AI 440 · Year 4 · Semester 1 · 3 credits · AI & Machine Learning

Causal Inference for Pricing & Claims

Beyond correlation: mastering structural graphs, quasi-experiments, and causal machine learning to make optimal underwriting, pricing, and claims interventions.

Full practice exam · 30 Q · 120 min

Season 1 · 8 episodes

The Counterfactual Ledger

When Sovereign Mutual's predictive pricing models trigger a catastrophic spiral of adverse selection and regulatory scrutiny, senior pricing actuary Marcus Vance must dismantle eighty years of correlational dogma to rebuild the company's entire underwriting and claims engine on causal foundations.

Protagonist · Marcus Vance, FSA, CERA — Lead Pricing & Causal ML Actuary at Sovereign Mutual

Setting · The actuarial floor and executive boardroom of Sovereign Mutual in downtown Chicago, alongside state insurance department hearing rooms.

Stakes · A 108% combined ratio threatening solvency, fifty million dollars bleeding into counterproductive retention discounts, and an impending regulatory audit that could revoke Sovereign's operating license.

Pick an episode

Episodes

Season finale: Marcus deploys a Double Machine Learning causal pricing engine that isolates genuine price elasticity from risk confounding, reversing the underwriting deficit and setting a new benchmark for ASOP 12 compliance before the state commissioner.

From the lecturer: Welcome to AI 440. In standard predictive actuarial modelling, our goal is prediction: estimating expected loss given observed covariates. But insurance management is inherently interventional: What happens to renewal lapse rates if we raise premiums by 5%? Does early physical therapy triage causally reduce workers' compensation settlement duration, or are healthy claimants simply more compliant? To answer these questions without costly or unethical randomisation, you must master causal identification. Keep a dedicated notebook divided into two halves: on the left, draw DAGs and identify backdoor paths; on the right, write potential outcome equations, identification proofs, and sample moment estimators. Although this advanced Year 4 course does not feed a traditional preliminary exam, causal inference and uplift modelling represent the leading edge of predictive analytics in modern CAS/SOA research, predictive rate filing defences under ASOP 12 and ASOP 56, and commercial data science roles. Work through every derivation by hand before checking the film solutions, code each estimator from first principles in Python/R, and never confuse a prediction model with an interventional policy engine.

Cold open

Marcus stares at the executive dashboard at 6:45 AM. Sovereign Mutual just raised auto rates by 12% across tier-two drivers to improve profitability, but loss ratios actually worsened by four points within sixty days. Chief Actuary Eleanor Vance demands an explanation before the 9:00 AM board meeting: did the rate hike cause safe drivers to flee, or were claims going to spike anyway?

preparing lecture…

Transcript

Formulate causal questions using the Neyman-Rubin Potential Outcomes framework and state the Fundamental Problem of Causal Inference.
Decompose naive observational differences in means into Average Treatment Effect on the Treated (ATT) and Selection Bias.
Define and evaluate the three core identification assumptions: SUTVA, Unconfoundedness (Conditional Exchangeability), and Positivity.
Calculate and interpret Average Treatment Effect (ATE), ATT, and Average Treatment Effect on the Untreated (ATU) in pricing and claims interventions.