Methodology & Data Quality

This page documents the data sources, definitions, and quality considerations behind every chart and table in the Auto Loan ABS Data Explorer. It is written for practitioners who need to evaluate whether the numbers meet their standard of care before citing them.

1. Data Source

All content is derived from SEC ABS-EE loan-level filings submitted to EDGAR under Regulation AB II (Rule 17g-7). These are regulatory disclosures that auto-loan securitization trusts are legally required to file on a defined schedule — typically monthly — containing approximately 100 fields per loan per reporting period.

• Issuers covered: 24 active auto-loan securitizers (Ally, Ameri­Credit, Bridgecrest, BMW, CarMax, Capital One Prime, Carvana (P), Carvana (N), Exeter, Fifth Third, Ford, GM Financial, Honda, Hyundai, Mercedes‑Benz, Nissan, Santander (SDART), Santander (DRIVE), Toyota, Volkswagen, World Omni, and CarMax Select, Exeter Select, World Omni Select). Carvana’s single shelf is shown as two programs — P (prime) and N (non-prime) — which carry distinct collateral and report on different days of the month; blending them would distort both. Retired trusts (California Republic, USAA) remain available for historical access.
• Time coverage: Issuance vintages begin in 2017, though individual loans in seasoned pools can extend back to ~2010. Most recent data is typically within a few weeks of the most recent filing deadline.
• Update cadence: The pipeline runs automatically twice daily on weekdays. New filings are fetched, parsed, and merged; aggregated CSVs and the dashboard HTML are regenerated. Typical data latency is <24 hours after SEC filing.

2. Credit Score Methodology

Analysis uses the consumer credit bureau score reported in SEC ABS-EE filings, filtered to the 300–900 range. Seven of twenty-five issuers explicitly identify this as a FICO score; the remaining eighteen report a generic “bureau score” label. Industry practice strongly favors FICO Auto Score for auto lending decisions, but we cannot independently verify the score model for all issuers. Loans without a reported score (~0–15% by issuer) are excluded from the analysis.

Commercial / dealer bureau scores (obligorCreditScoreType LIKE '%commercial%') are excluded because they use a different scoring scale than consumer scores and would contaminate the low-score bands of prime issuers.

3. Metric Definitions

Delinquency (Vintage Curves + Time Series)

• 60+ DPD rate: Loans with currentDelinquencyStatus >= 60 AND balance > $10, divided by total active loans (balance > $10).
• 30–59 DPD rate: Loans with 30 ≤ currentDelinquencyStatus ≤ 59 AND balance > $10, over same denominator.
• 30+ DPD rate: Combined 30–59 and 60+.
• Current rate: Loans with dpd < 30 (or null) AND balance > $10, over total.

DPD values capped at 999 in the data pipeline; some issuers use sentinel values (e.g., SDART uses 2577, DRIVE uses 1699) for already-charged-off loans; these are excluded from the DPD bucket counts.

Losses (Cumulative Net Loss by Vintage)

• Source: Monthly 10-D servicer distribution reports (EDGAR), which state each deal’s cumulative net loss directly — the authoritative pool-level figure.
• Cumulative Net Loss Rate: Cumulative net losses (charge‑offs net of recoveries, per the servicer) ÷ original pool balance, by deal‑issuance vintage.
• View: Selectable — By Vintage (pool‑weighted across deals of the same issuance year) or By Series (individual deals).

Repossessions (1st-Time Repo Rate)

1st-time repossession = the first reporting period in which a loan’s repossessedIndicator transitions to TRUE. Each loan is counted at most once, at its first-ever repo period.

• Numerator: Count of loans whose first-ever repo period is the current reporting period, within the selected (issuer × credit-score band × new/used) filter.
• Denominator: Count of all loans active in the current reporting period in the same filter bucket.
• Why 1st-time only: The raw repossessedIndicator can flip from TRUE back to FALSE (borrower redemption) and then to TRUE again. Counting every transition would double-count these edge cases; industry convention uses 1st-time only.

Credit Score vs Default (Kaplan-Meier Survival Analysis)

• Event: Loan charge-off (zeroBalanceCode = '4').
• Censoring: Loans that paid off (zeroBalanceCode = '1') or remain active are treated as right-censored at their most recent observation.
• Duration: Months between originationDate and event (or censor) date.
• Output: Kaplan-Meier cumulative default probability as a function of months-on-book, per credit score band.

4. Scope & Boundary Notes

Note on issuers with commercial-auto exposure: For issuers that securitize both consumer and commercial auto loans (Ford, Ally), we exclude commercial-bureau-scored loans from our “Below Prime” bucket. Commercial bureau scores use a different scale than consumer FICO (typically 0–100), and including them would mix fleet and small-business loans with subprime consumer loans. For a consumer credit analysis, the stricter exclusion is more defensible. The SQL filter is: LOWER(obligorCreditScoreType) NOT LIKE '%commercial%'.

Score-boundary note: Our “Below Prime” bucket is scores 300–659 inclusive, reflecting 10-point score bucketing. Definitions that treat 660 as below prime will show slightly higher counts; the difference is typically small (∼1–4% for small-volume issuers where a repo happens to land on exactly 660).

5. Known Limitations

• Repo indicator reporting lag: Some issuers (notably Exeter during 2023–2024) updated repossessedIndicator with delay, producing apparent anomalies in monthly rates. The cumulative count is still correct; monthly timing may be shifted.
• Score model ambiguity: As noted above, 18 of 25 issuers use a generic “bureau” label instead of explicitly naming FICO. A Vantage­Score 640 may not be risk-equivalent to a FICO 640.
• Non-FICO fields not reported: SEC ABS-EE does not require disclosure of down payment amount, borrower income (only payment-to-income), employment status, housing, or dealer identity. These variables affect default risk but cannot be analyzed from this dataset.
• Vehicle valuation varies by issuer: vehicleValueAmount may reflect wholesale, retail, dealer purchase price, or another basis depending on the issuer’s servicing system. Cross-issuer LTV comparisons should be treated as directional, not precise.
• Commercial purpose loans: Some consumer-bureau-scored loans may have commercial use (e.g., sole proprietors financing work vehicles). The loan tape does not distinguish these, but visible evidence of distortion is minimal.

6. Update Schedule & Audit Trail

• Daily pipeline: Runs twice daily on weekdays.
• Weekend behavior: Pipeline does not run; data is stable Saturday–Sunday.
• Build identifier: Each generated HTML includes a build timestamp in the page header (BUILD_ID).

This methodology page is regenerated with every pipeline run and is always in sync with the data currently served on the Explorer.