Federal Reserve Data Download Program Data

Federal Reserve Data Download Program - the DDP - is the machine room behind the Federal Reserve Board's statistical releases. Where each release page shows you one publication, the DDP gathers roughly eighteen of them behind a single bulk-packaging facility: select the series, and the Board assembles the package. The landing page announces itself plainly - "download data related to selected Federal Reserve Board statistical releases" - and that is precisely the job.

The flagships carry the shelf. H.8 reports assets and liabilities of US commercial banks weekly, seasonally adjusted and not, with hundreds of series splitting the loan book by type. Z.1, the Financial Accounts of the United States, ships 286 tables per quarter - the entire who-owes-whom matrix, with annual figures reaching back to 1919. H.15 posts selected interest rates on business days: federal funds, commercial paper, bank prime loan, discount window credit, Treasury bills and constant maturities out to thirty years.

Around them sit the rest of the eighteen: G.19 consumer credit, G.17 industrial production and capacity utilization, G.20 finance companies, H.6 money stock measures, H.3 aggregate reserves and the monetary base, H.4.1 factors affecting reserve balances, charge-off and delinquency rates, commercial paper, foreign exchange rates (G.5/H.10), household debt service ratios, the E.2 survey of terms of business lending, and the SLOOS and SCOOS senior-lofficer lending surveys.

The design choice that matters is the spine: every package, whatever the release, resolves to the same five fields - series ID, release, observation date, value, package format. Eighteen publication schedules, one spreadsheet-shaped answer. On the Asset Management & Custody Banks shelf this record is the breadth play, and it scores 9/10 on our rubric against a catalog average of 7.81 across all 1,744 datasets.

Three of the eighteen releases, exactly as the tidy panel arrives:

release          : G.19 Consumer Credit
series_id        : DTCTL_@%A_BA.M     # total consumer credit, percent change, SAAR
2026-04 : 4.66
2026-05 : -0.25
2026-06 : 3.30

series_id        : RIFSPBCICC_N.M     # commercial bank credit card plans, all accounts, NSA
2026-03 : ND               # no published value for the month yet
2026-04 : ND
2026-05 : 20.94

series_id        : RIFLPBCIANM72_N.M  # new-auto 72-month bank finance rate, NSA
2026-05 : 6.97

release          : H.15 Selected Interest Rates
series_id        : RIFSPFF_N.B        # federal funds effective, percent per annum
1954-07-01 : 1.13           # first print on file
2026-08-19 : 3.63

series_id        : RIFLGFCY10_N.B     # 10-year Treasury constant maturity yield
2026-08-19 : 4.65

release          : Z.1 Financial Accounts of the United States
table            : S124.1.s Mutual funds (levels)
series_id        : LM654090000.Q      # mutual funds; total financial assets
1945:Q4 : 1245            # millions of dollars, end of period

Read the blocks against each other once. Total consumer credit grew 4.66% annualized in April 2026, contracted 0.25% in May, rebounded to 3.30% in June - while the revolving component swung +10.47%, -4.71%, +6.02%. Card plans printed 20.94% in May after two months of ND, which is the program telling you honestly that no value existed rather than papering one in. And the H.15 pair brackets seventy-two years of policy in two rows: 1.13% on the first day the funds-rate series was ever kept, 3.63% now.

Flags travel with values, identifiers travel with rows, and nothing is renamed in transit.

Five fields form the spine, identical in every package regardless of release:

• Series ID - the identifier for one time series in RELEASE/SUBSET/SERIES.MNEMONIC form, such as H8/H8/H8_BCBML. Because the release is encoded in the identifier and the mnemonic survives display-name changes, it doubles as the warehouse join key.
• Release - the statistical family the series belongs to, the label every downstream cut starts from.
• Observation Date - Wednesday close-of-business for weeklies, quarter-end labels like 2026:Q1 for quarterlies, plain calendar dates for the dailies.
• Value - the reported figure in the units the release defines: billions of dollars, percent, rates. Published as-is, with ND wherever the Board published nothing.
• Filetype - the packaging format the row was resolved from.

Everything else folds out under additional fields on request: the charge-off and delinquency rate panels, SLOOS and SCOOS net-percentage responses, the E.2 terms-of-business-lending survey, commercial paper outstanding, G.20 finance company credit, G.5/H.10 exchange rates, household debt service ratios, H.3 reserve aggregates, H.4.1 reserve factors, H.6 money stock measures and G.17 industrial production. Name the families your workflow touches when you request a sample and they arrive as populated columns rather than promised ones.

