Special-Purpose Property Risk: What 1.28 Million SBA Loans Reveal, Industry by Industry

Some collateral is worth nearly as much the day a loan works out as the day it closed. A warehouse, an office suite, a strip of general retail can be re-let or resold to the next tenant in line. A car wash cannot. Neither can a bowling center, a funeral home with a crematorium, or a select-service hotel built to a single flag’s prototype. These are special-purpose properties, and lenders have always treated them as a category apart — more equity, tighter structure, an independent feasibility study before the file advances. The instinct is sound. What has been missing is the proof, industry by industry, and it turns out the proof has been sitting in public the whole time: in the SBA’s own record of every loan it has guaranteed since 1991.

This note does two things with that record. It shows how to compute default risk by industry from the SBA’s loan-level data in a way that survives a credit committee, and it explains why the two analyses already in circulation reach almost opposite conclusions about the same properties. The reconciliation is not a footnote. It is the analytical heart of special-purpose risk, and it is the difference between a lender who believes hotels are dangerous because a headline said so and one who can say precisely which special-purpose exposures fail often, which fail expensively, and why those are not the same list.

What makes a property special-purpose, and why it changes the risk

Start with the definition, because it is doing more work than it appears to. SBA SOP 50 10 8, effective June 1, 2025 with a technical update effective March 1, 2026, defines a special-purpose property as “a limited-market property with a unique physical design, special construction materials, or a layout that restricts its utility to the specific use for which it was built.” The Appraisal Institute’s Dictionary of Real Estate Appraisal (6th ed.) reaches the same place from the valuation side, defining a special-purpose (or special-design) property as one “with a unique physical design, special construction materials, or a layout that particularly adapts its utility to the use for which it was built,” and a limited-market property as one “that has relatively few potential buyers.”

Those phrases — few potential buyers, utility restricted to one use — are the entire theory of the risk. Thin resale markets mean extended exposure time; the Appraisal Institute’s Guide Note 14 treats exposure time as a function of property type and market conditions, and a single-use asset sits at the long end of that distribution. Single-use design means high re-tenanting or conversion cost when the operator leaves. And much of the asset’s value is going-concern or business-enterprise value bound up in the operation rather than the real estate, so when the business fails the collateral is worth far less than it was at origination because it cannot be readily repurposed. This is why appraisers of these assets often lean on the cost approach rather than sales comparison — comparable sales are scarce by definition — and why lenders demand more equity, going-concern appraisals that separate land, building, FF&E, and intangibles, and an independent feasibility study.

SOP 50 10 8 reinstates the full example list of roughly twenty-five special-purpose types, and it reads like a tour of the operating-intensive economy: amusement parks, bowling alleys, car washes, cemeteries, cold-storage facilities, dormitories, farms, funeral homes with crematoriums, gas stations, golf courses, hospitals and surgical or urgent-care centers, hotels and motels, marinas, mines, nursing homes including assisted living, oil wells, quarries, railroads, sanitary landfills, oil-and-lube and brake service centers with in-ground pits and lifts, sports arenas, swimming pools, tennis clubs, theaters, and wineries. One omission is as instructive as the list itself: the SBA treats restaurants, daycares, and mini-storage as multipurpose, not special-purpose, on the theory that they turn over often enough in the marketplace to have a real resale market. Restaurants fail frequently, but in the SBA’s own taxonomy they are not special-purpose — a distinction worth holding onto, because it separates default frequency from collateral risk, and the data will show those are two different things.

One dataset can settle it

The evidence lives in the SBA’s loan-level FOIA files for the 7(a) and 504 programs, published on data.sba.gov under the dataset “7(a) & 504 FOIA” (slug 7-a-504-foia), updated quarterly and mirrored in the federal catalog at catalog.data.gov as “SBA 7(a) and 504 Loan Data Reports.” The files cover every loan approved since fiscal 1991. The 7(a) program is split into decade files (FY1991–1999, FY2000–2009, FY2010–2019, and FY2020 to present) and the 504 program into two roughly twenty-year files, each with a companion data dictionary.

