Macroeconomic Data APIs — Official Sources and Aggregators

Four different readers end up here. A researcher who needs one series — CPI, the ten-year yield, unemployment — should start at FRED and stop there unless something specific pushes them on: it has the series, a free key and an Excel add-in. A developer who needs many countries behind one API wants DBnomics, which is free, keyless and keeps every provider's own series codes. Someone backtesting a strategy that reacts to macro releases needs vintages — what the number was on the day — and that eliminates most of the listing: FRED's ALFRED side, the ECB's includeHistory and the IMF's vintage snapshots are the three free ones that document one, and FXMacroData sells the same question as its product, with a release timestamp on every row. Someone outside the United States should go to their own statistical office or central bank first, then to the ECB, Eurostat, the OECD, the IMF or the World Bank depending on how many countries and how fresh.

• One US series, free, into a notebook or a spreadsheet — FRED.
• CPI, payrolls and PPI with the agency's own preliminary flags — BLS Public Data API.
• Repo, money-fund and hedge-fund aggregates, and a daily stress index — OFR Data, keyless.
• What a figure was on a past date — FRED for US and much else; ECB Data Portal for the euro area.
• When each number became public, for 22 currencies, with a calendar and COT — FXMacroData, from $50 a month; the free tier is US only.
• Dozens of national agencies behind one query shape — DBnomics.
• Euro reference rates, the euro yield curve and €STR, daily — ECB Data Portal.
• EU prices, national accounts and labour — Eurostat.
• Rich-country comparisons — OECD Data.
• Balance of payments, government finance, reserves, WEO projections — IMF Data.
• Every economy, annual, under a licence you can build on — World Bank Open Data.
• Model forecasts, 196 countries and a calendar, in a paid product — Trading Economics, from $199 a month for API access.
• Container-shipping indicators beside the macro series — EconDB, €24.99 a month.

If you want the step-by-step version — which call, which parameter — that is how to pull an economic data series. This page is about choosing the source.

Macro used to be a job inside market data APIs and moved out on 7 October 2026. Nothing listed here sells a price you can trade on, and a reader who typed "stock market data API" was being shown Eurostat beside Databento.

Official source or aggregator: who answers for the number

The two jobs in this category are not a quality ranking. They answer different questions.

An official source is the office that computes the series or the institution that publishes it — the BLS, the OFR, the ECB, Eurostat, the IMF, the OECD, the World Bank. It owns the definition, the seasonal adjustment and the errata. It is free, it is the first place the number appears, and it covers its own patch and nothing else: the BLS is United States only, the ECB is the euro area's monetary plumbing. The OFR — the Treasury bureau that collects data for the Financial Stability Oversight Council — publishes funding-market and hedge-fund aggregates: its own repo release, money market fund totals from Form N-MFP filings, Form PF hedge-fund series with blanks left so no single filer is disclosed, and a daily Financial Stress Index that is a CSV download, not part of either API. Some of what it serves is republished — New York Fed reference rates, primary dealer statistics, Treasury yields — so it is an aggregator for part of its catalogue.

An aggregator republishes other people's series behind one API. FRED belongs here, even though a Federal Reserve Bank runs it: its 851,100 series come from 126 sources — the BLS, the BEA, the Census Bureau, the EIA, the OECD, the BIS, a dozen central banks — under FRED's own series IDs. When a FRED figure looks wrong, the fix is at the agency that published it, not in St. Louis.

What an aggregator buys you, and what it costs, comes down to four things.

One query shape instead of six. This is the real product. Every official source here speaks a different dialect, and four of them are built on SDMX — a standard whose series keys are positional and dataset-specific, and which answers a wrong key with a bare 400, or at the OECD with a 403 telling you it expected 13 key values and got 12. Generic SDMX clients such as sdmx1 work against the ECB, Eurostat, the IMF and the OECD, which takes the sting out of the protocol but not out of reading each dataset's structure first.

