Methodology & sources

Data source

CFPB / FFIEC HMDA combined Modified LAR (loan-level public dataset, early release). We use the loan-level public file for 2025, the most recent year available. The universe is single-family 1-4 unit, site-built; home purchase + refinance + cash-out; originated & denied. HMDA is a U.S. government work in the public domain.

For 2025 we use the combined Modified Loan Application Register (LAR) — the early-release file the CFPB publishes months before the finalized Snapshot National Loan-Level Dataset. Because the Modified LAR carries raw reported codes rather than the Snapshot's derived fields, we reconstruct the single-family 1-4 unit, site-built universe from construction_method and total_units directly. Every metric is computed with the same logic used for prior Snapshot years, so the years remain directly comparable; 2025 figures will be reconciled when the 2025 Snapshot is published.

For each state we separate investor loans (the property is an investment property) from owner-occupant loans, then compute medians from the full distribution of reported values. National figures are computed from the pooled 50-state-plus-DC distribution, not as an average of state medians.

How each metric is defined

Investor rate premium

Median interest rate on investor (occupancy_type = 3) originations minus the median rate on owner-occupant (occupancy_type = 1) originations, in basis points.

Denial rate

Denied investor applications ÷ (originated + denied investor applications).

Investor share

Investor originations ÷ all single-family 1-4 originations (any occupancy) in the state.

DSCR / business-purpose share

Investor originations flagged business_or_commercial_purpose = 1 ÷ all investor originations.

Cash-out share

Investor originations with loan_purpose = 32 (cash-out refinance) ÷ all investor originations.

Median LTV

Median combined loan-to-value ratio (CLTV) on investor originations.

Coverage caveats

No dataset is perfect. These limits matter when interpreting the numbers — we state them plainly so the report can be cited with confidence.

Financed loans only

HMDA records mortgage applications and loans, so all-cash investor purchases — a large share of investor activity — are not captured. Read every figure as describing the financed market.

Business-purpose loans are partially covered

Under Regulation C (12 CFR 1003.3(c)(10)), a business-purpose loan secured by a dwelling is reported only when it is a home purchase, home improvement, or refinance. Most DSCR purchase and refinance loans qualify, but the full private-lending universe is undercounted. Treat the business-purpose share as a floor.

Agricultural loans excluded

Loans used primarily for agricultural purposes are excluded under 12 CFR 1003.3(c)(9), so farm and rural-ag investor loans drop out.

Privacy modifications

The public file rounds property value to the nearest $10,000 and bins debt-to-income and age. Interest rate and loan amount are disclosed, so the rate-premium analysis uses exact reported rates.

Small-filer exemptions

Some smaller institutions report partial data ("Exempt" fields). Rows missing a usable interest rate or loan-to-value are excluded from those medians; rate coverage exceeds 90% nationally.

Early-release vintage

The 2025 figures are built on the combined Modified LAR, the early-release HMDA product. It contains the same loan records as the later Snapshot but may shift slightly as institutions file late or resubmit. We refresh from the Snapshot once it publishes. Year-over-year comparisons hold the metric definitions constant across vintages.