This guide develops advanced property and casualty ratemaking foundations for candidates preparing for the Casualty Actuarial Society Advanced Ratemaking Examination (CAS Exam 8). Work through the concepts in order, then use the examples to check calculations and interpret assumptions. Each example uses a stated, simplified model; actual rating plans and contracts require their own definitions.
Classification ratemaking and generalized linear models
1. Comparable losses and exposures
A prospective loss cost requires losses and exposures measured on compatible coverage, valuation and time bases. Develop historical losses toward ultimate and adjust them for relevant cost trends before comparing classes. Separate changes in underlying risk from changes in policy limits, deductibles or claim handling. Exposure definitions must remain consistent throughout the calculation.
Worked example: Historical losses of 200,000 receive a development factor of 1.10 and a trend factor of 1.05. Adjusted losses are 231,000; with 1,000 comparable exposure units, the loss cost is 231.
Mistake to avoid: Comparing developed losses for one class with undeveloped losses for another.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
2. Frequency and severity decomposition
Expected loss cost per exposure equals expected claim frequency multiplied by expected severity, using matching claim definitions. This decomposition helps distinguish more claims from more expensive claims. Claims excluded from severity must also be excluded from the corresponding frequency measure. Separate models can reveal effects that cancel or reinforce each other in aggregate loss cost.
Worked example: A class has 80 claims across 1,000 exposures and average severity of 1,250. Frequency is 0.08, so its indicated loss cost is 0.08 × 1,250 = 100 per exposure.
Mistake to avoid: Combining all-claim frequency with severity calculated only from large claims.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
3. Normalizing classification relativities
Relativities express differences between classes; their average depends on the exposure mix. Divide each relativity by the exposure-weighted average when the objective is a weighted average of one. Adjust the base rate consistently if existing class prices must be preserved. Normalization changes the reference scale rather than the relative differences between classes.
Worked example: Classes have exposure shares of 60% and 40%, with relativities 1.00 and 1.50. Their weighted average is 1.20. Normalized relativities are 0.8333 and 1.25, whose weighted average is one.
Mistake to avoid: Using an unweighted average when the classes have different exposure volumes.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
4. Log links and multiplicative effects
With a log link, a generalized linear model specifies log(expected response) as a linear predictor. A coefficient therefore acts multiplicatively on the expected response: its factor is exp(coefficient). This differs from fitting the logarithm of observed losses by ordinary regression, which introduces a separate retransformation issue. Interpret coefficients on the scale the model actually uses.
Worked example: A log-link coefficient of log(1.20) gives a factor of 1.20. If the reference expected severity is 500, the indicated severity for that category is 600.
Mistake to avoid: Interpreting a coefficient of 0.20 as exactly a 20% increase.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
5. Exposure offsets in count models
A log-exposure offset fixes expected claim counts as proportional to exposure while allowing explanatory variables to model frequency. Its coefficient is fixed at one rather than estimated. Use positive exposure values and a count response consistent with that denominator. A rate response with exposure weights is a different specification and should not receive an offset automatically.
Worked example: At modeled frequency 0.04, accounts with 100 and 250 exposure units have expected counts of 4 and 10. The exposure offset supplies this proportional difference.
Mistake to avoid: Treating the exposure offset as an additional fitted rating factor.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
6. Frequency variance and overdispersion
A Poisson count model has conditional variance equal to its conditional mean. Insurance counts may show additional variation from unmodeled differences or dependence. Overdispersion can affect uncertainty estimates and model choice even when the mean specification is useful. Diagnose it after accounting for exposure and explanatory factors; raw portfolio variance also reflects differences between risks.
Worked example: Comparable fitted groups each have mean count 2, but their observed conditional variance is about 5. A Poisson variance assumption understates variability; the mean model and remaining heterogeneity both deserve examination.
Mistake to avoid: Diagnosing overdispersion solely from the variance of counts across unequal exposures.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
7. Severity models and proportional variability
A Gamma generalized linear model is useful for strictly positive, right-skewed severity when variance is proportional to squared mean. Under a common dispersion parameter, the coefficient of variation is constant. The response should represent compatible claim amounts, and observation weights should reflect the intended sampling unit when modeling averages rather than individual claims.
Worked example: With dispersion 0.25, mean severity 200 implies standard deviation 100. At mean severity 400, standard deviation becomes 200; the coefficient of variation remains 0.50.
