Converting COREP to XBRL-CSV: a walkthrough.

What is in the package

An XBRL-CSV filing is not one file. It is a package containing a metadata file and a set of data files, one per reported table, together with the reference to the taxonomy version the report is made against. The metadata file declares which taxonomy entry point applies, which tables are present and how each data file’s columns correspond to the taxonomy’s concepts and dimensions. The data files carry the figures in plain tabular form.

The consequence for a reporting team is welcome: the artefact you prepare looks like a table, because it is one. The consequence that surprises people is that the metadata file — not the tables — is what makes the package valid, and it is generated from the taxonomy rather than written by hand.

Tables, keys and dimensions

Each COREP table becomes a data file. Every row is identified by a key made up of the dimensions that apply to that table — the breakdowns the framework requires, such as exposure class, approach, counterparty sector, country or maturity band. The remaining columns carry the reported facts.

Two properties of this model cause most of the mapping work:

• Dimensions are not labels. A column headed with a business description has to resolve to a taxonomy member. Where a firm’s internal breakdown is finer or coarser than the framework’s, the mapping is an aggregation or an allocation decision, not a rename.
• Open tables grow rows; closed tables do not. Tables with an open axis — a list of counterparties, countries or instruments — take as many rows as the data requires and need a stable row identifier. Tables with fixed breakdowns take a known set of rows, and a missing one is a finding rather than an absence.

Filing indicators

Filing indicators state which templates the report covers. They are declared explicitly, and they are the most common cause of a package being rejected while every figure in it is correct.

Units, signs and decimals

Monetary facts carry a currency; ratios and percentages do not carry one and must not be given one. Sign conventions follow the framework’s instructions rather than accounting habit, so a figure that is negative in a general ledger may be reported positive, and the instruction — not the source system — decides. Decimals declare the precision of a figure, and declaring more precision than the data holds is itself a misstatement. Reporting in thousands while declaring units, or the reverse, produces figures wrong by three orders of magnitude that pass every structural check.

Taxonomy version alignment

A report package is made against one taxonomy version. Framework releases add, remove and redefine tables, dimensions and rules, so a mapping built against one release is not assumed to hold against the next — it is revalidated. When a reference date falls near a framework change, the version that applies is the one prescribed for that reference date, not the newest one available.

The conversion, step by step

1. Export the source figures per table in the shape your systems produce them. Resist reshaping in a spreadsheet at this stage; every manual reshaping step has to be repeated every period.
2. Confirm the reporting perimeter — entity, level and reference date — before mapping, because it determines which tables are in scope.
3. Map columns to concepts and dimensions, recording every aggregation or allocation decision. This mapping is the reusable asset; the files are not.
4. Set filing indicators deliberately, distinguishing templates that are filed and empty from templates that do not apply.
5. Set units, signs and decimals per the framework instructions.
6. Validate and read the full report, triaging structural and coded findings before arithmetic ones.
7. Generate the package and retain the validation report alongside it as part of the filing record.

Recurring failures

• Scale mismatch — source data in thousands, declared as units.
• Sign convention — ledger signs carried through unchanged into templates with their own instruction.
• Missing filing indicator — complete data for a template that the package never declares.
• Unstable row identifiers on open tables — identifiers regenerated each period, so cross-period comparisons fail.
• Stale mapping — a mapping carried across a framework release without revalidation.
• Manual edits to the output — changes made to the package rather than to the mapping, which are silently lost on the next run.

REGREP is an independent software provider. This record explains a reporting framework in plain language and is not legal, tax or regulatory advice. Confirm scope, thresholds and submission dates with your competent authority before you file.