Datasets:

OptimalSolution
/

databank-bea-io

license: other
pretty_name: US Input-Output Accounts (BEA)
tags:
  - input-output
  - supply-use
  - national-accounts
  - united-states
configs:
  - config_name: supply_detail
    data_files: datasets/bea_supply_detail/data/*.parquet
  - config_name: supply_sector
    data_files: datasets/bea_supply_sector/data/*.parquet
  - config_name: supply_summary
    data_files: datasets/bea_supply_summary/data/*.parquet
  - config_name: use_detail
    data_files: datasets/bea_use_detail/data/*.parquet
  - config_name: use_sector
    data_files: datasets/bea_use_sector/data/*.parquet
  - config_name: use_summary
    data_files: datasets/bea_use_summary/data/*.parquet
license_name: public-domain
license_link: https://www.bea.gov/help/faq/125

US Input-Output Accounts (BEA)

The US supply and use tables -- which industries make which commodities, and who consumes them -- at three levels of industry detail. Annual 1997-2023 at the Sector and Summary levels, plus the 2017 Detail benchmark.

Terms

Licence: public domain (US federal government work) (https://www.bea.gov/help/faq/125)
Attribution: Source: U.S. Bureau of Economic Analysis
Cite as: U.S. Bureau of Economic Analysis, Input-Output Accounts Data, 2024-12 release.
Source: https://www.bea.gov/itable/input-output

Redistributed by OptimalSolution LLC -- see DISCLAIMER.md. Redistribution only: no responsibility for the content, no endorsement or position, no affiliation with the publisher, no warranty. What was changed from the published data is recorded per dataset under bank.modifications and bank.changes.

Datasets

dataset	rows	coverage
`bea_supply_sector`	7,142	1997-2023
`bea_supply_summary`	41,740	1997-2023
`bea_supply_detail`	30,294	2007, 2012, 2017
`bea_use_sector`	12,665	1997-2023
`bea_use_summary`	121,974	1997-2023
`bea_use_detail`	159,538	2007, 2012, 2017

Read this before you aggregate

Both axes carry their own totals. T007, T017, T001, T005, T013-T019, VABAS and VAPRO are sums of other cells in the same table. Summing an axis without filtering *_is_derived double-counts. The manifest's default_filter drops them.
Not every non-industry code is a total. Final uses (F*), value added (V001, V003, the tax and subsidy rows), margins (Trade, Trans), taxes (TOP, SUB, MDTY) and imports (MCIF, MADJ) are real content, not aggregates. Filter on *_kind, not on 'is it in the scale'.
The published Sector tables are coarser than BEA's own Sector column. They merge manufacturing (31ND+33DG into 31G), finance (52+53 into FIRE), professional services (54+55+56 into PROF), education and health (61+62 into 6) and arts and hospitality (71+72 into 7). The scale carries both as sector and naics_sector; the build checks the merge numerically against the data.
The Sector and Summary tables differ in their final-demand columns too. Sector uses F020 and F100; Summary carries their components. Both are flagged, but they are not interchangeable.
Suppressed cells are dropped, not zeroed. BEA writes ... where a cell is not published; those rows are absent.
The Detail benchmarks spell the account codes differently. They pad to six characters (F01000 for F010, V00100 for V001) and upper-case the margins (TRADE, TRANS). Codes are stored as published, so a join across levels on the literal code will miss them; commodity_kind and industry_kind are level-independent and are the safe thing to filter on.
Other and Used are commodities only. Noncomparable imports, the rest-of-the-world adjustment, scrap and secondhand goods appear on the commodity axis with no matching industry.

Using it

library(arrow); library(dplyr)
d <- read_parquet("datasets/bea_use_summary/data/part-0.parquet")

# The core use matrix for one year, totals excluded.
d |>
  filter(year == 2023, !commodity_is_derived, !industry_is_derived,
         commodity_kind == "commodity", industry_kind == "industry") |>
  select(commodity, industry, value)