Auto loan credit standards from SEC asset-level filings

Summary statistics computed from Form ABS-EE asset-level data — the loan-level file that Regulation AB II requires for registered public auto-loan securitizations. Covers 28 shelves, filings dated 2026-06-15 to 2026-08-21, computed from 52,523 loan records sampled systematically across pools holding 1,056,360 loans.

Download the dataset (JSON) · method in full below · every row carries the CIK, accession, filing date and the direct URL of the asset file it was computed from.

By shelf

Credit score, payment-to-income and rate are distributions, not thresholds. p5 is the 5th percentile of loans in the sampled pool.

How this was collected

Source

Regulation AB II (17 CFR 229.1125, Schedule AL) requires sponsors of registered public auto-loan securitizations to file loan-level data for every asset in the pool. Each record carries that loan's credit score, payment-to-income ratio, original amount, term, interest rate, vehicle value and originator name, among other fields.

Selection

EDGAR publishes a quarterly index of every filing accepted. Eight quarters were parsed (2024-Q4 through 2026-Q3) — a census of accepted filings, not a search result, so nothing is ranked or truncated. That yielded 8,231 ABS-EE filings across 386 entities named for auto collateral. Lease trusts, floorplan and dealer-inventory trusts and operating companies were excluded; collateral type was confirmed from the XML namespace (absee/autoloan versus absee/autolease) rather than the entity name, because at least one lease trust carries no "lease" token in its name. The result is 329 retail auto-loan trusts in 35 shelf families, of which 28 yielded asset data.

Sampling

Every 20th record was retained, read sequentially across the entire file. Two constraints shaped that. SEC Archives does not honour HTTP range requests — a byte-range request returns the whole file from byte 0 with no Content-Range header — so partial reads are performed client-side. And reading only the beginning of a file was tested and rejected: on one pool the first 4 MB gave a median credit score of 563 against 597–600 for blocks drawn later in the same file. A controlled comparison across all 28 shelves found beginning-of-file sampling differed by a mean of +3.6 points at the 5th percentile, but with individual shelves diverging by as much as 76 points in either direction.

Computation

Percentiles are computed per pool by linear interpolation. Loan-to-value is original loan amount divided by vehicle value amount, where both are present and the ratio falls between 0 and 3. Scale is detected per pool rather than assumed: filers report payment-to-income and interest rate inconsistently — of these shelves, 24 file payment-to-income as a decimal and 4 file it already as a percentage — so each pool's basis is determined from its median and published alongside the figure.

Field meanings come from the filer

17 CFR 229.1125 names each Schedule AL field but publishes no code table. Each filing includes a small exhibit (EX-103) in which the filer documents what its own codes mean, and those differ between filers. That exhibit was harvested and stored with each pool; it was retrieved for 12 of 28 shelves. Examples that change how a figure reads: one filer applies income-verification code "3" to every loan whose income it verified, because it cannot distinguish levels 3, 4 and 5 without a physical file review; "vehicle value source" code 98 ("Other") is used by at least one filer when the vehicle's sales price is the basis, making that ratio loan-to-price rather than loan-to-value; payment-to-income is calculated against income reported by the obligor on the credit application. No code meaning is asserted without that filer's own note.

Deliberately not published

The underwriting indicator is defined in 17 CFR 229.1125 as whether the asset met the criteria for the first level of underwriting criteria used to originate it. Filers apply it inconsistently — one marks it negative on every loan in its pool, another on high-score loans carrying subvented rates — so it is recorded but is not published as an exception rate, because it cannot be read as one. Verification levels are not compared across filers, for the reason above. Loan-to-value is flagged cross-filer comparable for only some shelves, based on the reported vehicle-value source code.

Known limits

• One pool per shelf, one filing date. In a wider survey one shelf's share of loans at 0% APR ranged from 21.7% to 81.9% across its own pools. No figure here should be read as a lender's rate.
• 37 of the 329 trusts could not be resolved to an asset file within five filings.
• Shelf grouping is by name. Series that carry genuinely different credit characteristics were separated only where the difference was measured and large; series letters reflecting sequential issuance were not split.

Corrections

The underlying filings are public records, free at sec.gov. If any figure here misstates what a filing contains, we will correct it — the filing governs, not this dataset.

← Private credit standards