Everything we hold, and everything we don’t

Every HMDA public loan record we can obtain, plus the lender and geography aggregates built from it.

GSE single-family acquisitions by acquisition year.

Retail and wholesale 1-4 family originations, loans sold, loans held for sale, repurchases and mortgage-banking noninterest income, filed quarterly by banks on Schedule RC-P of the FFIEC Call Report. History from 2006Q3, the first quarter the schedule existed.

What is missing: Banks only, and only the ones the regulator requires to file: a bank needs $1B in assets or $10M of 1-4 family mortgage banking activity, which is roughly 539 origination filers a quarter out of a 4,336-bank panel. No nonbank mortgage lender files a Call Report at all — the nonbank side of the market is covered by HMDA, not here. The representation-and-warranty reserve split (RCONL191/RCONL192) is confidential in every quarter and is not published for any filer.

Purchase (PALI) and refinance (RALI, split rate/term vs cash-out) application index levels, weekly since 2004-01-09, on both a dollar and a count basis. The only weekly series in the product — every other origination figure is annual HMDA at a ~9-month lag.

What is missing: These are INDEX LEVELS rebased to 100 at the week ending 2004-01-09, never dollars and never loan counts: the _amt/_ct suffix names the index basis. Fannie publishes no state, lender or product cut, so this series is national only.

Purchase, IRRRL and cash-out loan counts and guaranty amounts by geography from the VA Loan Guaranty Service volume reports.

What is missing: VA reports on a FISCAL year running October to September, so a VA year is not a HMDA year and the two are never placed on one axis.

CRA small-business and small-farm origination counts and dollar amounts by reporting institution and county, 2021-2024, plus the county aggregate every lender is measured against. This is the peer and market-share half of a CRA lending test: for a given lender and county it answers how many loans, how many dollars, what share went to businesses under $1M revenue, and what share of the county the lender took.

What is missing: NOT MORTGAGE DATA. CRA reporting covers small-business and small-farm credit, a different programme from HMDA under the same regulator, and the two are never summed. Only CRA-reporting institutions appear — banks and thrifts above the annual asset threshold — so an independent mortgage bank has no CRA record at all, and an absent lender must never be read as a lender with no small-business lending.

All eleven published years, 187,462,446 records, each year reconciled to the digit against the published national totals. The pre-2018 era carries its own 45-column shape: era-specific code spaces are enforced by CHECK constraints, and the four demographic columns of the 2018+ rule are structurally absent rather than fabricated.

What is missing: Company linkage reaches records through the lender's 2017 filing identity: institutions that changed respondent identity before 2018 keep their earlier years under the raw respondent name, unlinked to the company page (the pre-2018 tab states this coverage per company). No demographic derivations exist for this era anywhere in the product.

Acquisition and performance files behind the GSE surfaces.

The FHA single-family portfolio snapshot.

The HECM half of the FHA book.

Early-default and servicing performance by lender.

Aggregate mortgage-performance statistics.

What is missing: No single authoritative public source; Ginnie transfer files cover only the Ginnie book.

Issuer, pool and loan-level disclosure for the forward MBS programme.

The HECM securities programme, alongside the forward files.

What entered the programme each month, as distinct from the outstanding book.

CAS deal disclosure aggregated to servicer-month.

STACR deal disclosure aggregated to servicer-month.

What is missing: History starts 2026-01 — the earliest month obtainable from the publisher, not a loading gap.

Registered non-QM securitizations assembled from the sponsors' own EDGAR filings — Form ABS-15G third-party due-diligence disclosures, 424B5 prospectuses and shelf registrations — with sponsor/depositor joined to the entity spine (unmatched kept raw) and the named third-party-review firms rolled up as their own dimension.

What is missing: There is no public loan tape for 144A/Reg D non-QM: asset-level (ABS-EE) disclosure does not apply to these deals, and an ABS-15G discloses no deal size, so the calendar carries no UPB and the surface states both gaps rather than estimating.

26 dimensions of single-family pool composition — origination channel, first-time homebuyer, down-payment assistance, origination year, MSA, both MIP rates, FHA partial claims, low/moderate-income area, re-pooled, rural, area-median-income band and re-performing — each as loan count, UPB and pool share.