Codes, kept or replaced. DBnomics keeps each provider's own codes, so the ECB's daily dollar rate is EXR/D.USD.EUR.SP00.A at the ECB and ECB/EXR/D.USD.EUR.SP00.A at DBnomics — a lookup worked out on the agency's site transfers unchanged. FRED and EconDB assign their own: US CPI at EconDB is CPIUS. Shorter, and nothing you learned at the source carries over.

Lag of its own. An aggregator refreshes on its own schedule, on top of the publisher's. DBnomics runs each fetcher once a day and says new data should appear within 24 hours; on 14 September 2026 its copy of the ECB's daily rate ran through 4 September, and the project's own FAQ says many fetchers sit in error status. EconDB's US CPI stopped at April 2026 when checked in mid-September, while the BLS had August's reading out by 19 September. A quiet series and a broken fetcher look identical until you check the last observation.

A licence layered on a licence. The aggregator's terms sit on top of the publisher's, never in place of them — see below.

A series count is not a coverage claim

Every source here quotes a large number, and none of them count the same thing.

FRED's 851,100 series come from 126 sources. DBnomics reports 1.73 billion series across 47,062 datasets from 93 providers. Trading Economics headlines 20 million indicators, and its own breakdown puts 20 million of them in COMTRADE bilateral trade records and 20 thousand directly from national sources. EconDB curates 5,170 named series over 13,834 raw datasets. FXMacroData counts 104 indicators and 484 currency-indicator pairs, 21 of them with no start date. The World Bank lists 29,544 indicator codes across 71 databases, of which World Development Indicators — the database people mean — holds 1,498.

The size that matters is how much of the catalogue is still being refreshed, and the agencies are not better at this than the vendors. Of Eurostat's 8,877 datasets, 4,066 stop at 2022 or earlier and answer queries exactly as readily as the live ones. The World Bank's Doing Business database was last updated in August 2021. At EconDB, 38.4% of all datasets had not synced in over a year. Nothing in any of these responses says the series is dead.

Revisions, vintages and release calendars

Macro series are revised, and a backtest that reads today's values is reading numbers nobody had at the time. How much of the past a source lets you see decides whether you can fix that.

Vintages on the ordinary endpoint. FRED's ALFRED is not a separate product: the same key and the same series/observations endpoint return past vintages through realtime_start, realtime_end and vintage_dates. The ECB does the same with includeHistory=true, which returns previous versions of the data alongside the current one, and updatedAfter, which returns only what was added, revised or deleted since a timestamp.

Snapshots of a publication round. The IMF publishes dated vintage dataflows — 119 of its 222 — so a result built on one round stays reproducible. For the World Economic Outlook, that history is the current round plus one predecessor.

Release timestamps, sold. FXMacroData is the paid aggregator built around the question. Every row carries an announcement_datetime, the time the number became public, and revisions are kept as separate entries rather than written over the first print, so a backtest can ask what was known at 08:30 on release day. Not every timestamp was observed: where no official calendar survives, the time is derived from the series' median publication lag, can land a few days off, and is flagged per row — 997 of 1,349 US CPI rows in the vendor's own example. One derived row turns a whole response's point_in_time_safe flag false, so a strict backtest filters row by row.

Nothing at all. The BLS request has no vintage parameter. Eurostat states that its database always holds the latest version, with no record of past ones. The World Bank restates history at every quarterly refresh and publishes the list of changes after the fact. The OFR labels its Form PF series "current vintage, as of last update" and corrected its stress index in place; only its repo release keeps preliminary and final apart. From these, the only point-in-time series is the one you saved yourself on release day.

How much moves is a property of the series, not the source. At the BLS, PPI is revised monthly for four months after publication; payrolls arrive preliminary, are revised twice and then benchmarked annually; unadjusted CPI is final on release, while seasonally adjusted CPI is restated for up to five years. The BLS puts the preliminary flag in each observation's footnote — the reason to read it at the source rather than second-hand.

Poll the calendar, not the series. These numbers change on known dates, and the sources say when. FRED has releases and releases/dates endpoints for the publication calendar. Eurostat refreshes at 11:00 and 23:00 CET when there is something new. The World Bank refreshed WDI on 27 January, 8 April and 1 July in 2026. Trading Economics shows an economic calendar on its website, but its calendar API is not in any self-serve plan; FXMacroData's forward release calendar is in its $50 one.

