SOA Exam SRM (Statistics for Risk Modeling) Exam Guide
Exam SRM is the Statistics for Risk Modeling examination on the Society of Actuaries (SOA) preliminary examination pathway toward the ASA designation. It introduces the analytic methods that underpin modern predictive modeling in actuarial work: statistical learning basics, linear and generalized linear models, time series models, decision trees and ensembles, and unsupervised learning. The syllabus states that Exam SRM is a prerequisite for the Predictive Analytics (PA) exam, which goes deeper into each of the covered techniques, and it assumes calculus, the probability of Exam P, and VEE Mathematical Statistics. Questions may present statistical output for interpretation, but the ability to program is not assumed. It is a professional examination administered by the SOA; it is not a state license.
Exam Facts
- Questions and time: 35 multiple-choice questions in 3.5 hours, delivered as a computer-based test. Each question has five answer choices, and a few pilot questions are placed randomly in the sitting and do not count in scoring. All unanswered questions are scored incorrect. Tables of the standard normal, t, and chi-square distributions are provided at the examination through an on-screen Exhibit. Unofficial pass/fail results are sent within about an hour of the appointment.
- Passing standard: results are reported on a 0-10 grade scale, with 6 the passing grade. The SOA does not publish a fixed passing percentage or raw cut score; the underlying standard is set per form through statistical scaling so that results are comparable across sittings.
- Eligibility: set by the SOA. The SOA publishes no degree, coursework, or experience prerequisite to sit Exam SRM; candidates register with the SOA and schedule their sitting, and sittings run several times each year. Requirements change, so confirm the current SOA guidance when you register.
- Fees: the SOA’s published Exam SRM fee is $357 (confirm the current fee on the SOA’s fee schedule when you register).
- The work: SRM is where actuarial practice meets data science — the regression, time series, tree-based, and unsupervised methods it tests are the daily tools of predictive modeling and risk analysis.
What the Test Covers
Basics of statistical learning (5-10% of the syllabus weighting); linear models (40-50%); time series models (10-15%); decision trees (20-25%); and unsupervised learning techniques — principal components and cluster analysis (10-15%).
How to Prepare
Work the topic drills portion by portion, giving the linear models material the largest share of your time as the syllabus does, then take the full 35-question simulated exam under the real 210-minute limit. Review every explanation – Exam SRM rewards candidates who can choose, diagnose, and interpret a model quickly and accurately, so practise until the standard setups feel routine.
Which actuarial examination is this?
This page covers SRM — Statistics for Risk Modeling, an examination of the Society of Actuaries (SOA) on the pathway to the Associate of the Society of Actuaries (ASA) designation. It is not a Casualty Actuarial Society (CAS) examination: the CAS maintains its own statistics examinations, MAS-I and MAS-II, under its own outlines, and no CAS facts appear on this page. Within the SOA pathway, SRM sits after Exam P and the VEE Mathematical Statistics subject, and before the Predictive Analytics exam, for which SRM is the prerequisite.
The real exam at a glance
| Fact of record | Detail |
|---|---|
| Credential pathway | Associate of the Society of Actuaries (ASA) — Exam SRM |
| Awarding body | SOA (Society of Actuaries) |
| Questions | 35 multiple-choice, five answer choices each |
| Time | 3.5 hours (210 minutes) |
| Delivery | Computer-based test (CBT); paper/pencil form at select centers |
| Syllabus | Statistics for Risk Modeling syllabus (December 2026) — 5 topics |
| Scoring notes | A few pilot questions are placed in the exam and are not used in scoring; all unanswered questions are scored incorrect |
| Results | Unofficial pass/fail results sent within one hour to the email address used to schedule the appointment |
| Passing standard | Pass mark set per administration; grades reported 0–10, with 6–10 passing — the SOA publishes no fixed passing percentage |
Who can take this examination?
SRM carries no occupational experience requirement; its gates are academic, and the SOA states them in the syllabus itself. The exam assumes a thorough knowledge of calculus, of probability as covered in Exam P, and of mathematical statistics as covered in the VEE Mathematical Statistics subject. Ability to solve problems using the R programming language is not assumed — but questions may present R output for interpretation, so reading output matters even though writing code does not. Specified calculators are permitted, and the statistical tables used on the exam (standard normal, t and chi-square) are provided during the sitting through an on-screen Exhibit, so candidates may not bring their own copies. Read the current SRM syllabus and exam rules at soa.org before you register.
