There were 4.6 months of housing supply in July. There were 4.6 months in June. There were 4.6 months in July of last year. Everything around that number moved a little and the number itself did not move at all.
In the same stretch, affordability got materially better. The Housing Affordability Index rose to 103.3 from 98.3 a year ago, improving across all four regions. Mortgage rates came down to a four-week low. And purchase applications, measured against the same week last year, are down one percent.
So the input improved, across an entire country, measurably, and the output did not respond. In any operating system that is not a disappointment. It is a completed experiment, and it has told you that the thing you relieved was not the constraint. Almost nobody reads the result that way, which is what this piece is about, and it is not really about houses.
The number that stayed exactly the same
Existing home sales ran at a seasonally adjusted annual rate of 4.06 million in July, down 1.7% from June and up 0.7% against a year ago, with year-to-date sales up 2.4%. Total inventory was 1.54 million units, down 1.9% on the month and down 0.6% on the year.
Both halves of the ratio moved, in the same direction, by almost the same amount. Which is why months of supply came out at 4.6 and sat exactly where it sat in June and exactly where it sat last July.
The median existing-home price was $434,100, up 2.0% year over year, and that was the thirty-seventh consecutive month of annual price increases. Single-family ran $440,300 and condos $371,800.
NAR's chief economist Lawrence Yun called it "remarkably stable, even amid the rising mortgage rate environment of the past few months." That is a fair description and we are not going to argue with it. It is also a description of a system at equilibrium, and equilibrium is a statement about balance rather than about health. A queue that never gets shorter is stable too.
The experiment already ran, and somebody else paid for it
Here is what makes this month unusually informative rather than merely flat. The thing everybody said was the problem got better, and we can measure exactly what that bought.
Affordability improved by five index points against a year ago, in every region. Rates fell: the Mortgage Bankers Association's survey put the 30-year contract rate at 6.77%, down from 6.81%, and the daily lender surveys had the 30-year at 6.71% on 14 August, which apart from one afternoon was the lowest in four weeks.
The response, from the MBA's week ending 7 August: total application volume up 3.6%, purchase applications up 3% on the week, refinance applications up 5% on the week. Good numbers, until you look at the comparison that matters. Purchase applications are 1% below where they were a year ago. Refinances are 22% below. The average refinance loan size fell to its lowest level since July 2025.
Joel Kan, the MBA's vice president and deputy chief economist, put it precisely and without spin: "The reprieve in rates supported an increase in both purchase and refinance applications over the week, although the pace of applications has fallen below last year's pace in recent weeks."
Read those two facts next to each other. Affordability is better than it was a year ago. Transaction demand is slightly worse than it was a year ago. You improved the input and the output went the other way.
The sentence after the quote
The most interesting thing in the NAR release is not the quote. It is the sentence that follows the affordability figure, which says that the decline in inventory could continue to limit buyers' options and provide support for home prices.
So the same document reports that affordability improved and then explains the mechanism by which that improvement will not turn into transactions. Both halves are correct. Only one of them was quotable.
Yun's suggested remedy is that the market could see stronger activity if rates return closer to 6%. Be fair to him: that is a real effect at the margin, and some buyers do transact at 6% who will not transact at 6.7%. But hold it against what the release itself identifies as the limiting factor. If supply is what binds at 4.6 months, cheaper money does not produce houses. It produces more bidders for the same houses.
Which is, more or less, what thirty-seven consecutive months of year-over-year price increases looks like from the inside.
We are not forecasting and we are not going to. The point is narrower and more useful: the two halves of a single press release disagree about which lever matters, and the industry has spent a month quoting the half that recommends pulling harder on the lever that just got pulled.
Relieving the wrong constraint is the most expensive mistake we see
This is why the piece exists, because the pattern is not a housing pattern. It is the single most common expensive error in the operations work we do, and it has a completely reliable signature.
A team identifies a bottleneck. It spends real money relieving it. Throughput does not move. And the response, essentially every time, is not "we were wrong about the bottleneck." It is "we did not relieve it enough." So they spend again, against the same theory, and the second round is usually larger than the first.
We have watched it in three shapes in the last two years. Headcount added to a queue whose actual constraint was a single approval gate, so the new people queued alongside the old ones. Compute bought for a pipeline whose actual constraint was a nightly batch window, so the jobs finished faster and then waited. Budget doubled on a channel whose actual constraint was the time it took anybody to ring the lead back, so the cost per lead held and the cost per sale rose.