Geography - the United States as national aggregates, throughout. The cuts run by holder type, maturity and instrument; there is no state, metro or institution-level resolution below the release level, so firm-granularity questions need a supervisory slice layered alongside rather than instead.

Temporal - depth follows the release, and the deep end is very deep. H.8 weekly observations begin 1973. Z.1 quarterly accounts begin 1945, with annual figures reaching 1919 - a flow-of-funds record that predates most of the institutions trading against it. H.15 daily observations begin July 1954. Histories default to complete runs unless a start and end are named for the cut, which suits one-off calibration loads and wastes nobody's patience on partial archives.

Granularity - one row per series per observation period, with weekly, monthly, quarterly and annual frequencies stacked in a single record. A desk marking a book daily and a strategist reading household leverage quarterly pull from the same dictionary, which is why results stay reproducible across teams.

Against the wider Datadory catalog - 1,744 datasets, average quality score 7.81 - this record scores 9/10, carried by its breadth-plus-depth combination: eighteen releases, seven decades, one spine. Its one deducted point reflects documentation captured from observed package structure rather than a published schema.

API, files, or your warehouse. Daily, weekly, or hourly.

Channel and cadence are settings, not projects. Files suit the team loading the full Z.1 archive once and joining it to their own portfolio tables. Structured feeds suit products surfacing a funds-rate print or a card-rate benchmark inside an app. Warehouse delivery suits analysts running bank-credit screens in SQL next to their own books. Every delivery carries the field dictionary above, the sample rows and the coverage profile mapped to the releases you named - so the schema in the sample is the schema you ship against. Revisions arrive as currently revised, flagged, so a backtest never mistakes a vintage for an error.

1. Rates and macro strategists read the whole complex - funds rate, curve, credit growth, bank credit - off one shelf instead of reconciling three release formats (investors & quants).
2. Bank strategy and treasury teams benchmark their own balance sheet weekly against the H.8 system aggregates, loan book split by type.
3. Feature engineers turn four cadences into model inputs off a single series vocabulary, with ND flags riding alongside values so missing months are never silently interpolated (data scientists).
4. Product builders stand Fed-backed dashboards on a spine that does not change shape between releases (developers & builders).
5. Consultants and market researchers cite holder-type splits and credit growth in client decks with the federal record behind every figure (market researchers).
6. Journalists, academics and students cite series identifiers rather than screenshots, which is what makes a chart defensible (journalists & academics).

Get a sample of this dataset scoped to the releases and periods your work actually touches.

Quant researchers and strategists come first: the Z.1 reaches 1945 and H.8 reaches 1973, so backtests run on the official record rather than a spliced reconstruction, and the H.15 ladder prices everything they discount. Data scientists and ML engineers inherit a single series vocabulary across four frequencies - rare enough in financial data to count as a feature. Developers building data products get the cleanest integration surface the Board offers: one spine, consistent typing, no per-release surprises. Market researchers and consultants anchor financial-sector narratives in citable aggregates, and journalists, academics and students get bulk researcher-friendly form for stories and papers where the citation needs to survive review.

Within Asset Management & Custody Banks, this record owns breadth: eighteen releases behind one spine. Neighbors own different jobs. The Z.1 Financial Accounts (quality 10) goes deeper on one quarterly snapshot family - the complete sector-by-instrument matrix - where the DDP trades that depth for span. The H.8 release (quality 9) is the fastest single cadence on the shelf, weekly bank balance sheets since 1973; the head-to-head is worked through in DDP vs H.8. And FRED Category 13 Banking Data (quality 8, St. Louis Fed) re-publishes much of this universe with broader aggregation but thinner Board-native packaging.

The honest limitation cuts both ways: this is a packaging layer over the releases, not new measurements - it will never be deeper than the releases themselves. What it buys is that the releases stop being eighteen separate integration projects.

Four things worth having in hand.

First, it is a packaging layer, not new data: the numbers are the Board's numbers, units exactly as published, ND wherever the Board published nothing. Anyone expecting cleaned or modeled values is holding the wrong record.

Second, the interface has an announced sunset: the Board has flagged removal of the interactive package-builder feature and eventual retirement of the DDP front end. The releases persist; the wrapper around them is explicitly time-limited - which is precisely the churn a delivery layer exists to absorb on your behalf.

Third, documentation confidence is recorded as inferred: the five-field dictionary was captured from observed package structure rather than a published schema, so it is pinned against live packages empirically when your sample is cut rather than promised blind.

Fourth, mixed frequencies are real: weekly H.8 columns sit in the same record as quarterly Z.1 and daily H.15 series. Aligning cadences before differencing is your model's job; the flags and labels to do it correctly ship with the rows.