Keys, limits, and "no published limit"

The official sources are free the way government statistics are free, not the way a freemium API is: there is no paid tier to upgrade to, and what limits exist protect the server. They are still worth knowing before you write a backfill.

• FRED — a free key from a free account, and 120 requests a minute before HTTP 429. The terms reserve the right to change that, so treat it as the current setting.
• BLS — keyless at 25 queries a day, 25 series and 10 years per query; a free registered key gives 500 a day, 50 series and 20 years. Both share 50 requests per 10 seconds, and the key expires every year.
• OECD — keyless, 60 data downloads an hour per IP, counting CSV exports from the browser on the same connection. VPN and anonymising traffic is not permitted, which matters for a cloud job.
• OFR — keyless; both API overviews say no tokens or registration are needed, and neither states a rate limit. Responses carry access-control-allow-origin: *, so a browser page can call it directly.
• ECB, Eurostat, IMF, World Bank — keyless, with no rate limit published. The limits that exist are per request: Eurostat queues anything between 500,000 and 5 million cells and refuses anything above with HTTP 413; the World Bank caps a page at 32,500 rows and a query at 60 indicators.
• DBnomics — keyless, with a fair-use cap it does not quantify. The paid tiers lift it.
• Trading Economics — no free key; the old guest key answers HTTP 410. The $199 Standard plan carries 500 API requests a month, which is a research allowance, and the hard limit is 2 requests a second.
• FXMacroData — keyless for US data only: the trailing 90 days, each release held back 15 minutes, 100 requests a day. The paid plans allow 1,000,000 requests a month at 300 a minute, and both start with a 14-day trial that needs a card.
• EconDB — a token issued without signup works on the older endpoints for about 24 hours; the v1 API answers 401 without a paid account's token.

"No published rate limit" does not mean unlimited, and on one card it means the opposite of permission: the IMF's terms prohibit bulk download by automated technology without explicit permission. Cache what you pull and pull it after a release, not on a timer.

A login is not always a key. The ECB portal account buys dashboards and notifications, not API access. The IMF's sign-in gates its Swagger console, not the data. The BLS v2 endpoint answers without a key and quietly drops the calculations and catalog metadata you asked for.

Public data still comes with a licence

Statistics paid for by taxes are not automatically free to resell, and the conditions are specific enough to break a product design. The redistribution question is the same one as for market data; the answers are different.

Open, with attribution. BLS data is public domain; the terms ask you to cite BLS and to state that it cannot vouch for anything derived after the data left its site. World Bank datasets it produces itself are CC BY 4.0 — the plainest licence in the category — with exceptions for some third-party datasets. The OECD permits commercial reuse with citation but does not warrant that it owns every right in every dataset. The OFR's own work, by federal employees, carries no copyright and its legal notice asks only for credit — but several of its datasets are other publishers' numbers, and for those the notice sends you to the copyright holder.

Free, if you tell your buyers it is free. The ECB requires that a sold document tell buyers, before they pay and each time they access it, that the information is free from the ECB. The IMF asks the same of anyone selling its data as a standalone product, then asks commercial reusers to email for permission, and bars use of its content for training language models without it.

Free, except for some rows. Eurostat's commercial permission excludes data for countries outside the EU, EFTA and the candidate list — its own examples are the United States, Japan and China. The euro-area-against-the-US chart a macro dashboard most wants is exactly what that clause covers if the dashboard is sold.

Free container, owned contents. FRED's legal notice says to contact the owner before using third-party series beyond personal use, and the S&P 500 series prohibits reproduction without written permission; copyrighted series carry the word Copyright in their notes. Any application using the FRED API must also display a notice that it is not endorsed by the St. Louis Fed. DBnomics publishes its datasets under the Open Database Licence and, in the same terms, leaves each series subject to its original provider's terms.

