Federal Reserve Balance Sheet Methodology

Source and scope

This route reads data/parquet/fed_h41.parquet, the on-disk normalization of the Federal Reserve Board's H.4.1 Factors Affecting Reserve Balances of Depository Institutions and Condition Statement of Federal Reserve Banks release. The raw bundle is the Fed Data Download Program zip for release H41; scripts/build_staged_small.py parses the SDMX-ML <Series SERIES_NAME=...><Obs .../> structure into long-format parquet with four columns: series_name, unit, date, value.

Every value used by /flows/fed-balance-sheet is a Wednesday level stored by the source in millions of dollars. The charts convert those observations to USD trillions, while cards and table cells use compact dollar formatting such as billions or trillions. Week averages and week-over-week change series are present in the parquet and deliberately ignored here.

Primary source pages:

• Federal Reserve H.4.1 current release: https://www.federalreserve.gov/releases/h41/current/h41.htm
• Federal Reserve Data Download Program, H.4.1 release: https://www.federalreserve.gov/datadownload/Download.aspx?rel=H41

Exact line mapping

src/lib/fedH41.ts keeps the mnemonic-to-line-item map in one place. The page uses the following raw H.4.1 series:

The asset chart also publishes two derived buckets:

• Loans and facilities = RESPPALD_N.WW + RESPPAAEH_N.WW + RESH4SCS_N.WW
• Other assets (residual) = RESPPMA_N.WW - Treasuries - Mortgage-backed securities - Loans and facilities

That residual "other assets" bucket is deliberate. It makes the displayed asset legs add exactly back to total assets, while keeping the page at four asset lines instead of reproducing every small H.4.1 component separately. In current H.4.1 terms the residual contains the remaining lines such as federal agency debt securities, unamortized premiums and discounts, repurchase agreements, float, gold stock, the SDR certificate account, foreign-currency-denominated assets, and other Federal Reserve assets.

Query shape

src/lib/fedH41.ts uses the shared DuckDB helper q() from src/lib/db.ts. The parquet is already registered as DuckDB view fed_h41, so the module only has to:

1. Filter to the fixed mnemonic list above.
2. Drop the DDP sentinel -999999.
3. Pivot the chosen series into one weekly row per date with max(CASE WHEN ...).
4. Keep only dates on which every mapped series has a non-sentinel observation.
5. Derive loansAndFacilities and otherAssets after the pivot.

The page never embeds raw SQL. All numbers flow from latestWeek() and series(from, to).

Freshness and verification

• Freshness on the page is the latest date on which all mapped series have valid observations, rendered as the latest weekly vintage carried in this checkout.
• scripts/verify/checks/fed_h41.mjs exports the latest RESPPMA_N.WW value and date so P-14 can compare the local parquet against the published H.4.1 headline.

Limitations

• This is a selected-lines view, not the full H.4.1 statement. The page intentionally omits many smaller asset and liability lines.
• The route does not fetch live data. If the on-disk parquet lags the Fed's website, the page honestly shows the older parquet vintage and stamps it as such.
• The liability-side chart uses four widely watched lines, not total liabilities. RESTBC_N.WW is the table 1 currency-in-circulation reserve factor, not table 5's Federal Reserve notes liability. Reserve balances are the reserve stock; currency, reverse repo, and the Treasury General Account are factors that absorb reserve funds.