Model risk monitoring

Model risk monitoring is part of the model development phase and it must be considered an essential part to approve and use a model. It represents an essential part in the MRM, and it is where the progress must keep momentum.

To fully understand the motivations to monitor the performance of the models, the first step is to define the model risk appetite. This element will have a powerful effect on the way the MRM is established and operates.

On this basis, why monitoring? Here are some valid reasons:

The model reaches its theoretical limitations

The model ceases to function correctly due to errors in the live environment e.g. due to poor data feeds or to system changes

The model is used incorrectly

The model could be substituted with a more effective and efficient solution.

Model risk monitoring usually focuses on the following elements:

Model performance rating / status

Key changes in products, exposures, activities, clients, and market conditions

Review of input data, assumptions, and limitations

Outcome analysis, model stability checks

Review of the overlays / management judgement

Benchmarking

Portfolio changes and evolution

Thresholds for model recalibration or rebuild

Of course, the model performance monitoring must be calibrated on the materiality of the model.

According to the points above, it is possible to create performance index specific for the individual model accounting for data quality, discriminative power, stability, back-testing, predictive error etc.