Model risk management and model governance

Manage model risk with Watson OpenScale. Financial institutions manage many complex and integrated areas of risk. Management of model risk is critical to meet regulatory requirements and to protect institutions from operational and reputational risk.

What is a Model?

The Federal Reserve and Office of the Comptroller of the Currency guidance SR Letter 11-7 defines a Model. It is “…a quantitative method, system, or approach that applies statistical, economic, financial, or mathematical theories, techniques, and assumptions to process input data into quantitative estimates.”

These types of models, either deterministic or probabilistic, raise different model risk management challenges.

What is Model Risk?

Model risk is a type of risk that occurs when a mathematical model is used in financial institutions to predict and measure quantitative information, and the model performs inadequately. Inadequate performance leads to adverse outcomes for the firm and operational losses in millions.

Model Development Cycle

There are many challenges with machine learning and deep learning models. For example, you need to face the lack of knowledge of methods that are used by Model Developers / Vendors along with inconsistent documentation and increased volume of model change.

Tests to be run on machine learning and deep learning models differ from straightforward application testing:

  • Drift: Any change in input data also known as Drift can cause the model to make inaccurate decisions, impacting business predictions
  • Bias: Training data can be cleaned to be free from bias but runtime data might induce biased behavior of model
  • Explainability: Traditional statistical models are simpler to interpret and explain
  • Missing Validation or Test Data: Model training data sets might not capture the range of data or combinations that are encountered in runtime. Validation and monitoring of AI models is necessary in addition to govern and manage risk.

Options

IBM offers a model risk management solution with IBM Watson OpenScale. IBM Watson OpenScale monitors and measures outcomes from AI Models across its lifecycle and validates models. Specifically, you can use the following configurations: