Appendix N — Data Card — U.S. Bureau of Economic Analysis (BEA)

Source slug: bea · Appendix C (Data Dictionary) · last touched 2026-05-28

Provider & what it is

The Bureau of Economic Analysis, part of the Department of Commerce, is the agency that builds the national accounts — GDP and everything that hangs off it. Where BLS measures prices and jobs, BEA measures output and income. Two BEA products do the work in this book. Regional GDP (the “GDP by state” and “GDP by county/metro” accounts) gives output by industry for each state, metro area, and county — the local-economy denominator that lets you ask whether a place is growing. Fixed Assets (the Fixed Assets Accounts, and the related capital-stock tables) give the value of the nation’s accumulated capital — equipment, structures, intellectual property products — by industry and by type, which is the macro counterpart to the firm-level investment a finance student usually studies. The reveal-the-trick point: BEA numbers are constructed estimates assembled from dozens of source datasets, not direct measurements, and they get revised heavily. A BEA series is a model output wearing a data costume.

Coverage

National GDP (NIPA) is quarterly and annual back to 1929 (annual) / 1947 (quarterly). Regional GDP by state is annual from 1963 and quarterly from 2005; GDP by county is annual and relatively recent (county estimates begin around 2001 [CHECK exact first county-GDP year]). Fixed Assets tables are annual, back to 1925 for some aggregates. Values come in both current dollars (nominal) and chained dollars / real terms, plus quantity and price indexes. Geography spans nation, BEA region, state, metro area, and county.

Key identifiers

BEA’s API is organized by dataset (Regional, FixedAssets, NIPA, etc.), and within a dataset you select a table (TableName, e.g. SAGDP2N for state annual GDP by industry), a LineCode (the specific row/industry within that table), a GeoFIPS (state/county/metro FIPS, with STATE, COUNTY, MSA wildcards), and a Year. So the joint key for a regional observation is roughly (TableName, LineCode, GeoFIPS, Year). GeoFIPS aligns with the Census FIPS codes — the same zero-padding-as-string discipline applies, and BEA county FIPS even reuses Census county codes, so a regional GDP table can be joined to ACS/CBP on FIPS.

Access path (API; key via env)

The BEA API lives at https://apps.bea.gov/api/data and requires a free UserID key (request it on the BEA site; it arrives by email). It is a GET with a method parameter (GetData, GetParameterValues, GetParameterList for discovering valid LineCodes).

import os, requests
key = os.environ["BEA_API_KEY"]  # env only; CONVENTIONS §5
params = {"UserID": key, "method": "GetData",
          "datasetname": "Regional", "TableName": "SAGDP2N",
          "LineCode": "1", "GeoFIPS": "STATE", "Year": "2022",
          "ResultFormat": "JSON"}
r = requests.get("https://apps.bea.gov/api/data", params=params).json()
# r["BEAAPI"]["Results"]["Data"] is the list of records

There is no first-party Python SDK we depend on; requests + pandas.json_normalize is the standard path. Many BEA headline series are also mirrored on FRED for quick exploration.

License (FREE / public) & note

BEA data are U.S. Government works in the public domain — free to use, redistribute, and republish with attribution; no row-level license and no GMU-only constraint, so they belong in a packet’s data/raw/. Cite as “U.S. Bureau of Economic Analysis, GDP by State, table SAGDP2N, retrieved [date].” The API’s terms ask for attribution and reasonable use (cache, do not hammer).

Gotchas

Heavy revisions and comprehensive (benchmark) updates. BEA revises quarterly GDP three times, then again in annual updates, then periodically rewrites the entire history in a comprehensive benchmark revision (methodology and base year change at once). The 2018 series you pulled in 2019 may not exist in today’s API. As with BLS, this is a vintage problem — record the snapshot date or risk look-ahead bias.
Nominal vs. real, and chained-dollar non-additivity. Chained-dollar (real) components do not sum to their chained-dollar total — a property of the chaining method that surprises everyone the first time. Sum nominal components; use the published real total.
LineCode is opaque. You must call GetParameterValues to learn which LineCode is “All industry total” vs. a specific NAICS sector. Guessing a LineCode silently returns the wrong industry.
County GDP is modeled and noisy. County estimates are statistically derived from state totals and are far less reliable than state numbers; treat them as indicative, not precise, and never as the dependent variable in a high-stakes design without caveat.
GeoFIPS formatting. Wildcards (STATE, COUNTY) and explicit FIPS behave differently; explicit county FIPS must be 5-digit zero-padded strings.
Table renames. Table names occasionally change across vintages; a stale TableName returns an error rather than data.

