Windy City to Host September Seminar on Predictive Modeling
By Ronald T. Kozlowski, Chairperson, Committee on Special Interest Seminars

Learn how insurance entities are using predictive modeling techniques for marketing, underwriting, pricing, and fraud detection at the CAS Special Interest Seminar on Predictive Modeling, September 19-20 at the Westin Michigan Avenue Chicago.

Predictive modeling is an idea whose time has come. Simply put, predictive modeling is a process by which one uses statistical analysis of data to make predictions about future events or otherwise gain actionable insights about multidimensional problems. Predictive modeling can therefore serve as an aid to human reasoning, and as such has nearly unlimited applications. Chain stores use it to select their next store's location; baseball teams use it to recruit "undervalued" players. Dot-com retailers use it to predict their customers' next likely purchases; governments use it to detect instances of corporate corruption. Insurers can also use predictive modeling to select and retain optimal portfolios of policyholders.

While predictive modeling has been around for a long time, notably in the form of regression analysis, recent advances in statistics and the proliferation of cheap computing power have dramatically increased its practicality and applicability. As a result, applying predictive modeling to the actuarial profession is limited only by actuaries' imaginations.

The purpose of this seminar is twofold. The first goal is to promote a better understanding of predictive modeling techniques in the actuarial community. The second is to further discuss current and future insurance applications of predictive models.

Basic and intermediate sessions will be offered covering such predictive modeling and analytic techniques as:

• Generalized Linear Models (GLM)
• Classification And Regression Trees (CART)
• Multivariate Adaptive Regression Splines (MARS)
• Neural Networks
• Generalized Additive Models (GAM)
• Clustering
• Principal Components Analysis
• Bootstrapping
• Model Validation

Perhaps the best-known application of predictive modeling in insurance is credit scoring, which will be the focus of the seminar. The successful utility of credit scoring has served as an early indication of the power of data mining and predictive modeling. Personal and commercial lines insurers are increasingly applying predictive modeling techniques in marketing, underwriting, rating, fraud detection, and retention and cross-sell analyses. These insurers are also using predictive modeling to identify appropriate case reserves for claims. Key topics of discussion for credit scoring will be data sources, business strategies behind predictive modeling projects, and model implementation.

This seminar is intended for actuaries and other insurance professionals at all levels who wish to learn about the potential uses of predictive modeling in their work. The seminar will benefit prospective users of all levels, from actuarial students and insurance analysts to insurance company management.