undata-sdmx

Five dataflows of the UNdata SDMX API of the United Nations Statistics Division (UNSD), complete as the API served them on 2026-10-06: the UNSD Energy Statistics, the UNSD energy balances, the national accounts main aggregates, and the SEEA air emission and physical energy flow accounts. 5,395,787 observations are published here, with the data structure definitions and codelists that describe them; 61,921 rows of NA_MAIN are left out because the organisation that compiled them does not allow redistribution under its own terms (see Licence).

Source: UNdata, United Nations Statistics Division, http://data.un.org (SDMX API, https://data.un.org/Host.aspx?Content=API). Reuse is allowed under UNdata's Terms of Use provided UNdata is cited; the terms do not mention commercial use (see Licence). NA_MAIN marks each row with the organisation that compiled it (Eurostat, the OECD, UN organisations, the IMF); a row is published only when that organisation's terms allow redistribution, commercial use included.

The data follows upstream corrections

The API overwrites its data in place and keeps no versions; UNSD's terms say it "periodically incorporates, without notice, revisions, updates and improvements". This dataset therefore holds one run, the newest, and is replaced as a whole when it is fetched again; older runs are not kept here. The current run is raw/2026-10-06/ (fetched 2026-10-06, 04:00 to 04:50 UTC). The date of the run is the date the values were read from the API, not a release date: the API states none.

Subsets

subset	rows	values	reference areas	time	what it is
DF_UNDATA_ENERGY	1,919,234	1,919,234	245	1990-2024	UNSD Energy Statistics (UNSD 1.2): production, trade and use of 75 energy commodities, 216 transactions, in physical units
DF_UNData_EnergyBalance	1,045,891	1,045,891	218	1990-2023	Energy balances (UNSD 1.0): 11 commodity groups by 53 balance items, in terajoules
NA_MAIN	1,161,845	1,137,010	159	1947-Q1 to 2025-Q3	National accounts main aggregates (Eurostat structure NA_MAIN 1.9), annual and quarterly; the rows compiled by the IMF (58 areas) and Eurostat's Kosovo rows are left out
DF_SEEA_AEA	134,248	17,148	7	1990-2022	SEEA air emission accounts (Eurostat structure SEEAAIR 1.3)
DF_SEEA_ENERGY	1,134,569	1,128,764	38	2000-2022	SEEA physical energy flow accounts (Eurostat structure SEEANRG 1.3)
codelists	9,907				every code of every codelist the five structures use, with its English label
components	120				each flow's dimensions, measure and attributes, with concept name and codelist
compilers	726				rows per flow, compiling organisation and reference area, whether they are published, and why

"rows" is the number of observations published. With the rows left out (compilers), each flow's total equals the count the API states in its available constraint (annotation sdmx_metrics). "values" is the rows whose OBS_VALUE is a number; the others are NULL (see below).

A reference area is a country, territory or aggregate. In the two energy flows REF_AREA is a numeric M49 code (392 is Japan); in the other three it is a two-letter code from the IMF/SDMX area list (JP), which also has aggregates such as B6 (EU27), I8 (euro area 19), 6O (OECD) and R22 (G20).

Columns

Each dataflow subset has the columns of its SDMX-CSV answers, in their order: DATAFLOW (for example UNSD:DF_UNDATA_ENERGY(1.2)), the dimensions, TIME_PERIOD, OBS_VALUE, then the attributes. Codes are kept as codes; labels are in codelists.

column	type
OBS_VALUE, CONVERSION_FACTOR, PRE_BREAK_VALUE	DOUBLE
UNIT_MULT, DECIMALS	INTEGER
every other column, including TIME_PERIOD (1990, 2020-Q1)	VARCHAR

Types come from the column names, not from inference; each value was checked against a regular expression (-?[0-9]+ for INTEGER, -?[0-9]+(\.[0-9]+)?([eE][-+]?[0-9]+)? for DOUBLE) before the cast, and one that did not fit would have stopped the build. An empty cell is NULL. In OBS_VALUE, "NA" and ":" are NULL too: the answers write "NA" (111,167 rows of DF_SEEA_AEA) and ":" (573 rows of DF_SEEA_AEA, 40 of DF_SEEA_ENERGY) for an observation without a value, and leave it empty in 5,360, 5,765 and 24,841 rows of DF_SEEA_AEA, DF_SEEA_ENERGY and NA_MAIN (24,835 of them published). Those rows are kept because their attributes may say why the value is missing (OBS_STATUS, for example L, M or Q in NA_MAIN). The raw files keep the tokens as sent.