What is missing: Loaded for a MEASURED MONTH WINDOW rather than the full archive: all 77 months on disk would be ~730-750M rows / ~150 GiB against a 317 GiB database, so scope is cut on the month axis and never on the dimension axis. Record types 01/02/03/22 are held in their own tables and 04/15 are parquet-only.

Which registered funds hold whose Ginnie Mae collateral, joined on CUSIP to the pool-factor file — the buy side of the agency-MBS market.

What is missing: A fund holds a CUSIP, not a house: allocating a position across the states a pool's collateral sits in would report a holding the fund never took, so there is deliberately no geography cut. N-PORT is filed on a ~60-day lag with a further quarterly public-release delay.

The Federal Reserve's own agency-MBS position, CUSIP by CUSIP — the largest single holder on the demand side of the market the N-PORT map covers the rest of. 6,989,541 holding rows are loaded.

What is missing: HELD IN FULL AND EXPOSED NOWHERE YET. There is no screen, no API route and no entry in the DataCloud SQL catalog, so nothing on this site reads these figures today; the surface decision — where a single-holder position belongs beside the N-PORT fund map — has not been made. This row is counted as partial rather than loaded-and-serving for exactly that reason.

Depository mortgage metrics from bank and credit-union call reports.

Credit-union financials and branch locations.

Which Credit Union Service Organizations provide which mortgage services — origination, servicing, 1-4 family residential real estate — and which credit unions each is associated with, from the CUSO Registry's annual filings under 12 CFR 712 / 741.222. The census is one public search; the credit-union edges are built by sweeping every NCUA 5300 charter through the public per-charter filter, the only public path to a CUSO's credit-union list.

What is missing: An edge is a REPORTED ASSOCIATION, not a service claim: the public per-charter filter matches a credit union that OWNS and/or BUYS FROM the CUSO and does not distinguish the two, so no row may be read as "this CUSO services this credit union". Only CUSOs that filed appear; an edge whose charter has left the latest 5300 cycle is kept but unmatched; and the registry states services offered, never volumes, so nothing here is a dollar figure or a loan count.

Branch-level deposit totals for insured institutions.

ACS 5-year county estimates — the demographic overlay.

The mortgage slice of the Consumer Complaint Database.

County natural-hazard risk scores.

What is missing: Expected risk only — this is not a realised-event feed.

Quotes and filing history for the listed mortgage names.

What is missing: Quote refresh depends on an upstream that blocks datacenter egress; stale rows stay labelled stale rather than being carried forward.

Public GSE disclosure context framing the acquisition views.

Form 5500 and 5500-SF plan-sponsor filings from the EFAST2 public disclosure programme, matched to lenders by sponsor EIN — the operating entity itself, plus the listed holding company's EIN through the SEC bridge — and rendered as the per-lender headcount column on the company directory.

What is missing: An active-participant count is a headcount PROXY, never an employee census: a lender that sponsors no plan is absent entirely, a holding-company plan spans every subsidiary rather than the originator alone, and each plan year fills in over roughly two years, so a lender shows at its most recent FILED year rather than the current one. On a complete plan year the panel reaches 76.0% of active lenders and 63.2% of origination dollars.

Quarterly counts of licensed mortgage companies, branches and mortgage loan originators by state agency, plus the national state-licensed / federally-registered composition, from the published NMLS Mortgage Industry Report workbooks.

What is missing: Published AGGREGATES only. The report carries no company, branch or individual licensee record, so it cannot answer who holds a licence in a given state — that is the separate "State licensing & NMLS registry" row, which is still not loaded. The machine-readable series also stops at 2025Q1: NMLS has not published a newer quarterly workbook, and every figure is labelled with that vintage rather than carried forward.

Enforcement actions published by state banking and financial-institution regulators against licensed mortgage lenders and servicers, matched to the entity spine where the respondent resolves.

What is missing: There is no national file — each regulator publishes its own list, several retain only recent years, and a state with no rows must never be read as a state with no enforcement.

Employer layoff notices filed under the federal WARN Act, collected from each state's own published list, with mortgage-lender matches made against the entity spine.

What is missing: No national file exists; the federal DOL page only indexes the state lists. States publish on different schedules and in different formats and several keep only a rolling window, so coverage is uneven by construction.

