Methodology

The SBA publishes loan-level disclosure files for 7(a) and 504 quarterly. We process each release. The 2026 report covers FY2010–FY2025 — 1,010,720loan records across both programs. Data dictionary lines up one-for-one with SBA's column spec.

FY2025 is the primary window for every section. Comparables run back to FY2020. We use named windows (“FY25”, “FY20–FY25”) instead of rolling windows so that every number in the report reconciles against SBA's own annual totals.

Every stat in the report is accompanied by an inline(program, FY, n, filter)citation. That is the denominator — not a marketing flourish. When two sections cite the same metric with different denominators, both are correct for their filter, and the reconciliation note explains the delta.

Spread means borrower note rate minus the US Prime Rate at origination date. Current Prime is 6.75%. We report in basis points (bps) where precision matters; 100 bps equals 1.00 percentage point.

Only 7(a) variable-rate loans are included in spread rankings. Fixed-rate 7(a) loans, 504 debentures, and Express lines use different pricing mechanisms — mixing them in one chart produces a number no borrower can actually act on.

The floor is not arbitrary. Below 50 loans, the median spread of a single lender bounces around so much on any given quarter that a ranked list starts telling you more about variance than skill.

We say “funding rate” — approvals that actually disbursed — rather than “approval rate.” Approval is step one. A borrower who gets approved but never closes funded zero dollars and signed zero notes. Every chart that looks like an approval rate is really a funding rate unless explicitly labelled otherwise.

• Loans from before FY2010. SBA's public dataset has gaps in the 2000s that make pre-FY2010 comparisons unreliable.
• Borrower-identifying details. We aggregate by lender, sector, state, and size band. Individual borrower records stay private.
• Conventional (non-SBA) lending. This report is scoped to the SBA 7(a) and 504 programs.