For One Quarter, SBA Showed Which Loans Were Late. Then the Window Closed.

• 

  SBA loan delinquency data snapshot

SBA's March 2026 loan file revealed delinquencies, guaranty purchases and liquidations for the first time. By the next release, the detail appears to have been folded back into a single "exempt" code. What the snapshot showed, and how easily it misleads.

For most of its history, SBA's public loan-level data has had a blind spot in the middle. A loan could be shown as approved, cancelled, paid in full or charged off. Everything in between, every loan that was disbursed and still on the books, whether perfectly current or months behind, carried the same label: EXEMPT.

In the quarterly release posted around April 28, 2026, carrying data as of March 31, that changed. According to the data firms that parsed the file, the single EXEMPT code split into seven, separating current loans from delinquent, past-due, deferred and liquidating ones, and from loans SBA had already bought back from lenders. The charge-off code split as well. For the first time, anyone could see where stress sat in the SBA 7(a) portfolio before it became a loss.

It did not last. Two independent sources report that the next release, data as of June 30, 2026, went back to the old five-code scheme. SBA has published no notice explaining either change, and its data dictionary still describes the 2018 codes.

This article reconstructs what the March snapshot showed, what it can and cannot support, and why the most widely quoted delinquency figures from it need careful handling. The central point is that one snapshot is a photograph, not a trend, and a photograph taken with definitions SBA never published.

What SBA publishes, and what the law holds back

SBA's loan-level data lives in one dataset on data.sba: four 7(a) files split by decade, two 504 files, and a data dictionary. SBA says the files update quarterly and are typically available about a month after each quarter ends. In practice the lag has been longer: the September 2025 data was posted on November 14, 2025, and the December 2025 data on February 6, 2026. The dataset page still shows a "last modified" date of April 28, 2026, even though it now lists files as of September 30, 2026, so that field cannot be used to date releases.

The EXEMPT code has a legal basis. SBA's Freedom of Information Act regulations list "loan status, other than charged-off or paid-in-full" among the categories of information generally exempt from disclosure, alongside non-statistical information on defaults and delinquencies. Data firms that opened SBA's dictionary report that it still defines EXEMPT as status withheld under FOIA Exemption 4, the protection for confidential commercial information. SBA's own page for the dictionary shows it was last updated on August 20, 2018.

That is the backdrop for March 2026. The release did not just add detail; it disclosed a category of information that SBA's standing rules treat as generally protected. We found no SBA notice, procedural notice, press release or dictionary update explaining why it did so, or why it appears to have stopped.

What the March 2026 file contained

The counts below come from LoanTape, a data vendor, which compared the December 2025 and March 2026 7(a) files. They have not been independently replicated from SBA's files, and the March snapshot is no longer posted on data.sba. The labels are the vendor's and the trade press's readings; SBA has not published definitions for the new codes.

Status code	Loans, March 31, 2026	Reported meaning
CURR	260,896	Current
DELINQ	5,502	Delinquent
PSTDUE	2,670	Past due
DEFERD	730	Deferred
LIQUID	8,502	In liquidation
PURCH (NOT C/O)	14,404	Purchased by SBA, not charged off
SOLDNC	2,083	Sold, not charged off
CLSLN	75,522	Closed after charge-off
SOLDCO	309	Sold after charge-off
CHGOFF	142,898	Charged off (down from 216,123 in December 2025)

The figures hang together. The seven codes that replaced EXEMPT sum to 294,787 loans, against 292,937 loans coded EXEMPT in December, a difference consistent with three months of new disbursements. CHGOFF, CLSLN and SOLDCO sum to 218,729, against 216,123 charge-offs in December. LoanTape reports that about a tenth of the newly coded CLSLN and SOLDCO loans had previously been coded as paid in full, which accounts for the small excess.

Two features of the table matter for anyone using older analyses. First, roughly 73,000 loans left the CHGOFF code in March without being any less charged off. A charge-off rate built on the CHGOFF code alone would have dropped by about a third overnight, purely through relabeling. Second, the 14,404 loans in PURCH (NOT C/O) are loans on which SBA had already paid the lender's guaranty claim but had not yet written off. Under the old scheme they looked like healthy EXEMPT loans.

