Announcement: SOA releases June 2019 Exam STAM passing candidate numbers and congratulates the new FSAs for August 2019.

Agenda and Presentations

The morning sessions focus on conceptual issues, while also going through examples. In the afternoon the participants will form groups to address challenges running models and interpreting results. The seminar concludes with a panel discussion and Q&A. Presenters will stay an additional hour for attendees who are able to continue.

Presentations

Wednesday, May 18
7:00 a.m. – 8:00 a.m.
8:00 a.m. – 8:10 a.m.

Presenter(s): Brian D. Holland, FSA, MAAA,PAF Section Chair; Ricardo Trachtman, FSA, MAAA, PAF Section Vice Chair; Vincent J. Granieri, FSA, MAAA, EA, PAF Section Council Member

The presenters will highlight recent public applications of predictive analytics in the life and annuity space, and place the seminar's contents in the context of actuarial practice. 

Session Coordinator(s)

Facilitator(s)

8:10 a.m. – 9:05 a.m.

Presenter(s): Jean-Marc Fix, FSA, MAAA

The most essential skills in R will be covered to review and round out the seminar's preparation material. These skills are data input/output; data frames in R; equivalent tools in R for familiar spreadsheet techniques like pivots tables; tools for data visualization and exploration; and running a GLM.

Session Coordinator(s)

Facilitator(s)

9:05 a.m. – 9:15 a.m.
9:15 a.m. – 10:50 a.m.

Presenter(s): Eileen Sheila Burns, FSA, MAAA; Matthias Kullowatz, MS

In this session, presenters will discuss questions of interest for life and annuity products, and the predictive model forms that are best suited to investigating them, including associated theoretical concerns that may arise in the modeling process. They will set the stage for the afternoon session to address more practical concerns by introducing several concepts such as identifying and dealing with outlier data values, accounting for missing values, using the step function for variable selection, identifying correlated variables, setting reference levels for factor variables, and testing and improving the model fit across the range of each covariate. We will also explore a technique to improve computational efficiency for logistic GLMs. Finally, we will discuss assessing overall model fit and comparison between two candidate models.

Session Coordinator(s)

Facilitator(s)

10:50 a.m. – 12:30 p.m.

Presenter(s): Satadru Sengupta

The presenter will start with generalized linear models and then will create a case for more advanced machine learning techniques. He will first introduce classification and regression tree (CART) - a basic building block of many advanced machine learning techniques. It can organically incorporate nonlinearity and interactions. Then he will discuss regularization in the context of underfitting and overfitting concepts. He will also introduce ensemble techniques (random forest and boosted trees) and regularization as a solution to deal with such issues. All concepts will be explained with illustrative examples.

Session Coordinator(s)

Facilitator(s)

12:30 p.m. – 1:15 p.m.
1:15 p.m. – 3:30 p.m.

Participants will divide into groups to work through the prepared examples, with assistance from presenters and revisiting presentations as needed.  Times are approximate.

Session Coordinator(s)

Facilitator(s)

3:30 p.m. – 4:00 p.m.
4:00 p.m. – 5:00 p.m.


 

Session Coordinator(s)

Facilitator(s)