CAS Exam MAS-II Guide
MAS-II is where the CAS statistics sequence turns into a real predictive-modeling and credibility exam: shrinkage, mixed models, statistical learning, and time-series interpretation in a property-casualty pathway.
What MAS-II Entails
- The second CAS modern actuarial statistics exam in the ACAS path.
- A deeper statistics and modeling step before later P&C credential requirements.
- Core practice: using official content outlines, tables, sample questions, and past exam materials.
CAS Exam MAS-II
Current MAS-II page and 2026 outline are mapped for format, domain weights, assumed knowledge, tables, and the reading list that drives credibility, mixed models, statistical learning, and time series preparation.
What the official PDFs establish
- Appointment length
- 4.5-hour appointment with a 4-hour exam duration.
- Scheduled break
- The appointment includes a scheduled 15-minute break plus tutorial/confidentiality/survey time.
- Assumed knowledge
- Calculus, probability, linear algebra concepts at the regression-prerequisite level, and mastery of MAS-I concepts are assumed.
Topic and domain coverage
| Topic | Weight | Source |
|---|---|---|
| Introduction to Credibility | 15-25% | |
| Linear Mixed Models | 10-20% | |
| Statistical Learning | 40-50% | |
| Time Series with Constant Variance | 15-25% | |
| Cognitive level: Remember | 5-10% | |
| Cognitive level: Understand and Apply | 55-60% | |
| Cognitive level: Analyze and Evaluate | 35-40% | |
| Cognitive level: Create | 0-5% |
Chapter and reading intelligence
- Tse
Credibility work is assigned from Nonlife Actuarial Models, covering classical, Buhlmann, Buhlmann-Straub, and Bayesian credibility sections in chapters 6-9.
- West
Linear Mixed Models: A Practical Guide Using Statistical Software is assigned across all chapters, excluding coding examples, with shrinkage notes called out separately.
- James et al., Salis, and GLM Monograph
Statistical learning is anchored to ISLR chapters 2.2, 4.4.2, 8, 10, and 12, Salis chapters 3 and 10, and Chapter 7 of Generalized Linear Models for Insurance Rating.
- Cowpertwait and Metcalfe
Time series preparation uses Introductory Time Series with R chapters 1-5 excluding selected sections, plus chapter 6 and sections 7.1-7.3.
Official files used by the map
- CAS content outlinecontent-outline
Primary source for domain weights, exam format, assumed knowledge, and official reading assignments.
Source: MAS-II Content Outline 2026
Quick Answer
MAS-II is the second CAS modern-statistics exam in the ACAS sequence. It follows MAS-I and pushes harder into credibility, linear mixed models, statistical learning, and time series before candidates move into PCPA and the operational P&C exams.
Official Format And Why MAS-II Feels Different
The current CAS outline uses a 4.5-hour appointment with a 4-hour exam duration and the broader CAS item-type mix rather than a plain multiple-choice format. Candidates may see multiple selection, point-and-click, fill-in-the-blank, matching, and other Pearson VUE item types, so MAS-II is partly a statistics exam and partly a platform-discipline exam.
The same outline also says MAS-II assumes calculus, probability, linear algebra at the usual regression-prerequisite level, and mastery of MAS-I concepts. That matters because the exam does not spend time reteaching GLM-style basics before asking you to interpret model output or credibility structure.
Where The Weight Really Sits
The content outline makes the center of gravity clear: statistical learning is the biggest domain, with credibility and time series each still large enough to matter, while linear mixed models create a separate interpretation lane many candidates underprepare for.
That shape is why MAS-II does not behave like a generic machine-learning survey. It is a targeted actuarial statistics exam that wants you to move between shrinkage logic, hierarchical structures, predictive-model comparison, and forecast interpretation without losing the P&C framing.
- Introduction to credibility: 15-25%.
- Linear mixed models: 10-20%.
- Statistical learning: 40-50%.
- Time series with constant variance: 15-25%.
What The Official Readings Are Actually Doing
The reading list is much more revealing than the exam name. Credibility is sourced from Tse. Mixed-model interpretation is anchored to West. Statistical learning is split between ISLR and the CAS GLM monograph chapter on predictive-performance measures. Time series comes from Cowpertwait and Metcalfe.
That stack tells you the exam is not just about memorizing algorithms. It is about recognizing what kind of modeling problem you are in, choosing the right tool family, and reading the output well enough to defend a modeling decision.