What the snapshot showed

At the program level. Using LoanTape's counts, delinquent and past-due loans together made up about 3.0 percent of the active 7(a) book by count, defined as current, delinquent, past-due and deferred loans. Adding loans in liquidation to both sides raises the share to about 6.0 percent. That second figure matches the 6.0 percent "active distress" rate reported by independent analyst Shane Pierson, who used the same definition.

By lender. LoanTape published a ranking of the 20 largest 7(a) lenders by delinquency, weighted by approved dollars. The top 20 together showed 2.2 percent. Individual rates ranged widely.

Lender	Delinquent and past-due share, March 31, 2026
Huntington	6.0%
Newtek Bank	4.3%
Celtic Bank	3.7%
Pinnacle Bank	3.0%
Byline Bank	2.6%
PNC	0.5%
Fifth Third	0.5%
Bank of America	0.0%
Top 20 combined	2.2%

The weighting matters. LoanTape reports KeyBank at 5.7 percent by loan count but 1.7 percent by dollars, a gap that reflects many small loans in distress. No lender we reviewed has publicly addressed its figure. Huntington's third-quarter 2026 earnings call did not mention the SBA data, but management said nonperforming assets had risen because of increases in government-guaranteed loan categories, which it described as carrying very low loss content. That is consistent in direction with the snapshot, though the two measures are not comparable.

By industry. LoanTape reported delinquency of 3.7 percent in trucking and 3.0 percent in accommodation and food services, with $829 million past due in accommodation. Pierson's distress measure ranges from 3.2 percent in finance to 10.9 percent in transportation. Lumos Data, using a different metric, an annualized default rate for the first half of fiscal 2026, puts transportation and warehousing at 7.6 percent. The measures differ, but every source puts transportation and restaurants and hotels at the top.

Then the window closed

The detail did not survive the next release. FinObservatory, an independent analytics site, parsed the June 30, 2026 7(a) file and found only five status values: paid in full (still written as "P I F" in the 7(a) files), EXEMPT, cancelled, charged off and committed. Its charge-off count of 220,688 is back near December 2025's 216,123, which suggests the closed and sold charge-offs were folded back into CHGOFF. Coleman Report's data service states plainly that SBA stopped publishing detailed portfolio status after March 31, 2026, and that it carries the last reported status forward as a floor.

We found no explanation. SBA's November 2026 fiscal year release does not mention it. We found no statement on the change from SBA leadership, NAGGL or the inspector general. Whether the September 30, 2026 files carry the detail could not be confirmed for this article, and is the first thing any analyst should check.

The reversion has practical consequences. Several data products and analyses still present the March figures as if they were current, and some anticipated that the next release would allow the first measurement of how loans move from current to delinquent over time. If the reversion holds, that comparison cannot be made from public data. Every delinquency figure from SBA's loan-level files describes a single day, March 31, 2026, which by now is more than seven months old.

Where delinquency data misleads

Even taken at face value, the March snapshot is easy to misread. These are the main traps.

The codes are undefined. SBA has not published what separates DELINQ from PSTDUE, or what qualifies a loan as deferred. They should not be mapped to 30, 60 or 90 days past due.

Dollars are approval amounts, not balances. The file carries the original approved amount, not the outstanding balance. A dollar-weighted rate gives an old, mostly repaid loan the same weight as a new one of the same original size, and count-based and dollar-based rates can diverge sharply, as KeyBank's figures show.

The denominator drops the worst loans. Most published delinquency rates exclude loans in liquidation, loans already purchased by SBA and loans closed after charge-off. Those are the most distressed loans in the portfolio. A rate that leaves them out describes the survivors.

Purchased is not the same as charged off. A loan coded PURCH (NOT C/O) is one where the borrower defaulted and SBA paid the lender's claim, but the loss has not yet been realized. Leaving these loans out of a default measure understates stress. Adding them to charge-offs overstates realized losses. The same ambiguity applies to loans sold before and after charge-off.