Values are in the unit of UNIT_MEASURE (UNIT in the balances) times 10 to the power UNIT_MULT: Japan's GDP at current prices for 2024 is OBS_VALUE 609642500 with UNIT_MEASURE XDC (national currency) and UNIT_MULT 6, that is 609,642,500 million yen. UNIT_MULT differs between flows and series (0, 3 and 6 occur), so read it on the row.

components has (dataflow, dataflow_version, component, role (dimension, time, measure, attribute), position, assignment (Mandatory or Conditional), attached_to (observation, series or dataset), concept_name, codelist_agency, codelist_id, codelist_version). codelists has (codelist_agency, codelist_id, codelist_version, codelist_name, code, name, description, parent). The flows use different versions of lists with the same id (CL_AREA 1.8 and 1.15, for example), so join on all three of agency, id and version, through components.

compilers has (dataflow, compiling_org (the COMPILING_ORG code, NULL where the flow or row has none), organisation, ref_area, rows, published, reason (the organisation's terms in a sentence, with their URL)), one row per flow, organisation and area, covering every row fetched, published or not.

-- DuckDB, straight off the Hub: Japan's 2023 energy statistics with labels
WITH lab AS (
  SELECT k.component, l.code, l.name
  FROM 'hf://datasets/yuiseki/undata-sdmx/parquet/components.parquet' k
  JOIN 'hf://datasets/yuiseki/undata-sdmx/parquet/codelists.parquet' l
    USING (codelist_agency, codelist_id, codelist_version)
  WHERE k.dataflow = 'DF_UNDATA_ENERGY'
)
SELECT d.TIME_PERIOD, c.name AS commodity, t.name AS transaction,
       d.OBS_VALUE, d.UNIT_MEASURE, d.UNIT_MULT
FROM 'hf://datasets/yuiseki/undata-sdmx/parquet/DF_UNDATA_ENERGY.parquet' d
JOIN lab c ON c.component = 'COMMODITY'   AND c.code = d.COMMODITY
JOIN lab t ON t.component = 'TRANSACTION' AND t.code = d.TRANSACTION
WHERE d.REF_AREA = '392' AND d.TIME_PERIOD = '2023'
ORDER BY commodity, transaction;

-- Japan's annual GDP at current prices, national currency
SELECT TIME_PERIOD, OBS_VALUE, UNIT_MULT
FROM 'hf://datasets/yuiseki/undata-sdmx/parquet/NA_MAIN.parquet'
WHERE REF_AREA = 'JP' AND FREQ = 'A' AND STO = 'B1GQ' AND PRICES = 'V'
  AND UNIT_MEASURE = 'XDC' AND TRANSFORMATION = 'N'
ORDER BY TIME_PERIOD;

Raw files

raw/2026-10-06/ holds the API's answers byte for byte, except the 59 answers with rows left out (58 NA_MAIN areas compiled by the IMF and Eurostat's XK), which are not uploaded; MANIFEST.json still lists them with their hashes:

• structure/dataflow.xml: the API's list of 15 dataflows
• structure/datastructure_<agency>_<id>_<version>.xml: each DSD with its concept schemes and codelists (?references=children)
• structure/availableconstraint_<flow>.xml: the codes in use, the time range and the observation count of each flow
• data/<flow>/<area>.csv: one SDMX-CSV answer per reference area (726 fetched, 667 uploaded)
• MANIFEST.json: for every file its URL, bytes, sha256, Content-Type and fetch time (the API sends no Last-Modified), and for every flow what the API stated about it

The data was requested as SDMX-CSV (Accept: text/csv), one request per reference area listed in the flow's available constraint, with the key fixing REF_AREA and leaving every other dimension open.

Things to know

• Who compiled the values. The data carries attributes for this in the three Eurostat-structured flows. In NA_MAIN, COMPILING_ORG is filled on every row: European Commission (including Eurostat) for 799,531 rows and 41 areas, OECD for 277,347 rows and 22 areas (among them the United States, Japan, China, India and the OECD and G20 aggregates), UN organisations for 88,848 rows and 97 areas, and the IMF for 58,040 rows and 58 areas. The IMF rows and Eurostat's 3,881 rows for Kosovo are not published (see Licence). DISS_ORG (disseminating organisation) is the same except that it is empty for the OECD rows. DATA_COMP is empty. In the two SEEA flows COMPILING_ORG and DISS_ORG are empty; DF_SEEA_ENERGY covers 38 areas: 34 European countries, the EU27 aggregate B6, Indonesia, Peru and Samoa. The two UNSD energy flows carry no such attribute; FOOTNOTES in the balances is empty. No structure or answer carries licence text.
• NA_MAIN mixes frequencies (FREQ A and Q), adjustments (ADJUSTMENT N and Y), price bases (PRICES) and units; filter on them before adding anything up. COMMENT_DSET is "Validated at 20220401_070001" on the 17,898 rows of the aggregate B5.
• DF_SEEA_AEA has 7 areas (Australia, Canada, Colombia, Indonesia, South Korea, New Zealand, Ukraine), and most of its rows have no value: the Indonesian and Ukrainian answers list every series and fill few. COMMENT_TS notes that Indonesia's accounts "include combustion related emissions only and do not include fugitive emissions from fuels" (1,221 rows) and, for other rows, "only cover energy-related emissions" (2,937).
• CONVERSION_FACTOR in DF_UNDATA_ENERGY is the factor UNSD gives with the value (for example the calorific value of a fuel), empty on 82,847 rows.
• Some balance values are floating-point residues such as -2.7284841053187847E-12, and some are "-0"; they are kept as sent.
• Every dimension code in the five subsets is in its codelist.

