The Census Bureau Says New Home Sales Cratered 10.5%. Its Own Footnote Says That Might Be Zero.

The Census Bureau just told you new home sales cratered 10.5%. Buried in the same release is a footnote saying that number might actually be zero.

Every housing outlet ran the same headline this week: new home sales fell 10.5% in July. HousingWire called it a "miss to estimates." Floordaily and Universe News Network both led with the same double-digit drop. Nobody led with the sentence sitting three paragraphs into the Census Bureau's own release: the 90% confidence interval on that 10.5% is ±14.0 percentage points.

Do the arithmetic and the actual range runs from -24.5% to +3.5%. That interval contains zero. In the Census Bureau's own words, printed at the bottom of every New Residential Sales release, "there is insufficient statistical evidence to conclude that the actual change is different from zero." The number that led every housing story this week isn't confirmed to have moved at all.

This isn't a one-off gotcha. It's how new home sales data works every single month, and almost nobody in housing media writes the sentence that matters: which of this month's numbers are real, and which are noise dressed up as news.

What the confidence intervals actually say

The July 2026 New Residential Sales release (CB26-128, published August 25) is a survey estimate, not a census count, and the Bureau publishes a 90% confidence interval next to every single percent change so readers can tell signal from sampling noise. Here's what happens when you actually apply that filter to July's numbers:

• National sales, -10.5% month-over-month (±14.0): not statistically significant. Could be flat, could be up.
• National sales, -6.3% year-over-year (±19.6): not statistically significant either.
• Northeast sales, +30.3% (±83.9): wildly not significant — the margin of error is nearly three times the headline number.
• South sales, -13.0% (±17.8): not significant.
• West sales, +6.2% (±34.2): not significant.
• Months' supply, +12.9% (±21.3): not significant.
• Median price, -2.3% (±7.4): not significant.
• Average price, +4.1% (±11.8): not significant.

Out of every headline figure in the release, exactly two clear the bar for statistical significance — meaning the confidence interval does not contain zero:

• Midwest sales, -42.7% (±13.4): real. The range runs from -56.1% to -29.3%. However you want to explain it, new home sales in the Midwest actually collapsed in July.
• New homes for sale (inventory), +1.9% (±1.2): real. Builder inventory genuinely grew, confirmed at 90% confidence, and it's the smallest headline number in the whole release — which is probably why nobody wrote about it.

That's the actual story hiding in a report every outlet covered: the loud national number is noise, a real regional collapse got buried under it, and the one quietly-confirmed data point — rising for-sale inventory — didn't make a single headline.

The Numbers

Straight from the Census Bureau and HUD's joint release, seasonally adjusted at an annual rate:

• New home sales: 607,000 units, down from June's 678,000 (revised)
• New homes for sale: 488,000, a confirmed +1.9% above June's 479,000
• Months' supply: 9.6, up from 8.5 in June (not statistically significant)
• Median sales price: $393,800, down from $403,100 in June (not statistically significant)
• Average sales price: $508,800, up from $488,900 in June (not statistically significant)
• Midwest sales: roughly 43,000 units (SAAR), down a confirmed 42.7% from June

HousingWire's Logan Mohtashami — to his credit, one of the only analysts who reprinted the confidence-interval language in full — points out that builders have kept new home sales in a stable range for a decade by buying mortgage rates down below 6%, a lever the existing-home market doesn't have. That's real context. It's also a separate claim from "sales cratered 10.5% this month," and the two got merged into one headline anyway.

Common Mistakes Investors Make Here

• Reacting to a single month's percent change as if it's a trend. The Census Bureau says it takes four months of data to establish a trend in new home sales — one print, especially a noisy one, tells you almost nothing.
• Treating "seasonally adjusted annual rate" swings from small regional bases as equivalent to national swings. The Northeast's headline-grabbing +30.3% comes from a sample so small (33,000 to 43,000 units) that an 84-point margin of error is mathematically inevitable — it looks dramatic and means almost nothing.
• Skipping the inventory line because it's the smallest number on the page. A confirmed 1.9% rise in homes for sale is a real, durable signal about builder behavior. A noisy 10.5% sales headline is not. Investors who chase the big scary number and ignore the small confirmed one are optimizing for the wrong data.
• Assuming every government data release is precise to the decimal. These are sample surveys with published sampling error for exactly this reason — the error bars are not a disclaimer to skip, they're the actual answer to "did anything change."

Where I Land

The Midwest number is real and I'd underwrite it that way — a confirmed 42.7% drop in new home sales in a single month is a genuine regional air pocket, not noise, and it deserves more attention than the national headline it got buried under. Everything else in this release that made headlines this week — the -10.5% national print, the price moves, the months'-supply jump — I'm treating as unconfirmed until it shows up again in the August release on September 24. I don't trade on a number the source itself says might be zero, and neither should you. The one thing I'd actually act on today is the confirmed inventory build: builders now have more completed and in-progress homes sitting for sale than they did in June, for real, and that's quiet, durable pressure on new-construction pricing power regardless of what the noisy sales print says. By the September 24 release, I'll call it now: the July "-10.5%" print gets revised toward something closer to flat, and the Midwest weakness shows up again as the one region still genuinely underperforming.

How to Use PropGPT for This

"Pull new construction inventory and days-on-market trends for [metro] over the last 6 months, and flag whether the change looks like a sustained trend or a single noisy month." This gives you a local read on whether the "confirmed" national inventory build (+1.9%) is showing up in your specific market, instead of reacting to the national headline.

"Show me active new-construction listings in the Midwest census region with the steepest month-over-month price cuts or longest days-on-market right now." Targets the one region the Census data says genuinely weakened — a real, confirmed pullback, not a noisy national average.

"Pull builder incentive and rate-buydown activity for new-construction listings in [metro], and compare effective buyer cost against comparable resale inventory." New-home sales are being propped up by builder-funded rate buydowns that resale sellers can't offer — this finds where that gap is widest, and where it's closing.

"Re-run my 5-year cash flow projection for [deal] using flat 0% sales-growth and flat 0% price-growth assumptions instead of trailing 3-month figures, and show me the delta." Strips out reliance on a monthly print that might not have moved at all, and forces your underwriting to survive a "nothing changed" scenario.

"Track new-construction for-sale inventory counts monthly for [metro] and alert me only when the month-over-month change exceeds [X]%, to filter out normal sampling noise." Builds your own confidence-interval discipline into a screener, so you stop reacting to swings too small to be real.

The Bottom Line

The housing data cycle runs on headline percentages, and this month's headline percentage — new home sales down 10.5% — comes with the Census Bureau's own admission that it might not have happened at all. The number that's actually confirmed, a 1.9% rise in homes for sale, didn't make a single story. Read the confidence interval before you read the headline, underwrite off what's statistically real, and let everyone else trade on the noise.

This is market analysis and commentary, not individualized investment, legal, or financial advice.