Separate the Permit Count From the Start Count Before Calling a Construction Slowdown

On 18 August 2026 the Census Bureau and the Department of Housing and Urban Development published the July residential construction report. Building permits rose 5.0 percent to a seasonally adjusted annual rate of 1,443,000 units. Housing starts fell 12.4 percent to 1,239,000. The two series describe the same month, the same builders and the same financing conditions, and they point in opposite directions.

That matters because the standard framing of housing as a leading indicator depends on an ordering: permits lead starts, starts lead construction payrolls, and payrolls lead the layoff announcements that reach the press. The lag is the entire reason the permit series is treated as an early warning device. In July the first link ran backwards, and the payroll report released eleven days earlier had already complicated the third.

The useful conclusion is not that the indicator failed, but that the chain is usually described wrongly. Permits and starts are decision flows. Construction employment is funded by a stock — the number of units physically under way — and a stock does not turn when a flow turns. Reading the two as one quantity produces false alarms in both directions, and the July release contains an example of each.

The July Print, Line by Line

All of the following is from the 18 August release. Levels are seasonally adjusted annual rates in thousands of units; bands are the Census Bureau's 90 percent intervals on the monthly change.

Series	July 2026	vs June	90% band	vs July 2025
Building permits, total	1,443	+5.0%	±9.5%	+3.1%
Permits, single-family	894	+2.5%	—	+1.1%
Permits, five units or more	490	—	—	—
Housing starts, total	1,239	-12.4%	±9.5%	-13.5%
Starts, single-family	808	-9.9%	±10.4%	-15.7%
Starts, five units or more	421	-15.6%	±20.2%	-7.1%
Completions, total	1,212	-9.1%	±10.2%	-16.8%
Authorized but not started	279	—	—	—
Units under construction	1,262	—	—	-6.0%

Two lines carry most of the information and neither is a headline. Units authorized but not started stood at 279,000, the widest backlog in over a year. Units physically under construction stood at 1,262,000, six percent below a year earlier. Single-family starts at 808,000 were the lowest since November 2022.

What the Confidence Interval Removes

The Census Bureau attaches a 90 percent band to every monthly change in this release. The bands are wide because the survey samples permit offices and building sites rather than counting all of them, and applied to July they do most of the analytical work.

• Starts fell 12.4 percent with a band of ±9.5 percentage points. The interval does not contain zero. The decline is statistically distinguishable from no change.
• Permits rose 5.0 percent with the same ±9.5 point band. The interval does contain zero. On the agency's own standard, there is not sufficient evidence that permits changed at all in July.
• Multifamily starts fell 15.6 percent with a band of ±20.2 points. That series is close to uninterpretable on a single month.

So the divergence that looks dramatic in the headline pairing is, statistically, one confirmed decline and one figure that cannot be separated from noise. Reading direction without reading dispersion builds a permit-recovery story out of a number the publishing agency declines to call a change. The reverse holds: one strong permit month is not evidence of a turn.

A Permit and a Start Are Not the Same Decision

A permit is an option: it costs a fee, establishes an entitlement, and in most jurisdictions stays valid for a defined window. A start exercises that option, committing capital, locking a construction loan draw schedule and putting a crew on a site with a payroll attached. The two decisions respond to different variables.

Permit filing responds to land inventory, entitlement timelines and the cost of keeping optionality alive; builders with land under control have reason to keep permits current even without intending to break ground, since re-entitlement is slower and dearer than renewal. Starts respond to the marginal economics of the unit: the rate the buyer faces, the price the builder expects, and the finished inventory already unsold.

The gap therefore has an interpretation of its own. When permits hold and starts fall, the queue of authorized-but-unstarted units widens — not a supply pipeline filling, but options held rather than exercised. That 279,000-unit backlog is the clearest expression of builder hesitancy in the release, and more informative than either headline change.

The Stock That Pays the Crew

A framing crew, an electrical subcontractor and a drywall crew are not paid by groundbreakings. They are paid by the number of units in the ground at their stage — the stock of units under construction, 1,262,000 in July, down 6.0 percent year over year while starts were down 13.5 percent.

The arithmetic is straightforward. The stock drains through completions and refills through starts, and completions ran at 1,212,000 in July, down 16.8 percent on the year. When both inflow and outflow fall, the stock erodes slowly, and residential payrolls erode with it — on a lag measured in quarters, not months.

