- Tech article: Explainability for machine learning models in MATLAB
Model Risk Management
A key trend in the finance industry is the increased reliance on models to inform decision making processes. The development and use of models introduce an additional source of risk. The workload required to manage the increasing number of models and their resulting model risk can become crippling if not managed effectively.
Common challenges we’ve observed across South African Banks in their model development and deployment workflows are:
- Slow implementation and deployment phases for new models – in extreme cases this time exceeds the validity-period of the model, which leads to outdated models being deployed into production.
- A lack of model transparency in review and deployment – it’s often not easy for non-technical staff to interrogate the decision process of deployed models.
- Time-consuming documentation processes – documentation is often decoupled from code, and requires significant effort to develop alongside the model code. Managing these decoupled document artifacts introduces additional risk.
Our services and product solutions range from providing enterprise level infrastructure through to the support of effective model risk management of individual model lifecycles. Our experience includes:
- Automation of models and model workflows to reduce prevalence of human error
- Improving the transparency of models to facilitate model validation and identifying the source of unexpected model behaviour
- Robust model development using software engineering best practices and unit test frameworks to ensure proper use of the model
- Managing model and data lineage and regulatory compliance, e.g. RDARR
- Model development frameworks for rapid and robust model development, review and deployment