Guide · Markets
Housing Starts and Building Permits Explained: How to Read the Monthly Construction Report
Each month the Census Bureau reports how many homes US builders began and how many permits they pulled. Here is what the numbers measure, why permits lead starts, how wide the error bars are, and what the September 2026 report, starts down 5.2% to 1.221 million, said about rates and homebuilder stocks.
The New Residential Construction report is published jointly by the Census Bureau and the Department of Housing and Urban Development around the middle of each month, at 8:30 a.m. ET, for the previous month. It contains three series. Building permits count the authorisations local governments issued for new privately owned housing units. Housing starts count the units on which construction began, which in practice means excavation for the footings. Completions count the units finished. All three are reported as seasonally adjusted annual rates, so a figure of 1,221,000 means that if the month's pace continued for a year, adjusted for the usual seasonal pattern, builders would start 1.22 million homes.
The split that matters is between single-family homes and buildings of five or more units. Single-family starts are the part that responds most directly to mortgage rates, because they are built for sale to households who need a loan; multi-family starts respond to rents, construction financing and the pipeline of projects approved years earlier, which is why apartment starts can fall for two years while single-family starts recover. Permits lead starts by one to three months, so a drop in single-family permits is the earliest warning that starts will fall, and a rise in permits with flat starts says builders are preparing to build but have not yet committed the money.
Here is a real example. On October 20, 2026, the Census Bureau reported that housing starts fell 5.2% in September to an annual rate of 1,221,000, from a revised 1,288,000 in August and against a consensus near 1.30 million; the rate was 7.4% below September 2025. Single-family starts fell 4.6% to 859,000, the lowest in a year, and starts in buildings of five or more units were 349,000, 25.9% lower than a year earlier. Total permits were almost unchanged at 1,400,000, but single-family permits fell 4.0% to 845,000. The release came a day after the NAHB builder-confidence index fell to 29, its lowest since 2012, and ninety minutes before the National Association of Realtors said pending home sales fell 7.3%, so the three reports told one story: mortgage rates above 7% were reaching construction and contracts, not only sentiment.
Two things about the report trip up readers. The first is the margin of error. The Census Bureau publishes a 90% confidence interval with every change, and for the September 2026 headline it was ±11.8 percentage points, which means the 5.2% decline was not statistically distinguishable from zero. Monthly changes in starts are noisy; the three-month average, and the year-on-year comparison, are what economists use. The second is revisions. The August figure was revised, as the prior month usually is, so part of each month's change is the previous month moving under it. Read the level, the trend and the single-family line before the headline percentage.
For markets, the report moves three things. Homebuilder stocks, through ETFs such as ITB and XHB, react to the single-family numbers and to what they imply for order books, but they react more to the Treasury market's reading of the data: on October 20, 2026 the homebuilder ETF rose 2.8% on a weak report because the 10-year yield fell to 5.12%. Bond yields react when the data change the picture of how much high rates are slowing the economy. And building-products and materials companies, from lumber to appliances, trade on the starts trend because each start is a bill of materials. A simple routine: compare the headline with consensus, check whether single-family permits confirm it, look at the confidence interval, and then watch the bond market, because that is where the stock reaction will be decided.