Federal Reserve H.8 release download: weekly bank balance sheets as delivered rows

Weekly balance sheets sit inside a five-record industry slice - Asset Management & Custody Banks - where every primary record comes from an official statistical publisher and all five ship through Datadory in the same typed-row format, scored against a catalog-wide mean of 7.81 across 1,744 datasets:

The division of labor matters more than any single record: H.8 answers what did the banking system do this week, N-PORT answers which funds held what at quarter end, and the Z.1 reconciles both against every other holder sector in one schema.

Five fields, each answering one question. Account answers what - the balance-sheet line being estimated, from bank credit down through securities in bank credit, loans and leases by category, cash assets, total federal funds sold and reverse RPs, deposits with large time deposits broken out, borrowings, and net due to related foreign offices. Bank subset answers who - all commercial banks together or one of the four tier views. Week ending / period answers when - a close-of-business Wednesday for weekly levels, calendar labels for averages and change columns. Level answers how much, in billions of dollars, seasonally adjusted or not. Percent change (annualized) answers how fast - month-, quarter- and year-over-year growth on a common footing.

The deliberate absence is as informative as the presence: no institution identifier appears anywhere. This is the system's record, not its members'.

# one week of deposits, shaped as delivered -- slot values fill in your sample
account       : deposits
bank subset   : all banks | large domestic | small domestic | foreign-related
week ending   : <Wednesday close-of-business>
level         : <billions of dollars>     sa_flag : adjusted | not adjusted
pct_change    : <month | quarter | year, annualized rate>

# the account roster repeats identically across all five views, so a tier spread
# lands as a join on account + week -- not a mapping exercise

Two structural properties reward attention before anything gets built on top. The hierarchy nests - credit cards inside consumer loans inside loans and leases inside bank credit - so account names carry parent-child structure worth preserving rather than flattening. And because the roster repeats identically across views, large-domestic-versus-small-domestic spreads become computed columns rather than manual stitches.

Coverage is drawn precisely, and the edges are part of the specification.

• Geography: the United States - domestic offices of commercial banks, offices on US military installations abroad, and US-located foreign-related institutions. Treated as foreign and therefore excluded: IBFs, Puerto Rico and Guam. Outside the universe entirely: thrifts, credit unions and consumer finance companies. If the question includes those institution classes, another record has to supply them - this one will not approximate them.
• Temporal: weekly observations from January 1973 to present, with monthly averages derived from the weeklies; each current cut also presents roughly the last five years of annual, quarterly and monthly columns beside recent weeks, so a quick read and a long build pull from the same record at different depths.
• Granularity: aggregate accounts crossed with five bank tiers at weekly frequency. No individual institutions, no sub-national geography. The trade that buys weekly speed is exactly this coarseness, and for liquidity and credit-cycle questions it is the right trade.

Four patterns recur where weekly aggregates earn their keep.

Liquidity and funding-stress monitoring. Weekly deposit levels by tier, beside borrowings and net due to related foreign offices, catch funding movement weeks before quarterly filings confirm it - which is why treasury desks accept the aggregate in exchange for 52 observations a year.

Credit-cycle tracking. Loans and leases divided into C&I, real estate by type, consumer, credit cards and auto, converted to annualized percent changes, turn "is lending accelerating" into a weekly measured answer. Contrasting the large-domestic tier against the small-domestic and foreign-related aggregations shows whether a swing concentrates in one group of institutions or washes across the system.

Nowcasts and model features. A fixed account-by-week-by-tier schema since January 1973 drops straight into feature pipelines; nothing needs reshaping week to week, and every cycle a model will encounter has happened at least once inside the record.

Citable scale numbers. Aggregate balances by tier put defensible magnitudes under client decks and published commentary on US banking-sector size and deposit trends - attributed as 'Federal Reserve Statistical Release H.8' - while joining the Z.1 Financial Accounts benchmarks the asset-management complex against the banking system and N-PORT adds which funds held what at quarter end.

Ranked by relevance scores in Datadory's persona research for this industry:

1. Investors & quants (relevance 3) - the staple high-frequency read on US bank liquidity conditions and funding stress, and a core input for rates and macro work; the workflow continues on the investors quants use cases page.
2. Data scientists & ML engineers (relevance 2) - deposit and lending features feeding models directly from January 1973 onward, with tier splits pre-aligned (data scientists use cases).
3. Journalists & academics (relevance 2) - the go-to citation for US commercial-bank deposit and lending coverage, attribution that survives editors and reviewers (journalists academics use cases).
4. Market researchers & consultants (relevance 2) - citable weekly aggregates anchoring banking-scale and deposit-trend narratives (market researchers use cases).
5. Developers & data-product builders (relevance 2) - a schema that has not changed shape since 1973 wires into monitoring pipelines without reshaping work (developers builders use cases).

One property separates H.8 from most long-history records: the figures served are current, revised in place as later weeks restate earlier ones, with no archived real-time vintages riding along in the release. Every week's restatement quietly invalidates last week's extract unless someone pinned the vintage.

That splits consumption into two modes, and Datadory ships both. Monitoring desks take the live feed - each new week as it settles onto the tape, so funding swings surface the day they print. Research desks take the opposite: a vintage-assembled panel in which past values pin once they settle, so a backtest never trains on numbers that were still moving. Which mode a team needs is a delivery configuration - decided once, changed later as a setting rather than a migration.

Files, feeds, or straight into your warehouse. Daily, weekly, or hourly - your call.

Deliveries arrive normalized to the field dictionary: the account hierarchy preserved rather than flattened, the five tier views aligned on one grid, seasonally adjusted and not-seasonally-adjusted levels kept side by side, monthly averages materialized as first-class series, and crosswalks onto Z.1 categories and FRED banking identifiers folded in when you name them. Scoping is the default - name the accounts, the tiers and the window, and the sample ships shaped to that frame in the exact production schema, so anything prototyped survives delivery intact. Get a sample of the H.8 slice.

Start with the anchor record and a sample cut to your tiers and window: Federal Reserve H.8 - Assets and Liabilities of Commercial Banks.

This page is one cluster inside the asset management custody banks data guide, which maps the industry's full nine-record landscape. The ranked shortlist sits on best asset-management-custody-banks datasets and the complete inventory on the asset-management-custody-banks data hub.