The Stories Holding Up Commercial Real Estate
Downtown Chicago has roughly a 24% office vacancy rate. The buildings holding that vacancy are still on bank balance sheets at values near pre-pandemic. The owners report cap rates from 2019. The auditors signed off. The credit-rating agencies rated the bonds. The pension funds hold them.
What is holding the buildings up is not the structure. The structure is fine. What is holding the buildings up is the story. Six people tell each other six things, and each story does specific work in keeping the number on the books and the loan on the balance sheet.
Coda's When the Middle Breaks mapped the cascade — five timers, $2.2 trillion in maturing CRE loans, the convergence window in late 2026 to mid-2027. The cascade has a script. This is the script.
I. We're returning to the office.
In January 2025, Jamie Dimon told JPMorgan's managing directors they would be back at their desks five days a week starting in March. The reasoning was given as productivity, culture, and mentorship. Amazon went the same way. Goldman did. Then a hundred imitators. The narrative settled: hybrid was a pandemic compromise; the office is coming back.
What the story does: justifies continued lease commitments for executives whose ego is in the building. Justifies the bank's willingness to carry the loan at par. Justifies the broker's "stabilizing" report. The story creates a future demand — the office will refill — that the math hasn't yet earned.
What the story doesn't say: occupancy data has stayed flat at 28-30% downtown vacancy in major markets a year after the mandates. The mandates are written; the seats are not filled. Workers come in two days a week, badge in for a third on Friday, and leave at noon. Several companies have quietly shifted from RTO mandates to presence expectations — a softer phrase that does not require the building to be 90% occupied to claim victory.
Inertia: the people who pushed hardest for return are the people who would lose face if remote won.
Pair: Suburban malls will come back. What JC Penney, Sears, and the REITs holding the anchor leases told themselves and each other in 2010-2015. The anchor stores leaving was supposed to be temporary. The mall's value was supposed to recover. Some did. Most didn't. The malls that survived were converted into something else.
II. Mark-to-market doesn't matter as long as we don't sell.
A regional bank holds a $200 million loan against an office building — pick any one of a thousand. The market value is now closer to $130 million. The bank doesn't have to recognize the $70 million loss until the loan defaults or it sells the asset. So the bank does not sell. The bank lets the loan sit at par on the balance sheet. The auditors accept it because there is no observable recent transaction to mark against.
What the story does: preserves tier-one capital ratios on paper. Keeps the bank above its regulatory minimum. Allows the dividend to continue. Keeps the stock from cratering on the next earnings call.
What the story doesn't say: the loan's value isn't determined by whether the bank acknowledges it. The borrower is going to default at refinancing. The story is buying time, not value.
Inertia: the moment any peer bank is forced to sell at market price, every bank's no observable transaction defense collapses simultaneously. Each bank's balance sheet depends on every other bank holding the same line.
Pair: There's no market for this paper, so we'll value it at our model price. What Bear Stearns, Lehman, and the holders of subprime CDOs told themselves in 2007. There was no market because there were no transactions because everyone was using model prices. The model prices held until one institution had to sell. Then they didn't.
III. AI will need offices anyway.
Both Anthropic and OpenAI have expanded substantial office footprints in San Francisco. The story formed quickly: AI labs need physical proximity. Researchers need to talk in hallways. The dense cluster of AI talent will refill the office buildings that the rest of the economy is leaving.
What the story does: gives the office sector a future tenant base that is plausible, prestigious, and well-funded. Justifies the brokerage report's tech demand returning slide. Justifies the lease at the rate it was originally underwritten.
What the story doesn't say: the AI sector is small. Anthropic and OpenAI together would not refill a single mid-size downtown. The largest AI companies have a few thousand employees combined. The mid-tier AI companies — the ones that took the next layer of leases — are exactly the cohort that Coda's Timer Four predicts gets squeezed by foundation-model commoditization in 2026-2027. The AI tenants are also at risk.
Inertia: the brokers who closed the AI leases are the same ones writing the demand reports.
Pair: The internet companies will need office space. True-ish in 1999, until the 2001 dot-com collapse left San Francisco's South of Market half-empty for years. The new tenant always justifies the building until the new tenant joins the cycle.
IV. Distressed debt is opportunity.
Private equity firms — Blackstone, Brookfield, Starwood, dozens of smaller ones — have raised tens of billions for distressed CRE opportunity funds. The story is that the cycle is already turning, that the bottom is near or here, that smart money buys when others are forced to sell.
What the story does: justifies the LP commitment. Keeps the dry powder deployed. Keeps the GP's fees flowing. Reassures the institutional investors whose pensions are riding on it.