Not included

• The other dataflows of the API: DF_UNData_UNFCC (GHG inventory), DF_UNDATA_WDI, DF_UNData_UIS and DF_UNDATA_MDG answered every query with HTTP 500 and hold other agencies' data; DF_UNDATA_COUNTRYDATA returned no records; the sector accounts NASEC_* were not taken.
• The SDG global database (DF_SDG_GLH): UNSD publishes it itself at https://unstats.un.org/sdgs/dataportal, which is the place to take it from.
• The UNdata databases that the API does not serve (Demographic Statistics, National Accounts Official Country Data, Environment Statistics and others).

Checked before upload

scripts/02_verify.py in the source repository checks:

• all 737 raw files have the size and sha256 in MANIFEST.json; every structure file is well-formed XML
• every SDMX-CSV answer, read with Python's csv module, has the header its DSD gives, the right DATAFLOW and REF_AREA on every row, and the row count in the manifest; every planned area has its file; no slice failed
• the rows of each flow add up to the observation count the API states (all five equal)
• every cell of each dataflow subset equals the CSV's, compared one by one in file and row order (text as text, numbers as Python reads them, empty, "NA" and ":" as NULL), counting only the CSV rows whose compiling organisation is published
• compilers has the CSV's row count and the export's decision for every flow, organisation and area
• every dimension code is in its codelist
• a DuckDB with extension autoloading turned off opens every file, and no row group has a bloom filter

Licence

UNdata, Terms and Conditions of use (https://data.un.org/Host.aspx?Content=UNdataUse, read on 2026-10-06):

Terms of Use

All data and metadata provided on UNdata’s website are available free of charge and may be copied freely, duplicated and further distributed provided that UNdata is cited as the reference.

Disclaimers

The data and metadata presented on UNdata are supplied as contained in the source database without any addition, subtraction, amendment, or modification by the United Nations Statistics Division.

Information about the quality or limitations of the data and metadata should be obtained from the organization responsible for the source database.

The API page says "The API is governed by UNdata Terms of Use". The terms name no standard licence and do not mention commercial use: they neither permit nor forbid it. They also say nothing about modified or derived works. The card declares license: other for this reason.

The terms speak for UNdata as a whole and do not mention the rights of the organisations that compiled the data. NA_MAIN names the compiling organisation on every row, so each row is also read under that organisation's terms (read on 2026-10-06, not legal advice), and published only when they allow redistribution, commercial use included:

organisation (COMPILING_ORG)	published	areas	rows	reason
UNSD (the four other flows, which carry no compiling organisation)	yes		4,233,942	UNSD's own data under the UNdata terms above
European Commission, including Eurostat (4D0)	yes	40	795,650	Eurostat statistical data may be reused, commercially too, with the source acknowledged (https://ec.europa.eu/eurostat/help/copyright-notice)
OECD (6O0)	yes	22	277,347	the OECD's terms allow reuse, including commercial, with attribution (https://www.oecd.org/en/about/terms-conditions.html)
UN organisations (1B0)	yes	97	88,848	UN data on UNdata, under the UNdata terms above
IMF (1C0)	no	58	58,040	the IMF's terms need permission for commercial reuse and for re-publishing a substantial amount (https://www.imf.org/en/about/copyright-and-terms)
European Commission, Kosovo (4D0, XK)	no	1	3,881	Eurostat's notice excludes from commercial reuse the data of countries other than EU members, EFTA members and official EU candidate countries; Kosovo is a potential candidate

The rule is compiler_status() in scripts/03_export_parquet.py of the source repository; an organisation it does not know is left out until its terms are checked. compilers gives every flow, organisation and area its decision and reason.

This dataset is not made or endorsed by the United Nations or UNSD. What was changed: the SDMX-CSV answers were concatenated per dataflow, typed, the rows of the organisations above that are not published left out, and written as Parquet as described above, and the codelists and components were read from the structure messages; no value was changed. The uploaded raw files are unchanged. LICENSE beside this file says the same.

Citation

As the terms ask, cite UNdata:

UNdata, United Nations Statistics Division. SDMX API, dataflows DF_UNDATA_ENERGY, DF_UNData_EnergyBalance, NA_MAIN, DF_SEEA_AEA and DF_SEEA_ENERGY, http://data.un.org, accessed 2026-10-06.

and, for NA_MAIN, the organisation in COMPILING_ORG (Eurostat or the OECD) as their terms ask.