Commercial & Residential Mortgage Finance Data: Collateral Values, Loan-Level Tapes and Agency Disclosure · Head-to-head
FHFA House Price Index (HPI) vs FRED Mortgage and Housing Data API
Which commercial & residential mortgage finance data: collateral values, loan-level tapes and agency disclosure data fits your job: FHFA House Price Index, or FRED Mortgage and Housing Data API. API, files, or your warehouse. Daily, weekly, or hourly.
FHFA House Price Index (HPI)
FRED Mortgage and Housing Data API
Coverage, side by side
| FHFA House Price Index | FRED Mortgage and Housing Data API | |
|---|---|---|
| Geographic | United States: national, nine census divisions, 50 states plus DC and Puerto Rico, MSAs and metropolitan divisions, CBSAs, counties, three-digit and five-digit ZIP codes, and census tracts; 400+ cities in summary tables | — |
| Temporal | Index history begins January 1975; all-transactions series publish before 1991; monthly purchase-only begins January 1991; annual files run to the most recent year | `observation_start` and `observation_end` bounds chosen per request |
| Granularity | One row per geography per period per index flavor with NSA and SA values where available; roughly 90,000+ observations in the consolidated master file | — |
What each contains
Pick by fit, not by loyalty.
| FHFA House Price Index | FRED Mortgage and Housing Data API | |
|---|---|---|
| Row key | `place_id` plus `place_name` - `DV_ENC` for the East North Central Division | `series_id` named per request - `MORTGAGE30US` |
| Period stamp | `yr` integer plus `period` integer (1-4 quarterly, 1-12 monthly) | `date`, one ISO observation date per row |
| Value | `index_nsa` and `index_sa` index levels, base varying by series | `value`, carried as a string in the payload |
| Frequency concept | `frequency` enum - monthly or quarterly; only USA and Census Division series are produced monthly | `frequency` reshape switch across d, w, m, q and a grains |
| History window | Every geography carried in one file back to January 1975 | `observation_start` and `observation_end` bounds chosen per request |
| Revision handling | `note` footnotes attached to specific observations | `vintage_dates`, `realtime_start`/`realtime_end` and `output_type` selecting as-first-reported or revised values |
| Uncertainty | `rstderr` relative standard error where published | Not represented |
| Unit shaping | None - values stand as index levels | `units` transforms - lin levels, chg, pc1 percent change from a year ago |
What each does better
FHFA House Price Index
Geography all the way down to the neighborhood. National and division series anchor the top, all 50 states plus DC and Puerto Rico fill the middle, and the annual files descend through counties and metropolitan divisions to three-digit ZIPs, five-digit ZIPs and individual census tracts, with more than 400 cities covered besides. Street-grain collateral work has no counterpart on the FRED side.
Seasonal adjustment is a column, not a guess. Every observation carries index_nsa beside index_sa - 100.00 and 100.87 sit side by side in the sample rows - so a model can test the adjustment rather than inherit it.
Precision travels with the estimate. The rstderr column reports relative standard error where FHFA publishes one, volatility parameter files accompany each index family, and footnote-style note values qualify individual observations.
Thin markets are labeled, not silent. Documented thresholds govern visibility: a metro needs roughly 1,000 total transactions and at least 10 in a quarter before an index value appears, so a gap reads as thin data rather than absent data.
FRED Mortgage and Housing Data API
A prepayment or affordability model wants all of these; HPI supplies none of them.
Revision history is first-class. vintage_dates, realtime_start and realtime_end, and an output_type switch select between as-first-reported and current values, so a backtest can be pinned to what was knowable on a given date instead of today's smoothed record.
Shaping lands with the numbers. units transforms - levels, period change, percent change from a year ago - and aggregation_method choices of avg, sum or end-of-period fold higher-frequency series to the grain a model actually uses.
The flagship thread is half a century long. The 30-year fixed-rate average spans April 1971 (7.33 percent) to August 2026 (6.65 percent) - the single most quoted number in American housing finance, unbroken.
The verdict
Verdict: sample both, pick by fit - they tie at 10 out of 10, so the question decides, not the leaderboard.
Valuation comps and buy-box screening, LTV drift on seasoned portfolios, index-based hedging, appreciation studies resolved to ZIP codes and census tracts - the investors quants use cases lean on exactly that shape.
Rate assumptions for prepayment models, payment-shock and affordability screens, delinquency monitoring, construction pipeline tracking through starts and permits - the data scientists use cases live here. Accept that a rate series knows nothing about the house itself.
Rule of thumb: if the question begins "what did this home sell for twice", take FHFA HPI; if it begins "what would the monthly payment be", take FRED. Both hang off the commercial residential mortgage finance data hub.
Sample both, pick by fit. See FHFA House Price Index · See FRED Mortgage and Housing Data API
Fair questions
Do the two datasets overlap anywhere?
On subject, barely; on schema, yes. Both document a key, a period stamp and one numeric value per row, so a normalized join is mechanical.
Which dataset reaches further back in time?
Depends on the series you name. FHFA HPI's index history begins January 1975, with monthly purchase-only indexes starting January 1991. On the FRED side the 30-year fixed-rate average begins April 1971, and Financial Accounts mortgage debt measures reach back as far as 1945 - the deepest single thread of the two.
Which dataset covers smaller geographies?
Its annual files descend to counties, five-digit ZIP codes and individual census tracts, with quarterly state, metro and three-digit ZIP series and more than 400 cities covered.
Can Datadory deliver both datasets together?
Yes. Either record arrives alone or merged onto one calendar: HPI appreciation by geography alongside the rate, application and delinquency series that explain it, aligned on period and geography before values mix, and delivered daily, weekly, or hourly - your call. Name the metros or ZIPs and the series you want beside them.