Every record carries what an industry default analysis needs: borrower and bank identifiers, gross approval and SBA-guaranteed approval, approval and first-disbursement dates, term, delivery method, NAICS code and description, franchise code and name, project geography, business type, and the three fields that matter most here — loan status, charge-off date, and gross charge-off amount. NAICS is stored at the full six-digit level, so the data can be cut at the two-digit sector or the six-digit industry, which is where special-purpose analysis has to operate.

A charge-off is identified by the loan-status field, which takes values including PIF (paid in full), CHGOFF (charged off), CANCLD (cancelled), and EXEMPT. A defaulted loan is status CHGOFF, corroborated by a non-null charge-off date and amount. The defensible way to state a charge-off rate is CHGOFF divided by the sum of PIF and CHGOFF — the resolved-loan, or cohort-default, basis — which excludes loans still in repayment, cancelled, or exempt. Counting still-performing loans in the denominator understates the true rate, which is precisely the error a naive query makes.

The benchmark before the breakdown

A rate means nothing without the book average behind it. On the resolved-loan-count basis, the 7(a) program’s lifetime charge-off rate is about 15.8 percent. PeerSense, which works from the FOIA files, puts it at roughly 15.8 percent across 1,283,073 resolved loans: “the average default rate is about 15.8%, measured as charge offs divided by fully resolved loans.” That is the number against which any single industry should be read — not zero, but roughly one resolved loan in six.

Change the denominator and the picture changes entirely, which is the whole lesson of this piece in miniature. On a disbursed-dollar basis over recent vintages, Forvis Mazars found charge-offs of about 1.5 percent of gross approved dollars and 3.9 percent of loan count across 660,221 originations from FY2009 through the first quarter of 2022. And the 504 program, secured by real estate in a 50 percent first-lien position, charges off dramatically less: Equalize Capital, citing SBA loan-program-performance data, reports cumulative net charge-offs on 504 loans below 1.0 percent for every fiscal year since 2015 and at or below 0.2 percent since FY2018. Same agency, same borrowers, radically different numbers depending on how the question is framed. Hold that thought.

Same data, opposite answers

Two analyses of this dataset dominate the search results, and they appear to contradict each other on exactly the special-purpose types a lender cares about. The contradiction is real, and it is entirely methodological.

The first, published by sbalenders.com (Darren King, March 2025), ranks roughly fifty industries by dollar-weighted charge-off rate over 1995–2024. Its headline is a paradox: “Hotels (except Casino Hotels) and Motels” show a charge-off rate of just 1.8 percent — $784.4 million charged off against $44.2 billion lent, 1,175 of 25,447 loans — ranking 49th of 50, among the very lowest. Car washes come in at 3.2 percent ($274.9 million against $8.7 billion; 969 of 9,645 loans) and gasoline stations with convenience stores at 3.5 percent ($625.9 million against $17.7 billion; 2,106 of 21,149 loans). Full-service restaurants register 5.9 percent and limited-service restaurants 6.6 percent. The worst performers on this basis are not special-purpose at all: shellfish fishing at 37.4 percent and commercial lithographic printing at 11.2 percent.

The second, PeerSense (Ed Freeman, updated 2026), analyzes roughly 2.1 million records, about 1.28 million of them resolved, on a count basis — and produces almost the reverse ordering for the same properties. Its niche table shows gasoline stations and convenience stores at 14.9 percent, car washes at 13.9 percent, full-service restaurants at 9.9 percent, childcare at 11.9 percent, gyms and fitness at 17.1 percent, and long-haul trucking at 15.2 percent, against a safe end of veterinary practices at 4.1 percent and dental offices at 4.6 percent. As PeerSense puts it, “professional healthcare niches default below 5% (veterinary 4.1%, dental 4.6%), while gyms, long-haul trucking, and gas stations exceed 14% — roughly four times higher.”

