Exam PA (Predictive Analytics) Practice Test

The official prerequisite is passing Exam SRM (Statistics for Risk Modeling). It is strongly recommended, as Exam PA builds directly upon the statistical learning concepts covered in SRM, applying them to practical case studies. Candidates must also have credit for the SOA's Validation by Educational Experience (VEE) requirements.

Exam PA is a 5-hour and 15-minute computer-based exam consisting of several tasks based on a real-world case study and dataset. You will receive the case study and data prior to the exam for review. The exam tasks require you to perform analysis, write code (in R), interpret results, and provide written recommendations. Grading is holistic, assessing the technical correctness, appropriateness of methodology, and clarity and effectiveness of written communication.

Proficiency in the R programming language is essential. The exam is administered in RStudio, and you must write R code to perform data manipulation, visualization, and modeling. The SOA provides a specific list of allowed R packages. Deep expertise in advanced R programming is not tested; rather, the focus is on using R effectively to conduct a standard predictive analytics workflow.

Exam PA is uniquely situated within the actuarial credentialing framework, emphasizing the business context, risk, and communication standards of the profession. While it covers similar technical ground (e.g., GLMs, decision trees), its assessment is deeply integrated with actuarial applications like insurance pricing and reserving. The evaluation heavily weights the communication of technical findings to a non-technical business audience, a core actuarial skill.

Successful preparation involves consistent practice in articulating modeling choices, interpreting diagnostic plots, and translating statistical output into business implications. Review past exam solutions and grader reports to understand expectations. Practice writing concise, clear, and actionable summaries that would be suitable for a manager or client, avoiding unnecessary statistical jargon while accurately conveying the model's insights and limitations.