Start with a thorough review of linear algebra basics

Matrix multiplication, eigenvalues, and eigenvectors appear in PCA and regression. If your linear algebra is rusty, spend a few days reviewing before starting the SRM syllabus.

Understand bias-variance tradeoff conceptually

Many exam questions test whether you understand why adding model complexity reduces bias but increases variance. Be able to explain this tradeoff in the context of polynomial regression, tree depth, and regularization.

Practice reading R output

The exam frequently presents R regression output and asks you to interpret coefficients, p-values, or diagnostic plots. Practice reading summary() output and ANOVA tables from real R sessions.

Know when to use each variable selection method

Understand the pros and cons of forward selection, backward elimination, and stepwise selection. Know that AIC and BIC can give different results and why (BIC penalizes complexity more heavily).

Study GLM link functions with worked examples

For each GLM family (Gaussian, Poisson, Binomial, Gamma), know the canonical link, how to interpret coefficients on the link scale, and how to convert predictions back to the response scale.

Practice time series model identification

Given ACF and PACF plots, you should be able to identify whether the generating process is AR, MA, or ARMA. Practice with many examples until pattern recognition becomes fast.

Understand cross-validation deeply

Know the difference between leave-one-out, k-fold, and repeated k-fold cross-validation. Understand why we use cross-validation for model selection rather than in-sample metrics like R-squared.

Learn tree methods by building small trees by hand

Work through a decision tree split calculation with Gini impurity or deviance on a 10-row dataset. This makes the algorithm concrete and helps you answer questions about pruning criteria.

Connect every method to an actuarial application

The exam tests practical understanding. Know that GLMs are used in insurance pricing, PCA for correlated risk factors, and clustering for market segmentation. These connections help with conceptual questions.

Review the SOA study note on predictive analytics

The SOA-authored study note is required reading and is the source for many exam questions. Do not skip it in favor of general statistics textbooks.

SRM Study Tips