This node maps the plumbing of macro monitoring: where the raw numbers actually come from, which aggregators repackage them, and the data-integrity traps (revisions, vintages, frequency mismatches) that quietly poison naive analysis. The core tension is that the most convenient sources (free aggregators, the "latest value" anyone sees today) are often the least trustworthy for backtests and decisions, because they silently overwrite history. Knowing the source hierarchy — primary statistical agency, central-bank archive, commercial aggregator — is what separates a defensible macro read from a confidently-stated artifact.

The source landscape

Macro data flows through three tiers:

Tier 1 — primary statistical agencies (the originators). In the U.S., the big four are:

• BLS (Bureau of Labor Statistics, bls.gov): the Employment Situation report (nonfarm payrolls, unemployment rate), CPI, PPI, JOLTS, average hourly earnings.
• BEA (Bureau of Economic Analysis, bea.gov): GDP and its components, the PCE price index (the Fed's preferred inflation gauge), personal income and spending, trade balance.
• Census Bureau (census.gov): retail sales, housing starts/permits, durable-goods orders, construction spending, wholesale/business inventories.
• Federal Reserve Board (federalreserve.gov): industrial production/capacity utilization, consumer credit (G.19), money supply, the H.4.1 balance sheet, and the FOMC's Summary of Economic Projections.

Each agency publishes on a fixed release calendar and is the authoritative source for its own numbers (see sibling node Economic Calendars).

Tier 2 — central-bank databases (the warehouse). FRED (Federal Reserve Economic Data, fred.stlouisfed.org), run by the St. Louis Fed's Research Division, is the dominant free warehouse. FRED documentation states it carries over 800,000 time series drawn from roughly 120+ public and private sources (BLS, BEA, Census, OECD, IMF, World Bank, and others) — the figure grows continuously, so treat any exact count as approximate. FRED offers charting, a well-documented REST API, and mature third-party clients (Python's fredapi/pandas-datareader, R, Stata's import fred, Excel add-in). ALFRED (Archival FRED, alfred.stlouisfed.org) is its companion that stores vintages — the data exactly as it appeared on a given historical date. This distinction is the single most important thing on this page (see Limitations).

Tier 3 — commercial aggregators & dashboards. Trading Economics (advertises 20M+ indicators across 196 countries), Investing.com, CEIC, Bloomberg/Refinitiv (paid terminals, the institutional standard, also the source of most published consensus forecasts), MacroMicro and Yardeni Research (curated chart dashboards). International originators include OECD, the IMF (World Economic Outlook database, released each April/October; data.imf.org), the World Bank, and Eurostat.

How they're used in practice

Most macro workflows split by purpose:

• Live monitoring / event risk: an economic calendar (Trading Economics, Investing.com, or the agency calendars) for when releases land, plus the consensus forecast so you can gauge the surprise versus actual (see sibling nodes Economic Calendars and Economic Surprise Indices).
• Charting and quick context: FRED, because it standardizes thousands of series under one charting interface, supports custom transformations (YoY %, recession shading via NBER, ratios of two series), and exports cleanly.
• Reproducible research / backtesting: ALFRED or the FRED API's vintage parameters, so the analysis only "sees" what was actually known at each point in time.
• Nowcasting: model-based real-time GDP estimates (Atlanta Fed GDPNow, NY Fed Nowcast) pull these feeds automatically (see sibling node Nowcasting / GDPNow).

The practical skill is matching frequency and lag to the question: payrolls are monthly with a ~1-week lag; GDP is quarterly with multiple revisions; the Fed balance sheet is weekly; market-implied gauges (yields, fed-funds futures) are real-time and revision-free.

Adoption, debate & evidence

FRED is effectively the de facto free standard for U.S. and cross-country macro series in academia, finance, and journalism — its API underpins a large open-source ecosystem, which is strong evidence of entrenched adoption rather than marketing. There is little genuine controversy about the raw agency data (BLS/BEA methods are heavily documented and peer-reviewed). The real debates are downstream:

• Revision magnitude. First payroll prints rest on an incomplete sample: a 2025 Congressional Research Service brief reports the CES first-preliminary collection rate averaged ~60% in 2024, rising to ~89% at the second estimate and ~91% at the third — so each release incorporates more responses than the last. (Note: this is the collection rate among committed respondents; the CES response rate has fallen further, to roughly 43% in 2024 per the same source.) First prints are then revised in the two subsequent months, plus an annual benchmark revision anchored to March QCEW unemployment-insurance records — the preliminary estimate is released in late August and the final benchmark with the January data in early February, revising about 21 months of history. BEA revises GDP twice after the advance estimate, then annually, then in a comprehensive (~5-year) revision. So "the number" you act on at release is frequently not the number that ends up in the history book.
• Aggregator reliability. Free aggregators occasionally mislabel units, frequencies, or source lineage. For anything load-bearing, the discipline is to trace back to the Tier-1 originator.

Strengths & limitations

Strengths: FRED and the agency calendars give near-comprehensive, free, well-documented coverage with reproducible APIs; ALFRED uniquely solves the point-in-time problem; commercial terminals add the consensus forecasts the free sources lack.

Limitations & the #1 misuse: The look-ahead / revision trap. FRED's default view always shows the latest vintage — fully revised, benchmarked data that did not exist when a decision would have been made. Backtesting a macro signal on FRED's current series silently injects future information and inflates apparent edge; for example, real-time payrolls are documented to overstate employment in downturns and understate it in recoveries, exactly when a signal matters most. Any historical macro study must use ALFRED/vintage data or it is, at best, an in-sample illusion. Secondary traps: confusing seasonally-adjusted vs. not-adjusted series, mixing frequencies without alignment, and trusting an aggregator's relabeled copy over the originator.

Sources

• FRED — Federal Reserve Economic Data, St. Louis Fed: https://fred.stlouisfed.org/ ; "What is FRED?": https://fredhelp.stlouisfed.org/fred/about/about-fred/what-is-fred/ ; FRED API/ALFRED docs: https://fred.stlouisfed.org/docs/api/fred/alfred.html
• ALFRED (Archival FRED) — vintage data: https://alfred.stlouisfed.org/ ; St. Louis Fed Open Vault on vintages: https://www.stlouisfed.org/open-vault/2021/august/using-the-alfred-database
• BLS economic news releases: https://www.bls.gov/bls/newsrels.htm ; CES benchmark technical notes (21-month revision window, QCEW/UI anchoring, Feb release): https://www.bls.gov/web/empsit/cestn.htm
• Congressional Research Service, "Current Employment Statistics" (IF13084) — CES collection/response rates by estimate stage: https://www.congress.gov/crs_external_products/IF/HTML/IF13084.html
• BEA release schedule: https://www.bea.gov/news/schedule
• U.S. Census Bureau economic indicators & calendar: https://www.census.gov/economic-indicators/
• IMF Data / World Economic Outlook: https://www.imf.org/en/data ; https://data.imf.org/en
• Trading Economics (aggregator + calendar): https://tradingeconomics.com/ ; https://tradingeconomics.com/calendar
• Citigroup Economic Surprise Index background (consensus-vs-actual framing): https://www.fpmarkets.com/education/trading-guides/what-is-the-citigroup-economic-surprise-index/
• Wikipedia, Federal Reserve Economic Data (series-count context, third-party clients): https://en.wikipedia.org/wiki/Federal_Reserve_Economic_Data

Flag: exact FRED series counts grow continuously (currently ~845k series / ~121 sources) and are reported as approximate. CES collection/response rates are year-specific (figures here are 2024 per CRS) and have trended down since 2020; specific benchmark-revision magnitudes are likewise year-dependent.