The unique IDs that firms assign to all important models typically appear in just three places: model documents, validation documents, and model inventory databases. Where the IDs do not, as a rule, appear is within the actual model source code. Incomplete model inventory information (including usage) is a chronic issue throughout the financial industry. Few firms can accurately answer such vexing questions as how many times each model in inventory was executed during the last year, which models exhibit significant seasonality, which models are used in each geographic region or legal entity, or whether any unvalidated models were used during the last year on any firm computer. This article will demonstrate that a root cause of model usage opacity is, unfortunately, that most models do not actually know who they are. This article will further explain how software-embedded model IDs can be leveraged to increase transparency and address some of the most difficult questions that may be posed about model usage.
To what extent do human explanations of model behavior align with actual model behavior?
