A model owner argues that a newly developed neural network model do…

Question

A model owner argues that a newly developed neural network model does not require formal model risk management treatment because it assists decisions but humans make the final call. How should a model risk officer respond?

Accepted Answer

Disagree — SR 11-7 applies to models whose outputs materially influence decisions, regardless of whether a human makes the final call — Here's the deal — you should disagree. SR 11-7 covers models whose outputs materially influence decisions, even when a human makes the final call. Human review doesn't take the model out of model risk management.

Answer

Agree — SR 11-7 only applies to fully automated decision-making systems

Answer

Agree — neural networks are exempt because they cannot be validated using traditional statistical methods, so SR 11-7's validation requirements only apply to simpler, fully transparent regression models.

Answer

Disagree — but only if the model processes more than one million records annually

A model owner argues that a newly developed neural network model does not require formal model risk management treatment because it assists decisions but humans make the final call. How should a model risk officer respond?

Short Explanation

Full Explanation