The outline, area by area
| Content area | Published weight |
|---|---|
| Basics of Statistical Learning | 5–10% |
| Linear Models | 40–50% |
| Time Series Models | 10–15% |
| Decision Trees | 20–25% |
| Unsupervised Learning Techniques | 10–15% |
The SOA publishes each topic’s weight as a range intended to apply to the large majority of exams administered; an individual exam may fall outside a published range, and some questions cover multiple learning outcomes.
What does Basics of Statistical Learning cover? (5–10%)
The vocabulary and logic of modeling: supervised versus unsupervised learning, regression versus classification, how model accuracy is assessed, the bias-variance tradeoff and its effect on method choice, and the resampling methods used for validation — training-versus-test splits, k-fold cross-validation and leave-one-out cross-validation.
What does Linear Models cover? (40–50%)
The heaviest topic by a wide margin: generalized linear models — ordinary least squares assumptions against GLM assumptions, the exponential family and its link functions, interpreting parameters in business context, diagnostic tests of fit and assumptions by graphical and quantitative methods, model selection across distributions, links, transformations and interactions using t and F tests, AIC, BIC and the likelihood ratio test, predicted values with confidence and prediction intervals, and how regularized regression (lasso and ridge) and k-nearest neighbors differ from the ordinary least squares approach.
What does Time Series Models cover? (10–15%)
Regression-based time series: the concepts and components of stochastic time series processes, including random walks, stationarity and autocorrelation; the specific model families the syllabus names — exponential smoothing, autoregressive models and autoregressive conditionally heteroskedastic models — and calculating and interpreting predicted values and confidence intervals from them.
What does Decision Trees cover? (20–25%)
Tree models in depth: how decision trees are constructed and optimally fit to training data, how they are pruned to mitigate overfitting, prediction from classification trees and regression trees, and the ensemble methods built on trees — bagging, boosting and random forests — with the hyperparameters that control them, closing with a comparison of decision trees against linear models, their uses and relative strengths.
What does Unsupervised Learning Techniques cover? (10–15%)
Structure without a response variable: principal component analysis — what principal components are, how they are calculated, and how to interpret an analysis through loading factors and the proportion of variance explained — and cluster analysis, comparing the k-means and hierarchical clustering algorithms and the methods for deciding how many clusters the data supports.
What changed in the current syllabus?
The December 2026 syllabus carries five topics. Earlier editions carried six: Principal Components Analysis and Cluster Analysis stood as separate topics, Decision Trees was weighted at only 10–15%, and Time Series Models at 12.5–17.5%. In the current syllabus, principal component analysis and cluster analysis are merged into the single topic Unsupervised Learning Techniques (10–15%), Decision Trees has roughly doubled to 20–25%, and Time Series Models sits at 10–15%. Study material that still teaches the six-topic split — or that splits Linear Models into separate “linear regression” and “GLM” weight blocks — is teaching a superseded syllabus. This guide and the PassProfessor bank are built to the current five topics.
Exam SRM facts — checked November 19, 2026
Facts checked November 19, 2026 against soa.org (the SOA’s December 2026 Exam SRM syllabus, the SOA Exam Schedule, and the SOA’s fee pages). Exam SRM presents 35 multiple-choice questions in three and one-half hours (210 minutes) by computer-based testing, with a paper/pencil form at select centers on the window’s first day. Each question has five answer choices. In the syllabus’s own posture, “a few pilot questions will be randomly placed in the exam (both paper and pencil and computer-based forms)” and “will NOT be used in the scoring of this exam” — the SOA publishes no pilot count, and this page prints none. All unanswered questions are scored incorrect. Standard normal, t, and chi-square tables are provided at the sitting through an on-screen Exhibit. Unofficial pass/fail results are sent within one hour to the email address used to schedule the appointment; official results follow through the SOA’s Online Transcript Access, usually 8–11 weeks after the examination date.
Passing, in the SOA’s own form: grades are reported on a 0 to 10 scale, and passing grades range from 6 to 10. The SOA publishes no fixed passing percentage in advance; CBT forms are graded using an item response theory (IRT) procedure, and multiple forms given over a few days are kept comparable by statistical scaling. For the January 2026 sitting, the SOA’s January 2026 Examination Results (dated March 13, 2026) report that an average of 63% correct was needed to pass, and that 71.2% of the 1,218 candidates who wrote the exam passed (73.5% effective passing). Under CBT each administration consists of multiple versions over several days, constructed and scored using item response theory (IRT), with the pass mark for each examination determined before the examination is given — so that 63% figure is the January 2026 sitting’s own labeled result, not a standing passing percentage for Exam SRM.
