CAS Research & Industry Advancements 3 ā Questions and Answers
Question 1: The CAS Working Party on Generative AI in Actuarial Science is most concerned with which limitation of large language models for actuarial tasks?
- High computational cost of training
- Hallucination and lack of auditability in numerical outputs (Correct answer)
- Inability to process text data
- Requirement for proprietary hardware
Correct answer: Hallucination and lack of auditability in numerical outputs
Hallucinationāgenerating plausible but incorrect outputsāand poor auditability are key concerns when applying LLMs to actuarial calculations requiring precision.
Question 2: In insurance research, 'loss development factor (LDF) selection' using machine learning differs from the volume-weighted average approach primarily because ML methods:
- Always produce lower IBNR estimates
- Can incorporate external covariates and detect structural breaks in development patterns (Correct answer)
- Require no historical paid loss data
- Eliminate the need for actuarial judgment entirely
Correct answer: Can incorporate external covariates and detect structural breaks in development patterns
ML-based LDF selection can leverage external predictors (e.g., legal environment, inflation) and identify non-stationarity that simple averages miss.
Question 3: Which research technique is used to assess the marginal contribution of individual predictors in a black-box insurance pricing model?
- Forward stepwise regression
- SHAP (SHapley Additive exPlanations) values (Correct answer)
- Credibility weighting
- Bornhuetter-Ferguson method
Correct answer: SHAP (SHapley Additive exPlanations) values
SHAP values decompose a model's prediction into contributions from each feature, providing interpretability for complex models.
Question 4: The CAS research area of 'resilience modeling' for catastrophe risk focuses primarily on:
- Reducing reinsurance premiums through captive structures
- Quantifying post-disaster recovery trajectories and their impact on insured losses (Correct answer)
- Eliminating parametric triggers from catastrophe bonds
- Standardizing policy language across jurisdictions
Correct answer: Quantifying post-disaster recovery trajectories and their impact on insured losses
Resilience modeling examines how quickly communities and economies recover after catastrophes, which directly affects loss development and business interruption claims.
Question 5: A key finding in CAS research on social inflation is that it is primarily driven by:
- Increases in medical cost inflation only
- Changes in jury verdicts, litigation funding, and plaintiff attorney tactics increasing loss severity (Correct answer)
- Federal interest rate increases affecting reserve discounting
- Expanded policyholder deductible buyback options
Correct answer: Changes in jury verdicts, litigation funding, and plaintiff attorney tactics increasing loss severity
Social inflation refers to rising claim costs driven by legal system changesānuclear verdicts, third-party litigation funding, and aggressive plaintiff strategies.
Question 6: When CAS researchers use a train/validation/test split in a predictive model for claim frequency, the test set's primary purpose is to:
- Tune hyperparameters during model training
- Provide a final unbiased estimate of model performance on unseen data (Correct answer)
- Select variables for inclusion in the model
- Calibrate credibility weights for sparse segments
Correct answer: Provide a final unbiased estimate of model performance on unseen data
The test set is held out until final model selection is complete, providing an unbiased estimate of how the model will perform in production.
Question 7: Which CAS research paper category addresses the application of stochastic simulation methods to estimate the full distribution of unpaid claim liabilities?
- GLM ratemaking papers
- Stochastic reserving and bootstrapping papers (Correct answer)
- Underwriting cycle analysis papers
- Premium auditing methodology papers
Correct answer: Stochastic reserving and bootstrapping papers
Stochastic reserving research uses bootstrapping and simulation to produce probability distributions of IBNR, not just point estimates.
CAS Research & Industry Advancements 3
The CAS Working Party on Generative AI in Actuarial Science is most concerned with which limitation of large language models for actuarial tasks?