Paid, and narrower than the free sources. Trading Economics grants a limited, personal licence to analyse; redistribution is an Enterprise feature. EconDB publishes no terms of use at all, so nothing is granted in writing either way. FXMacroData's plans licence internal and personal use only; showing its numbers to anyone else is a $10-a-month add-on per block of 100 users or 1,000 public followers, for attributed display only, with raw downloads and API pass-through excluded even then. Its terms name no jurisdiction.

Outside the United States

Coverage is where the official sources differ most, and frequency falls as country count rises.

• Euro area, daily and monthly — the ECB for rates, FX reference rates and €STR, current to the previous business day; Eurostat for HICP, national accounts and labour across the EU27, EFTA and candidate countries, monthly at best.
• OECD members, monthly at best — OECD Data: 38 members and five Key Partners, thinning fast beyond them, with most datasets revised once or twice a year.
• Most of the world, quarterly or annual — IMF Data for balance of payments, government finance, reserves and the WEO; World Bank Open Data for 217 economies, annual, and a year or more behind.
• Many agencies, one API — DBnomics mirrors 93 providers, from INSEE and the Bank of England to Statistics Japan, the ECB, Eurostat and the IMF.
• Breadth in a paid product — Trading Economics claims 196 countries; EconDB claims 242 geographies, though after the United States its best-covered countries are Poland, Sweden, Hungary, Denmark and Czechia.

Two traps on the cross-country sources. The IMF, the OECD and the World Bank publish the same headline indicators under different revision rules, and their figures for one country and year frequently disagree — pick a source per indicator and stay on it. And the IMF's WEO runs to 2031: those later years are IMF staff projections in exactly the same shape as 1980, marked only by a LATEST_ACTUAL_ANNUAL_DATA series attribute that the SDMX 2.1 CSV carries and the SDMX 3.0 CSV drops.

Old tutorials point at dead endpoints

Three of the official sources have moved or rebuilt their APIs since mid-2024, and most code you find online predates the move.

• The IMF retired its legacy portal on 5 November 2025; dataservices.imf.org no longer resolves, so old code fails at DNS rather than with a 404, and International Financial Statistics no longer exists as a dataflow.
• The OECD took stats.oecd.org offline on 1 July 2024. The bare host still answers a 301, which makes a dead integration look half-alive; the data paths return 404.
• The World Bank retired API v1 in November 2018.
• Eurostat retires dataset codes in place: prc_hicp_manr still answers 200 and stops at December 2025, and the live series moved to a dataflow with different dimension names.
• Trading Economics' guest key answers 410. Public API directories still say EconDB needs no authentication; both its endpoints answer 401.

The agencies ship no client libraries — not the BLS, the ECB, Eurostat, the IMF, the OECD, the World Bank or FRED — so the Python package with the obvious name is somebody else's. imf on PyPI is an unrelated stranger's stub. DBnomics and Trading Economics publish their own clients; EconDB's, prognosis, still targets its legacy API.

What "real-time" cannot mean here

Nothing in this category is market data, and the word has three meanings on these cards, none of them the one a trader has in mind.

Release lag. Monthly headline series arrive two to three weeks after the month; real GDP for the quarter that ended on 30 June 2026 first appeared on 26 August. The fastest official series are daily and run to the previous business day — the ECB's reference rates, yield curve and €STR.

A fixing, not a quote. The ECB's daily euro reference rates are a 14:15 CET fixing published for reference, and the ECB refuses liability for decisions taken on them. The euro yield curve is fitted, not quoted. FRED's SP500 is one daily close, no bid, no ask, no volume, and only ten years of it.

A vintage parameter. FRED's realtime_start and realtime_end choose which published version of a series you get. They are about history, not speed.

Where a vendor here does sell something live, it is market quotes on the side: Trading Economics labels instruments Live, Daily or Delayed without quantifying the delay, and its streaming is Enterprise-only. If you need a price that moves while you watch, that is market data APIs. If you want macro series beside equities and fundamentals on one screen rather than in a notebook, that is a research platform such as Koyfin. And OpenBB wraps FRED, the ECB, the IMF and the OECD behind one Python interface without being a source itself.