How It Fits After MAS-I And Before PCPA
MAS-I is the broader probability, statistics, and extended-linear-model base. MAS-II narrows that into a more advanced modeling exam. Then PCPA asks whether you can apply that modeling language in a practical property-casualty analytics workflow.
So MAS-II is not a side quest. It is the bridge from theoretical statistical fluency into the parts of the CAS path that start to look like real predictive-modeling work.
How To Study MAS-II Rationally
A practical plan is to split prep into four lanes: credibility, mixed models, statistical learning, and time series. Statistical learning gets the most hours, but mixed models and credibility are where many candidates lose easy points because the material feels less familiar and the output interpretation is easy to bluff until you are under time pressure.
The fastest way to make the exam feel coherent is to treat credibility, mixed models, and statistical learning as different forms of shrinkage and model comparison rather than isolated chapters. That mental link is not the whole exam, but it makes the material much easier to organize.
Original exam practice
6 questions built from syllabus outcomes and released-exam patterns. The prompts and answers are original, so they train the skill without copying official exam text.
MAS-II Modeling Readiness Drill
Original MAS-II checks for credibility, mixed-model interpretation, validation, regularization, unsupervised learning, and time-series diagnostics.
- Question 1/Flashcard
Largest MAS-II domain
Which MAS-II domain carries the largest official weight, and what should that change in a study plan?
Solution and grading points
Statistical Learning is the largest domain at 40-50%. A study plan should reserve the largest practice block for validation, tuning, regularization, trees, clustering, PCA, predictive-performance measures, and model-choice interpretation.
- Names Statistical Learning.
- Gives the 40-50% weight range.
- Connects the weight to model-choice and validation practice.
- Question 2/Calculation
Buhlmann-Straub credibility estimate
A territory has exposure 40 and observed loss cost 1,250. The collective mean is 1,000, EPV is 900, and VHM is 30. Using Z = w/(w+K) and K = EPV/VHM, find the credibility estimate.
Solution and grading points
K = 900 / 30 = 30. Z = 40 / (40 + 30) = 0.5714. The credibility estimate is 0.5714(1,250) + 0.4286(1,000) = 1,142.9.
- Compute K as EPV divided by VHM.
- Compute the exposure-based credibility weight.
- Blend the territory loss cost with the collective mean.
- Interpret the result as a shrinkage estimate.
- Computes K = 30.
- Computes Z near 0.5714.
- Reports a credibility estimate near 1,142.9.
- Explains why the answer lies between 1,000 and 1,250.
- Question 3/Written Answer
Random territory effect
A mixed model has fixed effects for vehicle age and coverage limit, plus random intercepts by territory. What does the random intercept represent?
Solution and grading points
The random intercept represents a territory-level deviation from the population intercept after the fixed effects are accounted for. It lets territories vary around the overall relationship while shrinking noisy territory estimates toward the population mean.
- Identifies the random intercept as a group-level deviation.
- States that fixed effects are already accounted for.
- Connects limited territory information to shrinkage.
- Question 4/Written Answer
Validation gap
Model A has training RMSE 820 and validation RMSE 1,410. Model B has training RMSE 930 and validation RMSE 1,050. Which model is more defensible for production pricing, absent other constraints?
Solution and grading points
Model B is more defensible because validation error is lower and the training-validation gap is smaller. Model A is likely overfit to the training sample.
- Chooses Model B.
- Uses validation error rather than training error as the main evidence.
- Identifies Model A as an overfitting warning.
- Question 5/Written Answer
Lasso versus ridge
Give one practical difference between lasso and ridge regularization in model selection.
Solution and grading points
Ridge shrinks coefficients toward zero but usually keeps all predictors. Lasso can shrink some coefficients exactly to zero, so it can perform variable selection.
- States that both methods shrink coefficients.
- Identifies lasso as capable of setting coefficients to zero.
- Connects lasso to variable selection.
- Question 6/Written Answer
Stationarity warning
A quarterly loss-cost index has a strong upward trend and positive autocorrelation. Why might a stationary ARMA model on the raw series be inappropriate?
Solution and grading points
The trend suggests the raw series is not stationary, so an ARMA model that assumes stable mean and autocovariance may be misspecified. A candidate should consider detrending, differencing, or modeling the trend before checking residual autocorrelation.
- Names the trend as a stationarity problem.
- Connects stationarity to stable mean or autocovariance.
- Gives at least one reasonable response such as detrending or differencing.