The pass rate for SOA Exam SRM (Statistics for Risk Modeling) was 67.7% at the May 2026 sitting and 71.2% in January 2026, each on a pass mark of 63% average correct, and the last four sittings have all landed between 67.7% and 71.2%. That's a higher pass rate than most SOA preliminaries. SRM is a 3.5-hour exam with 35 multiple-choice questions on the statistical and predictive modeling methods actuaries use, and it usually comes after Exams P, FM and FAM.
SRM is where the actuarial exams move from pure math to applied data science. A lot of it is what data scientists and quantitative analysts learn, applied to actuarial problems, so if you liked the statistics and modeling courses in your degree, it'll feel natural. Most questions hand you fitted output, like a regression table, deviance statistics or an ACF plot, and ask you to read it, compare models or spot what's wrong. That's a different skill from Exams P and FM, where you crank calculations from scratch, and it's why the reading list is anchored by a statistical learning text rather than a probability text.
What the exam covers
There are five topic areas on the SOA syllabus:
Scroll horizontally to view all columns.
| Topic | Weight | What's in it |
|---|---|---|
| Basics of statistical learning | 5-10% | Supervised vs unsupervised learning, regression vs classification, model flexibility and the bias-variance tradeoff, training vs test error, resampling (validation set, k-fold cross-validation) |
| Linear models | 40-50% | Simple and multiple linear regression, assumptions and diagnostics, residual analysis, multicollinearity, interpreting coefficients, R-squared and adjusted R-squared, ridge and lasso, K-nearest neighbors, and generalized linear models |
| Time series models | 10-15% | Random walks, stationarity and autocorrelation, exponential smoothing, autoregressive (AR) and ARCH models, forecasting |
| Decision trees | 20-25% | Classification and regression trees (CART), pruning, bagging, boosting, random forests |
| Unsupervised learning | 10-15% | Principal components analysis (eigenvalues, loadings, proportion of variance explained), K-means and hierarchical clustering, choosing the number of clusters |
Linear models are the biggest block by far, and the GLM half matters most, since it's how actuaries model insurance data. That means link functions (log, logit, identity), exponential-family distributions (Poisson, binomial, gamma, normal), logistic regression for binary outcomes, Poisson regression for counts, deviance, AIC and BIC for comparing models, forward, backward and stepwise variable selection, and overdispersion and how to handle it. I'd put at least 40% of your study time here.
People who studied machine learning in school sometimes rush through GLMs, figuring logistic regression is easy. The exam goes deeper than most intro ML courses: you need deviance residuals, link function choice and overdispersion diagnostics, not just the ability to fit a model. Decision trees are the second-largest area, so give them real time too. PCA and clustering are tested mostly on concepts and reading output.
The SOA updates the syllabus and reading list from time to time, so check the SOA's Exam SRM page for the current syllabus and readings before you lock in a plan. The weights above match the current version.
The main reference is An Introduction to Statistical Learning (ISLR) by James, Witten, Hastie and Tibshirani, and it's free from the authors' website. That's a real advantage, since reference texts for other actuarial exams run $100+. ISLR is well written and easy to follow. Read it carefully, work the conceptual exercises, and you'll have most of SRM covered. The one gap is GLMs, which ISLR doesn't cover as deeply as the SOA wants, so plan on supplementary material there, like the SOA study note on GLMs.
How I'd spend 13 weeks
Plan for roughly 250 to 300 hours, which works out to 20 to 23 hours a week over 13 weeks:
- Weeks 1 to 3: linear regression. Don't rush it, because every GLM concept extends from it. Practice reading regression output, checking assumptions and knowing what each diagnostic is telling you.
- Weeks 4 to 7: GLMs. Spend the most time here. Work through Poisson, logistic and gamma regression examples, calculate deviance, compare models with AIC and BIC, and practice spotting overdispersion.
- Weeks 8 to 10: time series, PCA, trees and clustering. These each weigh less than linear models, but they show up on every exam. Time series is the most calculation-heavy of the four, and PCA and clustering lean conceptual.
- Weeks 11 to 13: full-length practice exams (35 questions, 3.5 hours). You get 6 minutes per question on average, a full minute more than Exam FM's five, because the material is so interpretive. Use the extra time to read output tables closely and double-check your reasoning.
Plan on 700 to 1,000 practice questions before exam day. As rough benchmarks, you should be scoring 45 to 55% on mixed practice after 300 questions, 55 to 65% on practice exams after 600, and 65%+ consistently on full-length mocks when you're ready.
Where people lose points, and what to study from
The biggest mistake is underinvesting in GLMs and overinvesting in topics like PCA or clustering. Linear models, GLMs included, are 40 to 50% of the exam. The second is memorizing formulas without understanding them. Knowing the formula for AIC matters less than knowing when AIC points to a different model than BIC. The third is ignoring diagnostics, since the exam keeps asking what's wrong with a model or which diagnostic flags a problem, so practice reading residual plots and goodness-of-fit statistics. Some candidates study SRM like Exam P, drilling calculations, when what you need is to know what the numbers mean.
For free material, FreeFellow's Exam SRM practice has 1,028 questions covering all five topic areas with worked solutions and a free readiness score. Fellow adds focus practice on your weak topics and analytics. ISLR is the free primary reference, the SOA study note on GLMs fills in what ISLR leaves out, and the Exam SRM sample questions post has all 66 SOA-published sample items with worked solutions.