A failure to manage predictive models adequately usually results in a steady decline into model anarchy. This is certainly the case in some banks, where the overhead associated with regulatory requirements is inhibiting new model development, and in others there is even an unawareness of exactly how many models are being used.
Model management requires a number of components to be put in place:
- The management structures necessary to ensure model management is taking place, with responsibility for metrics that are important to the regulator and the organization.
- A technology platform capable of facilitating model management and delivering the tools necessary for its effective execution.
- The education and training of model developers so that models are documented adequately, and that concerns of the regulators are accommodated as models are being developed.
The tools and technology platform that facilitates model management needs to adequately cater for the following needs:
- Competitive need and technology developments mean predictive models are being developed at an increasing pace. In some firms this may mean that thousands of such models are already in use. Model management tools need to automate the management process so that large numbers of models can be adequately managed.
- Consistent, repeatable processes at every stage of the model lifecycle means that regulator requests can be responded to in a timely manner. A model management platform should provide an environment for the automation of model lifecycle management.
- Transparency is required within the business and to satisfy regulatory demands. As models become more sophisticated, so the need for clear documentation becomes more pressing. A model management platform should facilitate the efficient production of documentation.
- Prioritization of models is needed to identify those in most need of review and redevelopment.
- Automated workflow eases the documentation process and a model management platform should provide a workflow facility.
Of equal importance to the functionality of a model management platform, is the integration with model development tools and management reporting environments. Integration with model development provides a framework for developers, and allows managers to monitor productivity. And since predictive models affect business productivity, there should be some level of integration with management reporting systems. This level of integration is currently a hard call, since only a small number of suppliers have started to realise the importance of model management.
In reality model management is neither optional nor peripheral. It is actually the hub around which model development should take place, and will become a prime enabler of the use of decision automating technologies.
The previous article in this series is Financial Regulation and Predictive Models
Next article in this series is Predictive Model Management Business Benefits