“First 10 rows” — schema sketch (ILLUSTRATIVE, not real values)

Regional GDP by state, all-industry total (SAGDP2N, LineCode 1), current dollars:

Code	GeoFIPS	GeoName	TimePeriod	CL_UNIT	UNIT_MULT	DataValue
SAGDP2N-1	06000	California	2022	USD	6	3641000
SAGDP2N-1	48000	Texas	2022	USD	6	2356000
SAGDP2N-1	51000	Virginia	2022	USD	6	649000
SAGDP2N-1	36000	New York	2022	USD	6	2053000
SAGDP2N-1	50000	Vermont	2022	USD	6	38000

Fixed Assets, net stock by type (FAAt201, illustrative LineCode), current dollars:

Code	LineDescription	TimePeriod	UNIT_MULT	DataValue
FAAt201-2	Equipment	2022	9	8200
FAAt201-3	Structures	2022	9	25600
FAAt201-4	Intellectual property products	2022	9	5400

(Illustrative magnitudes only; do not cite. UNIT_MULT = 6 means values are in millions; the chained-dollar non-additivity caveat applies if you switch CL_UNIT to real.)

Which chapter / lab / capstone uses it

Capstone 5 — FRED Macro Event Study — GDP releases are major macro announcements, and BEA’s revision schedule is the textbook case of the vintage/look-ahead trap the capstone’s inference must respect.
Week 4, Ch 4.5 (Shift-Share) and macro controls — state/metro regional GDP supplies a local-output denominator and growth control; Fixed Assets by industry supplies the capital-stock context for industry-level shocks, complementing the Census/CBP shares and BLS shifts.
Week 6, Capstone 3 — Innovation from USPTO PatentsView — BEA Fixed Assets’ “intellectual property products” line is the macro mirror of the firm-level patent-value measure (KPSS 2017), useful as an aggregate sanity check on whether a constructed innovation measure tracks national IP investment.
Week 7, Ch 7.2 (Data Acquisition in Practice) — BEA API appears as a worked free-source pull (key-by-email, LineCode discovery, vintage discipline).

