Model risk regulation: where a validation team finds each item

Last updated 16 November 2026

What four model risk rules say about models from a supplier, and where GIAnalyzer and the Gamma Impact API publish what each asks for. This is our reading of each rule, for information; none of the regulators has reviewed it, and it does not replace your own validation.

Bank of Israel: Directive 369

Proper Conduct of Banking Business Directive 369, Management of Model Risk. Each clause is taken in detail on the Directive 369 page.

Directive 369What it asksWhere
§38, §93(b)Documentation that allows validation: components, design and intended useCoverage & models, one methodology page per area, and the function reference
§47The methodology, with its mathematical specification and sourcesEach methodology page names its models and ends with its references
§93(b)Test resultsA worked example for every function; in GIAnalyzer, the Manual Calculations workbooks
§93(b)Limitations and assumptionsThe "What the models do not do" section of each methodology page
§93(e)Sensitivity analysis and benchmarking where the code is not openEvery input and parameter returned with the result; sensitivities on any argument; models named, so they can be implemented independently
§93(g), §93(h)Data and assumptions from the supplierNone bundled: market data and histories come from you; defaults are stated on the methodology pages
§93(f)Adjustments made for the bankParameters the bank sets, returned with every result
§93(c)Ongoing monitoring and updatesA message to clients with each release; each client chooses when to upgrade; each methodology page states its version
§93(i)Contingency if the product is no longer supportedRuns and licenses offline; a licence keeps working until its expiry date, with support until then; source-code escrow on request
§93(d)Internal knowledge to continue if the relationship endsThe user guide, FAQ and glossary; training for clients

United States: interagency guidance, April 2026

Supervisory Guidance on Model Risk Management, 17 April 2026: Federal Reserve SR 26-2, OCC Bulletin 2026-13, FDIC FIL-15-2026. It replaces SR 11-7.

Section VIIWhat it asksWhere
Vendor productsModel risk principles apply even where the bank does not receive the vendor's code, data or methodologyThe methodology is published; every input and parameter is returned with the result
Vendor productsAn understanding of the model's conceptual soundness and designThe methodology pages and the function reference
Vendor productsAn understanding of its development dataThe pricing models are standard published models, not estimated from a data set: they are calibrated at each run to the market data you supply. Value at Risk is computed on histories you supply
Vendor productsAn understanding of its performanceWorked examples and manual calculations
Vendor productsOngoing monitoring and outcomes analysis by the bankResults run on your own positions and market data, with the inputs behind each one
Vendor productsCustomisations documented, justified and evaluatedEvery setting is a parameter, returned with the result

United Kingdom: PRA SS1/23

Supervisory Statement SS1/23, Model risk management principles for banks, updated April 2026.

SS1/23What it asksWhere
2.6(b)(i)The firm satisfies itself that vendor models are validated to its own standardsThe methodology pages, their limitations sections and the test evidence
2.6(b)(ii)The relevance of vendor-supplied data and assumptions is verifiedNo data bundled; defaults stated on the methodology pages
2.6(b)(iii)The firm validates its own use, with monitoring and outcomes analysis on its own outcomesResults run on your own data, with every input returned
3.5(a)Vendor documentation detailed enough to validate the firm's useThe methodology pages, the function reference and the inputs returned with each result
4.3(a)(i)Input data represents what the model is used forInputs come from you

Canada: OSFI Guideline E-23

Guideline E-23, Model Risk Management (2027), in effect from 1 May 2027.

E-23What it asksWhere
C.1, Appendix 1Vendor models recorded with their data sources and limitationsMarket data from you; limitations on each methodology page
D.2, model reviewReview of conceptual soundness and performance, including third-party modelsThe methodology pages, worked examples and manual calculations
D.2, model monitoringChanges in external dependencies, such as version updates, trackedA message with each release; you choose when to upgrade; each methodology page states its version
D.2, model monitoringContingency plans for model unavailabilityRuns and licenses offline; escrow on request
Response to consultationAdequate documentation from vendors, and controls on whether a vendor solution gives an appropriate responseThe documentation is public; every result comes back with the inputs that produced it