SBA Default Economics: Loss Rates by Industry, and Why the Headline Number Misleads

Ask how often SBA loans default and you will get two true answers that differ by a factor of three. Across roughly 2.1 million loans guaranteed since 1992, about 12.4 percent were charged off when counted one loan at a time, but only about 4.0 percent of the dollars were ultimately lost. Both numbers are correct. They describe the same portfolio. The gap between them is not a rounding difference or a quarrel between sources; it is the single most important fact in SBA credit, and it is the one most trade coverage gets wrong. This note is a reference for reading SBA default and loss rates the way an underwriter should: by industry, by loan size, by program, and always with the measurement basis stated out loud.

Two true numbers, three times apart

Start with the program-wide picture. Mazeka Labs, computing directly on the SBA's loan-level Freedom of Information Act file as of December 31, 2025, counts 2,136,870 loans (7(a) and 504 combined) approved since fiscal 1992, of which 228,228 have been charged off. That is a 12.4 percent count-based rate. Measured against dollars, the same file yields a 4.0 percent loss rate: $28.8 billion charged off against $727.0 billion approved. The average charged-off loan lost about $126,000; the average loan is about $340,000; and the average SBA guarantee is about 67 percent. The portfolio's status distribution is itself instructive: 58.9 percent paid in full, 16.6 percent active, 12.8 percent cancelled, 10.7 percent charged off, and 1.1 percent committed.

The count and dollar figures diverge for one structural reason, and it is the same reason that governs almost everything that follows: larger loans default far less often than smaller ones. Because a charged-off loan is on average smaller than a performing loan, counting defaults one at a time overweights the small loans that fail most, while weighting by dollars pulls the rate down toward the behavior of the large, better-collateralized credits that dominate the balance outstanding. Neither view is wrong. They answer different questions. Count-based rates describe how often a loan of a given type goes bad; dollar-weighted rates describe how much capital is actually lost. A lender sizing reserves cares about the second; a borrower or broker asking "will my loan survive" is closer to the first.

A third framing matters because it is the one that produces the scariest-looking headline. If you restrict the denominator to resolved loans only, that is, loans that have either paid in full or charged off, and exclude everything still active, cancelled, or exempt, the rate rises. PeerSense, computing on that resolved basis across more than 1,283,073 resolved 7(a) loans matched to 899 lenders, reports an average default rate of about 15.8 percent. That is not a contradiction of the 12.4 percent figure; it is arithmetic. Stripping still-performing loans out of the base necessarily lifts the rate. Feasibility-grade writing presents all three framings, count-based on disbursed loans, resolved-basis, and dollar-weighted, and labels each one. A number without a stated basis is not an answer.

Industry is the widest axis of risk

Once the measurement basis is fixed, the question becomes what actually moves the rate, and the answer is unambiguous: industry, identified by NAICS code, is the dominant axis. In the most citable public NAICS-level table, sbalenders.com's analysis of the 7(a) FOIA file for fiscal 1995 through 2024, the dollar charge-off rate ranges from about 1.6 percent for Veterinary Services to 37.4 percent for Shellfish Fishing. That spread, from under 2 percent to over 37 percent, is far wider than anything the state or the program dimension produces. If you know only one attribute of an SBA loan and want to predict its loss, know the industry.

The table below carries the categories a commercial-real-estate lender or feasibility analyst is most likely to underwrite, shown on both bases so the divergence is visible. Note how the count rate runs well above the dollar rate for the real-estate-secured categories: hotels charge off 4.6 percent of loans but only 1.8 percent of dollars, because the loans that fail are the smaller ones.

