Where are we in the CRE credit cycle?
One series, thirty-five years: the share of U.S. commercial real estate loans that banks report as noncurrent, dollar-weighted across every filer above $100M, straight from FDIC call reports. It covers three CRE downturns — the early-90s CRE and S&L crisis, the global financial crisis, and today — so a reading now has something real to be measured against, rather than a single resi-led episode.
As of 2026-03, the system noncurrent-CRE rate sits at 1.19% — the 58th percentile of the 141-quarter range since 1991, roughly the stable part of the cycle; year-over-year it is -0.01 pts, the climb has stalled rather than reversed. For scale, across three CRE downturns: the series peak was 7.61% (1991-06, the early-90s CRE crisis — worse than the GFC), the last trough 0.42% (2019-12). By collateral, nonresidential 1.29%, multifamily 1.07%, construction 0.92%. Spatially, the established-distress markets are rolling over from highs (New York) while a newer cohort is climbing (Rhode Island, Arizona, Utah) — the rotation, corroborated independently in bank call-report data. This is where the cycle is, calibrated against its own history — not a forecast of where it goes next.
The gap is the point
The grey bands are recessions, dated by the NBER — a fixed, third-party history we did not draw. The orange rails are something else: the stretches when commercial real estate credit itself was actually under stress, worked out from this series rather than painted on. For each era we take the peak within the recession and the three years after it, and mark the contiguous quarters that held above the midpoint between that peak and the calm before it.
Put them on the same axis and they do not line up. S&L — recession 1990-07 to 1991-03, CRE stressed 1991-03 to 1994-03.GFC — recession 2007-12 to 2009-06, CRE stressed 2009-03 to 2012-06.COVID — recession 2020-03 to 2020-06, CRE stressed 2020-06 to 2021-09. The recession is the shock; the rail is how long the buildings took to feel it and stop feeling it. That lag is why a headline saying the economy recovered tells you very little about whether a loan book has.
The S&L rail rests on an assumed baseline, not a measured one. This series begins in 1991 with the crisis already underway, so there is no pre-crisis calm in the record to measure against and we substitute a normal noncurrent level of 1%. The other rails take their baseline from the data. Read the S&L span as indicative; the two later ones are derived end to end.
The same quarter, by loan type
One headline rate hides three different books. Construction, multifamily and non-residential CRE do not turn together, and they have never had to.
Does the reading mean anything?
The honest way to ask that is to walk the clock forward from every quarter since 1991, calibrating it only on the data available at the time — no lookahead — and then check whether the rate actually moved the way the reading implied. Here is every cell, including the ones that argue against us.
| Reading | Horizon | Moved as implied | Quarters tested |
|---|---|---|---|
| Falling | next 4 quarters | 84% | 93 |
| Falling | next 8 quarters | 72% | 93 |
| Rising | next 4 quarters | 67% | 36 |
| Rising | next 8 quarters | 53% | 32 |
Read this the careful way: the percentage is the share of tested quarters in which the rate moved in the implied direction, over that many quarters of history. It is not a forecast accuracy, and the falling signal is both stronger and far better sampled than the rising one — which is exactly what you would expect, since recoveries are long and slow and turns are short and rare.
What this does not do
It does not forecast. It states where the cycle is and shows the series behind the statement. A percentile is a position, not a prediction, and nothing here says what the next quarter does.
It is the climate, not the weather. This is an aggregate read on the whole banking system. It cannot tell you anything about a particular building, a particular borrower or a particular loan, and a benign national reading is entirely compatible with severe distress in one market or one property type.
It is banks only. Loans in CMBS trusts and CRE CLOs are not in a call report. When banks offload CRE exposure, this series can read healthier without a single loan improving.
We also publish a state-by-state cut of this series internally and have deliberately kept it off this page. The FDIC assigns a bank's loans to its charter state rather than the property's location, so a large national lender can make a small state look like it is in trouble, and a thin state book can spike on a single loan. That cut is readable with the right guards attached; it is misleading without them, and a public map invites being screenshotted without its footnotes.
Where this comes from
Every U.S. bank files a quarterly call report with the FDIC, and has since long before anyone was worried about this cycle. The noncurrent CRE share is in it. We dollar-weight it across filers above $100M in assets, carry it back to 1991, and calibrate today against the full record rather than against living memory.
Nothing here is licensed, scraped or proprietary. The series is statutorily open and you can rebuild it yourself — the work is in making thirty-five years of it comparable.