Housing Starts Missed for a Third Straight Month at 1.22 Million — With December's Hike Priced In, Only Spec-Light Builders Keep Their Margin

September housing starts came in at 1,221,000, 5.2% below a revised 1,288,000 in August and 7.4% below September 2025 — a third consecutive monthly decline and the second-lowest reading since May 2020. Single-family starts of 859,000 were the weakest in a year, and single-family permits of 845,000 were the weakest since March 2023. Multifamily was the real hole: starts of 349,000 ran 25.9% below a year earlier. Total permits, at 1,400,000, sit at a three-and-a-half-year low, which means the pipeline for the first half of 2027 is being written down before it is written.

The detail that matters is the divergence between what builders finish and what they start. Completions rose 2.2% to 1,192,000 while starts fell 5.2% and single-family permits fell 4.0%. The industry is working off a construction pipeline rather than refilling it, which is what a rate regime that kills marginal demand looks like in the monthly data. NAHB's own survey said the same thing in words: buyer traffic fell to its lowest level since April 2012, with the prospective-traffic index dropping five points to 18 and the six-month sales expectation index to 32.

That wedge is why the December meeting matters more than the last round of easing. The policy rate is a 25-basis-point decision; the mortgage rate is what the 10-year and the spread decide, and it is 7.52% while the policy rate is 3.75%-4.00%. Homebuilders cannot change the second number. They can only change what they charge, what they give away and how much spec they carry into the reset — and the 2026 results have split the group precisely along that line.

The binding constraint is not the price of money at the short end. It is the monthly payment on a 30-year note, and the gap between what a seller already has and what a buyer would have to take on. With most pre-2022 mortgages written near 3%, the decision to move requires the new payment to look almost like the old one. At 7.52% it does not — and that is why pending home sales hit a record low in September, why existing-home sales ran at a 14-month low of 3.98 million in August with 1.62 million homes for sale and 4.9 months of supply, and why builders are paying buyers to show up.

Here is where the 2026 evidence separates the group. The spread between builders is no longer mainly about land quality or execution; it is about how much finished, unsold product each one has to discount into a 7.52% mortgage market. The cleanest measure is the ratio of homes under construction to the value of the sales backlog: below 1.2 times a builder is selling what it builds, and above 2.75 times it is effectively a merchant holding inventory. The peer group runs from 0.44 to 3.09.

Read the two ends of that table together. KB Home ran 74% of orders build-to-order and still printed a 16.5% gross margin on revenue 20% lower than a year earlier, because a home that has not been built does not have to be discounted. Lennar and D.R. Horton are carrying finished stock, and the discount is showing up as margin: 15.8% and 20.7%. NAHB's survey closes the loop — 43% of builders cut prices in October, the highest share of the 2020-2026 period, with the average cut holding at 6% for a seventh straight month. A builder with standing inventory cannot opt out of that.

Input deflation cannot fix a demand wedge. Lumber is 4.6% cheaper than a year ago and panels remain in surplus, yet the builder's gross margin is set by the incentive it pays, not by the price of a stud. That is the strange part of 2026: the cost side of the house is cooperating and the demand side is not, which means no amount of cost control rescues a builder carrying finished inventory into a 7.52% mortgage market.

Note what the appliance numbers are not saying. Completions rose 2.2% in September, so the appliance is still being bought for the house being finished — the one part of the chain that is working. What is missing is the second leg: replacement demand that comes from people moving, and moves have stopped. A frozen resale market means the appliance market gets one sale per new house and no replacement cycle, which is why Whirlpool is running a 13.5% trailing gross margin on 7.0x net debt to EBITDA.

This is why the housing complex trades on the 10-year rather than on the housing data itself. On Oct. 20 the market's response to a third consecutive miss in starts was to rally the builders — D.R. Horton up 2.5%, PulteGroup up 2.4% and Lennar up 3.6% — because weak data pulled the 10-year yield lower. That is a discount-rate trade, not a cash-flow trade, and it is the clearest warning in the complex: the equity market is priced for a pivot that the futures market says is not coming.

Facts: starts fell 5.2% to 1,221,000, single-family permits fell 4.0% to 845,000, builder sentiment hit 29, and a December Fed hike is roughly 89% priced with the 10-year near 5.08% and the 30-year at 7.52%. Inference: in that regime the binding constraint on a builder is not the size of its land bank but the number of finished homes it must discount, so the spread inside the group is an inventory story rather than a volume story. Speculation: if the 10-year breaks below 5% before the Fed moves, the complex re-rates on the discount rate and the inventory triage pays twice.

So the answer to the rate-lock question is uncomfortable for the biggest names in the industry. The builders that win a December hike are the ones that can walk away from a sale, and in 2026 that has meant build-to-order, low spec-to-backlog, and incentives held near 10% of price. KB Home and NVR are the cleanest expressions; Lennar and D.R. Horton carry the most inventory into the reset. In the supply chain, the same logic says prefer the supplier whose balance sheet can wait out a no-volume year, and be careful of the one carrying negative tangible book. If you are waiting for a Fed cut to fix housing, the futures market is telling you it is not coming in December — and the mortgage rate will not wait for it either.