The high end of the same table, outside the CRE categories, is populated by concept-driven retail and specialty trades: Commercial Lithographic Printing at 11.2 percent, Gift, Novelty, and Souvenir Stores at 10.6 percent, All Other Specialty Food Stores at 9.5 percent, Drycleaning and Laundry Services (except coin-operated) at 9.1 percent, Residential Remodelers at 8.6 percent, New Single-Family Housing Construction at 8.0 percent, Snack and Nonalcoholic Beverage Bars and Sporting Goods Stores both at 7.4 percent, Furniture Stores at 7.2 percent, and Offices of Chiropractors at 6.9 percent. The middle band holds much of the rest of the CRE roster: Fitness and Recreational Sports Centers at 5.4 percent ($8.2 billion lent), Drinking Places at 5.1 percent, Convenience Stores at 4.5 percent, Beer, Wine, and Liquor Stores at 3.2 percent, Home Health Care Services at 3.0 percent, Offices of Physicians at 2.8 percent, and Pharmacies and Drug Stores at 2.7 percent. Professional and medical uses cluster low: Offices of Dentists at 1.9 percent ($16.4 billion lent) sit just above hotels and veterinary services at the floor.

Why the biggest lending category loses the least

The most counterintuitive line in the table is the one a feasibility firm should internalize first. Hotels and motels are the single largest SBA lending category by dollars, $44.2 billion across 25,447 loans, and they carry one of the lowest dollar charge-off rates in the entire file, 1.8 percent. That is not because hotels are a safe business. It is because SBA hotel loans are large, real-estate-secured credits, and larger loans default at dramatically lower rates than small ones. Loan size is the second axis of SBA default risk, and within any given industry it is nearly as powerful as the industry choice itself.

Those figures, from Mazeka Labs' industry-by-loan-size heatmap on a count basis, hold within every sector: small loans fail many times more often than large loans of the same type. The practical consequence is that any honest by-industry table needs loan size as a column, or at least as a caveat, because an industry-average rate blends a high-risk small-loan tail with a low-risk large-loan core. A borrower financing a $6 million hotel and a borrower financing a $120,000 franchise unit in the same NAICS code face very different odds, and the blended industry rate flatters the first while frightening the second. This is precisely why the dollar-weighted view, which the large loans dominate, sits so far below the count view.

Franchise structure is a related, softer modifier. PeerSense and the reinstated SBA Franchise Directory both indicate that franchised concepts generally default less than independents within the same NAICS code, though the magnitude varies by brand and only about 40 percent of brands disclose Item 19 financial-performance representations, which limits how far the generalization can be pushed.

What the sectors say

Zooming out from four-digit NAICS codes to two-digit sectors confirms the ordering. Wilary Winn's April 2024 white paper, computed on the 7(a) FOIA file as of December 31, 2023 (1,786,705 loans, $517.5 billion in gross approvals), measures cumulative gross charge-off as a percent of gross approvals by sector. The three best performers are Management of Companies and Enterprises (below 1.5 percent cumulative), Utilities (2.42 percent at year-end 2023), and Finance and Insurance (declining from a 5.55 percent peak in 2011). The three worst are Retail Trade (which peaked at 6.37 percent in 2016), Construction (which peaked at 6.87 percent in 2015 and stood at 5.10 percent in 2023), and Information (5.68 percent in 2012, 4.73 percent in 2023). Construction and retail sit at the top of the risk table at every level of aggregation, which is consistent with the four-digit result that remodelers and specialty retail concepts default heavily.

Directional single-number ranges circulate too, and they are useful only if labeled as such. Crestmont Capital, for instance, offers directional industry guidance placing restaurants at 12 to 15 percent and healthcare under 1 percent, explicitly as guidance rather than measured rates. Headline single figures also travel widely through consumer-finance coverage such as NerdWallet and LendingTree. All of these can orient a first conversation, but none substitutes for a rate computed on a stated basis over a defined cohort, which is the standard this reference is built to.

State rankings, and why they move

Geography matters less than industry, and it matters in a way that is easy to misread. On a lifetime count basis, Mazeka Labs (FOIA as of December 31, 2025) ranks the highest-default states as Florida, Nevada, Georgia, Alabama, and a cluster of Southern and lower-Midwest states, and the lowest as a set of sparsely populated Mountain and Northern New England states.

