Difference between revisions of "List of Model Validation Questions"
From Open Risk Manual
Wiki admin (talk | contribs) (→Establish the Model Usage context) |
Wiki admin (talk | contribs) (→Establish the Model Usage context) |
||
Line 19: | Line 19: | ||
* Is the model [[Model Documentation | documented]] appropriately | * Is the model [[Model Documentation | documented]] appropriately | ||
* Are [[Model Usage | users]] aware of potential weaknesses | * Are [[Model Usage | users]] aware of potential weaknesses | ||
− | * | + | * Are [[Model Performance Measures]] being [[Model Monitoring | monitored]] appropriately? |
* Are instances of [[Model Failure]], [[Overrides]], recalibrations or adjustments documented and fed back into model redevelopment? | * Are instances of [[Model Failure]], [[Overrides]], recalibrations or adjustments documented and fed back into model redevelopment? | ||
* Is there a process of [[Periodic Validation]] | * Is there a process of [[Periodic Validation]] |
Latest revision as of 11:34, 15 September 2021
Contents
List of Model Validation Questions
Model Validation is a varied exercise and the precise elements involved depend significantly on the nature of the models being validated, the context and importance of Model Usage. The following list is a general blueprint.
Evaluate the Model Governance context
- Does the organization have a Model Governance framework? If not, what is applicable Internal Governance?
- Has the model been commisioned, developed and operated internally in accordance with that framework.
- What are the regulatory requirements (if any)
Evaluate the Model Development context
- The sourcing of data: Are the data used to built and/or operate the model comprehensive, with appropriate Data Quality, without bias?
- Is the concept / mathematical structure of the model well known and understood (e.g in literature)
- How much Intrinsic Model Risk is there (alternative possible models, hidden assumptions)
- Are there explicit quality requirements required before a model is accepted
- Is there benchmarking with alternative models
- Is there backtesting against historical data (when applicable)
Establish the Model Usage context
- Is the model documented appropriately
- Are users aware of potential weaknesses
- Are Model Performance Measures being monitored appropriately?
- Are instances of Model Failure, Overrides, recalibrations or adjustments documented and fed back into model redevelopment?
- Is there a process of Periodic Validation