Mistake to avoid: Including zero-loss observations in a Gamma response without changing the modeling approach.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
8. Aggregate loss models with zeros
Aggregate losses often contain an exact zero and continuous positive amounts. A Tweedie model with variance power between one and two can represent this through a compound Poisson-Gamma structure. Choose response, exposure treatment and weights together. A direct aggregate model can complement separate frequency and severity models, but agreement on average loss does not guarantee matching tail behavior.
Worked example: Two otherwise comparable policies record losses of zero and 900. A suitable aggregate model can accommodate both observations without adding an arbitrary constant before taking logarithms.
Mistake to avoid: Assuming equal predicted means imply equal probabilities of a claim or equal large-loss behavior.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
9. Interactions between rating factors
An interaction allows one rating factor's effect to depend on another factor. Without interactions, a log-link model applies the same multiplicative effect across all levels of other variables. Evaluate interactions using adequate joint exposure, interpretable risk mechanisms and validation. Sparse combinations can produce unstable estimates even when the overall categories have substantial volume.
Worked example: An age factor is 1.20 for one vehicle type but 1.50 for another. A vehicle-by-age interaction permits that difference; a main-effects-only model forces a common age factor.
Mistake to avoid: Adding every possible interaction without examining the data in each combination.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
10. Predictive validation and residual patterns
Validate predictions on observations excluded from estimation, using a split that reflects the intended future application. Repeated records from the same risk can require grouped splitting; prospective pricing may benefit from time-based validation. Examine residual patterns and observed-to-expected results by meaningful segments. Overall accuracy can conceal a consistently overpriced or underpriced class.
Worked example: A model has total observed-to-expected loss of 1.00, but a substantial region has 1.30 in a later validation period. The total masks regional underprediction that warrants investigation.
Mistake to avoid: Declaring a model adequate because its total fitted losses equal total historical losses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Experience rating and credibility
11. Expected losses reflect the risk mix
Experience rating compares actual experience with an expectation appropriate to the account's exposures and classifications. Calculate expected losses from the relevant exposure mix rather than using a portfolio-wide average indiscriminately. Differences in expected losses can arise from volume or composition. Comparing raw losses between accounts confounds those differences with evidence about individual risk quality.
Worked example: An account has 100 units in a class costing 2 per unit and 50 units in a class costing 4. Expected losses are 100 × 2 + 50 × 4 = 400.
Mistake to avoid: Interpreting higher total losses as worse experience without adjusting for expected losses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
12. Within-risk noise and between-risk differences
Credibility balances random fluctuation within an individual risk against persistent differences between risks. More observations help distinguish those components, but observation count alone does not determine credibility. When risks have similar underlying means, individual deviations provide less evidence of persistent differences. When genuine differences are large relative to noise, individual experience receives more weight.
Worked example: Two portfolios have equal account sizes. The portfolio with more persistent differences in account means, relative to annual randomness, supports greater individual credibility under an otherwise comparable credibility model.
Mistake to avoid: Assigning identical credibility solely because two accounts have the same number of years.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
13. Limited-fluctuation precision
Limited-fluctuation credibility evaluates whether experience is sufficiently precise relative to a selected tolerance and probability criterion. Under a Poisson claim-count assumption, the coefficient of variation of total count is one divided by the square root of expected count. Loss amounts require additional severity variability. The precision criterion is a modeling choice, not a universal mastery or rating threshold.
Worked example: Expected counts of 400 and 1,600 have Poisson coefficients of variation of 5% and 2.5%. Quadrupling expected count halves relative count variability.
Mistake to avoid: Applying a count-only precision calculation to aggregate losses with variable claim severity.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
14. Bühlmann credibility weighting
In the basic Bühlmann model, the credibility estimate is Z times the individual mean plus one minus Z times the collective mean. For n comparable observations, Z = n/(n + K). The constant K compares expected within-risk variance with variance of hypothetical means. The estimate concerns the modeled response; applying it to premiums requires compatible expense and exposure treatment.
Worked example: With n = 4 and K = 6, Z = 0.40. An individual mean of 140 and collective mean of 100 produce 0.40 × 140 + 0.60 × 100 = 116.
Mistake to avoid: Reversing the weights and giving the collective mean the individual credibility.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
15. Estimating structural credibility parameters
The expected process variance measures average random variation within risks. The variance of hypothetical means measures persistent differences in their underlying expected responses. Their ratio determines K in the basic Bühlmann model. Estimation must separate sampling noise from genuine heterogeneity; the raw variance of observed account averages includes both and can exaggerate individual credibility.
