SDMX (Statistical Data and Metadata Exchange)

SDMX is a shared standard, not a data source in itself. Seven international bodies (the BIS, ECB, Eurostat, IMF, OECD, UN, and World Bank) created it so that statistical agencies could exchange data and metadata in one common shape rather than each inventing their own format. What it defines is a consistent way to describe a time series: the observations themselves, the dimensions that slice them (country, year, indicator, unit), and the metadata that explains what each number actually measures.

In practice you rarely download SDMX in bulk. Instead you call an agency's SDMX-compliant REST API and get back either the older SDMX-ML (an XML format) or the newer SDMX-JSON. A request is built from a dataflow identifier plus a key that pins down the dimensions you want, for example a single indicator for one country over a date range. Libraries such as `pandasdmx` in Python turn those responses straight into tidy data frames, which saves you writing an XML parser by hand.

For RAG, the value is reach across sources. Because Eurostat, the ECB, the OECD, the IMF, and others all speak SDMX, learning it once lets you write a single ingestion path that pulls figures and their definitions from many agencies, then turn each series and its metadata into text passages your system can retrieve and cite. The structured metadata is what stops a bare number losing its meaning once it lands in a vector store.

The caveat is that SDMX is a protocol, so quality, coverage, licensing, and API quirks all belong to the agency serving the data, not to SDMX. Endpoints differ in how strictly they follow the spec, and rate limits vary. As an open standard it places no restrictions of its own, but always check the individual provider's licence before you reuse their figures. Think of SDMX as the connective tissue that sits alongside the statistics-and-economics sources in this catalogue rather than a dataset you consume directly.