near six is the clearest sign policy is restrictive in this sector.
• Weight caveat: with a ±14.0 percent interval on the headline, this release should not move rate expectations much and generally does not.
• Watch the revision — an average 5 percent revision is half the size of the reported change.

Impact on US Indices (ES / NQ / YM)

• Mixed, lean bearish — homebuilders face rising unsold stock and a softening entry-level buyer.
• A 9.6-month supply with median prices down 0.9 percent on the year points to discounting and incentives, which compress builder margins before they compress volumes.
• Building materials and home-improvement names share the exposure, though repair-and-remodel demand is more resilient than new construction.
• Median down while average up says the entry-level buyer is stepping back — read through to affordable-segment builders specifically.
• Offsetting: the West rose for the first time in three months and the Northeast hit its best level since December, so this is not a uniform national retreat.

Impact on Gold

• Slight bullish — marginal, via the housing-weakness-supports-easing channel.
• Housing is the sector where restrictive policy is most visibly biting, which keeps a cut on the table for 2027 if not September.
• Offsetting: nothing in this release is statistically distinguishable from no change, so it is thin support for any macro thesis.
• Conditional: August CPI on September 11 remains the release that sets the level.

What To Watch

• August new home sales — Thursday, September 24. Watch the revision to July as much as the new print; a 5 percent average revision could reshape this month entirely.
• Whether the Midwest reverses. A fall from 75,000 to 43,000 in the smallest-sample region is the classic setup for a bounce; if it does not bounce, the weakness is real.
• Months' supply and the for-sale count. These are the two figures with the tightest error bars — they are where a genuine deterioration would show up first and most credibly.
• August CPI — Friday, September 11. Shelter and owners' equivalent rent are where this morning's flat FHFA prices eventually land.
• FOMC decision and dot plot — September 15-16. Housing is the sector where the hike dissenters' case is weakest, and this release adds to that even without statistical significance.

TLDR

New Residential Sales, Census Bureau and HUD (July 2026, released August 25):

• New home sales: 607,000 SAAR (from 678,000) — down 10.5 percent, interval ±14.0, not statistically significant
• Versus July 2025: down 6.3 percent, interval ±19.6 — also not significant
• New homes for sale: 488,000, up 1.9 percent, interval ±1.2 — the only significant change in the release
• Months' supply: 9.6 months (from 8.5) — highest since January, versus roughly six for a balanced market
• Median sales price: 393,800 dollars, down 0.9 percent on the year
• Average sales price: 508,800 dollars, up 5.4 percent on the year — a mix shift toward higher-end sales
• Regional: Midwest 75,000 to 43,000, South down 57,000, Northeast up to 43,000, West up to 138,000
• 2026 run rate about 632,000 versus roughly 699,000 in the second half of 2025
• Census: it takes four months to establish a trend, and preliminary sales are revised about 5.0 percent on average
• Next release September 24

Every outlet will lead with new home sales falling 10.5 percent, and Census's own footnote says there is insufficient evidence to conclude the change differs from zero. Nine of the ten published comparisons in this release fail that test — the single exception is unsold homes rising 1.9 percent, which is also the least exciting number on the page. Strip out the noise and what July actually establishes is that builders are sitting on more unsold houses, months' supply is 9.6 against a balanced-market six, and the entry-level buyer is stepping back. Watch the September 24 revision, whether the Midwest bounce-back arrives, and August CPI on September 11.

_For informational purposes only. Not investment advice._