The lender is today's lender. The lender field reflects the institution a loan is currently assigned to. Loan sales and bank mergers move both good and bad loans between names.

The scheme changed twice in six months. Any time series that crosses December 2025, March 2026 and June 2026 has to collapse the expanded codes back into the old ones before comparing. Even then the series is not perfectly continuous, because some loans moved between paid-in-full and charge-off categories in the process.

The file has quality problems. FinObservatory reports 1,918 duplicate 7(a) rows, 6,984 loans with charge-off amounts above their approved amounts, and 58 charged-off loans with no charge-off date.

One snapshot, four answers

How much of the 7(a) portfolio was in trouble on March 31, 2026? Using the same reported counts, the answer depends entirely on which loans are counted as troubled and which are counted as the portfolio.

Definition of distress	Share of loans
Delinquent or past due, as a share of current, delinquent, past-due and deferred loans	3.0%
The same, counting deferred loans as distressed	3.3%
Delinquent, past due or in liquidation, with liquidating loans added to the base	6.0%
Delinquent, past due, in liquidation or already purchased by SBA, with both added to the base	10.6%

Source: Loan Analytics calculations from LoanTape's reported March 31, 2026 status counts, by number of loans.

Each line is defensible. The first is the narrowest measure of borrowers currently behind on payments. The last counts every loan that is behind, being liquidated or already in SBA's hands as a defaulted claim, which is closer to what a lender or investor means by a troubled book. The spread, from 3 percent to more than 10 percent of the same portfolio on the same day, is the strongest argument for publishing definitions alongside any figure from this data.

Delinquency, default and charge-off run on different clocks

The figures in circulation measure different events. Quoted side by side without definitions, they appear to contradict one another.

Measure	Figure	What it counts
Bank business-loan delinquency, all U.S. commercial banks	1.33% (Q1 2026), 1.27% (Q2 2026)	Federal Reserve series, all business loans, not SBA
Top 20 SBA lenders, delinquent and past due	2.2%	Share of active book by approved dollars, March 31, 2026 (LoanTape)
All 7(a), delinquent and past due	About 3.0%	Share of active book by count, March 31, 2026 (our calculation from LoanTape counts)
All 7(a), delinquent, past due or liquidating	About 6.0%	Same, including liquidation (our calculation; Pierson)
7(a) trailing-12-month default rate	4.8%	Loans entering default over a year to March 31, 2026, the highest since 2013 (Lumos Data)
Cumulative charge-off, recent vintages	2% to 4%	Realized losses to date; recent vintages incomplete

The SBA population is riskier than a bank's general business book by design, because 7(a) borrowers must show they cannot get comparable credit elsewhere. The bigger gap is timing. Delinquency is a snapshot of who is behind today. Default counts loans entering default over a period. Charge-off is the end of the process, and it arrives late. FinObservatory estimates that only about 10.5 percent of the charge-offs a 7(a) cohort will eventually record have occurred by its second year, and 62.9 percent by its fifth. That is why recent vintages look clean on charge-off data and why delinquency data, while it existed, was valuable: it showed stress years before losses would.

The oversight picture

The most authoritative performance finding of 2026 does not come from the loan file at all. SBA's Office of Inspector General, in Report 26-14 of September 17, 2026, found that 7(a) loans made by small business lending companies, the nonbank SBA lenders, significantly underperformed loans from other lenders on default rates for loans approved from fiscal 2016 through 2023, with early defaults rising substantially. It also found that SBA's credit risk office had not examined these lenders for safety and soundness since April 4, 2023, because of contractual issues. SBA's management largely disagreed with the draft but agreed with two of three recommendations and partially agreed with the third.