# Data Card — U.S. Bureau of Economic Analysis (BEA) **Source slug:** `bea` · Appendix C (Data Dictionary) · last touched 2026-05-28 ## Provider & what it is The Bureau of Economic Analysis, part of the Department of Commerce, is the agency that builds the **national accounts** — GDP and everything that hangs off it. Where BLS measures prices and jobs, BEA measures *output and income*. Two BEA products do the work in this book. **Regional GDP** (the "GDP by state" and "GDP by county/metro" accounts) gives output by industry for each state, metro area, and county — the local-economy denominator that lets you ask whether a place is growing. **Fixed Assets** (the Fixed Assets Accounts, and the related capital-stock tables) give the value of the nation's accumulated capital — equipment, structures, intellectual property products — by industry and by type, which is the macro counterpart to the firm-level investment a finance student usually studies. The reveal-the-trick point: BEA numbers are *constructed estimates assembled from dozens of source datasets*, not direct measurements, and they get revised heavily. A BEA series is a model output wearing a data costume. ## Coverage National GDP (NIPA) is quarterly and annual back to 1929 (annual) / 1947 (quarterly). Regional GDP by state is annual from 1963 and quarterly from 2005; GDP by county is annual and relatively recent (county estimates begin around 2001 [CHECK exact first county-GDP year]). Fixed Assets tables are annual, back to 1925 for some aggregates. Values come in both **current dollars** (nominal) and **chained dollars** / real terms, plus quantity and price indexes. Geography spans nation, BEA region, state, metro area, and county. ## Key identifiers BEA's API is organized by **dataset** (`Regional`, `FixedAssets`, `NIPA`, etc.), and within a dataset you select a **table** (`TableName`, e.g. `SAGDP2N` for state annual GDP by industry), a **LineCode** (the specific row/industry within that table), a **GeoFIPS** (state/county/metro FIPS, with `STATE`, `COUNTY`, `MSA` wildcards), and a **Year**. So the joint key for a regional observation is roughly (TableName, LineCode, GeoFIPS, Year). GeoFIPS aligns with the Census FIPS codes — the same zero-padding-as-string discipline applies, and BEA county FIPS even reuses Census county codes, so a regional GDP table can be joined to ACS/CBP on FIPS. ## Access path (API; key via env) The BEA API lives at `https://apps.bea.gov/api/data` and requires a free **UserID** key (request it on the BEA site; it arrives by email). It is a GET with a `method` parameter (`GetData`, `GetParameterValues`, `GetParameterList` for discovering valid LineCodes). ```python import os, requests key = os.environ["BEA_API_KEY"] # env only; CONVENTIONS §5 params = {"UserID": key, "method": "GetData", "datasetname": "Regional", "TableName": "SAGDP2N", "LineCode": "1", "GeoFIPS": "STATE", "Year": "2022", "ResultFormat": "JSON"} r = requests.get("https://apps.bea.gov/api/data", params=params).json() # r["BEAAPI"]["Results"]["Data"] is the list of records ``` There is no first-party Python SDK we depend on; `requests` + `pandas.json_normalize` is the standard path. Many BEA headline series are also mirrored on FRED for quick exploration. ## License (FREE / public) & note BEA data are **U.S. Government works in the public domain** — free to use, redistribute, and republish with attribution; no row-level license and no GMU-only constraint, so they belong in a packet's `data/raw/`. Cite as "U.S. Bureau of Economic Analysis, GDP by State, table SAGDP2N, retrieved [date]." The API's terms ask for attribution and reasonable use (cache, do not hammer). ## Gotchas - **Heavy revisions and comprehensive (benchmark) updates.** BEA revises quarterly GDP three times, then again in annual updates, then periodically rewrites *the entire history* in a comprehensive benchmark revision (methodology and base year change at once). The 2018 series you pulled in 2019 may not exist in today's API. As with BLS, this is a vintage problem — record the snapshot date or risk look-ahead bias. - **Nominal vs. real, and chained-dollar non-additivity.** Chained-dollar (real) components do *not* sum to their chained-dollar total — a property of the chaining method that surprises everyone the first time. Sum nominal components; use the published real total. - **LineCode is opaque.** You must call `GetParameterValues` to learn which LineCode is "All industry total" vs. a specific NAICS sector. Guessing a LineCode silently returns the wrong industry. - **County GDP is modeled and noisy.** County estimates are statistically derived from state totals and are far less reliable than state numbers; treat them as indicative, not precise, and never as the dependent variable in a high-stakes design without caveat. - **GeoFIPS formatting.** Wildcards (`STATE`, `COUNTY`) and explicit FIPS behave differently; explicit county FIPS must be 5-digit zero-padded strings. - **Table renames.** Table names occasionally change across vintages; a stale `TableName` returns an error rather than data. ## "First 10 rows" — schema sketch (ILLUSTRATIVE, not real values) Regional GDP by state, all-industry total (`SAGDP2N`, LineCode 1), current dollars: | Code | GeoFIPS | GeoName | TimePeriod | CL_UNIT | UNIT_MULT | DataValue | |------|---------|---------|-----------|---------|-----------|----------:| | SAGDP2N-1 | 06000 | California | 2022 | USD | 6 | 3641000 | | SAGDP2N-1 | 48000 | Texas | 2022 | USD | 6 | 2356000 | | SAGDP2N-1 | 51000 | Virginia | 2022 | USD | 6 | 649000 | | SAGDP2N-1 | 36000 | New York | 2022 | USD | 6 | 2053000 | | SAGDP2N-1 | 50000 | Vermont | 2022 | USD | 6 | 38000 | Fixed Assets, net stock by type (`FAAt201`, illustrative LineCode), current dollars: | Code | LineDescription | TimePeriod | UNIT_MULT | DataValue | |------|-----------------|-----------|-----------|----------:| | FAAt201-2 | Equipment | 2022 | 9 | 8200 | | FAAt201-3 | Structures | 2022 | 9 | 25600 | | FAAt201-4 | Intellectual property products | 2022 | 9 | 5400 | (Illustrative magnitudes only; do not cite. `UNIT_MULT = 6` means values are in millions; the chained-dollar non-additivity caveat applies if you switch `CL_UNIT` to real.) ## Which chapter / lab / capstone uses it - **Capstone 5 — FRED Macro Event Study** — GDP releases are major macro announcements, and BEA's revision schedule is the textbook case of the vintage/look-ahead trap the capstone's inference must respect. - **Week 4, Ch 4.5 (Shift-Share) and macro controls** — state/metro regional GDP supplies a local-output denominator and growth control; Fixed Assets by industry supplies the capital-stock context for industry-level shocks, complementing the Census/CBP shares and BLS shifts. - **Week 6, Capstone 3 — Innovation from USPTO PatentsView** — BEA Fixed Assets' "intellectual property products" line is the macro mirror of the firm-level patent-value measure (KPSS 2017), useful as an aggregate sanity check on whether a constructed innovation measure tracks national IP investment. - **Week 7, Ch 7.2 (Data Acquisition in Practice)** — BEA API appears as a worked free-source pull (key-by-email, LineCode discovery, vintage discipline).