US homebuilders — an industry-wide price war where the cheapest names are cheap on book and dear on earnings

Claim

Coverage caveat: this is a field read derived from one full single-name analysis (LEN) plus a peer snapshot and macro pull — not a screen sweep. It is written to fill an empty slot in the coverage map and to record what was learned once, so the next housing question starts here. Treat the LEN detail as high confidence and the cross-field generalisations as medium.

The field's condition. New-home months supply is 9.3 against 4.6 for existing homes. That single ratio is the whole macro story: builders compete against a resale market with half their inventory pressure and a locked-in low-rate seller base that does not need to transact, so they must buy the buyer down. 37% of builders were cutting price in July 2026, averaging −6%, and the median new-home price fell below the median existing-home price in April 2026 for the first time in five decades. The 30-year mortgage sat at 6.67% (2026-08-13) — higher than a year earlier. The rate-cut thesis embedded in 2026 builder multiples did not arrive. NAHB HMI at 34 marks a 15th consecutive month below 40, the longest run since 2012.

The screening trap, and it is uniform. Builders' earnings fall much faster than their book value, so a margin recession compresses P/E slowly and P/B fast. The result is that the names screening cheapest on book are the ones whose ROE collapsed most — and those same names carry the highest forward P/E. Low P/B plus low ROE is not a discount; it is the market correctly capitalizing a poor return. The justified multiple is (ROE − g)/(r − g): at 7% ROE with r=10%, g=3%, fair P/B is 0.61x, not 1x. Screen this field on ROE and reported gross margin first, then on P/B — never the reverse.

The differentiator is not land-light. See [[pattern-asset-light-can-retain-or-distribute-the-economics]]: the correlation between lots controlled and returns runs backwards across this group. What separates the winners is whether scale reaches the margin line at all.

Evidence — 2026-08-19

	LEN	DHI	PHM	NVR	TOL	KBH	MTH	TMHC (Berkshire take-out)
P/E fwd	~15.5	12.9	11.6	15.6	10.6	12.7	—	11.3
P/B	0.97	1.79	1.88	5.07	1.70	0.91	0.96	1.10
EV/EBITDA	11.2	11.7	9.3	11.1	8.4	12.7	—	8.1
Gross margin	16%	21%	26%	22%	25%	18%	18%	23%
ROE	7%	13%	15%	32%	16%	7%	6%	11%
% lots controlled	98%	76%	~60%	~100%	—	—	—	—

Reported gross margin by builder, most recent quarter: TOL 23.9% · PHM ~24.4% · NVR 21.2% · DHI 20.7% · LEN 15.6%. Lennar is the outlier at the bottom by 500bp against DHI, its closest structural comp, because its stated strategy is to hold volume and let margin absorb the shock (10-K: "we adjust prices, with our gross margin being a shock absorber … to generate increased market share"). TOL is the cleanest read on the luxury end holding better — Q3 FY26 deliveries −10% but contracts up 2,508 vs 2,388, incentives stable near 8% for three quarters.

Berkshire's revealed private-market value. Berkshire acquired Taylor Morrison outright at $72.50/share cash — $6.8B equity, $8.5B EV, a 24% premium, closed 2026-07-24; and raised Lennar Class A +29.8% to 6.23% of the class. TMHC went at 1.10x book / 1.19x tangible book. That is the most useful valuation anchor in the field — a cash transaction rather than a model — and it says a US builder is worth roughly book to 1.2x tangible book at this point in the cycle. Notably Berkshire paid a premium for the higher-margin builder rather than buying more of the cheapest one.

Structural positives worth carrying forward: a cumulative US housing shortage management sizes near 5M homes against ~1.5M annual demand; new homes insure ~53% cheaper than 10-year-old homes, a genuine and underrated relative tailwind for new construction versus resale as insurance costs rise.

Structural negatives: immigration enforcement against a construction workforce that is 34% immigrant (>60% in drywall, roofing, plastering), with 28% of firms reporting disruption — the highest-conviction threat, and asymmetrically dangerous to land-light builders who pay a cash option fee for every extra day of cycle time. Tariffs add $7,500–10,000/home. Insurance in FL/TX/CA is a demand tax on exactly the entry-level buyer.

Confirmed noise, do not re-research: modular/off-site construction is 3% of US single-family and falling — Katerra burned $2B+, Veev $600M, L&G £236M, Ilke £320M, all dead. iBuying is structurally dead. Builder "proptech" is SG&A, not a moat.

What would falsify this

New-home months supply converging toward the resale figure (below ~7) while incentive rates fall across multiple builders — that would mark the price war ending and flip the field from "value trap risk" to "cycle bottom." Equally, a builder demonstrating ROIC recovery above WACC on a land-light structure would falsify the returns-scepticism above.

Verdict

No entries taken. The field is interesting and no name is compelling at 2026-08-19 prices. LEN was analysed in full and rated WATCH, conviction 5.5, with entry at $70–78 versus a $87.30 price. If the field is entered, the evidence points to PHM or DHI over LEN — better margins, better ROE, and a lower forward multiple, which is the opposite of what the P/B screen suggests. TOL is the luxury-end alternative and the cheapest large builder on forward earnings. The field is also a genuine diversifier against an AI/software-concentrated book and is one of the few areas trading below its own historical band, against [[pattern-ai-levered-fields-trade-above-own-band]].

Related

• [[pattern-asset-light-can-retain-or-distribute-the-economics]]
• [[pattern-capitalized-option-fees-defer-cogs]]
• [[pitfall-vendor-gross-margin-buries-homebuilder-sga]] — LEN-specific vendor margin trap; DHI and PHM are mapped correctly
• [[pitfall-multiple-trim-inverts-on-peak-cycle-cyclicals]] — cyclicals need dollar trims
• [[principle-down-a-lot-is-not-cheap]]

History

• 2026-08-19 — established from /analyze LEN plus peer and macro pulls. First entry in the coverage map for housing. Not a screen sweep — a proper /screen of the field would upgrade confidence from medium.