What is model risk?

What it is about

Model risk is the potential loss an institution may incur from decisions based mainly on the results of internal models, because of errors in their development, implementation or use.

It is one of the activities through which the Club promotes a risk culture by exchanging ideas, research and best practices. Fernando Prieto and Leonardo Perozo take part as speakers, introduced by Francisco Rivillas, president of Némesis Risk, and Carlos Rijo, president of the Club.

Session summary

Two specialists present model risk: the risk of losses from decisions based on incorrect or misused models. The first starts from real cases, from Gaussian copula models and the 2008 crisis to losses attributed to errors in value-at-risk models, code or spreadsheet errors and control failures, to show that the risk is not only technical but also one of governance. He reviews the definition of a model, the sources of risk (data, methodology and use) and the US Federal Reserve guidance, with the principle of ‘effective challenge’, and how institutions set up a framework with scope, committees, policies, appetite, risk tiering, a model inventory and continuous monitoring.

The second returns to the copula crisis and stresses that the literature and regulators have focused on mathematical accuracy and little on governance, organization and people’s competencies. He insists on the role of the board of directors (model risk appetite, regular reporting, key indicators), on data integrity, on the fact that validating a model is not the same as managing model risk, on more frequent reviews in times of stress and on independence between those who build and those who validate. The questions covered whether model risk is unavoidable, the supervisors’ roadmap, the level of maturity in the Dominican Republic and recommended reading.

Topics covered

• Real cases of model risk and the Gaussian copula crisis
• Definition of a model and sources of risk: data, methodology and use
• Federal Reserve guidance and ‘effective challenge’
• Management framework: inventory, risk tiering, committees and monitoring
• Role of the board, data integrity and independence of validation
• Machine learning: more model risk and also new tools
• Questions: maturity in the Dominican Republic and the supervisor’s roadmap