Methodology

Source

All figures come from the Home Mortgage Disclosure Act loan-level register published by the Consumer Financial Protection Bureau. HMDA requires most mortgage lenders to report every application they receive, and the public release is modified to protect borrower privacy.

We use the most recent published year, filtered to loans that were originated — applications that were denied, withdrawn or closed for incompleteness are excluded, because these pages describe the terms borrowers actually received. The exact vintage and build date are stated at the top of this page.

Lender names come from the CFPB’s filer list for the same year, matched on Legal Entity Identifier.

What we compute

Median rate — the median of interest_rate across originated loans in the area.

Median lender fees — the median of total_loan_costs. This is what the lender charged to make the loan: origination, underwriting, points and related charges. It excludes third-party costs such as title and appraisal.

Median spread over APOR — the median of rate_spread, the difference between the loan’s APR and the average prime offer rate for a comparable transaction. It is the cleanest available measure of how far a loan sat above prevailing market pricing.

Paid discount points / received lender credits — the share of loans where those fields are populated with a non-zero amount.

Thresholds

Sparse data produces confident-looking nonsense, so we publish only what the records support:

• an area gets a page only with 300 or more originated loans;
• an individual metric is shown only with 100 or more non-missing values, otherwise it renders as a dash;
• a lender appears in a table only with 50 or more loans in that area.

These thresholds apply per metric, not per page. A rural county can have plenty of loans while fee reporting is much thinner — in that case the rate is shown and the fee column is blank.

Known limitations

No credit scores. HMDA does not publish them. Since credit score is the largest single driver of what a borrower is offered, none of these medians can be read as pricing, and differences between lenders may reflect differences in who they serve rather than what they charge for the same customer.

Fee reporting is incomplete. total_loan_costs is populated for roughly two-thirds of originations. Some institutions have partial exemptions and report Exempt; purchased loans often lack the field entirely. Each page states the share of loans with fees reported.

Rates outside 2%–25% are excluded. A small share of records carry rates that cannot describe ordinary mortgage lending — one lender’s filings put its median at 1.0%. These are usually subsidised programmes, down-payment-assistance second liens, or filing quirks. Left in, they turn a table of market rates into a statement that is simply untrue, and a footnote does not undo that. So such values are dropped before medians are computed. In Iowa this removes about 1.8% of records; where it leaves a group below the thresholds above, nothing is published for it.

Fees are published as filed. Unlike rates, we do not clean them: $0 in lender fees is entirely plausible at a credit union and is itself worth seeing. Only negative values are discarded. Individual outliers are therefore best treated with suspicion rather than as findings.

Annual data, published with a lag. The register covers a calendar year and is released the following spring. It is a picture of a completed year, not of today’s market.

Medians, not distributions. Half of borrowers paid more than the figure shown and half paid less. A median tells you nothing about the spread around it.

Corrections

If you are a lender and believe a figure here misrepresents your filings, write to corrections@apr.net with the page and the figure in question. Two things are worth knowing before you do.

We publish the public HMDA record as filed. If the underlying filing is wrong, the correction belongs with the CFPB first — we will pick it up at the next refresh. If our aggregation is wrong, we will fix it and say so.

Every figure states the number of loans behind it precisely so it can be checked. The source data is public and anyone can reproduce these numbers.

What these pages are not

They are not offers of credit, quotes, or recommendations, and they are not statements about any lender’s conduct or compliance. They are counts and medians of public records.

Differences in outcomes between groups of borrowers cannot be attributed to discrimination on the basis of this data — the variables that would be needed to support such a claim, credit score foremost among them, are not in it.