In all three the diagnostic was available and cheap, and in all three nobody ran it: a measurable improvement in the input that produced no measurable change in the output. That is not a bad quarter. It is a clean result with an unambiguous reading, and organisations throw it away because it arrives looking like failure rather than like information.
There is a second-order effect worth naming too, because it is where the money actually goes. Relieve a non-binding constraint hard enough and you do not get throughput. You get price inflation at the real bottleneck. In housing that shows up as the median sale price. In a business it shows up as the senior reviewer's calendar, or the one engineer who understands the billing system, or whatever your genuinely scarce resource is quietly becoming more expensive to access while you fund something else.
A small note on which rate you mean
Three numbers for the thirty-year rate crossed our desk inside seven days. The daily lender survey had 6.71% on 14 August. The MBA's weekly survey of actual applications had 6.77% for the week ending 7 August. Our own piece on 9 August, 110 Basis Points, cited 6.74%.
None of those contradict each other. One is a daily average of lender offers, one is an application-weighted weekly average, and they cover different days. But if you are building anything automated on top of "the rate," specify which rate, whose survey, and on what cadence, because we have watched teams alert daily on a number that only updates weekly and then spend a fortnight explaining the phantom moves.
Worth knowing how the last one actually moved, as well. The daily commentary attributes Friday's two-basis-point rise partly to traders closing out the week's positions rather than to the weaker retail sales report that preceded it. If your trigger fires on a two-basis-point change, some of what it is reacting to is position squaring on a Friday afternoon.
The playbook
Six moves, and the first one is a sentence.
1. Write down what you believe the binding constraint is before you spend. One sentence, dated, somewhere you will find it again. Almost nobody does this, which is why almost nobody can tell afterwards whether they were right.
2. State the output that should move, by how much, and by when. If you cannot name the expected magnitude in advance, no result can teach you anything, because every outcome will be explicable.
3. When you relieve it and nothing moves, bank the finding. You paid for the experiment either way. The finding is the only part you get to keep, and it is worth more than the relief would have been.
4. Look one step downstream for price. If relief did not produce throughput, check whether it produced cost at the real bottleneck. That is where non-binding spending reappears.
5. Specify the measurement before you specify the trigger. Which number, whose survey, what cadence, and what magnitude of change actually counts as a change.
6. Read both panels of your own reporting. NAR published the improvement and the reason it would not matter in one document. Most internal dashboards do exactly the same thing, and the reason nobody notices is that the two facts sit on different screens.
Our position
We picked housing because the data is public, unusually clean, and arrives monthly whether anyone likes the result or not. Very few businesses have anything that honest about their own operations.
But the argument transfers directly, and it is uncomfortable. Somewhere in your company there is a constraint you have been funding for two years on the assumption that it is the bottleneck. You do not need to run a new experiment to find out. Go and find the period when that constraint measurably eased, for any reason, including by accident, and look at whether output moved.
If it did not, you have been buying relief for something that was never the problem. That is expensive news and it is still the cheapest news you will get all year, because the experiment has already run and somebody has already paid for it.
Four point six months of supply, unchanged from a year ago, is what it looks like when an entire industry declines to read the result at the same time.
Sources
Every figure above is carried from the three reports below, credited in the body and linked here.
- Mortgage News Daily, "Existing-Home Sales Ease Slightly, Remain Above Year-Ago Levels", 14 August 2026. Source for the July existing-home sales pace and changes, the inventory figures, the 4.6 months of supply, the median prices, the thirty-seventh consecutive annual increase, the Housing Affordability Index, both Lawrence Yun statements and the inventory sentence that follows them. The underlying NAR release was read through this report rather than directly.
- Mortgage News Daily, "As Expected, Mortgage Apps Bounce in Response to Rate Reversal", 14 August 2026, reporting the Mortgage Bankers Association survey for the week ending 7 August. Source for the application volumes, the year-ago comparisons, the refinance share, the average loan size, the survey rates and the Joel Kan quotation.
- Mortgage News Daily, daily mortgage rate coverage, 14 August 2026. Source for the 6.71% thirty-year rate, the four-week-low observation and the explanation of Friday's move as partly the closing out of the week's trading positions.