What the story doesn't say: the cascade may not have a clear bottom. CRE distress in 2008-2010 looked like opportunity until 2011 looked worse than 2009. The vulture funds that bought too early lost their LPs more money than the funds that didn't deploy at all. There is a difference between the bottom and the floor; the bottom is wherever buyers stop being willing to take the next leg down, and that level is not knowable in advance.
Inertia: the fund needs to deploy. The fee structure rewards deployment. The LPs will replace the GP if the dry powder sits.
Pair: Subprime mortgages at 30 cents on the dollar are opportunity. What the early 2008 vulture funds told themselves and their LPs. They were wrong by another 50%. The bottom was not where the model said it was.
V. Class A is fine. Only B and C are in trouble.
The tier-flight story: the best buildings stay full while the marginal ones empty. Top-tier office space in major cities is supposedly insulated from the broader cycle because of the flight to quality — companies cutting space generally still want their HQ on Park Avenue or in the Salesforce Tower.
What the story does: lets owners of trophy assets sleep at night. Lets the major REITs continue their dividend narratives. Lets the analysts segment the market and say this part is fine.
What the story doesn't say: Class A vacancy is also rising. Slower, but rising. The flight to quality is partially real and partially a story that lets people generalize from one true case to a comfort that isn't general. The trophy buildings are losing tenants too — just to other trophy buildings — which means the absolute demand is shrinking, even when the relative tier holds.
Inertia: the people writing the tier-segmentation reports own the trophy buildings, or hold the loans against them, or manage the funds that hold the bonds.
Pair: Subprime is contained. What Ben Bernanke said in March 2007. He believed it. Most of the economics profession believed it. The reasoning was structurally similar: the bad cohort is segmented away from the good cohort; the contagion will not spread. The reasoning was wrong because the segmentation was a story, not a structure.
VI. The government won't let regional banks fail.
After the SVB collapse in 2023, the Treasury and FDIC stepped in to guarantee deposits beyond the $250,000 insurance cap. The intervention worked. Depositors were made whole. The narrative settled: the government will not let a major regional bank cause a panic.
What the story does: lets large depositors keep accounts at regional banks above FDIC limits. Lets bond holders price the credit as quasi-government. Lets the Treasury market function as if regional bank risk is sovereign-guaranteed.
What the story doesn't say: the SVB rescue saved depositors. Equity holders got wiped. So did subordinated debt holders. The government rescue was a deposit guarantee, not a balance sheet guarantee. The story conflates them. In a multi-bank cascade, the government's capacity to repeat the SVB rescue is not infinite — and the political cost of doing it for the third or fourth bank in a row gets worse, not better, as the public observes the pattern.
Inertia: every bank stock is priced as if SVB will be repeated. Every uninsured depositor is staying because they assume so. The pricing depends on the story holding.
Pair: Bear Stearns shareholders will be made whole. What Bear shareholders thought in March 2008. They were not. JPMorgan bought the company for $2 a share. The deposit insurance and the equity protection are not the same thing, and the market kept misreading which one applied.
The story that breaks first
The cascade does not need every story to break. It needs one to break and the others to follow.
The story that breaks first is not the most factually wrong. It is the most dependent on no one else changing their mind.
By that test, the most fragile story is we're returning to the office. Six fictions on this list, but five of them depend on professional-class behavior staying constant. The auditor doesn't change methodology. The PE LP doesn't pull capital. The Class A tenant doesn't downsize. The depositor doesn't move money.
The return-to-office story depends on a different cohort entirely — workers. Workers do not have to coordinate. Each one, individually, decides whether to come in on Tuesday. The aggregation is statistical, not strategic. If a critical mass of workers — even a quiet, untracked one — decides their effort is better spent on the work than on the commute, the occupancy data tells the truth before the executive memos do. The data is already telling it; only the framing is holding.
The other five fictions break later, and they break on the bank balance sheet. But the return-to-office story breaks earlier, and it breaks on the building itself.
When the first major employer makes hybrid permanent in a way that signals tier-acceptable — not Salesforce signaling weakness, but JPMorgan signaling pragmatism — the dam breaks. The office demand projection that all the other stories rest on collapses. The mark-to-market story can hold for another quarter, but only because the building no longer needs a tenant — it needs a buyer. The buyer needs a price. The price needs a market. The market needs comparable transactions.
The fictions are linked, and the worker is the first knot.
This is not a prediction with a date. The data is here. The math is here. The structure is here. What's missing is the moment one CEO at one company that everyone watches says let's stop pretending — and the rest of the room finally exhales.
— Linn April 2026