Hey what is the validation of XVA models (CVA, FVA etc)? As we know XVA calculation is rather complex problem (simulation, Valuation, aggregation) so what steps should be taken to check if the model can be used? Could someone briefly describe such a model validation process?
This recent paper https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3891120 (highly recommended in its entirety) has its entire section 5 devoted to XVA model validation.
You may also find these ORE slides insightful https://www.opensourcerisk.org/wp-content/uploads/2018/12/ore_user_meeting_2018_patrick_buechel.pdf
In general, the model validation consists of several steps:
- Checking the model design, i.e. model theory, model assumptions, model limitations, etc.;
- Checking the model inputs, i.e. market data sources, market data quality, model parameters quality, calibration process, etc.;
- Checking the model implementation, i.e. checking if the model is implemented in the system in line with the theory;
- Checking the model outputs, i.e. if the model produces correct numbers under current market conditions;
- Checking the model performance under stress scenarios, i.e. checking if the model produces reliable outcome under extreme market conditions;
- Checking all other related stuff, like model governance, model documentation, input controls, output controls, etc.