Investor Lending Data: Methodology, Sources & Limitations

Source

HMDA loan-level data — the loan-by-loan originations every covered mortgage lender must report annually under the Home Mortgage Disclosure Act — retrieved through the CFPB/FFIEC Data Browser API (ffiec.cfpb.gov/v2/data-browser-api). The current edition uses the 2025 vintage, the most recent annual release, retrieved October 9, 2026. (The CFPB rebuilds the combined file weekly as lenders resubmit and late filers come in, so a published vintage keeps moving after its release; alltwenty-six state tables are re-pulled together to a single retrieval date, so every figure reflects one consistent edition of the federal file.) Federal government data; public domain.

For a read of what one state's file actually shows, who lent on New Jersey investment property in 2025, ranked works through the New Jersey file lender by lender. It was written against the July 2026 edition — 18,316 loans and $7.88B — so its totals sit just below the current table above.

Filter definition

For each state we request originated loans (actions_taken=1) on 1–4 unit dwellings server-side, then filter locally to investment-property occupancy (occupancy_type=3), excluding open-end lines of credit and reverse mortgages. The result: closed-end loans on 1–4 unit investment properties — the rental-property loans an individual investor would actually get.

One implementation note we verified directly: the Data Browser API silently ignores an occupancy filter parameter in the request, so occupancy must be filtered locally from the full state extract. This is why the checked-in dataset files — not a live API query — are the source of truth for every table.

Lender names

HMDA identifies lenders by Legal Entity Identifier (LEI). We join each LEI to its institution name using the Data Browser's filers endpoint at the time the dataset is built. Where an LEI has no name on file, the table shows the raw LEI rather than a guess. Names appear as filed — a lender's HMDA filing name can differ from its consumer brand. A name can also change between editions without the lender changing at all: we report whatever the filer's current registered name is, so a familiar brand may appear under its legal entity name in a later refresh. The LEI is the stable identifier; match on that, not on the name.

Known limitations — read these

1. HMDA exempts temporary financing, so fix-and-flip, bridge, and hard-money lending are largely absent from these tables. A lender's rank here reflects reportable closed-end activity, not its full investor-lending footprint.
2. Lenders below the HMDA reporting threshold (roughly 25 closed-end loans per year) do not file and are invisible in this data.
3. The data is annual. The 2025 vintage is the current release; figures do not reflect activity since.

These are the same limitation strings the hmda_methodology tool on our read-only MCP server returns — one list, one wording, whether a person reads this page or an agent queries the data.

State totals (2025 investor 1–4 unit closed-end originations)

The same twenty-six files, grouped by Legal Entity Identifier rather than by state, answer the question these totals raise — who lends on investment properties across all twenty-six states, ranked by loan count with a states-reported column on every row.

Full per-state datasets (every reporting lender in each state, every underlying figure) are downloadable as CSV: CA · TX · FL · PA · NY · OH · NC · GA · IL · NJ · TN · MO · VA · MI · AZ · IN · OK · AL · MA · WA · MD · SC · CO · MS · CT · NM

Update cadence

HMDA data publishes annually. These tables update once per vintage: the 2026 data, expected in mid-2027, will be added alongside the current edition when it publishes.

A vintage is not frozen once it publishes, though. The CFPB rebuilds the combined file weekly against current submissions, so resubmissions and late filers keep raising a published year's counts for months afterward. We re-pull all twenty-six states together when that drift becomes material, which is why the retrieval date above matters as much as the vintage year.