There are more than 600 CAS MAS-I practice questions on FreeFellow, and every one has a step-by-step solution. The full original question bank is free with an account, subject to usage limits. Worked solutions, written lessons, mixed practice, and your readiness score stay free. Samples on the MAS-I practice page need no signup.
The MAS-I study guide has the schedule and study plan, and MAS-I vs Exam SRM covers what carries over from the SOA's exam. Below are the CAS outline, where to practice each section and four worked questions.
The syllabus and the sitting it applies to
The CAS calls its syllabus a content outline. As of November 1, 2026, its MAS-I page links the 2026 Content Outline, dated June 29, 2026 in the file's version line. The page lists the current window, October 28 to November 5, 2026, and the next one, January 27 to February 2, 2027, without saying whether January gets a newer version. If you're sitting in January, check the page again before you lock in your reading list. Later 2027 dates are on the CAS site.
MAS-I (Modern Actuarial Statistics I) is 4 hours with a 15-minute scheduled break. The outline doesn't give a question count, but the CAS's paid practice exams run 45 questions in 4 hours, and its FAQ says they give the look and feel of the real exam.
Besides standard multiple choice, the outline lists multiple-selection, point-and-click, fill-in-the-blank and matching items. FreeFellow's bank is all five-choice, single-answer, so it won't rehearse those formats. Pearson VUE's CAS page has a sample item demo for them.
The outline's sections and where to practice them
This table shortens the CAS's own sections and tasks, with FreeFellow's practice page for each section in the last column.
Scroll horizontally to view all columns.
| Outline section and weight | Tasks | FreeFellow practice page |
|---|---|---|
| A. Probability Models, 20-30% | A1 and A2 Poisson processes; A3 limited expected value; A4 survival and hazard rates; A5 joint life; A6 whole life and annuities | Probability Models: more than 130 questions |
| B. Statistics, 20-30% | B1 mean and variance estimates; B2 sufficient statistics; B3 and B4 hypothesis tests; B5 frequency and severity; B6 aggregate claims; B7 order statistics; B8 maximum likelihood; B9 censoring and truncation | Statistics: more than 150 questions |
| C. Extended Linear Models, 45-55% | C1 and C2 choosing the model, link and distribution; C3 evaluating models; C4 reading software output; C5 predictor types and interactions; C6 offsets; C7 AIC, BIC, deviance and R-squared; C8 diagnostic plots; C9 exploratory plots | Extended Linear Models: more than 320 questions |
Questions carry topic and task tags, but some bank questions cover material outside the assigned readings, including boosting and random forests, so use the CAS reading list to check scope. The 26 written lessons use the same task numbers. Drilling one section or task is part of Fellow, and free accounts get one preview drill a week on Probability Models.
Section A has two labels the CAS doesn't use. Its task list stops at A6, but the Ross reading still assigns sections 4.1 to 4.8 (Markov chains), 9.1 to 9.6 (reliability) and 11.1 to 11.2.2 (simulation), and the outline says CAS exam writers use the expected readings exclusively. FreeFellow covers that material as A7 and A8. I'd study both, even though no task names them.
If you're using older material, time series moved to MAS-II starting with Fall 2023. FreeFellow also carries the CAS's own sample questions and its Spring and Fall 2018 and 2019 papers, with attribution and free, and the questions from those papers that are off the current outline are hidden by default.
Four worked questions
The four below are bank questions, with the task tag in each heading. One is from Section A and three are from Section C, which is about half the exam. None is from Statistics, because those questions lean on notation that's easier to read on the Statistics practice page, which has samples you can try with no signup.
A merged Poisson process (task A1)
Two insurance branch offices receive claims independently. Office A receives claims according to a Poisson process at rate 8 per hour and Office B at rate 5 per hour. The streams are merged at a central processing center.
Given that exactly 3 claims arrived at the processing center during a particular 15-minute window, calculate the probability that exactly 2 of those claims came from Office A.
- A. Less than 0.30
- B. At least 0.30, but less than 0.36
- C. At least 0.36, but less than 0.42
- D. At least 0.42, but less than 0.48
- E. At least 0.48
The answer is D. Merging independent Poisson processes gives a rate of 8 + 5 = 13 an hour, and each arrival independently came from Office A with probability 8/13. So given 3 arrivals, the count from Office A is binomial: . The 15-minute window doesn't matter, because only the ratio of the rates does. The plain Poisson probability of 2 claims from Office A in 15 minutes, with a mean of 2, is about 0.271, which lands in A, but it ignores the condition that exactly 3 arrived.
A parameter table with a centered predictor (tasks C4 and C5)
A Poisson GLM with log link is fit to private passenger auto claim frequency data with the following estimated parameters (intercept plus three covariates):
Scroll horizontally to view all columns.
| Parameter | Estimate |
|---|---|
| Intercept | -1.30 |
| Urban territory (vs. Rural) | 0.40 |
| Driver age, centered at 40 (per year) | -0.025 |
| Vehicle value (per $1,000) | 0.010 |
Calculate the expected annual claim frequency for a 30-year-old driver in an urban territory whose vehicle is valued at $20,000.
- A. Less than 0.40
- B. At least 0.40, but less than 0.55
- C. At least 0.55, but less than 0.70
- D. At least 0.70, but less than 0.85
- E. At least 0.85
The answer is C. Urban adds 0.40. Age is centered at 40, so a 30-year-old enters as -10 and adds . Value is per $1,000, so $20,000 enters as 20 and adds 0.20. The linear predictor is , and the log link gives . Plugging in 30 instead of -10 gives 0.235, which is A, and so does flipping the sign on the centered age, at 0.39. B drops the urban term and lands at 0.43.
What an offset does (task C6)
Which of the following statements is TRUE regarding the use of an offset variable in a generalized linear model?
- A. Its coefficient is freely estimated from the data.
- B. It must be on the same scale as the response variable.
- C. Its coefficient is fixed at 1 in the linear predictor.
- D. It is centered and scaled before model fitting.
- E. It is added to the response, not the linear predictor.
The answer is C. An offset is a known quantity, usually log exposure under a log link, that enters the linear predictor with its coefficient fixed at 1 instead of estimated. That's what makes the predicted count scale with exposure, so a 6-month policy gets half the expected claims of an otherwise identical 12-month policy. A describes an ordinary predictor. B and E are the same mix-up, since the offset sits on the scale of the linear predictor and never gets added to the response.
An F-test when dispersion is unknown (tasks C3 and C7)
An actuary fits two nested gamma GLMs with log link to N = 400 commercial property claim severities. The reduced model has residual deviance 425.6 on 394 degrees of freedom. The full model, which adds three interaction parameters, has residual deviance 411.2 on 391 degrees of freedom. Because the gamma dispersion parameter is unknown, the actuary uses an F-test, estimating the dispersion as the full model's residual deviance divided by its residual df. The value of the F-test statistic is closest to:
- A. 1.05
- B. 4.56
- C. 4.80
- D. 13.69
- E. 14.40
The answer is B. The deviance falls by 425.6 - 411.2 = 14.4 on 3 parameters, or 4.80 each. Dispersion is 411.2 / 391 = 1.052, so . C stops at 4.80 and never divides by dispersion, which is easy to miss when dispersion is this close to 1. D divides by dispersion but not by 3, E is the raw drop in deviance and A is the dispersion itself.
FreeFellow is not affiliated with the Casualty Actuarial Society. CAS does not endorse, promote, review, or warrant the accuracy or quality of FreeFellow products or services.