The file everybody quotes has no column called default

Fifteen point eight percent. About 2 to 3 percent. Just over 1 percent. Four point eight percent in March 2026, against a portfolio average of 5.4. Google's answer box, asked the same question, offers 4.8 to 5.8 percent and adds that the historical range runs 2.5 to 6.

The Small Business Administration's own published figure for the same program, for fiscal 2024, is 0.55 percent.

None of those is a mistake. Each divides a different numerator by a different denominator, and the spread between the smallest and the largest is a factor of about fifteen. What almost nobody says is which one they used, and the reason that matters more here than in most credit statistics is that the public file underneath most of these calculations does not record default at all.

What is the SBA loan default rate?

There is no single official figure. The SBA publishes an annual charge off rate for its 7(a) program, which was 0.55 percent of the outstanding balance in fiscal 2024. Lifetime failure on a settled cohort of loans runs closer to 7 percent. Both are correct, and they answer different questions.

The 7(a) and 504 loan-level data is the closest thing the public has to a raw record. It carries every 7(a) loan approved since 1990, one row each, with the lender, the amount, the date, the industry code and the outcome. As of the 30 June 2026 file there are 1,961,455 of them.

Look at what the outcome field is allowed to say. The agency's own data dictionary lists the whole domain: cancelled, charged off, undisbursed, paid in full, and a fifth value meaning the loan was disbursed and has not been cancelled, paid off or charged off, and is therefore exempt from disclosure under the fourth exemption to the Freedom of Information Act.

That is the entire vocabulary. A loan ninety days past due, a loan whose guaranty the SBA bought last week, a loan sitting in liquidation with the collateral being sold, and a loan that has paid on time for six years all carry the same value in that column. The public file can tell you which loans ended badly. It cannot tell you which loans are going badly now.

Why does the public loan file have no default column?

Because delinquency is commercial information about an identifiable borrower and the SBA withholds it. The exemption covers commercial or financial information obtained from a person and privileged or confidential. So every loan currently in trouble carries the same status as every loan currently performing, and both are invisible.

This is the single fact that explains the spread. The largest published number, about 15.8 percent, is charge offs divided by loans that have finished. Run that on the full file and the answer is 220,688 charge offs against 1,400,048 finished loans, which is 15.76 percent. It reproduces almost exactly, and it is not a default rate.

It is not a default rate because it silently deletes 297,494 loans. Those are the ones still outstanding, and they are not distributed evenly through the life of a cohort. A loan that fails tends to fail early. A loan that succeeds takes years to pay itself off and leave the file. Divide by finished loans and you are dividing by a population weighted toward failure, because the successes have not finished yet.

The fiscal 2023 approval cohort makes the point cleanly. Measured against loans that have finished, its charge off rate is 17.40 percent. Measured against the loans that were actually disbursed, it is 3.46 percent. The difference is not a revision or a data update, and neither figure is more recent than the other. It is that 80.13 percent of that cohort is still outstanding, so the first measure is dividing by the fifth of the cohort that resolved earliest.

Push it one year forward and it collapses. The fiscal 2025 cohort shows 3.86 percent against finished loans and 0.14 percent against disbursed loans, on a cohort that is 96.35 percent outstanding. Neither figure means anything yet.

What is the lifetime default rate on a 7(a) loan?

For the five approval cohorts from fiscal 2010 to fiscal 2014, now between 95.5 and 98.4 percent finished, 14,620 of 210,808 disbursed loans were charged off. That is 6.94 percent by count. By money it is $2.31bn charged off against $72.05bn approved, or 3.20 percent.

That is the honest version of the question, and answering it takes patience, because the answer is only available for loans made more than a decade ago. Anything more recent is still running.

