31 March 2021
Over many years, credit scorecard modelling in banks and agencies has reached a high plateau of good-practice. Many principles introduced by early pioneers and practitioners are now well-tested and even required by the profession:
- Logistic regression
- Classing of input variables
- Over-weighting of defaults
- Reject inference
- Delta method for scorecard adjustment
- Use of AUC metrics and much more
These practices sometimes come with warnings or limitations in their use, or are recognised as approximations or to have potential biases or errors, but guidance is often simply cautionary — "it’s better to do it, but watch out in X circumstances!" and in need of quantification — "add some conservatism to cover Y". Determining when to stop using one classical approach or how to correct another, remains a material problem of model management.
In this talk, Alan will describe a geometric picture of scorecard adjustment and selection that helps address this difficulty. It aims to unify and simplify the complex menu of good practices as instances of geometric connection or confusion in a high dimensional model and data space.
This view is based on statistical information geometry, an active area of mathematical statistics started by Rao’s differential geometric approach to Fisher Information. Adapting this general theory to the case of scorecards — regressions on contingency tables — simplifies the mathematics greatly, providing a structure that allows practical calculation and insight.
To illustrate this, Alan will pick out three classical scorecard practices and quantify their geometry, model selection, the delta method in the presence of heavy correlation, and sample weighting. This gives fresh insight often confirming the classical approach is right, sometimes correcting, or simplifying it. It also makes this powerful theory accessible to scorecard builders to solve other similar problems.
If you are interested in attending, please email Yvonne Crichton at email@example.com
About The Speaker
Alan Forrest is an experienced Credit Risk and Marketing modeller, modelling team leader, and creative developer of advanced statistical methodologies in Banking Book Financial Services.
He has particular experience in:
- Credit Risk Basel modelling and validation
- Advanced statistical modelling in SAS
- Model Risk quantification and management
- Data management and migration
- Delivering professional training
- Sales and Marketing analytics
Alan currently leads the Model Risk Oversight team at Virgin Money UK PLC; a team of experienced quantitative specialists, whose purpose is to validate and review models throughout the Bank, independently of the model development teams.