The ranking is not stable across measurement choices, and that instability is the point. A separate GoSBA analysis of fiscal 2020 through 2025 7(a) FOIA data, measuring charge-offs on recent cohorts only and on a dollar basis, produces a nearly different list: Hawaii and New York highest at 2.4 percent and Nevada at 2.0 percent, with Alaska (0.2 percent), Puerto Rico (0.3 percent), and Maine (0.4 percent) lowest. Florida, the leader on the lifetime count list, does not top the recent-cohort dollar list at all. The two rankings disagree because one is lifetime and count-based while the other is recent-cohort and dollar-based, which is exactly the methodological lesson: state rankings shift with the measurement window and basis, and should never be quoted without both. Where state concentration is real, it is largely a proxy for industry mix, hospitality- and tourism-heavy states rank worse, and for cost-of-living pressure on operating margins, not evidence that a state line itself causes default.

504 versus 7(a): a structural difference in risk

The program dimension is narrower than industry but structurally the clearest. The 504 program defaults and loses materially less than 7(a), and the reasons are built into the product rather than into the borrower pool alone. A 504 loan is real-estate-collateralized, owner-occupied, fixed-rate CDC-debenture financing; a 7(a) loan is frequently working capital, acquisition goodwill, or equipment, secured by weaker collateral. The structural comparison, and its realized loss consequences, is summarized below.

Each row does work. The 504 project's roughly 50 percent bank first mortgage, 40 percent CDC debenture (100 percent SBA-guaranteed), and 10 percent borrower equity align incentives and put real estate under the loan; the mandatory owner-occupancy test (51 percent or more of the financed property) does the same. The fixed-rate debenture insulated 504 borrowers from the post-2022 rate shock that variable-rate 7(a) borrowers absorbed directly. And 504 borrowers skew more established, with tangible net worth under $20 million and average net income under $6.5 million. Industry commentary, not the official UPB-rate tables, puts 504 net charge-off near 0.8 to 1.5 percent against 7(a) near 2 to 4 percent in normal conditions, and Crestmont estimates 504 recovery at 50 to 70 percent versus 30 to 50 percent for 7(a), consistent with the collateral difference.

Scale frames the comparison. Per the SBA's fiscal 2025 activity report and its September 30, 2025 press release, the agency guaranteed 84,400 combined 7(a) and 504 loans for $44.8 billion in fiscal 2025, of which 6,750 were 504 loans for $7.8 billion and roughly 77,600 (some industry counts, such as Coleman and EBIT Community, cite 78,078) were 7(a) loans for about $37 billion; combined dollar volume of about $45.1 billion set a program record. Fiscal 2024's comparatives were 70,242 7(a) loans for $31.1 billion. The average 7(a) loan runs about $480,000 and the average 504 loan more than $1.2 million, a size difference that, by itself, predicts part of 504's lower loss.

Frequency versus severity: recovery and who bears the loss

A charge-off is not a total loss, and a default is not a charge-off. The 4.0 percent dollar rate is best understood as probability of default times loss given default times exposure, blended across the portfolio, and the machinery between default and final loss is where the guarantee structure does its work. On a 7(a) default, once the borrower is 60 or more days into an uncured default and business personal property has been liquidated, the lender may demand the SBA honor its guarantee. The SBA then pays the guaranteed portion, 75 percent for loans over $150,000, 85 percent for loans of $150,000 or less, and 90 percent for export loans and the new Made in America manufacturing guarantee. Charge-off is a separate, later accounting action; the remaining balance is referred to Treasury for cross-servicing and offset, including wage garnishment and tax-refund offset, and post-charge-off recovery continues for years. Because the FOIA file reports gross charge-off, not net of recovery, it overstates ultimate loss.

The distribution of that loss is the part lenders and CFOs actually price. The lender retains the unguaranteed 15 to 25 percent of the loan and bears first loss on that slice; the SBA, funded by borrower and lender guarantee fees under the program's zero-subsidy mandate, bears the guaranteed portion. About half of guaranteed portions are sold into the secondary market via SBA Form 1086, which converts the SBA's conditional guarantee to the lender into an unconditional guarantee to the investor, while the lender keeps servicing and the unguaranteed piece. Borrowers still owe the deficiency after the SBA pays the lender, though an Offer in Compromise can settle it for less.

