What Exam 8 Covers

CAS Exam 8 is a fellowship-level exam focusing on advanced ratemaking for property and casualty insurance. The exam covers two major areas: classification ratemaking (developing multivariate rate plans) and excess and reinsurance pricing (pricing for layers above the primary coverage). Like other upper-level CAS exams, Exam 8 uses a written-answer format that requires candidates to explain methodology, justify assumptions, and demonstrate both technical skill and professional judgment. This exam draws heavily on concepts from earlier CAS exams, particularly Exam 5 (basic ratemaking) and MAS-II (predictive modeling).

Classification Ratemaking Fundamentals

Classification ratemaking involves developing rate relativities for different risk categories within an insurance product. For personal auto insurance, classification variables might include driver age, vehicle type, territory, and coverage level. The goal is to set relativities that accurately reflect the expected cost differences between classes while satisfying regulatory requirements for fairness and transparency.

Univariate analysis examines each rating variable independently, computing the loss ratio or pure premium for each class and setting relativities based on the ratios. However, univariate analysis can produce misleading results when rating variables are correlated. For example, if young drivers tend to live in urban areas, a univariate analysis might overstate the effect of age (by attributing some urban territory effect to age) and similarly overstate the territory effect. Multivariate methods correct for these confounding relationships.

Generalized Linear Models for Rating Plans

GLMs are the standard multivariate tool for classification ratemaking on Exam 8. A Poisson GLM with a log link models claim frequency, while a gamma GLM with a log link models claim severity. The multiplicative structure of the log link means that the predicted pure premium for any combination of rating variables is the product of a base rate and the relativities for each applicable class.

For example, if the base frequency is 0.05 claims per year, the age relativity for a young driver is 1.8, and the territory relativity for an urban area is 1.3, then the predicted frequency for a young urban driver is 0.05 * 1.8 * 1.3 = 0.117. The GLM estimates these relativities simultaneously, accounting for correlations between variables. On the exam, you should be able to interpret GLM output, explain what each coefficient means in terms of rate relativities, and discuss model diagnostics that assess whether the GLM is fitting the data adequately.

Minimum Bias Procedures

Before GLMs became standard, actuaries used minimum bias procedures to develop multivariate rate plans. These iterative procedures adjust class relativities to minimize the bias (systematic error) across all classification dimensions simultaneously. The Bailey-Simon method and the Iterative Minimum Chi-Square method are two common minimum bias procedures covered on the exam.

In the Bailey-Simon approach, you start with initial relativities and iteratively update each variable's relativities by computing the ratio of actual to expected losses, weighted by exposure, holding the other variables' relativities fixed. The process repeats, cycling through all variables, until the relativities converge. Minimum bias procedures produce results similar to GLMs under certain conditions but lack the formal statistical framework (standard errors, hypothesis tests, deviance statistics) that GLMs provide. The exam may ask you to compare the two approaches and explain the advantages of GLMs.

Credibility in Classification Ratemaking

Some rating classes have limited data, making their raw relativities unreliable. Credibility weighting addresses this by blending the class-specific estimate with a prior estimate (often the overall average or a complement from a larger dataset). The credibility-weighted relativity is Z * (class relativity) + (1 - Z) * (prior relativity), where Z reflects the volume of data in that class.

For GLMs, regularization techniques like LASSO and Ridge regression serve a similar purpose by shrinking coefficient estimates toward zero, which effectively pulls small-volume class relativities toward the base level. The exam may ask you to explain how credibility and regularization achieve similar goals through different mechanisms, and when each approach is appropriate.

Increased Limits Factors and Loss Layering

Excess and reinsurance pricing begins with understanding how losses distribute across different policy limits. Increased limits factors (ILFs) express the expected cost of providing coverage at a higher limit relative to a base limit. For example, if the base limit is $100,000 and the ILF at $500,000 is 1.85, then coverage at the $500,000 limit costs 85% more than at $100,000.

ILFs are computed from the loss size distribution. If F(x) is the cumulative distribution of individual loss amounts, the ILF at limit k relative to base limit b is: ILF(k) = E[min(X, k)] / E[min(X, b)], where E[min(X, k)] is the limited expected value at limit k. Common distributions for modeling loss severity include the lognormal, Pareto, and mixed exponential. The exam tests your ability to compute ILFs from given distributions and to explain how loss trends and development affect ILF calculations.

Reinsurance Pricing Methods

Reinsurance pricing uses two primary approaches: experience rating and exposure rating. Experience rating uses the cedant's historical loss experience in the reinsured layer to estimate future expected losses. This approach requires credible historical data at the layer being priced, which may not be available for high-excess layers with few historical losses.

Exposure rating uses the cedant's premium and loss size distribution to estimate the expected losses in each layer, regardless of whether historical losses have actually penetrated those layers. The key formula is: expected loss in layer (a, a+h) = subject premium * expected loss ratio * [ILF(a+h) - ILF(a)] / ILF(policy limit), where a is the attachment point and h is the layer width. Exposure rating is particularly useful for high-excess layers where experience is sparse. The exam frequently presents scenarios where you must choose between experience and exposure rating, or combine both approaches using credibility weighting.

Study Tips for Exam 8

Practice writing clear, structured responses that walk through your methodology step by step. For classification ratemaking, work through GLM examples including interpreting coefficients, computing predicted values, and comparing model fit statistics. Review minimum bias procedures and understand how they relate to GLMs conceptually. For excess and reinsurance pricing, practice computing ILFs from severity distributions and pricing reinsurance layers using both experience and exposure methods. Time management is essential in the written-answer format, so practice past exams under timed conditions. Allocate approximately 400 to 500 study hours and focus on understanding the reasoning behind each method rather than memorizing formulas in isolation.

CAS Exam 8: Advanced Ratemaking