Both are correct. They answer different questions. Dollar-weighting asks what fraction of dollars was written off; the count basis asks what fraction of loans failed. Hotels borrow in very large tickets that mostly pay in full, so their dollar-weighted charge-off is tiny even when individual failures are painful; capital-intensive consumer operations like gas stations and car washes take many smaller loans that fail more often, so they look dangerous on a count basis. The share of dollars written off and the frequency of failure are simply different lenses, and an analysis that reports one without the other is not so much wrong as half-told.

Reading both bases side by side

The cut that no one currently publishes cleanly — and the one a feasibility firm should — sets each special-purpose industry beside a standard-use benchmark and reports both bases at once, at the six-digit NAICS level. The table below assembles the published secondary figures for the industries where both bases are available; on a live engagement these are recomputed from the raw FOIA files rather than taken from either vendor.

The paired view also disciplines the vocabulary. “Special-purpose,” in the SBA’s own taxonomy, is not a synonym for “highest charge-off.” Hotels are special-purpose and default rarely, at least on a dollar basis; restaurants default far more often and are not special-purpose at all. The honest thesis is not that special-purpose assets fail more frequently — some do, some do not — but that when they fail, the recovery is worse, because the collateral is thin-market and single-use. Frequency is a count-basis question; severity is a collateral question. Special-purpose risk lives mostly in the second.

Why seasoning decides the answer

The single largest analytical trap in this dataset is time. Recent vintages have not had the years they need to default, so a fast-growing industry with a young loan book looks artificially safe next to a mature one. Glennon and Nigro established the shape of the problem two decades ago: in discrete-time hazard models of SBA 7(a) loans (2005, Journal of Money, Credit and Banking 37(5):923–947, and Journal of Financial Services Research 28:77–111), default risk is time-dependent — the hazard rises after origination, peaks around year two, then declines. Compare a mature industry’s fully seasoned book against a newer industry’s under-seasoned one and the comparison is biased before it begins.

The fix is vintage control: restrict the headline comparison to fully matured cohorts — approvals through roughly FY2015, read against maturity dates — so that every industry has had the same opportunity to fail. Two further confounders belong in the same discussion. Glennon and Nigro also documented that the SBA guarantee percentage is positively correlated with default probability, a moral-hazard effect; and DeYoung, Glennon and Nigro (2008, Journal of Financial Intermediation 17(1)) showed that borrower–lender distance and the adoption of credit scoring both move performance. Because hotels, gas stations, and car washes are heavily franchised, a franchise-versus-independent split using the franchise code and franchise name fields is not optional — franchising materially changes default rates, and lumping the two together muddies every one of these industries.

One more caveat governs everything above: the FOIA file gives gross charge-off, not net loss. The gross charge-off amount is the money that went bad; it is not netted of recoveries, so a true loss rate — dollars lost after recovery — is a different and lower number the file cannot fully produce. For special-purpose property, where recovery is exactly the weak point, that gap matters, and it should be named rather than papered over.

What the capital markets say independently

The SBA record does not stand alone. Three independent bodies of evidence point the same way, and a lender-facing note should stack them, because each closes a different escape route.

The first is CMBS distress by property type. Lodging and retail have consistently posted the highest delinquencies through stress. AG Mortgage Investment Trust’s FY2020 10-K (SEC) records the COVID shock plainly: “Hotels were most impacted, with delinquencies rising from 1.5% to 24% and ending the year at 20%... Retail was negatively affected, with delinquencies rising from 4.4% to as high as 18% and ending the year at 13%,” and, counting loans in special servicing or on watchlists, “approximately 70% of all securitized loans in that space showed some level of distress at their peak in 2020.” Trepp’s official figures put the lodging CMBS delinquency peak at 24.30 percent and retail at 18.07 percent, both in June 2020, against an overall CMBS peak of 10.32 percent. The pattern holds in calmer conditions: per Trepp via Multi-Housing News (Emily Yue, updated June 24, 2026), “the Trepp CMBS Delinquency Rate increased by 41 basis points to 7.55 percent in March 2026,” with office highest, lodging at 5.94 percent (below its April 2025 peak of 7.85 percent), and industrial at just 0.67 percent. Operating-intensive, special-use lodging carries structurally more distress than passive industrial in good times and bad.

