Quantification of model risk and tiering of models for a risk sensitive approach

The views and opinions expressed in this article are those of the thought leader and not those of CeFPro.

By Fares Triki, Head of Model Risk Management, MUFG Securities.

In your opinion, what are the benefits of a risk sensitive approach to model risk management?

Models need to be rated or classified by the level of risk. It is a prerequisite for model risk reporting and monitoring. It might be useful to briefly describe the benefit of using this classification in the day to day risk management. With the ever increasing volume of models making their way to the inventory, if all models are to be treated the same, the model risk management would struggle to cope. This realistically would result in (i) either generating a bottleneck with delays in time to production, potentially creating an incentive to hide models, (ii) or lowering the standard of validation and monitoring for all models including the riskier ones. Therefore a risk sensitive approach is difficult to overlook. It is also only good management practice to allocate time and resources according to the risk.

In practice, that amounts to (i) have a different revalidation frequency for models in each risk class (ii) make the validation depth also function of the risk (iii) use the resources to better monitor the riskiest models. The scope of use of this model classification can also be extended to governance and technology. All of this does rely on the quality of model risk measurement and model tiering approach. It’s even more crucial in a framework designed in such a way that the less risky models get the least scrutiny. If the model risk measurement is not done right, the firm would be focusing all effort and attention to the wrong models while the model risk is elsewhere, in models less frequently revisited and for which the performance is poorly monitored.