CAS Exam 8 common mistakes
CAS Exam 8 · Understand the traps and how to avoid them.
On excess-of-loss and increased-limits calculations, the candidate applies the basic limit loss elimination ratio to the wrong layer, or forgets to convert the limited expected value back to an unlimited basis before building the excess ratio.
Candidates memorize the LER formula for a single limit but do not internalize that excess ratios require (LEV at higher limit minus LEV at lower limit) divided by (expected unlimited loss minus LEV at the attachment), and they lose track of whether the denominator is limited or unlimited.
What to do instead
Before computing, write the layer as [attachment, attachment plus limit] and restate every LEV in the same units. Build a one-line template: excess ratio equals (LEV(att plus lim) minus LEV(att)) divided by (E[X] minus LEV(att)). Practice with a severity distribution where the layer sits entirely above the data cutoff so you must extrapolate the tail.
The candidate computes a trended loss ratio or indicated rate change but mixes incurred and on-level bases, e.g., applying the on-level factor to developed losses that were already on-level, or trending from the wrong average date.
On-leveling, development, and trending are each simple alone, but exam questions chain them and candidates do not track which adjustments have already been applied to a given number. The average accident date is often stated in prose buried in a paragraph.
What to do instead
Draw a flow diagram for every multi-step problem: reported losses, then development to ultimate, then on-level to current rate level, then trend to the future period. Label each arrow with its factor and write the date range it covers. Recompute the average accident date explicitly rather than assuming calendar-year midpoint.
When selecting excess loss factors or increased limits factors from a table, the candidate interpolates linearly between limits instead of using the appropriate interpolation or curve, and ignores that the factor is per-occurrence rather than aggregate.
Exam tables often give factors at discrete limits, and candidates default to straight-line interpolation without checking whether the underlying severity is heavy-tailed, where linear interpolation understates the factor. They also conflate per-occurrence and aggregate (per-year) limits.
What to do instead
State in writing whether the limit is per occurrence or aggregate before touching the table. If interpolating, check the shape of the underlying distribution and use log or power interpolation when the source indicates it. Sanity-check that the increased limits factor rises with the limit and approaches the inverse of the expected excess ratio.
In experience rating, the candidate plugs the credibility factor into the wrong slot, e.g., uses Z times the actual loss ratio plus (1 minus Z) times the expected loss ratio but forgets to apply the ballast or the expected loss ratio modification, or misreads whether the mod is multiplicative on manual premium.
The plan formulas look symmetric, and candidates memorize the credibility-weighted average without tracking what each term represents. Modifications to expected losses (trend, rate level, off-balance) are easy to skip under time pressure.
What to do instead
Write the plan structure as a sentence first: modified expected losses, actual limited losses, credibility Z, then the mod equals the weighted average divided by expected. List every required adjustment (trend, on-level, development, ballast) as a checklist and tick each off before computing. Verify the mod moves in the expected direction when actual losses exceed expected.
The candidate treats classification ratemaking as pure loss ratio math and skips the off-balance, credibility, and complement-of-credibility structure, producing a indicated differential that ignores the current differential.
Classification differential problems often present a table of class loss ratios and expect the candidate to credibility-weight the observed differential against the current differential, then apply off-balance. Candidates compute the raw loss ratio and stop.
What to do instead
For each class, write the three inputs: observed loss ratio relative to the base class, credibility at the class's exposure, and current differential. Compute the indicated differential as credibility times observed plus (1 minus credibility) times current, then apply the off-balance factor. Check that the base class differential stays at 1.000.
On catastrophe or excess modeling questions, the candidate reports a pure premium from a simulated or parametric layer without accounting for the difference between ground-up and limited losses, or double counts the ALAE loading.
Cat model outputs are often given as ground-up loss costs by return period, and candidates forget that the reinsurance layer attaches above a retention, so the layer loss cost must be computed as the difference of expected losses at two attachment points.
What to do instead
Convert every model output to a common basis (ground-up expected loss, then layer expected loss) before combining. Treat ALAE as either included in the loss or added as a separate loading, never both. Write the layer expected loss as E[min(X, att plus lim)] minus E[min(X, att)] and confirm units.
The candidate misuses trend periods by counting months inclusively, e.g., trending a 12-month period from 7/1/2020 to 7/1/2023 as 2 years instead of 3, or applying a annual trend factor to a partial year without exponentiating.
Trend is often quoted as an annual factor, and candidates multiply by the number of years rather than raising the factor to the power of the fractional number of years. Inclusive month counting is a common arithmetic slip.
What to do instead
Compute the trend period as (future date minus historical date) in years, using exact month differences divided by 12. Apply the factor as (1 plus trend) raised to that period, not multiplied. For partial years, keep the exponent fractional and round only at the end.
In deductible and retention analyses, the candidate computes the deductible credit as the expected loss below the deductible but forgets to reflect the expense and profit loadings, or applies the credit to the wrong premium base.
Deductible credits are expressed as a percentage of manual premium and depend on the loss elimination ratio, which must be converted through the expense ratio. Candidates apply the LER directly to premium.
What to do instead
Write the deductible credit as (LER times expected loss ratio) divided by (1 minus variable expense ratio), or follow the exam's stated formula exactly. Confirm the base is manual premium before the credit and that the credit is subtracted, not multiplied.
The candidate answers a ratemaking question with a general description of the method instead of performing the required calculation or stating the specific numeric result, losing all points on a computational item.
Under time pressure candidates write what they remember about the concept rather than committing to a number, especially when the arithmetic is messy or they are unsure of a table value.
What to do instead
Budget time so every computational item gets a written formula, substituted values, and a final boxed number even if approximate. If a table value is uncertain, state the assumption and proceed; partial credit favors a documented attempt over a vague essay.
The candidate confuses the roles of the loss provision, the fixed expense provision, and the variable expense provision when converting a loss ratio to a rate change, producing an indicated change that is directionally wrong.
The rate change formula is a ratio of projected loss and expense to projected premium, and candidates invert terms or omit the variable expense provision because they treat expenses as a flat add-on.
What to do instead
Lay out the formula as (projected loss ratio plus fixed expense ratio) divided by (1 minus variable expense ratio), then subtract 1 for the indicated change. Test direction: if the projected loss ratio rises, the indicated change must rise. Recompute with a simple round-number example to confirm the formula before using exam data.