Point in Time Macro Data API for Backtests

Q: What does as_of actually do?

It filters by knowledge time. Ask for unemployment as_of a past date and you get the value that was published as of that day, not the revised figure released later.

Q: Are first releases kept after revisions?

Yes. Every vintage is stored. First prints, revisions and final values all remain queryable forever.

Q: Where does the data come from?

Official and public domain sources only, including FRED, BLS, BEA, SEC, ECB, Eurostat, OECD, BIS, CFTC and EIA. Every row carries provenance back to the release.

Q: Is there a free tier?

Yes. The free plan covers the full dataset with a trailing 24 month window and 2,500 calls a month, no card required. Point in time correctness is included on every plan.

Q: Can AI agents query it?

Yes. Paid plans include a hosted MCP server, so agents can query point in time macro data with no glue code.

What you get

What is included?

Everything ships on the same observation model as the rest of the API, with provenance on every row.

as_of on every endpoint

Pass a date and every observation returns the value that was published as of that moment.

Revisions kept forever

New releases never overwrite old ones. First prints, second estimates and final revisions all stay queryable.

Provenance on every row

Each value traces to its official source, the source URL, the raw payload reference and the ingestion run.

Official sources only

FRED, BLS, BEA, SEC, ECB, Eurostat, OECD, BIS, CFTC, EIA and the central banks. No scraped vendor feeds.

Calendar with consensus

A macro calendar with consensus estimates and first releases, so surprise is computable historically.

JSON, SDK and MCP

Cursor pagination, stable error codes, a typed TypeScript client and a hosted MCP server for agents.

The problem

Why does point in time matter?

Macro numbers get revised. A strategy backtested on today's revised series is trading on information nobody had at the time, and the result is lookahead bias that inflates every metric. The only honest way to backtest a macro signal is against the values that were actually published on each historical day.

This API separates the period a number describes from the moment it became known. Both timestamps live on every observation, so as_of queries reconstruct any past information state exactly.

GET/v1/public/observations/latest?type=macro&country=usa&indicator=unemployment_rate200 OK

{
  "data": [{
    "observationId": "obs_fred_unrate_2026-06-01",
    "indicatorId": "us_unemployment_rate",
    "country": "USA",
    "period": "2026-06-01T00:00:00.000Z",
    "actual": 4.2,
    "previous": 4.3,
    "unit": "percent",
    "vintage": "2026-07-02",
    "revised": false,
    "sourceUrl": "https://fred.stlouisfed.org/series/UNRATE",
    "attribution": "Federal Reserve Bank of St. Louis FRED/ALFRED"
  }],
  "meta": { "api_version": "v1", "pagination": { "limit": 1, "has_more": false } }
}

A real row from the live API. period is the month the number describes, vintage is when this value became known. Query any past date with as_of and the same row answers as it did then.

Coverage

Which datasets are point in time?

All of them. Macro indicators across the covered countries, the economic calendar with consensus, COT positioning percentiles, FOMC statement sentiment and daily valuation multiples built from fundamentals as they were reported on the day.

The same guarantee holds at every price tier, including free. Point in time correctness and provenance are the reason to trust the data, so they are never gated.

Questions

Common questions

The short answers, straight from the documentation.

What does as_of actually do?

Are first releases kept after revisions?

Where does the data come from?

Is there a free tier?

Can AI agents query it?

Get started

Start with the macro data, ship the rest faster

Free tier, no card. One key, one API and every macro number, with a hosted MCP server for your agent.

Get your free API key

Macro data as it was known at the time