FRED - Federal Reserve Economic Data (St. Louis Fed)

It is the reference library of American macroeconomic statistics, kept by the Research Department of the Federal Reserve Bank of St. Louis and online since 1991. The Fed's own description runs to 'hundreds of thousands of economic data time series from scores of national, international, public and private sources', and the collection is organized so every series stays anchored to its paperwork: is browsable by category, release, source, popularity and tags, and every series page carries its metadata sheet - originating source, release schedule, units, native frequency, seasonal-adjustment status, a last-revised timestamp and the next release date.

The span is the point. Money-market fund assets from the Federal Reserve Board's Z.1 Financial Accounts sit beside bank balance-sheet aggregates, interest rates, employment, prices, output and housing series, plus international collections contributed by other statistical agencies. Within Datadory's application-software shelf this is the macro-context record - the denominator layer our analysts reach for when a software or fintech question needs an economy attached to it. Our catalogers filed it here because its heaviest working companions are product and market datasets rather than other macro feeds.

Two captures from the verified August 2026 research pull - one modern Flow of Funds reading, one from the deep-history end of the shelf:

series_id : MMMFFAQ027S
title     : Money Market Funds; Total Financial Assets, Level
period    : Q1 2026
value     : 8289569
units     : Millions of U.S. Dollars, Not Seasonally Adjusted
frequency : Quarterly, End of Period
updated   : Jun 11, 2026

series_id : GNPCA
date      : 1929-01-01
value     : 1065.9
units     : Billions of Chained 2009 Dollars
frequency : Annual

series_id : GNPCA
date      : 1932-01-01
value     : 794.1
units     : Billions of Chained 2009 Dollars
frequency : Annual

Read aloud: American money-market funds held $8.29 trillion in total financial assets at the end of Q1 2026, on a quarterly series sourced from the Z.1 Financial Accounts and stamped with its revision time. Below it, real output per the chain-weighted national accounts falls from 1,065.9 billion in 1929 to 794.1 billion in 1932 - roughly a quarter of real output gone in three years, preserved at original observation dates. One flat schema carries both the current-money-market story and the Depression; deliveries keep that shape exactly.

Fourteen documented fields, definitions verified against live records during the August 2026 research pass. They split into four jobs: identity (series_id, title, notes), the observation pair (date, value), the window and description fields (observation_start, observation_end, realtime_start, realtime_end, frequency, units, seasonal_adjustment) and the bookkeeping pair (last_updated, popularity). Two design details matter in practice. Values arrive as strings with missing readings marked by a single dot, so gaps survive parsing instead of silently becoming zeros. And the real-time period bounds ride along per observation, which is what makes vintage analysis - what a figure looked like on a past date - possible without improvisation.

Geography - United States first and deepest, with international series contributed by national, international, public and private sources layered alongside. The US core is where the continuous daily-to-annual histories live; the international collections extend thematic reach where domestic series alone would stop.

Temporal - deeper than almost anything else on the shelf. Aggregate defaults reach to July 4, 1776, individual series start where their measurement begins - the chain-weighted output capture above opens in January 1929 - and every observation carries real-time period bounds, so historical vintages are part of the record rather than a reconstruction.

Granularity - one observation per series per period, at the series' native frequency: daily, weekly, biweekly, monthly, quarterly, semiannual or annual. Deliveries can resample onto a single grain - averages, sums or end-of-period readings - and can append derived columns such as change, percent change or year-ago values, so mixed-frequency series still land in one joinable table.

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

Pick the channel that fits the workload: an API for tools that want fresh observations on demand, files for batch work, or direct warehouse load when FRED dimensions need to sit beside your own positions. Cadence is yours too - daily suits rate and market series, weekly covers most research rhythms, hourly matters around release windows. Start smaller than a feed: get a sample cut to the series you actually model, check the rows against your question, then scale.

• Macro features and backtests - long, regularly-keyed official series drop straight into models as exogenous regressors, with revision timestamps telling you exactly how much restatement risk a window carries (data scientists x application software).
• Fund and treasury benchmarking - money-market fund assets, mutual fund flows and bank balance-sheet aggregates size the custody and cash-management addressable market quarter by quarter (investors and quants x application software).
• Market-sizing decks with a citable denominator - employment, prices and output series give consultants figures clients cannot argue with (market researchers x application software).
• Citation-grade journalism and research - quotable national statistics with named provenance behind every number (journalists and academics x application software).
• Product and fintech context layers - rate series and household credit aggregates turn a product dashboard into macro-aware commentary (competitive intel and product teams x application software).

Investors and quant researchers lead the fit: century-deep official series with explicit revision timestamps are the cleanest backtest substrate there is. Market researchers and consultants come next for denominators - every market-sizing slide needs an economy behind it. Data scientists and ML engineers get a stable identity-keyed panel that joins without reshaping. Journalists, academics and students inherit the citation default of record for US macro numbers. Competitive intelligence and product teams use rate, credit and fund-flow series as the context layer beneath product metrics.

Stated plainly, because they shape the analysis:

• It is aggregates, not microdata. Every row is a series-period observation. There are no firm-level or household-level records beneath the totals, so distributional questions route to other sources.
• Values revise, and the revision trail is part of the dataset. Official series restate as source agencies revise; the last_updated stamp and real-time bounds let you separate a movement from a restatement instead of discovering the difference in a chart.
• Mixed frequencies need deliberate alignment. A daily rate series and a quarterly Flow of Funds series describe different clocks; deliveries can resample both onto one grain, but the choice is analytical, not automatic.
• Popularity measures attention, not accuracy. The popularity field ranks what other users pull most; it is a discovery signal, never a proxy for series quality.
• Publication lags differ by series. Some series print near-real-time, others arrive months behind the period they describe; treat cross-series comparisons at the leading edge accordingly.

Because the raw artifact is a library - hundreds of thousands of series, each with its own frequency, units and revision rhythm - and every serious question wants a normalized table with the dictionary intact. Datadory collapses selected series into the fourteen-field schema above, keys every row on series and period, resolves units and seasonal-adjustment labels into joinable columns, applies upstream revisions before anything reaches you, and keeps delivering on your schedule so the archive becomes a living panel rather than a folder of exports. Browse the rest of the shelf on the application software data hub, the best application software datasets ranking, or the full catalog.

Application Software catalog neighbors that answer the questions this one raises. GitHub REST API (Repositories & Search) covers developer adoption - who builds the software - while FRED supplies the economy it sells into; the scored pairing lives at GitHub vs FRED. Similarweb Top Websites & App Intelligence and StatCounter Global Stats measure attention and platform share - demand-side context that pairs naturally with the supply-side macro here. Deeper FRED slices live elsewhere in the catalog by domain: Category 13 banking data, mortgage and housing series, housing starts and construction, and insurance-related categories. Together they cover who builds, who buys and the economy both operate in.