Worked example: Expected process variance is 900 and variance of hypothetical means is 100, so K = 9. With three observations, Z = 3/(3 + 9) = 0.25.
Mistake to avoid: Treating all variation between observed account averages as permanent risk differences.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
16. Exposure-weighted credibility
Bühlmann-Straub credibility accommodates observations with different exposure weights when conditional variance decreases in proportion to weight. The individual estimate uses an exposure-weighted mean, and total weight replaces observation count in the credibility expression. Weight units and structural parameters must match. Large exposure receives more influence only under the model's variance assumptions.
Worked example: With total weight 90 and a compatible K of 60, Z = 0.60. An exposure-weighted individual mean of 140 and collective mean of 100 yield a credibility estimate of 124.
Mistake to avoid: Changing exposure units without rescaling the structural parameter consistently.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
17. Bayesian frequency updating
A Gamma prior for Poisson frequency produces a Gamma posterior after observing claims and exposure. Using the shape-rate parameterization, posterior mean frequency is (prior shape + claims)/(prior rate + exposure). This combines prior information with experience on a common exposure scale. Distinguish the Gamma rate parameter from its reciprocal scale parameter before calculating.
Worked example: A Gamma prior has shape 3 and rate 30. After 8 claims over 50 exposure units, posterior mean frequency is (3 + 8)/(30 + 50) = 0.1375.
Mistake to avoid: Using a scale parameter in a formula defined for the rate parameterization.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
18. Separate credibility for primary and excess losses
Splitting losses into primary and excess components can assign different credibility to claim incidence and large-loss severity. Define the split consistently for actual and expected losses. Apply each component's credibility before recombining. This prevents a single unusually large claim from automatically receiving the same evidential weight as a broadly observed pattern of smaller claims.
Worked example: Primary actual and expected losses are 120 and 100, with Z = 0.60; excess amounts are 300 and 200, with Z = 0.20. Credibility-adjusted losses are 112 + 220 = 332.
Mistake to avoid: Applying the primary credibility factor to the entire loss amount.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
19. Comparable multiyear experience
Combining experience across years requires a common prospective cost and exposure basis. Development addresses incomplete emergence; trend addresses changing cost levels. Their roles differ, and their factors should reflect the relevant periods. Weight annual results using compatible exposures rather than assuming every year contains equal information. Identify coverage or operational changes before interpreting an account's trend.
Worked example: Already developed losses of 90 and 100 require two and one years of 10% annual trend. They become 108.90 and 110.00; with equal exposures, their average is 109.45.
Mistake to avoid: Treating development and inflation trend as interchangeable adjustments.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
20. Experience modifiers and eligible premium
An experience modifier translates a credibility-adjusted actual-to-expected relationship into a premium adjustment. In a simplified plan, M = [Z × actual + (1 − Z) × expected]/expected. Apply it only to the premium components designated by the plan. Fixed charges or excluded coverages may remain outside the modification, so multiplying the entire invoice can misstate premium.
Worked example: Actual losses are 150, expected losses 100 and Z = 0.40, giving M = 1.20. Eligible premium of 200 becomes 240; an excluded fixed charge of 30 makes total premium 270.
Mistake to avoid: Modifying an excluded fixed charge along with the eligible premium.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Retrospective rating plans
21. The retrospective premium equation
A retrospective plan links final premium to defined losses during the covered period. A simplified equation is premium = (basic premium + loss conversion factor × subject losses) × tax multiplier, followed by contractual minimum and maximum provisions. Every term needs a consistent definition. Contractual sequencing controls whether expenses, taxes and limits apply before or after each adjustment.
Worked example: With basic premium 100, subject losses 400, conversion factor 1.10 and tax multiplier 1.05, uncapped retrospective premium is (100 + 440) × 1.05 = 567.
Mistake to avoid: Assuming every plan applies its minimum and maximum at the same calculation stage.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
22. Loss conversion factors
A loss conversion factor converts specified losses into a broader cost amount, commonly by allowing for defined loss-related expenses. Its interpretation depends on which costs are already in the loss measure. Expenses included in subject losses should not be loaded again through an overlapping conversion factor. Separate the converted loss amount from fixed basic-premium charges.
Worked example: Subject indemnity losses of 200 multiplied by 1.10 produce converted losses of 220. The additional 20 is the modeled loss-related expense allowance under this simplified definition.