Windows and fees: Exam SRM is offered three times a year — January 5–11, 2027 (registration deadline December 8, 2026), May 21–27, 2027 (deadline April 20, 2027), and September 7–13, 2027 (deadline August 10, 2027). The SOA’s fees are $357 (Tier 1), $286 (Tier 2), $214 (Tier 3), and $286 for eligible full-time students. Those fees are the SOA’s, collected by the SOA: registration, fees, scheduling, and administration belong to the SOA. PassProfessor’s prep program is a separate purchase at $59. Assumed knowledge: calculus, probability as covered in Exam P, and mathematical statistics as covered in VEE Mathematical Statistics. R posture: the ability to solve problems using R is not assumed; questions may present R output for interpretation.
Study plan for the January 5–11, 2027 sitting
This plan is built backward from the January 5–11, 2027 window — registration deadline December 8, 2026 — and allocates time at the December 2026 syllabus weights. How many hours a week it takes depends on your starting point; treat the week counts as a structure, not a promise, and adjust by evidence from your own practice scores.
Weeks 1–2 — foundations and survey. Work through Basics of Statistical Learning completely (5–10% of the exam, and the vocabulary every other topic uses), then survey Linear Models end to end so the season’s largest topic is mapped early.
Weeks 3–6 — Linear Models, deeply. At 40–50% this topic earns the largest block: OLS estimation, assumptions and inference; GLMs, the exponential family and links; diagnostics and model selection (t and F tests, AIC, BIC, likelihood ratio); prediction, regularization, and k-nearest neighbors. Pair every reading block with timed drills the same day, and log every miss as a knowledge gap, a misread, or a wrong method choice.
Weeks 7–8 — Decision Trees. At 20–25%: construction, pruning and cost-complexity, classification versus regression trees, then bagging, boosting, and random forests, closing with the trees-versus-linear-models comparison the syllabus names.
Week 9 — Time Series Models (10–15%). Stationarity, autocorrelation, random walks; exponential smoothing, autoregressive and ARCH models; forecasts with intervals.
Week 10 — Unsupervised Learning Techniques (10–15%). Principal component analysis (loadings, proportion of variance explained) and cluster analysis (k-means, hierarchical, choosing the number of clusters).
Final stretch — full timed practice and taper. Sit full 35-question simulations on the real 210-minute clock, review every explanation, and spend the remaining days on your two weakest topics and on logistics. Register by December 8, 2026 — the SOA’s deadline for the January window.
An evergreen 12-week plan for the May and September windows
For the May 21–27, 2027 window (deadline April 20, 2027) or the September 7–13, 2027 window (deadline August 10, 2027), stretch the same structure over roughly twelve weeks: two weeks of foundations and survey, four weeks on Linear Models, two on Decision Trees, one each on Time Series Models and Unsupervised Learning Techniques, and a final two weeks of mixed timed practice and full simulations at the syllabus weights. Keep the error log from the first week; from the halfway point, mix topics within each study day so retrieval stops depending on knowing the topic in advance.
If you need to retake Exam SRM
Exam SRM is offered three times a year, so a candidate who does not pass in the January 5–11, 2027 window can sit again in the next window, May 21–27, 2027 (registration deadline April 20, 2027), and again in September if needed. SOA examination fees are non-refundable and are payable again at each registration. Build the retake plan from your result: the standard applies to the examination as a whole, so weight the repair toward the topics where your practice evidence is weakest — for most candidates that means returning to Linear Models first, since at 40–50% it taxes the whole result — and re-sit full timed simulations before the window so the retake is paced, not just better informed.
Frequently asked questions
How many questions are on SOA Exam SRM, and how long is the sitting?
Exam SRM presents 35 multiple-choice questions in three and one-half hours (210 minutes), delivered by computer-based testing, with a paper/pencil form at select centers. A few pilot questions are placed randomly in the exam and are not used in scoring — the SOA publishes no pilot count — and all unanswered questions are scored incorrect.
What score do you need to pass SOA Exam SRM?