The two figures differ because small loans fail more often than large ones, and the gradient in the file is clean. Within those same five cohorts, loans originally approved under $50,000 charged off at 8.37 percent by count. Loans of $1m and up charged off at 5.37 percent. By dollars the gap is wider, 6.03 percent against 2.61 percent. The bands in between, at $50,000 to $150,000, $150,000 to $350,000 and $350,000 to $1m, sit in order at 6.85, 6.60 and 5.89 percent by count.

A caveat belongs here rather than at the end, because it changes how the dollar figures should be read. The file records the amount charged off and the amount originally approved, and those are measured at different moments in the life of a loan. A borrower who paid down half the balance over four years and then failed contributes a smaller charge off against a larger approval, so the dollar rate is a loss rate on money lent rather than on money outstanding when things went wrong. The count is the cleaner measure of how often a loan fails. The dollars are the cleaner measure of what the failures cost.

What does a bad cohort actually look like?

Fiscal 2007. Of 88,270 loans disbursed that year, 32,439 were charged off, which is 36.75 percent by count and 23.91 percent of approved dollars. Fiscal 2006 came in at 32.14 percent and fiscal 2008 at 30.29 percent. Nothing since has been close.

Those three years are the reason the historical ranges quoted around this question are so wide, and they are also the reason a single lifetime number misleads. Set against them, the 6.94 percent of the fiscal 2010 to 2014 cohorts is a program operating in a good decade rather than a program with a fixed failure rate.

One measurable thing had changed by then. The average disbursed 7(a) loan in fiscal 2007 was approved at $143,000. By fiscal 2013 it was $383,000, and by fiscal 2021 it was $712,000. Given the size gradient above, a portfolio that moves up-market is a portfolio that should fail less often per loan, and it did. Why the pre-crisis cohorts were written the way they were is a question the file does not answer, and this piece will not invent one.

What does the SBA's own charge off rate measure?

Charge offs recorded during a fiscal year, divided by the unpaid principal balance of the whole 7(a) Regular portfolio at that year end. It is an annual flow over a stock rather than a share of loans, and that construction is why it lands near half a percent while cohort measures land near seven.

The loan program performance tables print the series back to fiscal 2016: 1.82, 0.80, 0.51, 0.68, 0.38, 0.36, 0.42, 0.48 and 0.55 percent, then 0.37 percent for the nine months to 30 June 2025. The denominator in fiscal 2024 was $116.25bn and the numerator $643.5m.

That is a real and useful number for anyone thinking about the portfolio as a balance sheet. It is close to useless for anyone asking how likely a small business loan is to fail, which is what people typing the question are usually asking. It is also the number a reader is least likely to meet, because it sits in a zip file of eleven PDFs on a legacy domain, last refreshed on 15 September 2025 with data stopping at 30 June 2025. The loan-level file is a year fresher than the agency's own summary of it.

There is a second definitional trap in the same tables. The published charge off amount counts the guaranteed portion of each loan only. The loan-level file counts both portions. In fiscal 2024 that is the difference between $643.5m and $850.1m for the same set of failures, a ratio of 75.7 percent, which is about what the statutory guaranty percentages would predict. Two SBA publications, two dollar totals, one set of events.

When is an SBA loan legally in default?

The regulations do not say. Section 120.520 sets the point at which a lender may demand that SBA honor its guarantee: default on any installment for more than 60 calendar days, uncured, and only once all business personal property collateral has been liquidated. That is a purchase trigger, not a definition.

Section 120.10 is the definitions section for the entire part governing the 7(a) program. It defines acceptable risk rating, associate, and dozens of other terms of art. The word default does not appear in it once.

So the widely repeated line that the SBA treats a loan as defaulted at 60 days past due is not what the regulation says. Default is a condition of the note between the borrower and the lender. Sixty days, plus an uncured default, plus liquidated personal property collateral, is the earliest the lender may send the bill to the government. A loan can be in default for months before any of that happens, and it is invisible in the public data the entire time.

