Small Business Default Rates by Industry: A 2026 Sector Benchmark

Executive Summary

Industry concentration limits are usually set against a sector default rate. This analysis provides those rates for nineteen NAICS sectors and finds three things that complicate how they should be used.

The rates come from SBA 7(a) performance history rather than from survey or aggregate data. That record is used here because it carries thirty years of loan-level outcomes with term, sector, lender, and result attached to each loan, a linkage that does not exist in public bank data. What it is not is a cross-section of all business credit, and the FAQ below says where that distinction matters.

Every sector deteriorated. Between 2019 and 2025, all nineteen sectors large enough to measure showed higher default rates and lower prepayment rates. There were no exceptions, which means no industry mix avoided the direction of travel.

Severity varied enormously. Default rates in 2025 range from 2.34% in Health Care to 6.69% in Transportation and Warehousing, and the decline in prepayment ranges from 0.4 points to 4.3. Industry mix therefore governs how hard a book was hit, not whether it was hit at all.

The third finding matters most for how these numbers are used. A sector default rate is a blend of two very different books. Within the same industry, loans written on terms of twenty-five years or longer default at a fraction of the sector's full-book rate, a small one in most, and that holds in all sixteen sectors where both can be measured. The margin is wide in most and narrow in Agriculture, where six defaults among 256 long-term loans separate 2.54% from 3.05%. Transportation defaults at 6.69% across its full book and 1.08% on its long-term loans. A concentration limit set on sector alone is set on the weaker of the two variables a concentration limit can be set on.

The Benchmark: 19 Sectors, 2019 Against 2025

Annual default rate is an exposure-adjusted annual rate: defaults are measured against loan-months at risk and annualized, so a loan booked in December is not treated as exposed for a full year. Prepayment is constructed the same way. Both are loan-weighted, meaning each loan counts once regardless of balance, and both are on a calendar basis. This runs somewhat above a simple defaults-over-loans share, which is why a figure drawn from another source may not match. Our guide to how Lumos measures default and prepayment sets out each construction and reconciles them. Sectors are two-digit NAICS groupings, restricted to those carrying 200 or more loans in both comparison years.

Rates are calculated from unrounded figures and displayed to two decimals, so the change column can differ by 0.01 from subtracting the two printed cells.

Utilities, Mining, and Management of Companies carry fewer than 600 loans each, so year-over-year movements in those rows are noisy and should not be read as trends. Management of Companies recorded no defaults in 2019, so no ratio is computable for it. Set those three aside and the rest still carry more than the table shows on its face.

Every Sector Moved the Same Way, by Very Different Amounts

Nineteen industries with different customers, cost structures, collateral, and cycles all moved the same way on both measures at once. Whatever happened to this population was not sector-specific.

The comparison year matters, and less than it might. Measured against each sector's own 2015 to 2019 average rather than against 2019 alone, defaults are higher in eighteen of nineteen sectors and prepayment lower in fifteen. The exceptions on prepayment are Finance and Insurance, Utilities, Mining, and Management of Companies; on defaults it is Management of Companies. The result is weaker on that basis and it does not reverse.

The severity is another matter. Transportation and Warehousing deteriorated to nearly three times the rate of Health Care, so industry mix governed how hard a book was hit even though it protected no one from being hit.

Prepayment tells the same story with a different distribution. Real Estate lost 4.3 points of prepayment and Health Care 3.5, while Finance and Insurance lost 0.4 and Utilities 0.7. Since prepayment is how a healthy borrower exits, through sale, payoff, or refinancing, a large decline means borrowers who would once have left are staying. Rising defaults alongside falling prepayments is what a closing exit looks like.

Every sector moved the same way. Between 2019 and 2025 all nineteen NAICS sectors carrying 200 or more loans recorded higher defaults and lower prepayments. Against 2015 to 2019 averages the counts are eighteen and fifteen. Dot size reflects loan count. Source: Lumos analysis of loan-level SBA 7(a) performance history.

Sectors that write longer paper generally saw the larger prepayment declines, though with nineteen sectors the pattern is a signal rather than a settled result. Term keeps surfacing in these numbers, and it carries more information than the industry label does.