The headline numbers come from press coverage of the full report, which we could not access directly. Inc. reported default rates of 14.87 percent for these lenders against 9.79 percent for others, and early default rates within 18 months of 5.38 percent against 2.62 percent, covering 11,068 loans worth about $9.5 billion, of which 1,657 had defaulted. Two of those figures do not quite reconcile: 1,657 divided by 11,068 is 14.97 percent, not 14.87, a difference that could reflect a different denominator or dollar weighting. Funder Intel reported that two unnamed lenders made 69 percent of the group's loans but accounted for 84 percent of defaults. SBA's response, written by credit risk deputy director Edward Ledford, argued that the remaining nonbank lenders outperformed traditional lenders, and that one large historical lender heavily influenced the combined numbers.

Two related findings complete the picture. In March 2026, OIG Report 26-07 found that SBA's risk-mitigation framework did not fully screen 7(a) loans approved from August 2023 through December 2024 for three of six eligibility requirements, leaving limited assurance on 73,302 loans totaling about $32 billion. SBA agreed to flag all of them for review if they come to guaranty purchase. In August, OIG Report 26-12 found SBA made final purchase-review decisions on 16 of 32 sampled loans without sufficient evidence, about $11.5 million in potential improper payments.

The March data also reached Congress. In a June 4, 2026 letter to Administrator Kelly Loeffler on the failure of Community Bank & Trust–West Georgia, Senate Small Business Committee Chair Joni Ernst cited SBA's March 31, 2026 loan data for the bank's 7(a) volume, which rose from $28.4 million in fiscal 2024 to $218.3 million in fiscal 2025. Her figure of more than $70 million in noncurrent 7(a) loans at the bank was sourced to American Banker rather than to SBA's file.

How to check the September files yourself

Because the March snapshot is no longer posted, the most useful next step is to confirm what the current files contain. It takes an analyst about an hour.

Download and record. From the 7(a) and 504 FOIA dataset on data.sba, download the 7(a) files for fiscal 2010 to 2019 and fiscal 2020 to the present, along with the data dictionary. Record each file's as-of date, which appears in a column inside the file.

List every status value. Read only the loan status column, strip extra spaces, and count each distinct value. If the file shows only paid in full, EXEMPT, cancelled, charged off and committed, the expanded scheme has not returned. If CURR, DELINQ or PSTDUE appear, it has, and the same calculations shown above can be rerun on fresh data.

Check the dictionary. Open the data dictionary and look for definitions of the expanded codes. If any appear, they would be the first SBA-authored definitions of what delinquent and past due mean in this data.

Keep the series comparable. Before comparing any two releases, collapse the expanded codes into the old ones: current, delinquent, past-due, deferred, liquidating, purchased and sold-not-charged-off loans back into EXEMPT, and closed and sold charge-offs back into CHGOFF.

If the March files are needed for replication, archived copies may exist with data vendors or web archives, but they should be checked against the file's own as-of date column before use. A FOIA request to SBA for the March 31, 2026 files and the internal record of the change is the authoritative route.

What this means

For analysts and lenders: the March 31, 2026 snapshot is a useful benchmark and a poor basis for anything else. It can show where stress sat on one date. It cannot show whether stress is rising or falling, and any figure from it should carry its date, its numerator, its denominator and its weighting.

For readers of lender rankings: a lender's delinquency rate depends heavily on choices the publisher made, including which statuses count, whether dollars or loans are counted and whether liquidations are excluded. Treat single-source rankings as indicators, not findings.

For SBA lenders: the snapshot, while it lasted, let anyone compare a lender's book with its peers' on a common basis, something previously possible only through regulators or a lender's own filings. Lenders whose figures circulated have not responded publicly, and a lender's own call-report and 10-Q disclosures remain the better guide to its portfolio. But the episode shows how quickly third parties will build rankings once status data is public, and lenders should expect it again if SBA restores the detail.

For borrowers and investors in SBA-backed businesses: the most consistent signal across every measure is sector risk. Transportation and hospitality lead on delinquency, default and charge-off alike. That agreement, across different data and different definitions, is more reliable than any single rate.