The zero-subsidy design is worth stating plainly, because it is the backdrop to the policy story below. For fiscal 2026 the estimated credit-subsidy rate for 7(a), 504, and SBIC is 0 percent, by statutory design, against 18.75 percent for disaster loans and 9.37 percent for microloan intermediaries. A negative subsidy means the program is built to generate budgetary net income rather than to cost taxpayers, which is exactly why the fiscal 2024 shortfall registered as an alarm.

Cohorts, and why recent vintages look safe

Any point-in-time default rate hides a vintage story, and reading SBA data without the seasoning curve is the fastest way to reach a wrong conclusion. Grouping loans by approval cohort makes the pattern obvious.

The 1.5 percent figure for 2020-and-later loans is not evidence of better lending. It is right-censoring: those loans are simply too young to have defaulted yet, because peak default occurs in years two through five, with a median time to default of 4.1 years. Never present a recent-cohort rate without that caveat. The seasoned cohorts show what a full loss curve looks like: the 2000 to 2007 vintage, which ran headlong into the financial crisis, ultimately defaulted at 23.0 percent by count. Gross 7(a) charge-offs from that era peaked at $2.04 billion in 2010, dipped to $0.68 billion in 2013, then spiked again to $2.06 billion in 2015 as lagged distress worked through, and some 2007 and 2008 origination segments showed 15 to 20 percent default by 2012.

Two more distortions sit in the recent data. The COVID era masked distress from both directions: Section 1112 of the CARES Act had the SBA pay principal, interest, and fees for six months (later extended, capped at $9,000 a month) on existing 7(a), 504, and microloans, funded by $17 billion plus $3.5 billion in appropriations, while separate, forgivable PPP lending and automatic deferments further suppressed 2020 and 2021 default signals. Then the rate environment reversed: prime peaked at 8.00 percent in September 2024 before easing to 6.75 percent by December 2025, and variable-rate 7(a) borrowers saw debt service climb sharply while fixed-rate 504 borrowers were insulated. Any current read of the file has to net these effects out rather than take the raw recent rates at face value.

The FY2024 reset and the July 2026 decoupling

The policy environment as of July 2026 is the most consequential in five years, and it bears directly on which cohorts a lender should watch. After a stretch of loosened underwriting and waived fees, the 7(a) program deteriorated visibly in fiscal 2024. Per the SBA's fiscal 2024 Annual 7(a) Risk Analysis Report (cited by Tax Guard in August 2025), the 7(a) default rate rose to 3.7 percent in 2024, the highest since 2012, with early-default outcomes at their worst since March 2020. American Banker reports SBA guaranty purchases of defaulted loans climbing from $733 million in fiscal 2022 to $1.1 billion in fiscal 2023 and $1.6 billion in fiscal 2024, the highest since the pandemic. The program posted about $397 million in negative cash flow in fiscal 2024, described by the SBA's March 27, 2025 press release as its first negative cash flow in over a decade and by its April 22, 2025 release as the first in 13 years, and the agency projected that more than $460 million in upfront lender fees went uncollected from 2022 to 2024.

The causal narrative the SBA attaches to that deterioration, that a "Do What You Do" posture loosened underwriting and waived fees, originates substantially in agency communications under the current administration and is politically framed; the $397 million deficit and the 3.7 percent default rate are presented here as reported, while the causal claim is attributed rather than asserted. One program within the portfolio ran hotter than the rest: Community Advantage exhibited a default rate above 7 percent, more than double the portfolio average, with several participating lenders generating early problem-loan rates above 30 percent, which prompted a May 2025 moratorium on new Community Advantage lenders.