Mistake to avoid: Applying an expense allowance to losses that already include the same expenses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
23. Minimum and maximum premium regions
A minimum premium creates a flat lower region, and a maximum premium creates a flat upper region, in the premium-versus-loss relationship. Between them, the premium responds according to the plan's slope. Identify the region before calculating the effect of another loss. A contractual maximum premium does not necessarily eliminate all insured costs associated with a severe event.
Worked example: A plan has uncapped premium 100 + losses, minimum 400 and maximum 800. Losses of 200 produce premium 400; losses of 500 produce 600; losses of 900 produce 800.
Mistake to avoid: Applying the uncapped premium slope after the maximum premium has been reached.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
24. Per-occurrence loss limitations
A per-occurrence loss limitation caps each defined occurrence before subject losses enter the retrospective calculation. It reduces the influence of severe events without necessarily changing the underlying insurance limit. Correct occurrence grouping matters: several claims may arise from one occurrence. Compare actual and expected loss measures using the same limitation.
Worked example: Two separate occurrences generate losses of 300 and 70. With a retrospective loss limitation of 100 per occurrence, subject losses are 100 + 70 = 170.
Mistake to avoid: Confusing the retrospective loss limitation with the policy's maximum covered payment.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
25. Aggregate limitations and calculation order
An aggregate limitation applies to a defined total, while an occurrence limitation applies separately to events. If a plan uses both, calculate the occurrence-limited amounts and then apply the aggregate provision in its prescribed order. These operations answer different questions. The aggregation period, included expenses and occurrence definitions determine which losses contribute to the total.
Worked example: Two occurrences each produce 60. An occurrence cap of 100 leaves 120 in total; an additional aggregate cap of 100 reduces subject losses to 100.
Mistake to avoid: Applying the aggregate cap separately to every occurrence.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
26. Expected premium under nonlinear limits
Calculate expected retrospective premium by applying the full plan to each loss scenario and probability-weighting the resulting premiums. Applying the plan only to expected losses generally fails when minimums or maximums make the function nonlinear. Scenario probabilities must describe the same covered period and loss basis as the plan.
Worked example: Uncapped premiums of 100, 500 and 900 have probabilities 0.20, 0.50 and 0.30. With bounds of 300 and 700, expected premium is 60 + 250 + 210 = 520.
Mistake to avoid: Capping the expected uncapped premium of 540 and treating it as expected final premium.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
27. Development and retrospective adjustments
Retrospective adjustments can change as losses mature or are revalued. Compare valuations on the plan's specified loss basis and apply its adjustment provisions. An increase in estimated subject losses does not always translate into the uncapped premium increment because minimums and maximums may intervene. Separate payments already made from the revised ultimate premium estimate.
Worked example: Subject losses increase from 550 to 600. With conversion factor 1.10 and tax multiplier 1.05, the uncapped premium increase is 50 × 1.10 × 1.05 = 57.75.
Mistake to avoid: Calling the full premium increase immediately payable without checking adjustment terms and prior payments.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
28. Fixed and loss-related expense allowances
A retrospective plan can distinguish expenses that arise regardless of losses from expenses modeled as proportional to losses. A fixed basic-premium component and a loss conversion factor serve different functions. Check whether the basic premium also contains other plan-specific charges before interpreting it as pure expense. Expense allocation should reconcile with the total modeled premium.
Worked example: With fixed expense allowance 70, losses 200 and conversion factor 1.20, premium before other charges is 70 + 240 = 310. The modeled expense allowance is 70 + 40 = 110.
Mistake to avoid: Treating the entire converted loss amount of 240 as an expense charge.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
29. Insurance charges for limited losses
When the insured's loss-sensitive contribution is limited, the insurer may retain costs beyond that limitation. An expected excess-loss charge measures one part of that retained cost. It differs from a risk load for uncertainty and from ordinary expenses. Identify the loss distribution and limitation before estimating the charge; a mean alone cannot determine expected excess losses.
Worked example: Loss is 50 with probability 0.80 and 500 with probability 0.20. Above a loss limitation of 100, expected excess loss is 0.20 × 400 = 80.
Mistake to avoid: Adding a risk load and calling it the expected cost of excess losses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
30. Comparing guaranteed and loss-sensitive premiums
Compare plans across a shared set of loss scenarios, considering expected premium and the distribution of possible premiums. A lower expected retrospective premium may come with greater payment variability. Include any charges outside the premium formula and keep coverage equivalent. Different loss distributions can change the comparison even when expected losses are unchanged.
Worked example: A bounded retrospective plan produces premiums of 400 and 800 with equal probability, giving expected premium 600. A guaranteed premium of 650 costs more on average but removes this modeled premium variability.