For policymakers: SBA briefly showed that loan-level status can be disclosed without identifying borrowers' private information beyond what the files already contain. Whether the change was intentional, and why it stopped, is a question SBA has not answered. A FOIA request for the March 31, 2026 files and the internal record of the change, or a direct question to SBA's chief data officer, is the most direct way to get an answer.

What this analysis can and cannot say

It can say that SBA's March 31, 2026 7(a) file exposed delinquency, past-due, deferral, liquidation and purchase statuses, as reported by multiple data firms; that the counts reported from it are internally consistent; that two independent sources report the June 2026 file reverted to five codes; and that SBA has published no explanation or definitions.

It cannot give any SBA-defined delinquency rate, any trend in delinquency, or any delinquency rate for the 504 program, which never received the new codes. It cannot independently confirm vendor figures for individual lenders or industries. And it cannot confirm the status scheme in the September 30, 2026 files.

Where the published record goes wrong

This series documents where commonly repeated claims fail against the data. For this topic:

• SBA did not release delinquency data in July 2026. The expanded statuses appeared in the March 31, 2026 snapshot, posted around April 28. The July date most likely traces to a paid industry webinar about the March data.

• Claims that SBA "now publicly discloses" loan-level delinquency are out of date. Two independent sources report the June 30, 2026 file reverted to five codes.

• Our own October analysis of early defaults in this series described the expanded codes as SBA's current scheme. On the evidence now available, they were in place for one quarterly release.

• LoanTape describes the change as five codes becoming thirteen with eight new codes, but lists nine new codes. The arithmetic only works if EXEMPT dropped out.

• The inspector general's nonbank default rate of 14.87 percent does not reconcile with the reported counts, 1,657 of 11,068 loans, which give 14.97 percent.

• Coleman Report's "53 percent higher" default rate for nonbank lenders uses rounded inputs. The unrounded figures, 14.87 against 9.79 percent, give about 52 percent.

• An SBA table, as reported by Coleman, shows an overall fiscal 2023 7(a) cumulative default rate of 11.9 percent, while charge-off-based measures for the same vintage run 2 to 3.5 percent. They measure different events at different points, and the SBA figure's definition is not public.

• Delinquency rates from the March file are often quoted beside bank delinquency rates, SBA default rates and charge-off rates as if they measured the same thing. They do not.

Sources:

• U.S. Small Business Administration, 7(a) and 504 FOIA loan-level data and data dictionary, data.sba

• U.S. Small Business Administration, 7(a) and 504 Loan Data Dictionary page, last updated August 20, 2018

• 13 CFR Part 102, Subpart A, Appendix A

• U.S. Small Business Administration, news release 26-113, November 2, 2026

• SBA Office of Inspector General, Report 26-14, Small Business Lending Companies' Performance in the 7(a) Loan Program and SBA's Oversight, September 17, 2026

• SBA Office of Inspector General, Report 26-07, SBA's Screening of 7(a) Loan Applications Under its Risk Mitigation Framework, March 11, 2026

• SBA Office of Inspector General, Report 26-12, SBA's 7(a) Loan Guaranty Purchase Denial Review Process, August 27, 2026

• SBA Office of Inspector General, Report 27-01, Top Management and Performance Challenges Facing the SBA in Fiscal Year 2027, October 16, 2026

• Senator Joni Ernst, letter to SBA Administrator Kelly Loeffler on the failure of Community Bank & Trust–West Georgia, June 4, 2026

• Federal Reserve Bank of St. Louis, FRED series DRBLACBS, delinquency rate on business loans, all commercial banks

• Huntington Bancshares, third quarter 2026 results and earnings call, October 22, 2026

• LoanTape, "SBA Loan-Level Delinquency Is Public for the First Time," April 30, 2026

• FinObservatory, SBA loan data analysis

• Coleman Report, SBA Data Portal and coverage of OIG Report 26-14, September 29, 2026

• Inc., coverage of OIG Report 26-14, September 24, 2026

• Funder Intel, coverage of OIG Report 26-14, September 25, 2026

• Lumos Data, SBA 7(a) default rate analysis, July 2026

• Shane Pierson, The State of SBA Lending, mid-2026 edition