This is why the simple ordering fails. Starts do not lead residential employment by a fixed interval, because the transmission runs through a stock with its own drain rate. A one-month drop in starts changes the refill rate of a pool holding 1.26 million units; it does not empty the pool.

The Payroll Print Says Something Different From the Headline

The Bureau of Labor Statistics published the July employment report on 7 August. Total nonfarm payrolls fell by 23,000, the unemployment rate held at 4.1 percent, and revisions took 66,000 off May and 37,000 off June, a combined 103,000. Construction in that same report added roughly 22,000 jobs on a base of about 8.34 million, with an industry unemployment rate of 3.7 percent. As an aggregate it was one of the stronger sectors in a contracting month. The composition says the opposite.

Component, July 2026	Monthly change
Nonresidential specialty trade contractors	+15,400
Nonresidential building	+4,200
Heavy and civil engineering	+400
Residential specialty trade contractors	+2,600
Residential building	-500

Residential construction employment therefore rose by roughly 2,100 in July, the first monthly gain in four, against a six-month average of about 5,350 jobs lost per month. Year over year it is down 44,200, the seventeenth consecutive month of annual decline. Almost the entire aggregate gain came from nonresidential work, where data-center and industrial construction has been absorbing trades.

The consequence is that the aggregate construction payroll line cannot confirm or deny a housing-driven slowdown. Two sub-sectors with different demand drivers are added together, and a process watching only the headline registers nothing while nonresidential offsets residential.

The Demand Ceiling That Sets the Conversion Rate

If permits are options, the variable deciding whether builders exercise them is finished inventory. The June new residential sales report, published 24 July, put sales at a 628,000 annual rate, up 1.6 percent on the month and down 5.6 percent on the year, with 485,000 new homes for sale at month end — a 9.3 months' supply. The median new home price was $398,300, down 2.7 percent on the year.

Within that 485,000, the stage split is the operative number: 113,000 not started, 254,000 under construction, 118,000 completed. Completed unsold units carry the highest holding cost of the three, and builder behaviour has historically tightened as that figure approaches roughly 120,000. The current reading sits just under it.

Survey data is consistent with that constraint. The NAHB/Wells Fargo Housing Market Index printed 35 in August, one point above July and the sixteenth month below the neutral 50 line, with components of 39 for current sales, 43 for six-month expectations and 23 for prospective buyer traffic. Thirty-five percent of builders reported cutting prices by an average of 6 percent; 63 percent reported using sales incentives.

Financing has not been the swing factor. Freddie Mac's weekly survey put the 30-year fixed rate at 6.67 percent on 13 August, against 6.69 percent the prior week and 6.58 percent a year earlier — nine basis points of annual change. The binding constraint is inventory and price, not rates.

Regional Dispersion Is Wider Than the National Change

The national 12.4 percent decline averages four census regions that moved very differently: the Midwest fell 27.1 percent, the South 12.6 percent and the West 13.8 percent, while the Northeast rose 17.1 percent on multifamily even as its single-family starts fell 27.1 percent. Regional monthly figures carry the widest bands in the release, so one month alone is close to unusable.

The dispersion still matters, because residential construction employment is geographically immobile in the short run, and national averaging conceals the concentration that produces the regional layoff headlines the simple framing treats as final confirmation.

The Objection That Survives Scrutiny

The strongest argument against the above is that it over-reads a single month of a noisy series, and that objection is largely correct.

July 2026 is one observation. The Census Bureau's own bands say the permit increase is not distinguishable from zero, and the same standard applies to the divergence itself: one significant change paired with one insignificant one is thin evidence that the permit-to-start relationship has structurally altered. The queue of authorized units has widened and narrowed before without a downturn following. A backlog of 279,000 is a level, and levels become signals only once a direction holds across several quarters.

There is also a mechanical explanation requiring no behavioral story. Permits and starts come from different collection processes with different seasonal factors, and a large multifamily project is permitted in one month and started many months later. A month in which several such permits landed is enough to produce the pattern.

The frame that survives is narrower than the one it replaces. It is not that permits have stopped leading starts. It is that the permit-to-start conversion rate, the stock of units under construction and the composition of construction payrolls are three separate quantities, and using one as a proxy for the others produces errors in a predictable direction. That claim is testable across quarters and does not require July to be a turning point.