Establishment counts, employment and average weekly wages for NAICS 5222 and 5223 — the mortgage industry's own payroll — by state and metropolitan area.

What is missing: State and metro only in this database: the county extract is a separate BLS file and is not loaded, so no county figure is derived from it.

Presidential major-disaster, emergency and fire-management declarations with their designated areas — the servicing-forbearance and collateral-damage trigger.

What is missing: A declaration carries both state-level and county-designation records; counting the latter as declarations reports 5,424 for Texas beside a table of 112, so the two are distinguished everywhere they are used.

What is missing: Obtainable; not yet loaded.

GLEIF Level-2 relationship records: direct parent, ultimate parent and international-branch links (plus fund-structure links), keyed on the LEI. CC0 public domain, refreshed from the monthly golden-copy bulk file. Joined to the entity spine on lender_master.lei to render the corporate family — parent, siblings and subsidiaries — on a company page.

What is missing: GLEIF only holds a relationship where at least one side registered it, so companies that never filed a parent disclosure show no family here, and the file carries no legal names (those live in the Level-1 data we do not ingest) so an off-spine relative shows only its LEI. Of ~483k global relationships only ~3,315 touch our mortgage spine, and ~582 of 6,563 spine lenders have a filed direct/ultimate parent.

Mortgage balances and delinquency from the FRBNY/Equifax panel.

House Price Index series.

What is missing: The state and metro slices publish on a longer lag than the national series — treat sub-national HPI as the older vintage.

Successor to the Monthly Interest Rate Survey.

Interest-rate series from the St. Louis Fed.

Building Permits Survey.

What is missing: Licensed. Not obtained — no redistribution right today.

Who subservices whose portfolio, each row carrying its evidence tier, source document and last-checked date. Nobody publishes this joined.

What is missing: A relationship is two companies and the document that ties them: the corpus holds no property, borrower or collateral column, so there is no geography lens on it anywhere in the product.

Which issuer hands a freshly issued pool's servicing to which buyer on day one, derived from Ginnie's PIIT / new-issuance transfer disclosure.

What is missing: A derivation with no loader of its own — its freshness is the Ginnie issuance load's. The transfer disclosure carries no property state, so there is no geography cut.

Who is acquiring and who is shedding Ginnie servicing, and how fast — an issuer-to-issuer graph with per-issuer velocity, derived from the servicing-transfer tape.

What is missing: The tape is keyed on CUSIP, pool and the two issuer ids with no property geography and no join to one, so the graph has no state or county lens.

Which homebuilders and brokerages own or joint-venture the lender on their own transactions, each row evidenced to the SEC filing that discloses it.

What is missing: An affiliated-business relationship is a corporate STRUCTURE, not a location: the corpus holds no state, city or county column. The captives' actual lending appears under HMDA at the affiliate's own LEI.

Which state housing finance agency runs its own Ginnie servicing and which hands it to a master servicer, with the naming document, evidence tier and last-checked date on every row.

What is missing: An HFA that is its own Ginnie issuer of record is self-evidencing; an outsourced one needs a document, and the agencies with none are shown as unresolved rather than guessed.

Federal public enforcement actions — consent orders, civil money penalties and settlements — as published by the CFPB, matched to companies on the entity spine and shown in the company regulatory record.

What is missing: FEDERAL ACTIONS ONLY, and distinct from the state mortgage-regulator orders catalogued separately: a company with no CFPB action may still carry state orders, and vice versa. The file records actions the Bureau made public; supervisory activity that never becomes a public action is by definition not in it.

Mortgage servicing complaints normalised by the servicer's own serviced-loan book — a rate rather than a count, so a large servicer is not penalised for being large.

What is missing: Honest ONLY per servicer. The numerator (CFPB complaints) carries a state and the denominator (the Ginnie issuer book) does not, so a state-filtered rate would be a rate for no population and none is offered.

What a lender looks like from outside its filings: whether it ships a consumer app and how that app is rated, how many originators it lists publicly, and whether it appears on published workplace lists.

What is missing: A reputation SIGNAL and never a measure of loan quality. Collected from public pages under each publisher's terms — derived counts and ratings only, no re-hosted listings or page content.

OPEN REQUESTS

No open requests yet. Yours would be the first.