December 2017

Society of Actuaries and Predictive Analytics

Stuart Klugman

It should be clear by now that the SOA is investing considerable resources into expanding the ability of members and candidates to employ predictive analytics techniques. In this article I will summarize three projects: predictive analytics certificate, the first predictive analytics symposium, and enhancements to the ASA Curriculum.

Predictive Analytics Certificate

At its March 2016 meeting, the SOA Board of Directors approved offering a pilot certificate program with the subject being predictive analytics. The purpose of the pilot was to ensure that the SOA could create a program that would meet the needs of actuaries and their employers and be worthy of granting an SOA-endorsed certificate. The certificate was not to convey a designation (no letters after your name) but rather that the learning objectives of the program were met, as demonstrated via an assessment.

To that end, a call for participants yielded almost 100 applicants from which 30 were selected for the pilot offering. For five months the participants proceeded through six e-Learning modules (Predictive Analytics Tools; Effective Problem Definition and Project Management; Data Design; Transformation and Visualization; Data Exploration; Feature Generation and Selection; and Model Development and Validation). Along the way they had numerous opportunities to practice using RStudio and had access to a discussion forum to comment on questions posed to the group as well as seek help from each other and the team that developed the curriculum.

The pilot concluded with a two-day seminar. The first day was devoted to presentations, group discussions, and further practice. The second day provided six hours during which participants were given a data set and a business problem and asked to provide a solution. A team of graders used standard SOA exam grading techniques to determine which participants had demonstrated knowledge of the key concepts and would receive a certificate.

Based on candidate and employer participant surveys and a review of the content, the SOA’s Professional Development Committee (PDC) recommended that the program be made permanent and that additional certificate programs be established. The SOA Board approved this recommendation at its October 2017 meeting. There will be at least two sessions of the Predictive Analytics Certificate program offered in 2018 with up to 50 participants in each. Registration details will be published here in early 2018. During 2018 the PDC will be determining the topic for the next certificate program.

Predictive Analytics Symposium
The first SOA symposium devoted exclusively to predictive analytics was held in Chicago on Sept. 14 and 15. It is expected that this will become an annual event, so be on the lookout for the 2018 edition. There were 241 attendees, exceeding expectations. Most time slots featured three sessions, each with a different targeted audience: beginner/intermediate, manager/supervisor, and advanced practitioner.

There was one session that was strictly focused on General Insurance, fittingly titled “General Insurance Applications of Predictive Analytics.” The presenter was Peter Wu, a managing director at Deloitte Consulting. His slides are available through this link.

Other sessions covered analytics topics that apply across practice areas, such as deep learning, clustering techniques, introduction to machine learning, dangers of overfitting, ordinal logistic modeling, TensorFlow, tidy data, visualization and many more.

I had the pleasure of moderating several of the sessions and was pleased to see an engaged audience contribute to the discussion at each one. I’m looking forward to next year’s conference.

ASA Enhancements

The SOA previously announced major changes in the ASA curriculum effective July 1, 2018. One of the main drivers was to add more predictive analytics content. The latest information on these changes is always available here. I will summarize where we currently are with respect to the two new predictive analytics components.

For the Statistics for Risk Modeling (SRM) Exam, the learning objectives and readings have been published. This multiple-choice exam will first be given in September 2018. It covers regression (including GLMs), time series, principal components analysis, decision trees, and classification. The two selected texts are

  • Regression Modeling with Actuarial and Financial Applications, Edward W. Frees, 2010, New York: Cambridge
  • An Introduction to Statistical Learning, with Applications in R, James, Witten, Hastie, Tibshirani, 2013, New York: Springer. A PDF version of the text can be downloaded at here.

The Predictive Analytics Assessment is scheduled for December 2018. Details are still being worked out, but at this time we envision that candidates will go to an exam center equipped with computers configured to provide access to Word, Excel and RStudio (with a pre-defined list of packages installed). Relevant files will also be pre-loaded. Candidates will not have internet access. They will be presented with a business problem and a data set and then will have five hours to analyze the data and prepare a report to be submitted electronically. The focus will be on applying the techniques covered in the SRM Exam with additional education provided regarding communicating results and use of the R packages.

A natural question is how this will affect the Applications of Statistical Techniques module for the General Insurance fellowship track. Once most candidates have learned the basics via the above exams, we can eliminate the introductory material from the module and go into more specific general insurance applications of those techniques. But that is few years in the future.

Stuart Klugman, FSA, CERA, Ph.D., is a senior staff fellow for the Society of Actuaries. He can be contacted at