What is the SOA Exam SRM exam?
Exam SRM is the SOA preliminary exam covering statistical learning as actuaries use it. It runs from the basics of supervised and unsupervised learning through linear and generalised linear models, time series, decision trees and ensembles, and principal components and clustering.
It is noticeably less about calculation than P, FM or FAM and more about knowing which model applies, what its assumptions are and how to read its output. R is not examined as a coding skill, but R output is presented for interpretation, so being comfortable reading a regression summary matters.
SRM is a required prerequisite for the Predictive Analytics exam, which revisits the same methods in much greater depth. Its pass rates are among the highest of any SOA preliminary exam, generally close to 70 percent.
Who sits it
- Actuarial students on the ASA pathway, for whom SRM is required
- Candidates preparing for the SOA's Predictive Analytics exam, which requires SRM first
- Analysts moving into modelling and data science work inside insurance
Exam format at a glance
- Questions
- 35 multiple-choice
- Duration
- 3.5 hours
- Delivery
- Computer-based at test centres, with a paper option where no centre is available
- Exam windows
- Three sittings a year: January, May and September
- Fee (USD)
- $357 per attempt
- Assumed knowledge
- Calculus, probability at the level of Exam P, and VEE Mathematical Statistics
- Passing standard
- Pass mark set by expert panel and adjusted for observed difficulty; reported on a 0–10 scale with 6 a pass
- Recent pass rate
- 71.2% (January 2026), the highest of any SOA preliminary exam
Exam-body facts last checked September 2026. Fees, dates and formats change; confirm the current details with Society of Actuaries (SOA) before registering.
SOA Exam SRM pass rates by sitting
| Sitting | Pass rate |
|---|---|
| May 2025 | 70.6% |
| September 2025 | 67.9% |
| January 2026 | 71.2% |
The SOA publishes the percentage passing and the pass mark for each sitting, but its results page shows only the most recent one, so the history here covers the sittings whose figures are still published. The pass mark for each of these was 63% of scored questions.
Syllabus and topic weights
Weights from the SOA's 2026 Exam SRM syllabus.
1.Basics of Statistical Learning5–10% · 8 lessons · 5 questions · 66 cards5–10%8566
- Types of Modeling Problems and Methods
- Assessing Model Accuracy
- The Bias-Variance Tradeoff
- Resampling and Cross-Validation
2.Linear Models40–50% · 14 lessons · 23 questions · 122 cards40–50%1423122
- OLS vs GLM: Model Assumptions
- Exponential Family and Link Functions
- Interpreting Model Parameters in Context
- Model Fit and Diagnostic Tests
- Model and Variable Selection
- Prediction and Confidence Intervals
- How GLM Approaches Differ from OLS
3.Time Series Models10–15% · 6 lessons · 10 questions · 53 cards10–15%61053
- Stochastic Time Series Concepts
- Time Series Models: Smoothing, AR, and ARCH
- Time Series Forecasts and Intervals
4.Decision Trees20–25% · 8 lessons · 17 questions · 73 cards20–25%81773
- Building Decision Trees
- Classification and Regression Trees
- Bagging, Boosting, and Random Forests
- Decision Trees vs Linear Models
5.Unsupervised Learning Techniques10–15% · 8 lessons · 16 questions · 71 cards10–15%81671
- Principal Components Analysis
- Interpreting PCA Results
- K-Means and Hierarchical Clustering
- Choosing the Number of Clusters
Lesson, question and flashcard counts are StudyOptima's content written so far for each topic. Expand a topic to see its subtopics.
How to prepare for SOA Exam SRM
- 1
Linear models are up to half the exam on their own. Ordinary least squares, generalised linear models, link functions, diagnostics, model selection by AIC and BIC, and the lasso and ridge penalties all need to be second nature.
- 2
Decision trees are the next largest block at up to a quarter. Know pruning, bagging, boosting and random forests, and be able to say when each beats a linear model.
- 3
Learn to read R output rather than write R. Questions present a fitted model summary and ask what it implies, so recognising coefficients, standard errors, deviance and information criteria at a glance is worth more than coding practice.
- 4
The statistical tables are provided on screen and you may not bring your own, so practise with the supplied versions.
- 5
Do not skip unsupervised learning. Principal components and clustering are up to 15 percent and are quick to learn relative to that weight.
SOA Exam SRM on StudyOptima
We have lessons and flashcards across all five SRM topics and a question bank in progress, none of it released yet. Our Exam P material, which is free with any account, covers the probability the SRM syllabus assumes, and our Exam FM material is live if you still need that exam.
SOA Exam SRM FAQ
Do I need to know how to program in R?
No. The syllabus does not require you to write R code, but it does present R output for interpretation, so you need to be able to read it.
Why is the SRM pass rate so much higher than Exam P or FM?
Most candidates reach SRM having already passed several exams and having covered much of the material in a university statistics course, so the population sitting it is stronger than for the entry-level exams.
How does SRM relate to the Predictive Analytics exam?
SRM is a prerequisite for it. Predictive Analytics takes the same families of model and goes considerably deeper, with a project-style assessment rather than multiple choice.
When will StudyOptima cover Exam SRM?
Lessons, flashcards and a question bank exist across all five syllabus topics, though the question bank is smaller than for our released exams. Email support@studyoptima.com if you would like to be told when it launches.