Book-wide stress testing under model-risk governance
Updated
In short
How do you run a macro stress test that a model validator will accept?
Kelvyn stresses the whole book on official macroeconomic scenario paths — EBA and ESRB, and Banca d'Italia — transcribed from the primary annexes with their source citations, and propagates the result through to a full set of stressed financial statements. Scenarios move through draft, shared and approved, where the approver must be someone other than the author. Runs are immutable and fingerprinted, can be sealed, and sealed runs can be backtested against realised actuals.
The scenarios are transcribed, not invented
GDP, unemployment, HICP, residential and commercial real-estate prices and sovereign yields come from the published scenario annexes, with per-annex source citations held against the seeded data and pinned by tests. Baseline and adverse Italy paths are there because Italy is the delivered jurisdiction.
Your own scenarios are first-class. Elasticity coefficients are editable data rather than constants in code, and an edit carries a justification note.
Change control an examiner recognises
Scenario governance is a fail-closed chokepoint, not a convention. A scenario moves draft to shared to approved, and the approver must differ from the author. Retirement is a lifecycle state rather than a delete.
The calculation engines are version-frozen and registered, so a run records which engine version produced it. Re-running last quarter's scenario next quarter does not silently pick up a different model.
Runs you can point at later
A run is immutable and fingerprinted. It can be sealed, and a sealed run can be assembled into a board pack. Figures are re-derived before they render, so a number in a summary is recomputed from the run rather than copied from it.
Backtesting compares sealed runs against realised actuals. It is gated to sealed runs only and fails closed with a stated reason if that condition is not met, which is the behaviour that makes the result worth showing a validator.
Scope
What this does not do
A stress run produces scenario-stressed loss on a single adverse path. It is not a probability-weighted IFRS 9 expected credit loss and the product will not label it as one.
There is no hedging capability. Rate and FX scenario curves exist for pricing debt facilities; hedge instruments, hedge accounting and hedge optimisation do not exist in the product at all.
An immutable run cannot be undone. That is the point of it, but it means a mistake is corrected by a new run rather than by editing the old one.
Questions
- Which official scenarios are seeded?
- EBA and ESRB paths, and Banca d'Italia, covering GDP, unemployment, HICP, residential and commercial real-estate prices and sovereign yields, each carrying its annex-level source citation.
- Can the same person write and approve a scenario?
- No. The approver must differ from the author, and the rule is enforced at the governance chokepoint rather than left to process.
- Can a stress result be checked against what actually happened?
- Yes, for sealed runs. Backtesting compares a sealed run against realised actuals and declines, with a stated reason, if the run is not sealed.
Related