The SOA reports Exam SRM grades on a 0–10 scale, and grades from 6 to 10 are passing grades. The SOA publishes no fixed passing percentage or raw cut score: pass marks are set per sitting and form, with multiple CBT forms made comparable through item response theory (IRT) scaling. Practice simulations in this program are scored against a labeled 70% PassProfessor practice benchmark only, which is PassProfessor’s own marker and never the SOA’s standard.
What was the pass rate for SOA Exam SRM at the January 2026 sitting?
For the January 2026 sitting, the SOA’s January 2026 Examination Results (dated March 13, 2026) report that an average of 63% correct was needed to pass, and that 71.2% of the 1,218 candidates who wrote the exam passed (73.5% effective passing). Under CBT each administration consists of multiple versions over several days, constructed and scored using item response theory (IRT), with the pass mark for each examination determined before the examination is given — so that 63% figure is the January 2026 sitting’s own labeled result, not a standing passing percentage for Exam SRM. No other sitting’s figure is printed on this page without the same full label — exam, sitting, population, and SOA source.
What topics are on SOA Exam SRM and how are they weighted?
The December 2026 syllabus has five topics: Basics of Statistical Learning 5–10%, Linear Models 40–50%, Time Series Models 10–15%, Decision Trees 20–25%, and Unsupervised Learning Techniques 10–15%. This program’s 1,500-question bank follows the same five topics: Linear Models 675 questions, Decision Trees 337, Time Series Models 188, Unsupervised Learning Techniques 187, and Basics of Statistical Learning 113.
When is SOA Exam SRM offered in 2027, and what does the SOA charge?
Three windows: January 5–11, 2027 (registration deadline December 8, 2026), May 21–27, 2027 (deadline April 20, 2027), and September 7–13, 2027 (deadline August 10, 2027). The SOA’s fees are $357 (Tier 1), $286 (Tier 2), $214 (Tier 3), and $286 for eligible full-time students. Registration, fees, scheduling, and administration belong to the SOA; PassProfessor’s prep program is a separate $59 purchase.
Do you need to know R programming for SOA Exam SRM?
No. The December 2026 syllabus states that the ability to solve problems using R is not assumed. Questions may present R output — parameter tables, diagnostics, fitted values — for interpretation, so reading output fluently is a scoring skill even though writing code is not tested.
What is the difference between SOA Exam SRM and CAS Exam MAS-II?
Society and pathway. SRM is the Society of Actuaries’ Statistics for Risk Modeling examination on the ASA pathway, built on its five-topic syllabus (Linear Models 40–50%, Decision Trees 20–25%, and the rest as weighted in this guide). MAS-II is the Casualty Actuarial Society’s own statistics examination under a separate outline, on the CAS pathway. The two societies set their own syllabi, fees, and grading; no fact about one exam transfers to the other.
What happens if you fail Exam SRM — when can you sit again?
You can sit at the next window: a candidate who does not pass in January 5–11, 2027 can register for May 21–27, 2027 (deadline April 20, 2027), with further windows each September and January. SOA fees are non-refundable and payable again at each registration. Grades are reported on the SOA’s 0–10 scale; use the result and your practice evidence to weight the retake plan toward your weakest topics.
Does the real exam use five answer choices while this bank uses four?
Yes. The real Exam SRM presents five answer choices (A–E) per question; the PassProfessor practice bank uses four answer choices per question in the PassProfessor house format, and that difference is disclosed on the program and product pages. Practice simulations are scored against a labeled 70% PassProfessor practice benchmark only — PassProfessor’s own marker, never the SOA’s standard.
What does the free sample include?
The free sample is a 20-question set served inside the program’s free sample lesson: five easy, seven medium, five hard, and three stretch questions, spanning all five syllabus topics (Linear Models 10, Decision Trees 4, Basics of Statistical Learning 2, Time Series Models 2, and Unsupervised Learning Techniques 2), with an explanation for every answer and no questions about the exam itself.
Try the questions
Start with the program’s free sample: read about the SOA Exam SRM sample questions, or take the sample as a quiz in the free sample lesson. Then work the published SOA Exam SRM practice questions with full explanations, and see the full program on the SOA Exam SRM (Statistics for Risk Modeling) Exam Prep program page — 1,500 original practice questions built to the December 2026 syllabus, with an explanation for every answer, at $59 with 6 months of access.
