Question 1
A Poisson GLM with a log link models claim frequency using territory and vehicle use as categorical predictors. Which of the following statements is true?
The Exam SRM syllabus: five topics and their weights
Exam SRM has five topics. Linear models carry the most weight, at 40 to 50% of the exam, and decision trees come next at 20 to 25%.
| Basics of Statistical Learning | 5–10% of the exam |
|---|---|
| Linear Models | 40–50% of the exam |
| Time Series Models | 10–15% of the exam |
| Decision Trees | 20–25% of the exam |
| Unsupervised Learning Techniques | 10–15% of the exam |
Basics of Statistical Learning
- Supervised and unsupervised learning, regression and classification
- Measuring model accuracy, and the bias-variance tradeoff
- The validation set, k-fold and leave-one-out cross-validation
Linear Models
- Assumptions of least squares and generalized linear models
- The exponential family and link functions
- Interpreting parameters in the business context
- Diagnostics of fit, by graph and by test
- Model selection: distributions and links, transformations and interactions, t and F tests, AIC and BIC, and the likelihood ratio test
- Predicted values, and confidence and prediction intervals
- Ridge regression, the lasso and K-nearest neighbors, compared with least squares
Time Series Models
- Random walks, stationarity and autocorrelation
- Exponential smoothing, autoregressive and ARCH models
- Predicted values and confidence intervals
Decision Trees
- How trees are fit to training data and pruned
- Predicting with classification and regression trees
- Bagging, boosting and random forests, and their hyperparameters
- Trees compared with linear models
Unsupervised Learning Techniques
- Principal components: how they are calculated and how to read loadings and variance explained
- K-means and hierarchical clustering
- Choosing the number of clusters
The format
35 multiple-choice questions in 3.5 hours, each with five choices, on a computer. A few are unscored pilot questions you cannot tell apart. An unanswered question counts as wrong, so answer every one.
Normal, t and chi-square tables are provided on screen under an Exhibit button. You may not bring your own tables.
Writing R is not assumed, but questions may show R output for you to interpret.
What it assumes
SOA's syllabus assumes calculus, probability as on Exam P and mathematical statistics as in VEE Mathematical Statistics. The readings are Frees, Regression Modeling with Actuarial and Financial Applications, and James, Witten, Hastie and Tibshirani, An Introduction to Statistical Learning (2nd edition), both chapter by chapter.
Try 3 questions
Question 2
Which of the following statements about bagging regression trees is true?
Question 3
An actuary runs a principal components analysis on five standardized rating variables for auto policyholders. The table gives the loadings of the first two components and one policyholder's standardized values.
| Variable | PC1 loading | PC2 loading | Standardized value |
|---|---|---|---|
| Driver age | 0.5694 | −0.1968 | 0.23 |
| Years licensed | 0.5438 | −0.2674 | 0.54 |
| Prior claims | −0.5647 | −0.0448 | 0.16 |
| Annual mileage | −0.1608 | −0.6726 | −1.45 |
| Vehicle value | −0.1880 | −0.6599 | −0.13 |
Calculate this policyholder's score on the second principal component.
Next step
The same page for Exam P · The same page for Exam FM · The same page for Exam FAM
Questions people ask
What topics are on Exam SRM?
Five: Basics of Statistical Learning (5 to 10%); Linear Models (40 to 50%); Time Series Models (10 to 15%); Decision Trees (20 to 25%); and Unsupervised Learning Techniques (10 to 15%).
How many questions are on Exam SRM?
35 multiple-choice questions in 3.5 hours, six minutes each.
Is R needed for Exam SRM?
SOA's syllabus says the ability to solve problems in R is not assumed, but questions may present R output for interpretation.