Mistake to avoid: Choosing a plan from its expected premium alone without examining adverse scenarios.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Increased limits and deductibles
31. Limited expected value
Limited expected value at limit u is E[min(X, u)] for a nonnegative ground-up loss X. It measures expected indemnity under a simple payment cap. Equivalently, it is the integral of the survival function from zero to u. Coverage conditions and expense treatment must be modeled separately when they differ from this simple capped-loss definition.
Worked example: Loss is 40 with probability 0.75 and 200 with probability 0.25. At limit 100, limited expected value is 0.75 × 40 + 0.25 × 100 = 55.
Mistake to avoid: Using min(E[X], u), which applies the cap after averaging.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
32. Increased limit factors
A loss-only increased limit factor compares limited expected values at a requested limit and a base limit. It converts the base expected indemnity cost to another limit under the same severity distribution. Total premium factors may differ when expenses and risk loads vary by limit. Maintain consistent deductible, coverage and loss-trend assumptions in the numerator and denominator.
Worked example: For losses of 40 or 200 with probabilities 0.75 and 0.25, limited expected values at 100 and 200 are 55 and 80. The loss-only factor is 80/55 ≈ 1.4545.
Mistake to avoid: Assuming the loss-only factor must apply unchanged to every premium component.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
33. Layer costs as limited-value differences
For attachment a and layer limit l, simple layer payment is min(max(X − a, 0), l). Its expected value equals limited expected value at a + l minus limited expected value at a. This identity supports consistent pricing across adjoining layers. It assumes the attachment and limit apply to the same loss unit without additional contractual restrictions.
Worked example: If limited expected values at 100 and 250 are 60 and 90, expected cost of the 150-wide layer above 100 is 90 − 60 = 30.
Mistake to avoid: Subtracting the attachment from expected ground-up loss instead of applying the layer to each loss.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
34. Ordinary deductible payments
Under a simple ordinary deductible d, insurer payment is max(X − d, 0), assuming no upper limit. Expected payment equals E[X] minus limited expected value at d. The deductible removes all small losses and the first d of larger losses. Adding an upper limit requires specifying whether it caps payment or ground-up loss.
Worked example: Losses of 40 and 200 have probabilities 0.75 and 0.25. With deductible 50, expected insurer payment is 0.75 × 0 + 0.25 × 150 = 37.50.
Mistake to avoid: Subtracting the deductible from every loss and allowing negative insurer payments.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
35. Franchise deductibles
A franchise deductible eliminates losses below a specified threshold but pays the full eligible loss once the contractual threshold condition is met. It therefore differs from an ordinary deductible, which subtracts the deductible from larger losses. The treatment of a loss exactly at the threshold must follow the stated contract. Franchise pricing requires the probability and size of qualifying losses.
Worked example: Under a franchise paying full losses exceeding 50, losses of 40 and 200 with probabilities 0.75 and 0.25 give expected payment 50, compared with 37.50 under an ordinary deductible.
Mistake to avoid: Subtracting the franchise threshold from a qualifying loss.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
36. Per-claim and aggregate deductibles
A per-claim deductible applies to each claim; an aggregate deductible applies to a defined combined amount over a period. Identical deductible amounts can therefore produce different payments. Determine whether the aggregate operates on ground-up losses, already reduced losses or another contractual measure. Frequency and dependence matter when modeling aggregate deductible exhaustion.
Worked example: Claims are 40 and 80. A deductible of 50 per claim produces insurer payment 0 + 30 = 30. A single aggregate deductible of 50 on their ground-up total produces 70.
Mistake to avoid: Applying an annual aggregate deductible separately to each claim.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
37. Payment frequency and conditional severity
A deductible changes both payment frequency and average severity among positive payments. Expected loss cost can be calculated from ground-up frequency times expected payment per ground-up claim, or payment frequency times conditional positive-payment severity. The two approaches agree when definitions are consistent. Conditional severity alone does not measure the deductible's total effect.
Worked example: There are four expected ground-up claims, and 25% exceed the deductible. Expected payment count is one. If positive payments average 150, annual expected payment is 150.
Mistake to avoid: Multiplying ground-up claim frequency by severity conditional on a positive payment.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
38. Trend through fixed limits
Trend ground-up severity before applying a fixed monetary limit or deductible. Multiplying an already limited expected loss by the ground-up trend factor generally gives a different result because claims can cross the coverage boundaries. The difference depends on the severity distribution around those boundaries. Distinguish monetary inflation from changes in claim frequency.
