In April I wrote that SBA loan-level delinquency was public for the first time. The March 2026 FOIA release split the old EXEMPT status into seven sub-statuses, so for one quarter you could see which 7(a) loans were delinquent, past due, in liquidation or deferred.
That lasted one release. The June 2026 file went back to the old codes, and the September 2026 file stayed there. Of the 8,199 7(a) loans coded DELINQ or PSTDUE in March, 6,644 (81%) are back to plain EXEMPT in September.
This post covers what changed, what it does to any trend built across these releases, and one finding the March window still supports.
What the 7(a) status field looks like now
I ran the raw LoanStatus values for every 7(a) loan in the last four releases through the same pipeline.
| Status | Dec 2025 | Mar 2026 | Jun 2026 | Sep 2026 |
|---|---|---|---|---|
EXEMPT |
292,466 |
– | 297,047 |
297,788 |
CURR |
– | 261,613 |
– | – |
PURCH(NOT C/O) |
– | 14,434 |
– | – |
LIQUID |
– | 8,536 |
– | – |
DELINQ |
– | 5,523 |
– | – |
PSTDUE |
– | 2,676 |
– | – |
SOLDNC |
– | 2,085 |
– | – |
DEFERD |
– | 733 |
– | – |
CHGOFF |
215,825 |
142,896 |
220,387 |
221,994 |
CLSLN |
– | 75,630 |
– | – |
SOLDCO |
– | 309 |
– | – |
PIF / P I F |
1,144,528 |
1,166,717 |
1,176,265 |
1,185,965 |
CANCLD |
226,132 |
231,598 |
236,000 |
238,808 |
COMMIT |
22,901 |
23,645 |
21,018 |
21,420 |
None of the nine codes added in March appear in June or September. The seven EXEMPT sub-statuses summed to 295,600 loans in March, which sits between December's 292,466 and June's 297,047 EXEMPT count. The loans are still in the file; the labels that separated them are not.
One March change did survive. Paid-in-full 7(a) loans are still coded P I F, with embedded spaces, in both later releases. A filter on 'PIF' still matches zero 7(a) rows.
The same loans, followed from March to September
To confirm this is a relabelling and not a change in the loans, I matched each March 7(a) loan to the September file on program, location ID, borrower name, approval date and approved amount. 99.2% of March loans matched.
| March status | Loans | EXEMPT in Sep |
CHGOFF in Sep |
P I F in Sep |
|---|---|---|---|---|
CURR |
261,613 |
92.2% |
0.1% |
6.4% |
DELINQ + PSTDUE |
8,199 |
81.0% |
7.3% |
10.0% |
LIQUID |
8,536 |
74.6% |
16.5% |
6.3% |
PURCH(NOT C/O) |
14,434 |
87.8% |
8.5% |
2.3% |
DEFERD |
733 |
94.3% |
– | 4.1% |
SOLDNC |
2,085 |
99.3% |
– | – |
CLSLN |
75,630 |
– | 99.5% |
– |
SOLDCO |
309 |
– | 99.7% |
– |
Every active sub-status went back into EXEMPT. Every post-charge-off sub-status went back into CHGOFF. Rows do not add to 100% because a small share moved to other statuses or did not match.
What this does to a charge-off trend
In April I wrote that the March drop in the CHGOFF count was a labelling effect, and that a like-for-like series needed CHGOFF + CLSLN + SOLDCO. The June release reversed the effect.
| Release | CHGOFF only |
CHGOFF + CLSLN + SOLDCO |
|---|---|---|
| Dec 2025 | 215,825 |
215,825 |
| Mar 2026 | 142,896 |
218,835 |
| Jun 2026 | 220,387 |
220,387 |
| Sep 2026 | 221,994 |
221,994 |
A series built on CHGOFF alone shows a 34% fall in March and a 54% rise in June. The combined series moves by less than 1.5% per quarter across all four releases. If your model picked up either swing, the combined definition removes both. The same applies to any active-book count: EXEMPT in December, June and September lines up with the sum of the seven March sub-statuses.
The 504 file
The March 504 file did not use the 7(a) codes. It carried its own plain-English statuses:
| March 2026 504 status | Loans |
|---|---|
CURRENT |
57,812 |
IN CATCH-UP |
1,115 |
PURCH(NOT C/O) |
1,144 |
1 MONTH PAST DUE through >9 MONTHS PAST DUE |
687 |
ASSET SALE |
337 |
IN DEFERMENT |
101 |
The 504 past-due status came in ten buckets by months delinquent, which is finer than anything on the 7(a) side. Paid-in-full 504 loans were split into PREPAID IN FULL (104,294), PAID IN FULL (5,213) and PAID IN FULL (LIQ) (3,844).
From June, 504 went back to EXEMPT, PIF, CHGOFF, CANCLD and NOT FUNDED, plus one new label, CLOSED. All 687 past-due 504 loans from March are EXEMPT or PIF in September, apart from one that did not match. The 339 loans coded CLOSED in June are almost exactly March's ASSET SALE loans: 337 of them carried that label in March.
What the March window still shows
The March labels are a single cross-section, but they can be followed forward. Two quarters later, 7.3% of 7(a) loans that were DELINQ or PSTDUE in March had charged off. For loans coded CURR in March, the figure was 0.08% (198 of 261,613). Loans in LIQUID charged off at 16.5% over the same six months.
That gap is the reason the March detail mattered. A 7(a) loan's status in March sorted loans by near-term charge-off by roughly two orders of magnitude, and the current file cannot make that cut. The chargeoff and paid-in-full dates in later releases still show which March delinquencies resolved and how, so the March snapshot keeps its value as a fixed cohort even with no successor.
The lender ranking from April cannot be refreshed. Huntington's 6.0% dollar-weighted delinquency rate was a March number, and there is no June or September equivalent to compare it against.
What to watch
I would check three things in the December 2026 release:
- whether the 7(a) sub-statuses or the 504 month-bucket labels come back
- whether
P I Fkeeps its spaces - whether 504
CLOSEDstays a distinct status or folds back intoCHGOFForPIF
Until then, a series that spans March 2026 is comparable only if the March sub-statuses are mapped back to the legacy codes. The current loan-level file is part of the SBA loan data subscription. Program-level coverage is on the SBA 7(a) and SBA 504 pages.
Methodology
All counts come from the SBA 7(a) and 504 FOIA bulk files with as-of dates 2025-12-31, 2026-03-31, 2026-06-30 and 2026-09-30, loaded through our pipeline. I loaded the June files on August 5 and the September files on November 2. Counts are loans as stored in the pipeline, so March figures here differ by less than 0.5% from the counts in the April post, which were read from the raw files directly.
Loans were matched across releases on program, canonicalised location ID, borrower name, approval date and gross approval amount. 99.2% of March 7(a) loans matched to September. Charge-off shares in the follow-forward table are counts of loans, not dollars, measured on the September status.