CRE Debt REITs: Office Exposure and Non-Accruals

The Cushion Is Somebody Else’s Equity

A commercial mortgage REIT does not own buildings. It lends against them, and it sits above the owner in the queue. When an office tower loses a third of its value, the owner’s stake disappears first and the lender is still whole. That ordering is the whole business, and it is why a CRE lender is read on different numbers from a landlord.

Starwood Property Trust and Blackstone Mortgage Trust are the two listed reference points. Both write senior loans against offices, flats, hotels and warehouses; both earn a spread over their funding cost; both take the loss when a borrower fails and the collateral is not worth the debt.

The cushion also explains why these companies are financed so differently from the agency mortgage REITs. An agency book holds government-guaranteed bonds that a lender will fund at a small haircut and sell in an afternoon, so it runs at seven or eight times equity. Nobody funds a portfolio of office loans on those terms, because there is no guarantee and no same-day market for the collateral. CRE lenders therefore run a fraction of that leverage. The leverage definitions guide works through why the two numbers cannot be ranked against each other.

Loan-to-Value, Then and Now

Loan-to-value is the loan divided by the building’s appraised value. At origination it measures the cushion. After that it measures nothing, unless someone re-appraises the building.

Take a $70 million senior loan against a tower appraised at $100 million: a 70% LTV, with $30 million of borrower equity underneath it.

Two things fall out of that table. The lender is unharmed through a fall that wipes the owner out completely, which is why a 70% LTV book can look serene while its borrowers are in real distress. And past the cushion, losses arrive fast: the last two rows are a 10-point move in the value of the building and a 15-point move in the loss on the loan.

The second column is the one that ages. A 2021 appraisal on a half-empty office building dates fast, and the loan written against it can be reported at a 65% LTV that has not been true for years. Blackstone Mortgage carries the office property it has repossessed at around half the value those buildings were appraised at when the loans were made. That is the scale of re-marking involved. So when a filing gives you a weighted average LTV, find out whether it is the LTV at origination or against a current valuation. They are different measurements wearing the same label.

Office: Not the Story It Was in 2023

Office is where the losses of this cycle came from, and the reflex is to assume conditions are still worsening. They are not. US office vacancy was 18.3% in the second quarter of 2026, 30 basis points lower on the quarter, with nine straight quarters of more space being taken than given back behind it (CBRE). Demand is growing again, from a far lower base than 2019.

What has not healed is the split underneath the average. Prime buildings ran 12.3% vacant against 18.3% for the market. Tenants that stayed traded up: less space, better building. Older stock in weaker locations did not get that call, and for much of it there is no route back to full occupancy, because the rent it can win will not pay for the fit-out a tenant demands before moving in.

That second half is precisely the lender’s exposure, and it is a sharper problem for a lender than for a landlord. A landlord with a building that cannot fund its own fit-out owns a bad asset. A lender against that building owns a loan the borrower cannot refinance, because no new lender will write against a value that only exists on paper. The office question for a CRE debt REIT is therefore “can my borrower fund the leasing capital, and if not, who takes the building”. For how the two property types diverge on that capital bill, see industrial versus office REITs, and for how the collateral itself reprices, cap rates by property type.

Office Exposure: Always Ask What It Is a Percentage Of

Office percentages in earnings decks look like they can be ranked. They cannot, unless the base is the same.

Starwood reports US office at 8% of its diversified asset base, its lowest ever share, measured across $30.7 billion of undepreciated assets at the end of 2025. That denominator is the whole company: commercial lending, infrastructure lending, property it owns outright, servicing. A lender that instead quotes office as a share of net loan exposure is dividing by a much smaller number, so the same dollars of office produce a much larger percentage.

Two rules follow. Never compare two office percentages without reading the footnote that defines each denominator. And where you can, convert to dollars, because a dollar of office loan is comparable across filers and a percentage is not.

Three Words That Decide What You Are Reading

Almost every credit disclosure at a CRE lender runs through these.

Non-accrual. The lender has stopped recognising interest income on the loan, because it no longer expects to collect it. The call comes before any loss is realised. That makes it the most useful early-warning line a lender publishes. It is also the line management controls, so a book with no non-accruals and a lot of extensions is telling you something.

Modification and extension. When a loan matures and the borrower cannot repay or refinance, the lender can extend the term, cut the rate, or take a partial paydown in exchange for more time. The loan stays performing on paper throughout. Sometimes that is the right commercial answer, because a building part-way through a leasing programme is worth more finished than sold. Sometimes it is a loss deferred. A high performing percentage is only reassuring once you know how much of the book was extended to keep it there.