Worked example: Equally likely losses of 80 and 160 have expected payment 90 at limit 100. After 25% severity inflation, losses become 100 and 200, giving expected payment 100 rather than 112.50.
Mistake to avoid: Applying ground-up severity inflation directly to capped payments.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
39. Censored loss observations
A payment recorded at a policy limit may reveal only that the underlying loss reached or exceeded that limit. Treating it as a fully observed ground-up severity understates the tail. Distinguish censoring from truncation, where some observations are absent entirely. Modeling higher limits requires defensible tail information rather than assuming capped data disclose the full loss distribution.
Worked example: A claim pays the full limit of 100. Its ground-up loss could be 100, 180 or much larger; the payment record alone cannot determine a higher-limit payment.
Mistake to avoid: Using limit payments as exact uncensored losses when fitting a ground-up severity model.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
40. Expense treatment within coverage limits
Claim expenses may be subject to different coverage provisions from indemnity. A model must specify whether expenses erode a limit, sit outside it or receive another treatment. Apply those provisions before calculating expected costs or limit factors. The same indemnity loss can generate different insurer payments under different expense definitions.
Worked example: Indemnity is 120, covered defense expense is 30 and the limit is 100. If defense is outside the indemnity limit, payment is 130; if both share one limit, payment is 100.
Mistake to avoid: Assuming defense expenses always have the same limit treatment across contracts.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Catastrophe modeling and risk loads
41. From hazard to insured catastrophe loss
A catastrophe model connects event occurrence and hazard intensity to exposed assets, physical vulnerability and financial coverage. Each stage answers a different question. Building characteristics can change damage without changing hazard; deductibles can change insured loss without changing damage. Exposure location and values must match the portfolio and valuation basis being priced.
Worked example: Modeled physical damage is 300,000. Under a simplified covered property loss with deductible 50,000 and payment limit 200,000, insured payment is 200,000 rather than 300,000.
Mistake to avoid: Treating modeled physical damage as insured loss before applying coverage terms.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
42. Average annual loss
Average annual loss is the expected total loss during a year under the model. In an event-rate representation, sum each event's expected annual occurrence rate multiplied by its loss contribution, with the appropriate financial treatment. Expected totals can be added without assuming independence, although annual loss distributions and aggregate contractual terms require more structure.
Worked example: Event types have annual rates 0.01 and 0.04, with insured losses of 10 million and 2 million. Their average annual loss contribution is 100,000 + 80,000 = 180,000.
Mistake to avoid: Calling average annual loss a prediction of what a typical individual year will cost.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
43. Occurrence exceedance probability
Occurrence exceedance probability measures the chance that the largest defined event loss in a year exceeds a specified amount. It describes the annual maximum rather than the annual sum. Keep event definitions and loss bases consistent when comparing curves. One catastrophe can generate many claims while still constituting a single modeled event.
Worked example: In 1,000 equally weighted simulated years, 30 have at least one event exceeding 20 million. Estimated occurrence exceedance probability at 20 million is 30/1,000 = 3%.
Mistake to avoid: Counting every exceeding event as a separate exceeding year.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
44. Aggregate exceedance probability
Aggregate exceedance probability measures the chance that total modeled annual loss exceeds a threshold. Multiple moderate events can produce an aggregate exceedance even when no individual event does. Apply annual aggregate financial provisions when the target is a net insured annual total. An aggregate curve and an occurrence curve describe different random variables.
Worked example: A simulated year contains event losses of 30 million and 40 million. It exceeds an aggregate threshold of 50 million, although its largest event does not exceed 50 million.
Mistake to avoid: Using the largest event loss as the annual aggregate loss.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
45. Return periods and annual probabilities
A return period is commonly expressed as the reciprocal of annual exceedance probability under the specified model. It does not schedule events at regular intervals. Multi-year probabilities require assumptions about dependence and stability over time. The occurrence or aggregate basis must accompany the return-period description so the loss measure remains clear.
Worked example: An annual exceedance probability of 2% corresponds to a 50-year return period. With independent, unchanged annual probabilities, the chance of at least one exceedance over ten years is 1 − 0.98^10, approximately 18.29%.
Mistake to avoid: Interpreting a 50-year return period as a guarantee of 49 event-free years.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
46. Process and parameter uncertainty
Process uncertainty describes random events and losses conditional on model parameters. Parameter uncertainty describes uncertainty in quantities such as event rates or vulnerability relationships. The law of total variance separates expected conditional variance from variance of conditional means. A simulation with fixed parameters may capture process variability while omitting parameter uncertainty and structural model limitations.