Two structural responses frame the vintages to come. First, SOP 50 10 8, effective June 1, 2025, re-tightened underwriting, so post-June-2025 cohorts should, if history holds, perform better than the 2023 and 2024 vintages that are now the ones to watch. Second, and pointing the other way on exposure, the July 4, 2026 decoupling (SBA News Release 26-52 and Policy Notice 5000-879058, announced May 18, 2026 by Administrator Kelly Loeffler) raised the combined 7(a)-plus-504 ceiling to $10 million, allowing up to $5 million in 7(a) plus up to $5 million in 504, the highest cap in agency history. Shifts of this kind have historically flipped cohort default performance, which is the strongest argument for tracking new vintages against the seasoning curve rather than trusting a portfolio-wide average.

How to read an SBA default rate correctly

The underlying evidence is public and computable, which is what makes disciplined reading possible and sloppy citation inexcusable. The SBA's Office of Capital Access publishes the 7(a) and 504 loan-level FOIA files at data.sba.gov, four 7(a) files by decade and two 504 files by 20-year blocks, updated quarterly and accompanied by a data dictionary, covering Credit Reform Act loans (those approved on or after October 1, 1991) forward, some 2.1 million records in all. The fields that matter for default analysis include the program, gross and SBA-guaranteed approval amounts, approval and first-disbursement dates, term, NAICS code and description, franchise code, borrower and project state and county, business type, loan status, charge-off date, and gross charge-off amount. As of July 2026 the SBA began, for the first time, disclosing delinquencies, guaranty purchases, and liquidations at the individual 7(a) loan level (per the Coleman Report), a material expansion of what can be computed directly.

Correct computation means stating, every time, whether a rate is count-based on disbursed loans, count-based on resolved loans only, or dollar-weighted, and whether it is a cross-sectional snapshot or a cohort tracked by approval fiscal year to maturity. Cohort-based, dollar-weighted rates on fully seasoned vintages are the gold standard for feasibility work. The known limitations should be disclosed rather than hidden: recent cohorts under-count eventual defaults through right-censoring; roughly 202,825 records lack a NAICS code and about 236,519 lack a NAICS description in one widely used extract, with older records worse; interest rate and first-disbursement date are missing on large shares of the file; the choice to include or exclude cancelled and committed loans moves the denominator; and gross charge-off overstates ultimate loss because it is not net of recovery.

One authoritative gap remains, and honesty about it is part of the credibility. The SBA's official "Loan Program Performance" report, 11 tables updated quarterly, most recently as of June 30, 2025 and published September 15, 2025, contains the exact charge-off, purchase, and recovery rates as a percent of unpaid principal balance by program. Those are the definitive official figures, but they are distributed as robots-blocked PDFs and a ZIP archive that automated tools cannot read, so program-level rates in circulation, including several cited here, rely on third-party FOIA computations and industry commentary until the official tables are transcribed directly. And the third-party totals differ by scope, not error: Mazeka counts 2,136,870 loans (7(a) and 504, fiscal 1992 forward); Wilary Winn counts 1,786,705 (7(a) only, through year-end 2023); PeerSense matches about 2.1 million loans to 899 lenders. Always cite the as-of date and the scope.

What this means for underwriting

For a lender or a feasibility analyst, the reference collapses to a short discipline. Lead every rate with its basis, count, resolved, or dollar, and its window, because the same portfolio supports a 4 percent story and a 16 percent story and the difference is entirely in the framing. Underwrite the industry first and the loan size second, since those two axes span the full range from under 2 percent to over 37 percent, far more than geography or program. Treat 504 as structurally safer than 7(a) for the concrete reasons, real estate, owner-occupancy, fixed rate, larger loans, rather than as a matter of taste. And read every rate against the cohort clock and the current policy inflection: the 2023 and 2024 vintages carry the fiscal 2024 stress, the post-SOP-50-10-8 cohorts should improve, and the July 2026 decoupling has pushed the ceiling to a record $10 million, which will reshape the exposure of the vintages now being written.

None of this is exotic. It is the difference between quoting a number and understanding one. A study that states its basis, names its source and as-of date, underwrites industry and size before geography, and reads the cohort in the context of the policy moment has said something a credit officer can rely on. A single figure without any of that, the 12 percent that is really 4 percent, or the 4 percent that is really 16 percent, has said almost nothing at all.