FCAS exams

Exam 8

Advanced Ratemaking

5 discussions

Study strategy

Exam 8: Validate the rating algorithm after fitting the model

A fitted model and an implemented rating plan can differ because of rounding, normalisation, constraints or interactions. Validation should check the entire algorithm and its portfolio impact, not just coefficient significance.

Worked example or practice scenario

A fitted relativity is normalised against one reference class, while a production table uses another. Copying coefficients directly can shift the overall rate level. Test representative risks and reconcile aggregate indicated versus implemented premium on a common exposure set.

Try this next

Prepare an original test set covering boundaries, missing values, rare classes and interactions. Record the off-balance and the selected normalisation. The current Exam 8 outline emphasises classification and individual-risk applications; use the sitting’s actual outline for scope.

Reading sources

Official syllabus and resources

ActNet editorial guide · October 1, 2026 · Original illustrative examples.

0 repliesOct 1

Explanation

Exam 8: A deductible option changes both payments and reported experience

A deductible removes small payments and changes the insurer’s severity and payment frequency. Whether policyholder behaviour changes is a separate modelling question. Do not treat a higher deductible as a uniform percentage reduction in every loss.

Worked example or practice scenario

For a loss of 300, raising an ordinary deductible from 100 to 200 reduces payment from 200 to 100. For a loss of 3,000 with no other limit, payment falls from 2,900 to 2,800. The payment percentage reduction differs by loss size.

Try this next

Define the loss distribution and calculate expected payments under each deductible. Distinguish per-loss and per-payment severity. Explain what data could be missing when historical records include only payments above the existing deductible.

Reading sources

Official syllabus and resources

ActNet editorial guide · October 1, 2026 · Original illustrative examples.

0 repliesOct 1

Explanation

Exam 8: Individual experience rating needs credibility and comparability

An individual risk’s past results can be informative, but changes in exposure, limits and loss maturity must be addressed before blending them with a broader benchmark. Credibility is not a reason to average incompatible bases.

Worked example or practice scenario

An account has adjusted own expected cost 140 and comparable benchmark cost 100 per exposure. With an illustrative credibility weight .30, blended cost is .30×140+.70×100=112. The weight is assumed here, not estimated from the single quoted observation.

Try this next

Describe the data and model needed to justify a weight. Recalculate after an exposure or coverage change and explain why applying the old blend to a different coverage basis can be misleading.

Reading sources

Official syllabus and resources

ActNet editorial guide · October 1, 2026 · Original illustrative examples.

0 repliesOct 1

Explanation

Exam 8: A layer cost is a difference of limited expected values

For a layer of width u above attachment d, payment is min((X−d)+,u). Its expectation equals E[min(X,d+u)]−E[min(X,d)] under a nonnegative loss model. Ground-up, limited and excess data have different meanings.

Worked example or practice scenario

For a toy exponential loss with mean 1,000, a 500-wide layer above 500 has expected payment 1,000(e^−.5−e^−1)≈238.65. The expected payment is not simply 500 times the probability of exceeding 500 because some losses only partly use the layer.

Try this next

Draw the payment function and integrate the survival function across the layer. Then explain how censored policy-limit observations affect fitting the underlying severity distribution. State whether limits apply per claim, per occurrence or in aggregate.

Reading sources

Official syllabus and resources

ActNet editorial guide · October 1, 2026 · Original illustrative examples.

0 repliesOct 1

Explanation

Exam 8: Class relativities need a common comparison basis

A raw segment loss ratio can reflect several correlated rating variables and different premium adequacy. Classification ratemaking must account for the intended rating structure. A univariate comparison is not automatically a new relativity.

Worked example or practice scenario

Suppose one vehicle class has more high-mileage drivers than another. Its higher observed claim frequency may partly reflect mileage mix. A multivariate model can separate effects under its specification, but confounding, sparse data and changed mix still need review.

Try this next

Compare raw summaries with an adjusted model result and explain the difference. Identify the benchmark category, normalisation and exposure basis. Check the prospective effect when inserting a relativity into the full rating algorithm.

Reading sources

Official syllabus and resources

ActNet editorial guide · October 1, 2026 · Original illustrative examples.

0 repliesOct 1