The nearest official proxy for default is therefore the guaranty purchase, and the SBA prints that rate too. On 7(a) Regular it ran 0.57 percent of active balance in fiscal 2021, 1.00 percent in fiscal 2023, 1.43 percent in fiscal 2024 and 1.37 percent in the nine months to 30 June 2025. In dollars, purchases in those nine months came to $1.626bn, which is already more than the $1.607bn the agency purchased across the whole of fiscal 2024.

Is the 7(a) portfolio getting worse in 2026?

More loans are being charged off and fewer dollars are going with them. In the nine months to 30 June 2026 the loan-level file records 3,763 charge offs worth $454.4m. At the same point in each of the seven preceding fiscal years the count was lower, and from fiscal 2023 onward the dollar total was higher.

The like-for-like nine-month count reads 3,680 loans in fiscal 2019, then 2,406, 1,886, 3,217, 2,417, 2,607, 3,417 and now 3,763. On dollars over the same windows it reads $699.3m, $353.7m, $340.8m, $453.1m, $483.3m, $613.1m, $614.5m and $454.4m. Failures are up in number and down in size, which is what the size gradient would predict if the growth in lending is concentrated in smaller loans.

Two limits on that reading, both real. Charge off recognition lags the event by months, and a file compiled on 30 June will not contain every charge off made in June, so the most recent window is the least complete of the eight. And a charge off is the end of a long process rather than the start of one, so a rising count in 2026 is describing loans that stopped paying well before 2026.

What does the government think the losses will cost?

Nothing. The fiscal 2027 budget request puts the 7(a) subsidy rate at 0.00 percent for fiscal 2025 actual, fiscal 2026 estimate and fiscal 2027 estimate, on program levels of $33.4bn, $35.5bn and $40bn.

A zero subsidy rate is not a claim that nothing defaults. It is a claim that borrower and lender fees, plus what comes back after a failure, are expected to cover what does. The recoveries side is published as well, and it is substantial. Of what the SBA paid out on 7(a) guaranties purchased in fiscal 2016, it had recovered 36.98 percent by 30 June 2025. The fiscal 2017 through 2020 purchase years sit at 34.33, 37.23, 38.60 and 37.22 percent, and collections were still running on all of them nine years later.

That is the number that reconciles the rest. A program can charge off close to 7 percent of its loans over their lifetime, recover better than a third of what it pays on the failures, price the remainder into fees, and cost the Treasury nothing. Every figure in that sentence is published. None of them is the default rate.

Frequently asked questions

What is the average SBA loan default rate? The most defensible answer is roughly 7 percent of loans over the life of a cohort, based on 6.94 percent for the fiscal 2010 to 2014 approval years once those cohorts had substantially finished. By dollars the same cohorts lost 3.20 percent of what was approved. Recent cohorts cannot be measured yet.

Why do published SBA default rates differ so much? Because the denominators differ and are rarely stated. Charge offs over finished loans gives 15.76 percent, charge offs over disbursed loans in settled cohorts gives 6.94 percent, and charge offs over the outstanding portfolio balance in a single year gives 0.55 percent.

Does the SBA publish a default rate? Not under that name. It publishes a charge off rate and a guaranty purchase rate, both as percentages of unpaid principal balance, in its loan program performance tables. Those tables were last updated on 15 September 2025 with data through 30 June 2025.

How many days late is an SBA loan in default? The regulation sets no day count for default itself. It allows a lender to demand purchase of the guaranty after more than 60 calendar days of uncured default on an installment, once business personal property collateral has been liquidated.

Can I look up whether a particular SBA loan defaulted? Only if it has already been charged off or paid in full. The public file marks every loan that is disbursed and still running as exempt from disclosure, which on the fiscal 2024 cohort is 89.8 percent of the loans.

Do bigger SBA loans default less often? In the fiscal 2010 to 2014 cohorts, yes, and at every step. Charge off rates by original approval size ran 8.37 percent under $50,000, then 6.85, 6.60 and 5.89 percent through the middle bands, and 5.37 percent at $1m and above.