What Is ABS-EE and Why Does It Exist

structured-finance

consumer-credit

methodology

A plain-English introduction to SEC EDGAR’s ABS-EE dataset — loan-level data from auto loan securitizations, available to anyone with the patience to parse XML.

Published

April 25, 2026

The SEC requires issuers of certain asset-backed securities to file loan-level data under a schema called ABS-EE. The filings contain performance data on the underlying collateral — in the case of auto loan ABS, that means individual car loans: origination date, current balance, days past due, payment status.

This data is public. It’s sitting on EDGAR. Most people don’t know it exists, and the ones who do know it exists are usually paid to care.

What gets filed

An auto loan ABS trust is a legal entity that holds a pool of car loans and issues securities backed by those cash flows. Every month, the servicer reports on each loan in the pool — how much is owed, whether payments are current, whether the borrower has missed payments and by how much.

Under ABS-EE, that reporting gets filed with the SEC in a machine-readable format. The format is XML. The XML is… extensive.

What the data lake does with it

The bens-data-lake pipeline downloads these filings from EDGAR’s master index, parses the XML into a structured schema, and writes Parquet files partitioned by filing month to Cloudflare R2.

SEC EDGAR
  → master.idx (index of all filings)
  → ABS-EE XML filing
  → Python parser (bens-data-lake/pipelines/)
  → Parquet (abs_ee/curated/asset_class=autoloan/)
  → R2
  → this website

The pipeline is incremental — it checks what’s already on R2 before downloading anything, so re-running it doesn’t re-parse filings that haven’t changed.

What it can and cannot tell you

Can tell you:

What percentage of loans in a given pool are 90+ days past due in a given month
How delinquency rates have changed over time across the filed universe
Which issuers are reporting, and at what loan volume

Cannot tell you:

Anything about auto loans that aren’t in a filed ABS trust (bank-held loans, captive finance company loans not securitized)
Default severity (recovery rates, loss given default) — that’s separate ABS reporting
Anything about loan-level characteristics that servicers don’t report under ABS-EE

The ABS-EE universe is a sample of the consumer auto market, not the whole thing. Keep that in mind when reading the dashboard.

Why bother

Because the data exists, it’s free, and the only cost is the time to parse it. That, apparently, is my Friday night.

--- title: "What Is ABS-EE and Why Does It Exist" date: 2026-04-25 description: "A plain-English introduction to SEC EDGAR's ABS-EE dataset — loan-level data from auto loan securitizations, available to anyone with the patience to parse XML." categories: [structured-finance, consumer-credit, methodology] type: explainer --- The SEC requires issuers of certain asset-backed securities to file loan-level data under a schema called ABS-EE. The filings contain performance data on the underlying collateral — in the case of auto loan ABS, that means individual car loans: origination date, current balance, days past due, payment status. This data is public. It's sitting on EDGAR. Most people don't know it exists, and the ones who do know it exists are usually paid to care. ## What gets filed An auto loan ABS trust is a legal entity that holds a pool of car loans and issues securities backed by those cash flows. Every month, the servicer reports on each loan in the pool — how much is owed, whether payments are current, whether the borrower has missed payments and by how much. Under ABS-EE, that reporting gets filed with the SEC in a machine-readable format. The format is XML. The XML is... extensive. ## What the data lake does with it The `bens-data-lake` pipeline downloads these filings from EDGAR's master index, parses the XML into a structured schema, and writes Parquet files partitioned by filing month to Cloudflare R2. ``` SEC EDGAR → master.idx (index of all filings) → ABS-EE XML filing → Python parser (bens-data-lake/pipelines/) → Parquet (abs_ee/curated/asset_class=autoloan/) → R2 → this website ``` The pipeline is incremental — it checks what's already on R2 before downloading anything, so re-running it doesn't re-parse filings that haven't changed. ## What it can and cannot tell you **Can tell you:** - What percentage of loans in a given pool are 90+ days past due in a given month - How delinquency rates have changed over time across the filed universe - Which issuers are reporting, and at what loan volume **Cannot tell you:** - Anything about auto loans that aren't in a filed ABS trust (bank-held loans, captive finance company loans not securitized) - Default severity (recovery rates, loss given default) — that's separate ABS reporting - Anything about loan-level characteristics that servicers don't report under ABS-EE The ABS-EE universe is a sample of the consumer auto market, not the whole thing. Keep that in mind when reading the [dashboard](../../dashboards/abs-ee.qmd). ## Why bother Because the data exists, it's free, and the only cost is the time to parse it. That, apparently, is my Friday night.