Appendix ` — Data Card — FRED (Federal Reserve Economic Data)

Provider & what it is

FRED is the Federal Reserve Bank of St. Louis’ free public database of economic time series. Here is the trick to using it well: FRED is mostly an aggregator, not the original source. It gathers series that other agencies actually produce — the unemployment rate from the Bureau of Labor Statistics (BLS), GDP from the Bureau of Economic Analysis (BEA), Treasury yields from the U.S. Treasury, mortgage rates from Freddie Mac — and re-publishes them under one clean, stable API with one consistent date index. That convenience is exactly why CONVENTIONS §6 makes you cite the underlying agency, not just “FRED”: FRED is the delivery truck, BLS/BEA is the factory. What you get per series is a tidy two-column table — a date and a value — at the series’ native frequency, plus rich metadata (units, seasonal adjustment, frequency, the source). FRED also has a sibling, ALFRED (ArchivaL FRED), which serves vintages: the value of a series as it was reported on a past date, before later revisions. ALFRED is the difference between a reproducible point-in-time study and an accidental look-ahead.

Coverage

Hundreds of thousands of U.S. and international macro/financial series. Frequencies range from daily (e.g. Treasury constant-maturity yields, DGS10) through weekly (e.g. the 30-year mortgage rate, MORTGAGE30US) to monthly (e.g. the unemployment rate, UNRATE; CPI) and quarterly/annual (e.g. GDP, GDP). History depth is per series — some monthly series reach back to the 1940s–50s; many newer indicators start far later. Each series page states its own span, units, and seasonal-adjustment status; read it rather than assuming.

Key identifiers

The one identifier you need is the series ID — a short uppercase string such as UNRATE, MORTGAGE30US, DGS10, CPIAUCSL, or GDP. It is the primary key you query on and the thing you record in your code and log. There is no cross-record join inside a single series (it is just a time series); the crosswalk that bites is the one to another dataset — e.g. lining FRED’s monthly index up to a daily price panel, or merging a county-level FRED series to NOAA/FEMA county codes. That alignment (frequency conversion, date joins) is a Chapter 7.4 task and out of scope for this card.

Access path (Python; key via env)

Two clean paths. The lightweight one needs no key; the fuller one needs a free API key, read from the environment (never hard-coded — CONVENTIONS §5):

# Path A — pandas-datareader, no API key required.
import pandas_datareader.data as web
unrate = web.DataReader("UNRATE", "fred", start="2000-01-01", end="2026-05-15")

# Path B — fredapi, free key from the environment.
import os
from fredapi import Fred
fred = Fred(api_key=os.environ["FRED_API_KEY"])   # export FRED_API_KEY=... in your shell
mortgage = fred.get_series("MORTGAGE30US", observation_start="2000-01-01")

# Point-in-time / reproducible: pull a VINTAGE via ALFRED, not the latest value.
gdp_vintage = fred.get_series("GDP", realtime_start="2026-05-15", realtime_end="2026-05-15")

Get the free key from the St. Louis Fed’s API page; set it once with export FRED_API_KEY=... (or a git-ignored .env). If os.environ["FRED_API_KEY"] raises KeyError, that is the system refusing to run until you supply the key out-of-band — exactly the behavior you want.

License & GMU-infra note (§5)

Free and public — the opposite of the licensed WRDS sources in this appendix. FRED carries no GMU-infrastructure restriction: you may pull it on a laptop, cache it, and (unlike CRSP or OptionMetrics) commit the cached raw file to git if you wish. The discipline of CONVENTIONS §5 still applies uniformly anyway — pin the snapshot/vintage, cache to data/raw/fred_*.parquet, and log the pull in logs/pulls.jsonl with a content hash — because reproducibility, not licensing, is the point. Two honest caveats remain: keep the API key in an env var (§5, it is a credential), and honor the redistribution courtesy — when you re-publish a FRED series, cite the underlying source (BLS, BEA, Treasury, Freddie Mac), per §6.

Gotchas

• Series get revised. GDP, payrolls, and many others are restated months after first release. A “pull the latest” today gives different numbers next quarter. For any point-in-time or event- study design, pull an ALFRED vintage, not the current value — otherwise you are quietly using data that did not exist on your event date (look-ahead bias).
• Cite the factory, not the truck. “Source: FRED” alone violates §6. Cite BLS/BEA/Treasury and give the FRED series ID for reproducibility.
• Frequency and seasonal adjustment. UNRATE is monthly and seasonally adjusted; DGS10 is daily; mixing frequencies without explicit resampling silently misaligns your panel. The same concept often exists in both SA and NSA versions under different IDs — pick deliberately.
• Rate limit. Roughly ~120 requests/minute per key [CHECK current published limit]; batch your series and cache so a re-run reads from disk, not the API.
• Discontinued/renamed series. Series can be discontinued or superseded by a new ID; a script hard-coding an old ID will silently return stale or empty data. Pin and log the ID you used.

“First 10 rows” schema sketch — UNRATE, monthly (illustrative — not real data)

date	UNRATE
2024-01-01	3.8
2024-02-01	3.9
2024-03-01	3.8
2024-04-01	4.0
2024-05-01	4.0
2024-06-01	4.1
2024-07-01	4.2
2024-08-01	4.2
2024-09-01	4.1
2024-10-01	4.1

(Values are fabricated for shape illustration only — do not cite these as real unemployment data.)

Which chapter / lab uses it

This is the worked example for the data-card template in Lab 7 (Your Data, Reproducibly), the headline free source in Chapter 7.2 (Data Acquisition in Practice), and the dataset behind Capstone 5 — FRED Macro Event Study (Week 8), where macro/monetary-announcement series drive a proper event-study inference. It is Priya’s macro-control source for her Week-7 climate-and- insurance DiD (NOAA storm events + FEMA declarations + a FRED/CPI insurance-cost proxy), and the generic macro-control layer for any cast member’s capstone. Pull it the §5 way: pinned (ALFRED vintage when point-in-time matters), cached, logged — and cite the underlying agency.