Worked example: Two equally likely parameter settings give mean losses 100 and 300, each with conditional variance 400. Total variance is 400 + 10,000 = 10,400.
Mistake to avoid: Assuming more simulations resolve uncertainty about whether the underlying model is appropriate.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
47. Spatial dependence and portfolio concentration
Catastrophe losses can be dependent because a single event affects many insured locations. Geographic diversification should therefore be assessed through joint losses rather than location counts alone. Covariance enters aggregate variance alongside individual variances. Even when exposures are geographically separated, shared hazard mechanisms or model assumptions can create dependence.
Worked example: Two portfolio losses have standard deviations 100 and 200 and correlation 0.50. Aggregate standard deviation is sqrt(10,000 + 40,000 + 20,000), approximately 264.58.
Mistake to avoid: Assuming losses are independent merely because the insured properties have different addresses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
48. Variance-based risk loads
Under a stated variance pricing principle, a risk load equals a chosen coefficient times loss variance. Expected loss remains a separate component. The coefficient must have units that convert squared monetary units into money. This is an analytical pricing choice rather than a universal requirement, and its result depends on the distribution and aggregation basis used.
Worked example: Loss variance is 1,000,000 in squared monetary units. A loading coefficient of 0.0001 per monetary unit produces a risk load of 100, added separately to expected loss and expenses.
Mistake to avoid: Adding variance directly to expected loss without a coefficient or compatible units.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
49. Marginal risk loads and covariance
Under a portfolio variance principle, adding risk Y changes variance by Var(Y) + 2Cov(existing portfolio, Y). Its incremental risk load can therefore differ from its standalone load. This captures the effect of dependence on the pricing decision. State the portfolio and allocation convention because sequential incremental loads do not automatically form a unique full allocation.
Worked example: A new risk has variance 1,000,000 and covariance 500,000 with the existing portfolio. At coefficient 0.0001, its incremental load is 0.0001 × 2,000,000 = 200.
Mistake to avoid: Using only standalone variance when pricing the risk's modeled portfolio impact.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
50. Cost-of-capital risk allowances
A simplified annual cost-of-capital allowance multiplies required capital by its specified annual carrying-cost rate. Capital is a stock supporting risk; the allowance is a period cost. Define capital, horizon and any investment-income treatment consistently. A multiyear allowance requires projected capital by period and appropriate discounting rather than repeatedly charging the initial capital without analysis.
Worked example: Required capital of 2 million at an 8% annual carrying-cost rate gives 160,000. With expected loss 300,000 and expenses 40,000, modeled annual price is 500,000.
Mistake to avoid: Adding the full 2 million of capital principal to a one-year premium.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Individual risk rating and specialty applications
51. Schedule rating and experience rating
Schedule rating adjusts for specified characteristics of an individual risk, while experience rating responds to its observed loss history relative to expectation. The methods can address different information, but overlapping adjustments may count the same effect twice. Relate each adjustment to a defined characteristic and check whether the base classification already recognizes it.
Worked example: The base classification already reflects a documented protective feature. Adding another schedule credit solely for that same feature would duplicate the modeled benefit unless the plan explicitly distinguishes an additional effect.
Mistake to avoid: Treating every favorable characteristic as new information outside the base rate.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
52. Layer burning cost
Burning cost estimates historical layer loss per exposure or another defined volume measure. Apply the layer to each relevant loss before aggregation, and bring losses and exposures to a comparable prospective basis. Sparse large-loss experience can make the estimate unstable. An observed burning cost is a starting point, not automatically a complete price including expenses and uncertainty.
Worked example: For a layer attaching at 100 with limit 200, ground-up losses of 80, 180 and 400 produce payments of 0, 80 and 200. Across 20 exposure units, burning cost is 280/20 = 14.
Mistake to avoid: Applying the layer to the sum of losses when it operates per occurrence.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
53. Occurrence and claims-made time bases
Occurrence and claims-made coverage can associate claims with different time bases. In a simplified comparison, occurrence coverage focuses on when the covered event occurred, while claims-made coverage focuses on when the claim was first made, subject to its other provisions. Pricing must reflect the relevant reporting pattern and any retroactive-date or reporting conditions stated in the contract.
Worked example: An event occurs in December and the claim is first made the following March. Those dates belong to different annual cohorts under simplified occurrence and claims-made classifications.
