Exam SRM: Statistics for Risk Modeling Study Guide
Regression, time series, principal components, decision trees, cluster analysis, and predictive modeling.
Overview
Exam SRM (Statistics for Risk Modeling) tests statistical learning methods that actuaries use to build predictive models. It covers regression analysis, generalized linear models, time series, principal components analysis, decision trees, and clustering. The exam emphasizes understanding when and why to apply each method, not just computation.
Exam Format
- Duration
- 210 minutes (3.5 hours)
- Questions
- 35 multiple-choice questions
- Pathway
- SOA
- Format
- Computer-based testing (CBT)
Topic Breakdown
Linear Models
25-30%Simple and multiple linear regression, variable selection (stepwise, AIC, BIC), diagnostics, residual analysis
Generalized Linear Models
20-25%Exponential family, link functions, deviance, overdispersion, Poisson and logistic regression
Time Series
10-15%AR, MA, ARMA, ARIMA models, stationarity, ACF and PACF, model identification and forecasting
Principal Components Analysis
10-15%Eigenvalues and eigenvectors of covariance or correlation matrices, variance explained, dimension reduction
Decision Trees and Ensemble Methods
15-20%Classification and regression trees, pruning, bagging, random forests, boosting
Cluster Analysis
5-10%K-means, hierarchical clustering, distance metrics, choosing the number of clusters
Recommended Study Approach
- 1
Start with a thorough review of multiple linear regression, including diagnostics, multicollinearity, and variable selection.
- 2
Understand the bias-variance tradeoff deeply, as it underlies many exam questions about model selection.
- 3
Master GLMs by studying the link functions for Poisson, Binomial, and Gamma families and interpreting coefficients on the link scale.
- 4
Practice interpreting R output (coefficient tables, ANOVA, AIC/BIC comparisons) since many problems present computer output.
- 5
For tree-based methods, know how to compute Gini impurity and how pruning via cross-validation works.
- 6
Study PCA by working through eigenvalue decompositions of small covariance matrices by hand.