How a source-and-vintage stamp replaces the vacancy cutoff

For the same quarter, the same country, the spread is more than six points. That isn’t a rounding argument. It’s a sourcing problem wearing a data costume, and it needs a rule that names the source and the vintage next to every number, not a finer label.

6.1 points · Office vacancy, Q2 2026 – 14.0% at CoStar, 18.3% at CBRE, 20.1% at Cushman & Wakefield¹²³

The fast-no already stopped trusting the label – office is the holdout

The whole point of a yield-side fast-no is to screen on income, not on the word above the door. A stabilised medical-office building and a half-empty tower are both “office”; only one of them clears a cash-flow gate. So I retired most asset-class labels and replaced them with occupancy floors, covenant tests and cap-rate reads.

Office resisted. It kept a label-level shortcut – a single vacancy cutoff – because the headline felt decisive enough to fast-no the whole category.

Then CoStar’s revised read puts national office vacancy at just under 14% for the same quarter¹. One number suggests a recovery story, another points to a crisis, and the third sits in between – and my old rule would have picked whichever I happened to open first.

The rule: a source-and-vintage stamp on every figure

The fix isn’t a cleverer taxonomy. It’s a discipline: no vacancy or cap-rate figure enters the screen without the name of who measured it and the quarter it belongs to, and no two are compared unless both stamps match. Here is how that runs through a single deal file.

1. Stamp the source and the as-of date before the number does any work. – a different cut again, and one that only means something once I know it’s prime-versus-all, not tracker-versus-tracker.

1. Split office by class instead of trusting one headline. The opportunity the blended number hides is in the class split, not the category.

1. Apply the same stamp to every other cap rate. Multifamily Class A and B cap rates compressed five basis points to an average of 5.2% in Q2 2026⁶ – but that figure comes from one mortgage-broker site, repeated across two of its own pages⁷, which is repetition, not confirmation. Industrial averaged 5.2% in the same quarter with 58 million square feet of net absorption⁸. Retail sale cap rates held at 6.59% in April 2026, 21 basis points below a year earlier⁹ – relayed on LinkedIn from a Crexi report I couldn’t confirm at the primary source, so it too carries a single-source flag.

1. Treat multifamily vacancy as contested, not settled. The same relay reports multifamily vacancy improving to 15.90% in April 2026 from 17.80% in March, a 190-basis-point move¹⁰. Another compilation puts multifamily vacancy at 6.2% to 7.2% for Q2 2026¹¹ – roughly ten points below the first, with no bridge between them. That gap is wider than the office spread that started this, and it is unreconciled.

1. Check the vintage even on Census-grade data. The Census Bureau’s June 2026 release authorises multifamily permits – buildings of five or more units – at a seasonally adjusted annual rate of 445,000¹². A FRED-based compilation the same summer puts the June figure at 449,000, against 468,000 in May¹³: same direction, different absolute count, because one rounds or revises against the other.

1. Watch the same divergence on totals. GlobeSt reports US home permits down 2.2% year over year to 1.42 million for the year ended May 2026¹⁴. Two Census-based analyses put the same May total at 1.413 million and the annual change at just 0.2%¹⁵¹⁶. Same month, same underlying survey, and a decline that reads either as material or as flat depending on the compiler.

On that last point, I’ll be plain about a gap: I found no US Census document that states a single “annual home permit drop” percentage for May 2026 in headline form. The primary release gives monthly and year-over-year changes; the tidy annual figure is something the secondary sources assemble, each their own way.

Stated plainly, the rule is one line

Every vacancy and cap-rate number in the screen carries two tags – who measured it, and as of when – and nothing gets compared across mismatched tags. That’s it. It is less elegant than a category cutoff and far harder to game, because a six-point disagreement can no longer hide inside the word “office.”

The strongest counter-case: direction beats level

The best argument against all this is that I’m chasing precision that doesn’t matter. What matters, this case runs, is direction and granularity – not the decimal.

There’s real weight here. Office vacancy declined in 49 of 92 major US markets in Q2 2026¹⁷, so the trend reads the same regardless of which tracker you trust. Morgan Stanley argues the next 12 to 24 months will be driven more by granular market- and asset-level dynamics than by broad macro or label-level calls¹⁸¹⁹²⁰. Deloitte’s property-type rankings for the next 12 to 18 months are stable, with no type moving more than four ranks year over year²¹. JPMorgan has multifamily and industrial strong, retail steady, and office recovering in some metros for 2026²²²³.

If the market is that granular, the objection goes, argue the asset in front of you and stop auditing trackers.

Granularity relocates the problem, it doesn’t dissolve it

But granularity is exactly what makes the source rule non-optional. The moment I drop from “office” to a specific building in a specific metro, I need a level. A covenant test and an occupancy floor take a precise number; a directional arrow is useless.

And the number is the thing three trackers can’t agree on. “Vacancy is falling in most markets” is true and useless at the point of underwriting a single asset, because it doesn’t tell me whether my floor is 14% or 20%. Direction survives the disagreement; the fast-no does not.

The problem runs deeper than the label

So the red-flag list changes shape. Cap-rate compression, occupancy floors, covenant strength stay as the screens – but each now runs against a number that carries the name of who measured it and the quarter it belongs to, and mismatched stamps don’t get compared at all.

One honest limit on my own thesis: I looked for a case where the same figure was used to justify both a yield play and an appreciation play, and I didn’t find one. What I found were the structural conditions for it – the office spread, the ten-point multifamily vacancy gap, the permit totals that read as 2.2% or 0.2% depending on the compiler. The ambiguity is sitting there, unfired.

The asset-class label was a distraction. The source was always the problem, and a finer category can’t fix a number three people measured three different ways.

This is observation from my own screening practice, not investment advice: no buy, sell or hold sits anywhere in it, and every figure above carries the name of who measured it so you can weigh the source yourself.