What does a “verifiable” risk model actually mean? In risk management, it is often said that models should be “capable of being validated.” What's interesting: Being capable of validation does not mean that a model has to predict the future exactly. A risk model is capable of being validated when you can meaningfully check, after the fact: • Which risk drivers were identified? • Which assumptions were made? • Which causal relationships were assumed? • Where were we right, and where were we not? • What do we learn from this for the next planning round? A model must not remain a riddle. Perhaps this is exactly a helpful shift in perspective: not asking whether a number is “correct,” but whether a model is explainable, checkable, and capable of learning. That changes the view of risk management, and makes it connectable to steering, review, and further development.
This post was originally published on LinkedIn. View and join the discussion there