Model risk audits provide an independent evaluation of the organization’s automated decision systems, ensuring that algorithms are functioning correctly and tracking data accurately.
Core Model Risk Audit Checkpoints
- Data Lineage Auditing: Tracing the data inputs used by the model back to their original source systems to confirm the information has not been distorted or corrupted during processing.
- System Limit Verification: Checking whether the model is operating outside its designed parameters. For example, a model built for stable market conditions may break down during sudden volatility, requiring human oversight.
- Overfitting Detection: Verifying that machine learning models have not been trained too tightly on historical data, which can leave them unable to adapt to new, real-world market movements.
Â