Methodology: reported SBA change-of-ownership approvals

Analysis date: September 17, 2026. Source snapshot: June 30, 2026. Scope: Reported classification subset, not a census of completed acquisitions.

Sources

• Official dataset page
• Downloaded 7(a) FY2020-present CSV
• Official data dictionary

CSV SHA-256: 6c1e9132b5141a19f82bdc8ccafb86c9a01662461cad41ddb36a3cf409d8a4fe.

The retrieved file contains 388,338 rows and 42 columns. Every row has AsOfDate 2026-06-30. The dataset page’s resource path changed from the older 7-a-504-foia pattern to 7a-504-foia; the actual download linked above was used, not the stale cached March snapshot.

Unit and classification

The unit is a row in the published approval extract. The 42 fields do not include a dedicated unique loan identifier. BusinessAge includes the exact value Change of Ownership; 40,555 rows in the full extract carry that value before period/status selection.

The dictionary describes BusinessAge as a categorical description. We observed its values directly and do not claim that this category captures every qualifying acquisition consistently. We did not infer acquisition status from names, business age in years, NAICS, loan size, or a search keyword.

Filters

1. Source is the 7(a) FY2020-present resource; no 504 records or other files are joined.
2. FY2025 comparison window: ApprovalDate from 2024-10-01 through 2025-06-30, inclusive.
3. FY2026 comparison window: ApprovalDate from 2025-10-01 through 2026-06-30, inclusive.
4. Exclude LoanStatus == CANCLD at the June 30, 2026 snapshot.
5. Require BusinessAge == Change of Ownership exactly.
6. Retain all other statuses, including undisbursed COMMIT.
7. Sum GrossApproval, not SBAGuaranteedApproval. Compute the median of GrossApproval in the selected rows.

ApprovalDate values are ISO dates in the source. Dollar values are numeric. The P I F label in the file corresponds to paid-in-full terminology in the dictionary; source status labels are preserved in the JSON outputs.

Results

Percentage change is (later / earlier − 1) × 100. Count: −12.4406%; gross amount: −10.1370%; median: +6.2960%. Round to one decimal in prose. COMMIT shares use selected record count as denominator: 5.1307% and 33.3258%.

Duplicate handling

There are 391 excess exact duplicate rows in the full file. They are reported, not automatically removed: without a unique loan identifier, identical disclosed fields need not prove a duplicate loan. Neither selected ownership-change subset contains exact duplicate rows, so exact-row deduplication does not change the article’s headline counts or totals.

The wider all-classification totals are raw record summaries with those source rows retained; they are not represented as independently deduplicated loan counts.

Limitations and prohibited interpretations

• Both cohorts use status at June 30, 2026. The older cohort has more time to disburse, cancel, repay, or charge off. This is a same-snapshot comparison, not a matched-age cohort analysis.
• Non-canceled approvals are not necessarily disbursed. Disbursed records are not independently verified completed business acquisitions.
• The classification’s completeness is not established. Do not label the selected subset “all SBA acquisitions.”
• No application-denial population is provided. Do not infer approval rates or rejection rates.
• GrossApproval is loan amount, not purchase price, project cost, down payment, or buyer resources.
• BankName is the bank currently assigned to the record per the dictionary; it does not establish origination or lender fit for a new buyer.
• EXEMPT does not mean a loan is certified as performing well; it is the dictionary’s FOIA status category.
• The extract cannot establish the cause of changes, conventional/private-credit activity, or the full business-acquisition market trend.
• JobsSupported is self-reported in the dictionary and is not used in this investigation.

Reproduction files

• Detailed aggregate JSON
• Selected subset CSV
• Broader period context CSV
• Reusable analysis script
• Reader-facing article

The reproduction script expects a local copy of the linked official CSV and checks its hash before computing the selected subset. It uses Python’s standard library and does not publish or output borrower names or addresses.