Mistake to avoid: Assigning both forms to the event year without checking their coverage definitions.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
54. Reinstatement provisions in layer pricing
A reinstatement can restore exhausted coverage under specified conditions, potentially creating additional premium. Model both the additional recoverable loss and the premium payment using the actual contract's count, timing and amount rules. A reinstated layer is not automatically unlimited coverage. Reinstatement provisions can materially change net cash flows in years with multiple events.
Worked example: A contract restores a fully exhausted 100-unit limit once, charging 20% of original premium 50. The additional premium is 10; a later covered loss of 60 can use the restored limit.
Mistake to avoid: Modeling restored recoveries while omitting the corresponding contractual reinstatement premium.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
55. Payment timing and present-value loss cost
When a pricing analysis uses discounting, calculate present value from expected payment amounts and their timing on a consistent basis. Equal ultimate losses can have different present values if settlement patterns differ. Keep expected claims, discount assumptions and risk allowances distinct. Whether discounting is appropriate for a particular pricing application must follow its applicable framework.
Worked example: An expected payment of 121 due in two years has present value 121/(1.10)^2 = 100 at a stated annual discount rate of 10%.
Mistake to avoid: Discounting an entire ultimate loss amount as though every payment occurs on one date.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
56. Fixed expenses, variable expenses and premium margins
When variable expenses and a target margin are percentages of premium, solve for premium rather than adding those percentages to loss cost. In a simplified model, P = (expected loss + fixed expenses)/(1 − variable expense ratio − margin ratio). The denominator must be positive. A premium margin is a percentage of premium and differs from return on capital.
Worked example: Expected loss is 600, fixed expenses 100, variable expense ratio 20% and margin ratio 10%. Premium is 700/0.70 = 1,000, comprising 600, 100, 200 and 100.
Mistake to avoid: Calculating percentage-of-premium expenses as percentages of expected losses.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
57. Premium on-level adjustments
Historical premium must reflect a comparable rate level before it supports a prospective loss-ratio analysis. Apply relevant rate changes to the exposure and timing portions affected, while keeping classification and coverage bases consistent. A single overall factor is valid only when it correctly represents the historical mix. Loss trend and premium on-level adjustments serve different purposes.
Worked example: For unchanged exposures entirely subject to a 25% rate increase, historical premium 800 becomes 1,000 on level. Losses of 1,000 then imply a loss ratio of 1.00 rather than 1.25.
Mistake to avoid: Applying the full rate change to premium portions that were already written at the new level.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
58. Specialty exposure-unit consistency
An exposure measure should connect to the risk mechanism and remain comparable across the experience used for pricing. Changing units alters the numerical rate even when total expected loss is unchanged. Specialty risks may require measures of activity, value or capacity rather than policy count. Evaluate whether the relationship remains proportional over the relevant exposure range.
Worked example: Expected losses of 10,000 across 100 original units imply 100 per unit. If each original unit is redefined as two units, the same risk has 200 units and rate 50.
Mistake to avoid: Interpreting a lower numerical rate after a unit change as evidence of reduced risk.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
59. Deposit, audit and minimum premiums
A deposit premium is an initial amount, while an audited premium reflects the contract's final exposure calculation. A minimum premium can constrain the final result. These amounts need not coincide. Calculate final premium under the stated provisions, then reconcile the deposit and prior adjustments. Refundability and adjustment timing depend on the contract rather than the arithmetic alone.
Worked example: A contract has deposit 80 and minimum 60. Audited formula premium of 100 leaves 20 additional premium; formula premium of 50 becomes final premium 60, leaving a potential 20 refund under the stated terms.
Mistake to avoid: Treating the deposit as the final premium regardless of audited exposure.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
60. Selection effects in negotiated pricing
A price change can alter which risks accept or retain coverage, changing the portfolio's expected loss cost. Historical averages do not automatically describe the post-change book. Analyze acceptance and retention alongside risk characteristics, while recognizing that observational differences do not establish causation. A pricing scenario should distinguish the direct price effect from the assumed change in mix.
Worked example: Low- and high-cost risks have expected costs 100 and 200. An equal mix costs 150 on average; a hypothetical shift to 20% low-cost and 80% high-cost raises the average to 180.
Mistake to avoid: Evaluating a negotiated discount while assuming the retained risk mix cannot change.
Reference: Exam 8 — Advanced Ratemaking | Casualty Actuarial Society
Sources
Exam identity verification:
Browse all study guides