Commercial & Residential Mortgage Finance Data: Collateral Values, Loan-Level Tapes and Agency Disclosure · Head-to-head
Freddie Mac Single-Family Loan-Level Dataset vs American Housing Survey (AHS) National Microdata
Which commercial & residential mortgage finance data: collateral values, loan-level tapes and agency disclosure data fits your job: Freddie Mac Single-Family Loan-Level Dataset, or American Housing Survey National Microdata. API, files, or your warehouse. Daily, weekly, or hourly.
Freddie Mac Single-Family Loan-Level Dataset
American Housing Survey (AHS) National Microdata
Coverage, side by side
| Freddie Mac Single-Family Loan-Level Dataset | American Housing Survey National Microdata | |
|---|---|---|
| Geographic | United States, identified by property state, postal code and Metropolitan Statistical Area or metropolitan division | United States national sample, plus separately published metropolitan public-use files and summary cuts for selected states and metros from 2011 |
| Temporal | Originations January 1999 through March 2026 (Release 47); monthly performance disclosed through the same cutoff | Biennial odd-year surveys since 1973; current release is the 2023 National PUF v1.1 (January 2025); samples redrawn 1985, 2015 and 2025; continuous model planned from the 2025 sample onward |
| Granularity | One origination record per loan per vintage quarter plus one performance record per loan per month; Standard and Non-Standard splits | Housing-unit/household-level microdata; one row per mortgage record per unit, joined on CONTROL |
What each contains
They tie on 1 attribute. Pick by fit, not by loyalty.
| Freddie Mac Single-Family Loan-Level Dataset | American Housing Survey National Microdata | |
|---|---|---|
| Publisher | Freddie Mac | US Census Bureau, sponsored by HUD |
| Subject lens | Loan-level credit performance disclosure: origination terms, monthly servicing status, disposition and actual loss | National housing survey microdata: household, person, project and mortgage tables describing America's housing stock and its financing |
| Geography | United States, identified by property state, postal code and Metropolitan Statistical Area or metropolitan division | United States national sample, plus separately published metropolitan PUFs and Table Creator cuts for selected states and metros |
| Temporal reach | Originations January 1999 through March 2026 (Release 47); monthly performance disclosed through the same cutoff | Biennial odd-year surveys since 1973; current release is the 2023 National PUF v1.1, with a continuous-collection model planned from 2026 |
| Granularity | One origination record per loan per vintage quarter plus one performance record per loan per month; Standard and Non-Standard splits | One row per housing unit per survey wave, with a separate mortgage table joined by CONTROL; longitudinal panel of the same addresses |
| Scale | Approximately 56 million mortgages; multiple GB per year of quarterly tape; 50,000-loan sample per vintage year | 55,669 household records and 16,834 mortgage records in the 2023 national file; about 410 MB uncompressed at its largest table |
| Documented fields | 33 documented fields | 11 documented field groups |
| Definition confidence | Verified | Verified |
| Formats delivered | TXT, ZIP | CSV, SAS, ASCII |
| Rubric rating | 10 out of 10 | 9 out of 10 |
| Best for | Credit performance: delinquency, prepayment, modification and loss given default on first-lien GSE mortgages | Household finance: mortgage burden, HELOC exposure, refinancing behavior and the full universe of US housing debt |
Or take both in one feed
They stack as ledger plus context, never as a row-level merge. There is no shared key - one side anonymizes loans with identifiers like F26Q50000001, the other anonymizes households with CONTROL codes - so the combination works by calibration: estimate state-level delinquency and severity on the tape, then weight it by the survey's portrait of who holds mortgages, what they pay and what they owe on top. An analyst sizing second-lien risk can take HELOCBAL distributions from the AHS and ask which first-lien vintages on the tape would sit behind them; a modeler can benchmark aggregate cash-out volumes implied by REFICSHAMT against the tape's refinance purposes.
Join discipline, not luck: respect the different units - loan-months against interviewed housing units - reconcile the periods, monthly performance against biennial waves, and remember the two populations differ wherever government-insured and portfolio loans matter. Datadory delivers both records already reconciled to the same industry slice - sample the pair and keep whichever answers faster. Or take both in one feed.
API, files, or your warehouse. Daily, weekly, or hourly.
Fair questions
Is the Freddie Mac Single-Family Loan-Level Dataset better than American Housing Survey (AHS) National Microdata?
They score 10 and 9 out of 10 on Datadory's rubric, but the point spread measures documentation depth, not relevance. Freddie Mac wins whenever the question is credit performance: delinquency transitions, prepayment, modifications and realized loss on 56 million loans observed month after month since 1999. Sample both and match each to the question.
Do the two datasets cover the same mortgages?
No, and the overlap is thinner than the shared subject suggests. The Freddie Mac tape covers only loans the company bought or guaranteed, filtered further into Standard and Non-Standard sets by CRT-like eligibility, with no borrower interview behind any row. The AHS covers a national sample of housing units regardless of who holds the paper - including FHA, VA, portfolio and private-label loans that never touch a GSE tape - reported by the people paying them. The populations meet at the conceptual middle, first-lien mortgages on owner-occupied homes, not at the row level; there is no shared key.
Which dataset has interest rates and which has losses?
Both carry interest rates; only one carries losses. The Freddie Mac dictionary records the original note rate plus the current rate in effect each reporting period, then follows distressed loans through net sales proceeds, MI and non-MI recoveries, expenses and final actual loss at disposition. The AHS records INTRATE as the householder's report of their rate alongside payment amount and frequency, but its dollar fields stop at balances, payments, escrow and HELOC limits - a survey never observes a default, let alone a loss severity.
Which one is better for studying home equity lines of credit?
The AHS, and it is not close. Its mortgage table classifies each record as a regular loan, a home-equity lump-sum loan or a line of credit, and reports outstanding HELOC balance and credit limit, top-coded only near the top of the distribution. The Freddie Mac tape is a first-lien disclosure: subordinate financing appears solely as the gap between combined LTV and LTV at origination, and no HELOC draws, limits or balances exist anywhere in its 33 documented fields.
Can Datadory deliver both datasets together?
Yes. Either record arrives alone or both land aligned on one calendar, delivered daily, weekly, or hourly - your call. Name the vintage years, states and field families when you request the sample and it arrives pre-cut, with field definitions and coverage profiles attached. Or take both in one feed.