Advancing ASA Education

The SOA updates the ASA pathway to include advanced topics in predictive analytics (and data science) Craig DeAlmeida and Stuart Klugman

Photo: iStock.com/metamorworks

As part of the Society of Actuaries’ (SOA’s) continuous updating of the pathways to associateship and fellowship, a new component titled Advanced Topics in Predictive Analytics was added in 2022 to upgrade the analytics education of those seeking the ASA designation. This component is also part of the new Data Science for Actuaries micro-credential. The overall goal is to enhance the skill set reflected in the Statistics for Risk Modeling and Predictive Analytics exams that were introduced in 2018. This article describes what candidates who master this component will know and be able to do, and how the approach to learning and assessing this material differs from the related exams.

SOA research has emphasized four areas in which an enhanced analytics experience could benefit all candidates seeking the ASA designation. While including “Advanced Topics” in the title of this new component is meant to distinguish it as building on the current Predictive Analytics exam, the material covers far more than just advanced techniques. The four main topic areas are:

1. Ethical Foundations

Candidates will learn the ethical framework that also supports the SOA’s Ethical and Responsible Use of Data and Predictive Models certificate program. The framework centers on three principles:

  1. Fairness
  2. Safety
  3. Transparency and accountability

Candidates then will be introduced to relevant regulations and standards of practice, which are applied to specific situations in later sections.

2. Working With Data

Prior courses assume that data is from a single source and relatively clean, but this course addresses the many challenges of working with data as it actually arrives. Candidates will learn how to combine data sets, assess accuracy and quality, address outliers and missing observations, and apply the ethical framework when governing the overall data pipeline.

3. Advanced Models

This is where the “advanced” part of the course comes in. Candidates will expand their analytics toolkit by learning about additive models, linear mixed models, neural networks and Bayesian techniques to varying degrees of sophistication. They then will be able to combine models via blending and stacking.

With this larger analytics toolkit, candidates also will learn more about how to assess a model’s suitability for a given analytics objective. There is also a return to ethics with further coverage of the risks that arise from having too many variables, fitting too many different models and proxy discrimination.

4. Explainability and Communication

While communication is important in all aspects of actuarial work, predictive models pose special challenges. The connection between input and output is far from transparent for many powerful models. Special techniques are needed to convey this relationship and, more generally, build trust with those who rely on these models. Various interpretation techniques will be covered along with advice on writing reports, memos and presentations when communicating modeling activities.

Learning and Assessment

Two aspects differentiate the assessment for this Advanced Topics in Predictive Analytics component from the two prior analytics exams. First, candidates may use any analytics platform. The modules provide code on all covered techniques in R and for most techniques in Python, but candidates are not restricted to those two options and may use any tools they prefer. Second, the assessment is an at-home activity structured similarly to the one used for the Fundamentals of Actuarial Practice course. This aligns with how actuaries leverage analytics in the field, providing sufficient time and access to resources such as the modules, additional texts and the internet. After being given a data set, a business problem and a series of prompts, the deliverable is a written report that shows output but not code.

We are excited about this new phase in actuarial education.

Craig DeAlmeida, FSA, CERA, MAAA, is VP, R&D, Annuities Business Analytics, at Lincoln Financial Group.
Stuart Klugman, FSA, CERA, Ph.D., is senior staff fellow, Education, at the Society of Actuaries.

Statements of fact and opinions expressed herein are those of the individual authors and are not necessarily those of the Society of Actuaries or the respective authors’ employers.

Copyright © 2022 by the Society of Actuaries, Schaumburg, Illinois.