January 24 2012- Luncheon

Assistant professor in the Department of Actuarial Science, Risk Management, and Insurance at the University of Wisconsin-Madison.

Abstract:

The past decade has seen a vigorous embrace of statistical and machine learning techniques within the actuarial community.  Classic examples of actuarial statistical modeling include the use of Generalized Linear Models to rate personal insurance along multiple dimensions; constructing credit and non-credit scoring models to refine pricing/underwriting business processes; and using statistical modeling to more rigorously forecast aggregate losses  for planning and capital allocation purposes.  While Generalized Linear Models are now solidly entrenched in actuarial practice, comparatively little attention has been given to the Multilevel/Hierarchical modeling framework.  This talk will sketch some classic actuarial modeling problems that can benefit from the application of Multilevel/Hierarchical models.  Indeed, Multilevel/Hierarchical models can be viewed as an extension of a cornerstone of actuarial science known as credibility theory.  Mainstream statistical notions such as "borrowing strength", "pooling", and "shrinkage" all have analogs in the actuarial credibility vernacular.  After sketching the relevant history and making the conceptual connections, a detailed case study will be presented to illustrate the power of Multilevel/Hierarchical models in actuarial work.