The second is pricing. Special-purpose and operating-intensive assets trade at higher cap rates, which is the market quoting a higher required yield for higher perceived risk. CBRE’s H2 2025 U.S. Cap Rate Survey (via Clearhouse) has cap rates ranging “from roughly 5.0% for Class A industrial properties to 8.5% or higher for select-service hotels,” with 2025 apartment cap rates averaging 5.7 percent (Class A at 4.74 percent). Matthews’ 2025 Cap Rate Recap notes that car washes, even amid heightened investor demand, “often trade at slightly higher cap rates, reflecting their more specialized nature.” The capital markets are pricing exactly the collateral risk the appraisal theory predicts.

The third is business survival. Per the U.S. Bureau of Labor Statistics Business Employment Dynamics series, about 78.7 percent of new businesses survive their first year but only about half survive to year six, and accommodation-and-food and arts-and-entertainment establishments underperform that already sobering baseline — the operating-intensive sectors where special-purpose collateral concentrates. Default data, delinquency data, pricing data, and survival data converge on the same conclusion from four independent directions.

Where the number lands: the feasibility file

All of this ends up in one place in the underwriting file. SOP 50 10 8’s elevated requirements for special-purpose property are the system’s structural response to precisely the risk the SBA data quantify. A special-purpose designation triggers a higher equity injection — 15 percent under a 50/35/15 stack, or 20 percent under a 50/30/20 stack for a special-purpose startup — and, for going-concern situations, an appraisal by a Certified General appraiser experienced in that specific property type, with value allocated across land, building, FF&E, and intangibles. The SOP requires Certified Development Companies to address in their credit memorandum whether a project property is limited or special purpose. Per QuickRead’s analysis of SOP 50 10 8, special-purpose going-concern appraisers must have completed at least four going-concern appraisals of similar special-use businesses and comply with USPAP.

The independent feasibility study sits at the center of that response. The SOP lists feasibility studies among the independent reports that can mitigate identified credit weaknesses, and in practice a third-party study is most consistently expected for exactly these asset types. The logic is direct: higher equity cushions the lender against thin recovery, the experienced appraiser values the going concern honestly, and the feasibility study independently tests whether this specific business, at this scale and this debt load, can actually service the loan. Each control answers a different piece of the special-purpose problem, and the charge-off data are what justify having them.

What the record tells an underwriter

Read carefully, the SBA’s own book says something more useful than “special-purpose is risky.” It says that default frequency and default severity are different exposures that happen to overlap. Some special-purpose types — gas stations, car washes — fail often on a count basis and should be underwritten to that frequency. Others — hotels — fail rarely and cheaply on a dollar basis, yet still carry the collateral problem that makes a workout ugly when it comes. And some of the most failure-prone businesses in the entire book, the restaurants, are not special-purpose at all in the SBA’s taxonomy, which is a warning against using the label as a proxy for the risk instead of measuring the risk directly.

For a lender or investor, the operational takeaway is to insist on the cut that no headline provides: a six-digit, dual-basis, vintage-controlled comparison of the specific special-purpose industry against a standard-use benchmark, with the franchise split shown and the gross-versus-net charge-off caveat stated. That is the number that survives a credit committee, and it is the number a feasibility study is built to stand behind. The data have been public since 1991. The differentiation is in reading them honestly — both ways at once, seasoned, and tied back to the collateral that has to be sold if the business does not make it.