Why a Sector Rate Hides Two Different Books

A sector default rate blends loans with very different structures, and the blend hides more than it reports.

Restricting each sector to loans written on terms of twenty-five years or longer, and comparing against that sector's full book on an identical measure, the long-term loans perform better in all sixteen sectors where both populations clear 200 loans. There are no exceptions.

Information recorded no defaults at all on its 231 long-term loans, which is a small enough population that the zero should be read as a low rate rather than a precise one. Mining, Utilities, Management of Companies, and Public Administration carry fewer than 200 long-term loans each and are not shown.

Portfolio-wide the same comparison gives 3.83% against 0.89%.

Default rates by sector, 2025 performance year. The long-term book is not made up of smaller, safer credits: those loans average about $1.31 million against $544,000 across the portfolio, so loan size does not explain the gap. Source: Lumos analysis of loan-level SBA 7(a) performance history.

A reader familiar with SBA lending will already have an explanation, and it is not that these are small, safe credits. Twenty-five year amortization generally signals a real estate purchase, which means a larger balance against an asset that holds its value. The loan sizes bear that out. Loans in the twenty-five year book average roughly $1.31 million on a loan-weighted basis against $544,000 across the whole portfolio, so the better-performing population is the larger-balance one.

That raises the question of which variable is doing the work. Across sectors, loan size has no relationship with default rates once term is controlled for, while term itself has a strong one. Size travels with term in this portfolio without driving the outcome.

Two structural mechanisms separate long-term paper from the rest, payment arithmetic and collateral. Stretching the same balance over twenty-five years rather than ten cuts the monthly payment substantially, leaving more of the borrower's cash flow uncommitted and blunting the effect of any rate rise. Long amortization points to real estate, as above, while shorter terms fund equipment, working capital, and acquisition, where the asset depreciates faster than the debt amortizes and there is less to recover.

For a lender, this means the term composition of a sector exposure carries more information than the sector label. Two books with identical NAICS concentration and different term profiles are not carrying the same risk, and the published sector rate describes neither of them.

Sector Explains Less Than the Lender Does

The variation among lenders exceeds the variation among sectors, and it is not close. Across 199 SBA lenders, 2025 default rates ranged from zero to 14.86%, with 24 lenders above 7%. That spread is far wider than the gap between the best and worst industries in the table above, and portfolio size does not order it. Sector mix explains part of a book's performance. Lender-specific factors, credit box, and term composition explain more.

Concentration limits set on industry alone are therefore set on the weakest of the three variables at this level of aggregation, weaker than term structure and weaker than whatever separates one lender from another. The sector rate is a useful reference and a poor ceiling.

The difference between those two uses is worth drawing out, because it points at the better question. A sector rate used as a ceiling asks whether an industry is too risky to hold, and answers it with the variable that explains least. A sector rate used as a baseline asks how a book's own performance in an industry compares with the market's performance in that same industry. That holds industry constant and isolates the thing the dispersion above says actually moves outcomes, which is the lender.

One scope note. These are two-digit groupings and backward-looking rates on completed years. Finer industry definitions carry more information than broad ones, and a forward-looking estimate is a different object from a historical rate. Neither is measured here.

For why aggregate bank data cannot produce these numbers at all, see our analysis of what C&I aggregates can and cannot measure. Prepayment behavior across the portfolio, rather than by sector, is covered in our analysis of SBA 7(a) prepayment speeds, the current headline default reading in our analysis of SBA default rates, and the underlying method in our guide to small business credit risk. Lumos Portfolio Insights scores an existing book loan by loan against this population, and Prime+ does the same at origination.

Figures current as of the 2025 performance year, drawn from loan-level SBA 7(a) history through May 2026. Updated annually.

Frequently Asked Questions

Sources and Methods

Conventions. All rates are loan-weighted and calendar-basis. Sector names follow the labels used in the exhibits, which omit the serial comma in three compound titles; the text otherwise uses it throughout. The 200-loan minimum is applied to both comparison years. The competing-risks result holds 19 of 19 on a calendar basis and 17 of 19 on an SBA fiscal-year basis; the calendar figure is reported because every other measure here is calendar-derived.