EBA NPL data tapes, validated field by field
Updated
In short
How do you validate a loan tape against the EBA NPL templates?
Kelvyn checks an uploaded loan tape against 134 EBA field definitions, each carrying its own mandatory flag, validation rule and EBA reference, and returns a validation report before the data reaches a portfolio. The check runs at ingestion rather than at review, so a tape with missing mandatory fields or out-of-range values is caught while the seller can still fix it — not after it has been priced, modelled and taken to a committee.
What gets checked
The 134 field definitions are stored as data, not as code: each one carries whether it is mandatory, the rule its values must satisfy, and the EBA reference it comes from. A validation report lists what failed and against which definition, so the finding is arguable against the source rather than against our implementation.
Tapes arrive as Excel or CSV and are parsed server-side. Nothing is accepted into a portfolio on the strength of a filename or a column header that merely looks right.
Why validation belongs at ingestion
A data-tape defect found during pricing is expensive. A defect found after a bid is submitted is worse, because the number that has to move is one you have already put in front of a counterparty.
Running the check at ingestion means the tape either enters the system in a known state or does not enter it. Everything downstream — exposure-level PD and LGD, recovery curves, cluster valuation — inherits that state instead of assuming it.
What the tape becomes
A validated tape populates exposure records that the rest of the product works from: recovery and collection curves, expected recovery, cluster valuation, and the amortised-cost schedule that follows from the curve you author.
The same exposures feed the stress engine, so a macro scenario runs against the book you actually loaded rather than against a summary of it.
Scope
What this does not do
Validation is against the EBA NPL transaction data templates. It is not a general-purpose data-quality suite and it does not check a tape against your internal conventions unless those are expressible as EBA field rules.
Passing validation means the tape is well-formed against the templates. It is not an opinion on whether the underlying data is true, and no automated check can be.
Questions
- How many fields are validated?
- 134 field definitions, each with a mandatory flag, a validation rule and its EBA reference.
- What file formats are accepted?
- Excel and CSV, parsed server-side.
- What happens to a tape that fails validation?
- It does not enter a portfolio. The validation report identifies each failure and the field definition it failed against, so it can be returned to the counterparty as a specific, referenced list rather than as a general objection.
Related