Where the frame genuinely fails is a rate-driven episode. If mortgage rates fell two hundred basis points over two quarters, the authorized queue could convert quickly, the under-construction stock could stabilize within a year, and residential payrolls could turn without the slow-drain arithmetic applying at all. The stock logic governs the downside path, not a sharp reversal.

Two Paths From Here, Weighted Evenly

Neither path below is a forecast. Both are consistent with the data published to date, and the same July release supports each.

Path A — the queue drains into starts. Rates hold near 6.7 percent or drift lower, the 118,000 completed unsold units clear through the incentive and price-cut activity 63 percent and 35 percent of builders respectively report using, and the 279,000-unit authorized backlog converts through the autumn. Starts stabilize, the under-construction stock flattens, and residential payroll losses slow toward zero without a visible layoff event. July then reads, in hindsight, as sampling noise around a flat trend.

Path B — the queue widens further. Completed unsold inventory pushes through 120,000, builders defer starts to protect margin, and the authorized backlog keeps climbing through the autumn releases. The under-construction stock keeps falling at six percent or faster, the six-month average of residential payroll losses reasserts itself after July's small gain, and the annual decline extends past 44,200. Nonresidential hiring keeps the aggregate construction line positive for several more months, and the first widely covered layoff headlines arrive well after the data has turned.

The observable that separates them is not starts. It is the authorized-but-not-started series read alongside completed unsold inventory: a narrowing queue with falling completed inventory points to Path A, a widening queue with rising completed inventory to Path B.

What to Watch Next Week

• New Residential Sales, July — Census Bureau, 25 August. The line that matters is the stage-of-construction split of homes for sale: whether completed units move above or below 118,000, and whether months' supply moves off 9.3.
• Construction Spending, July — Census Bureau, 1 September. Residential versus nonresidential put-in-place value tests directly whether the sector split seen in the payroll data is holding.
• New Residential Construction, August — Census Bureau, 17 September. This release carries the July revision as well as the August figures. Given a ±9.5 point band, the revision to July's 12.4 percent decline is as informative as the new print.
• Employment Situation, August — BLS, 4 September. Read residential building and residential specialty trade contractors separately, not the aggregate, and treat the first estimate as provisional given the 103,000 two-month revision just recorded.
• NAHB/Wells Fargo Housing Market Index, September. Buyer traffic at 23 is the leading component. A headline move without a traffic move is not a demand signal.

Concrete Framework — What Comes First

The ordering below replaces the permits-to-starts-to-payrolls chain. Each step names the series, the field to read, and the condition that would move the assessment.

1. Read the band before the direction. Every monthly change in the New Residential Construction release carries a 90 percent interval. If it contains zero, the change is not evidence on its own. In July that disqualified the permit figure and retained the starts figure.
2. Read the conversion, not the permits. Track authorized but not started, currently 279,000, against the trailing three-month average of single-family starts. A widening ratio is deferral; a narrowing ratio with flat permits is conversion.
3. Read the stock, not the flow, for labor. Units under construction, currently 1,262,000 and down 6.0 percent year over year, is the variable that funds residential payrolls. Its rate of change, not the starts figure, is the quantity to map against employment.
4. Split construction payrolls before using them. Residential building and residential specialty trade contractors are the housing lines; in July they netted about +2,100 while the aggregate added roughly 22,000.
5. Anchor the demand side on completed unsold inventory. Completed new homes for sale at 118,000 and months' supply at 9.3 cap how fast the authorized queue can convert; builders have historically throttled output as that figure approaches 120,000.
6. Treat regional layoff coverage as lagging confirmation. With 1.26 million units draining through completions running at a 1,212,000 annual rate, the interval between a starts decline and a visible employment event runs in quarters.
7. Re-check the ordering if rates move more than a hundred basis points in a quarter. The slow-drain logic governs a deferral cycle; a financing shock in either direction changes the conversion rate of the authorized queue fast enough to invalidate the lag structure above.

The July 2026 release is a useful test case precisely because it embarrasses the simple version of the indicator. Permits rose and starts fell in the same month; construction payrolls rose while residential payrolls fell for the seventeenth year-over-year month running. None of that is contradictory once the flows and the stock are kept apart. All of it is, if they are not.

This article is analysis of published economic data and is not investment, financial or professional advice. Figures cited are as published by the issuing agencies and are subject to revision.