STAT 203 · Year 2 · Semester 2 · 4 credits · AI & Machine Learning

Regression & Statistical Learning (Exam SRM)

From linear foundations to generalized models and modern ensembles: mastering statistical learning for actuarial science and SOA Exam SRM.

Full practice exam · 30 Q · 120 min

Season 1 · 8 episodes

The Telematics Audit

When Meridian Specialty Mutual faces regulatory receivership over a hemorrhaging commercial fleet portfolio, pricing actuary Maya Chen must rebuild their statistical learning engine from ordinary least squares to regularized ensembles before the insurance commissioner shuts them down.

Protagonist · Maya Chen, ASA, newly appointed Lead Predictive Modeler at Meridian Specialty Mutual.

Setting · Meridian Specialty Mutual's headquarters in Chicago, amid an emergency regulatory rate filing and an impending state solvency hearing.

Stakes · A 118 percent combined ratio on a five-hundred-million-dollar commercial fleet book threatens Meridian's solvency; failure to produce an ASOP-compliant, statistically sound pricing and severity pipeline means immediate regulatory takeover.

Pick an episode

Episodes

Season finale: Maya defends a fully unified, bias-variance optimized predictive framework before the State Insurance Commissioner and the Board, turning the portfolio profitable while establishing a transparent compliance benchmark under ASOP 56.

From the lecturer: Welcome to STAT 203. This course sits at the exact intersection of classical actuarial theory and modern machine learning, directly targeting the syllabus of SOA Exam SRM (Statistics for Risk Modeling) while preparing you for predictive modeling in pricing, reserving, and capital management. To get the most out of these lecture films, never watch them passively. Keep an engineering notebook and an active R or Python session side by side with the screen. When a derivation begins—whether deriving the normal equations via matrix calculus or computing the Iteratively Reweighted Least Squares (IRLS) step for a Tweedie GLM—pause the video and attempt the next algebraic leap yourself before letting me reveal it. When we transition from mathematics to empirical data, reproduce the diagnostic plots and cross-validation curves in code immediately. Exam SRM demands both conceptual fluency and rapid computational precision: you must know not only why the Lasso penalty sets coefficients to zero while Ridge shrinks them smoothly, but also how to calculate information criteria and deviance changes on timed exam problems. Let us build your statistical modeling intuition from the ground up.

Cold open

At six forty-five on a rainy Monday morning, Chief Actuary Elena Vance drops a two-hundred-page regulatory notice onto Maya's desk: the state insurance commissioner has rejected Meridian's commercial fleet rate filing due to unexplained residual patterns and massive severity underestimations, giving them sixty days to overhaul their linear baseline model.

preparing lecture…

Transcript

Derive the ordinary least squares estimator in matrix form and state its finite-sample sampling distribution under classical Gauss-Markov assumptions.
Decompose total variation to construct the ANOVA F-test, coefficient of determination, and adjusted R-squared for nested model comparison.
Diagnose violations of linear regression assumptions using residual plots, Q-Q plots, leverage values, and Cook's distance in insurance pricing contexts.