CECL reserve. Short for current expected credit losses, the accounting rule that makes a lender estimate the losses it expects over the life of its loans and deduct them from book value now, before anything goes wrong. It is an estimate the lender makes about itself, and it can be released back into earnings if the loans recover. Blackstone Mortgage released $33 million net in 2025 on resolutions that came in above carrying value.

The consequence is that reserves do not read the way instinct suggests. A large reserve says the lender has already put the expected pain into book value. A small one can mean a clean portfolio, or a management that has not marked yet.

The Two Books, Side by Side

Read the problem-asset row on its own and Starwood looks like the sick one: about a billion dollars on non-accrual against under $90 million of impaired loans at Blackstone Mortgage. That reading is wrong, and it is the trap this table exists to spring.

Charging a loan off removes it from the impaired balance. Blackstone Mortgage took $434 million of reserve charge-offs in the fourth quarter of 2025 alone and resolved roughly $600 million of impaired loans in the same three months. The small balance at the year end is what remains after the losses have run through the book, and its full-year distributable earnings say so plainly. Starwood’s larger non-accrual stack is stress that has been identified and reserved against but not yet crystallised.

Which is the safer book is a real question. The point is that the problem-asset line alone cannot answer it, and the reserve row is what makes the two comparable: both companies are carrying reserves worth roughly a tenth of book.

Distributable Earnings and the Dividend

CRE lenders declare dividends against distributable earnings, a cash-basis measure that strips out depreciation and unrealised marks but does include losses once they are realised. It is the closest thing this part of the sector has to a payout test.

Starwood paid 23 cents a share more than it earned, a shortfall it can fund for a while and not indefinitely. Blackstone Mortgage’s two rows are the same year measured either side of the credit losses, and the gap between them is $3.29 a share. On roughly 170 million shares that is over half a billion dollars of loans written off in one year, against a book value of $20.75.

That is the question a CRE lender asks its shareholders. Was the charge-off year a clear-out, in which case the $1.86 is the earnings power and the dividend is roughly covered, or is it the run rate, in which case the dividend is being paid out of book value.

Worked Mini-Example: What a Discount Would Be Saying

Take an illustrative CRE lender with $20.00 of book value per share, $1.75 a share of CECL reserves already inside that book, and a share price of $17.80, about 0.89× book.

A discount to book at a lender is an estimate: the market’s view of how much further book value has to fall. So test it rather than accept it.

1. How much of the problem is already reserved? The 11% discount is $2.20 a share, against $1.75 a share already reserved. So the price implies roughly that much again, on top of what management has taken.
2. What would have to be true for the price to be right? Another year of charge-offs on the scale of a bad one, or office collateral marking down further than the reserves assume.
3. What would have to be true for it to be wrong? Distributable earnings returning to their pre-charge-off level, with the reserve released rather than consumed.

The reserve row in the table above is what to run this test against for a real lender: $1.76 a share inside Blackstone Mortgage’s $20.75 book, and $1.84 inside Starwood’s $19.25 undepreciated book.

None of those is settled by the multiple. A cheap-looking book at a lender whose reserves are too small is expensive; an expensive-looking book at a lender that has already marked everything down can be the better buy. Compare the agency names in book value per share, whose book is read on economic return, and the difference in what a multiple means becomes obvious.

Screening CRE Debt REITs

1. Loan-to-value, and whether it is measured at origination or against a current appraisal. Only the second one tells you where the cushion is today.
2. Office in dollars, not percentages, until you have read what each filer divides by.
3. Non-accruals and foreclosed property against the reserves held on them. The pair is informative; either alone is not.
4. Maturities and extensions. How much of the book comes due in the next two years, and how much of what is performing today got there by being extended.
5. Distributable earnings against the dividend, on the same basis in each year, so a charge-off year stays visible.
6. The discount to book as a hypothesis about future marks, tested against the reserves already taken.

For where CRE lending sits in the wider mortgage REIT sector, see agency versus hybrid mortgage REITs. Prepayment speeds and agency leverage ratios have no application here.

What Would Signal the Turn

The credit cycle at a CRE lender ends quietly, and it is visible in the filings before it is visible in the earnings. Watch for repayments at par on loans that had been extended, because a borrower who can refinance has found a lender willing to write against a current valuation. Watch for reserve releases funded by resolutions above carrying value rather than by reclassification. And watch the foreclosed property balance fall through sales rather than through further write-downs. Those three, together, are what a clear-out looks like from the outside. Distributable earnings recovering without them is a lull.