Cape Cod Assumptions and Diagnostic Checks

Why Cape Cod Needs Diagnostics

A Cape Cod estimate combines observed losses with an expected-loss component for losses not yet reported or paid. Its pooled expected loss ratio is estimated from the same portfolio experience used to form the estimate. As a result, a calculation can be arithmetically correct yet misleading if premiums and losses represent different periods, if development weights are implausible, or if one accident year dominates the pooled result.

The purpose of a diagnostic is not to force a preferred answer. It is to find assumptions or data that need investigation, quantify the effect of a reasonable alternative, and distinguish a supported selection from an unexplained adjustment. Cape Cod results depend on the selected loss basis, premium basis, valuation date, and development weights; those choices must be understood before interpreting the estimate.

Definition: For accident year $i$, let $P_i$ be premium on a basis consistent with the losses, $C_i$ the cumulative reported or paid losses at the valuation date, and $w_i$ the proportion of expected ultimate losses represented by those cumulative losses. For a common loss basis, the Cape Cod expected loss ratio is $$ \widehat{ELR}_{CC}=\frac{\sum_i C_i}{\sum_i P_iw_i}. $$ The estimated ultimate loss for year $i$ is $$ \widehat{U}_i=C_i+(1-w_i)P_i\widehat{ELR}_{CC}. $$ These expressions assume that $P_i$ and $C_i$ refer to compatible accident-year experience, the selected weights are appropriate for the loss basis and valuation age, and a common expected loss ratio is reasonable for the included years. They do not correct mismatched periods or an unsuitable loss basis.

What the Formula Implies

The denominator $\sum_i P_iw_i$ is not simply total premium. It is premium adjusted for the portion of expected ultimate loss assumed to have emerged by the valuation date. The numerator is the observed cumulative loss total. Thus the pooled ratio compares observed losses with the exposure-weighted portion expected to have emerged.

The ultimate estimate adds the observed amount $C_i$ to the expected unobserved amount $(1-w_i)P_i\widehat{ELR}_{CC}$. This structure also explains why period alignment matters: a premium from one exposure population cannot sensibly scale the expected losses for a different population merely because the accident-year labels look similar.

Justification: Under the model, the expected ultimate loss for year $i$ is $P_i ELR$. If the fraction $w_i$ is expected to have emerged by valuation, the expected observed loss is $P_i ELR\,w_i$. Across the selected years, equating the total observed losses to their modeled expectation gives $$ \sum_i C_i=ELR\sum_i P_iw_i. $$ Solving for the common $ELR$ gives the Cape Cod pooled ratio. The expected unobserved amount for year $i$ is the remaining fraction $1-w_i$ of its expected ultimate loss, which yields the ultimate formula above. This is a modeling justification, not a claim that each year's observed loss must equal its expectation.

Diagnostic 1: Match Premium and Loss Periods

Start by writing down what each input measures. A loss triangle organized by accident year should ordinarily be paired with premium that measures exposure to those same accident years. Premium may need to be earned, on-leveled, or otherwise adjusted to a common basis, depending on the analysis. The loss amounts must also use a consistent definition and valuation date—for example, paid losses should not be mixed with incurred losses without an explicit, appropriate method.

A year label alone does not establish consistency. Calendar-year premium can include policies and exposure periods that differ from the accident-year losses in a triangle. Similarly, changing the premium basis for one year can change both the pooled ratio denominator and that year's expected unobserved losses. A numerical result can still be calculated, but it no longer has the intended interpretation unless the mismatch is addressed.

Worked Example: Checking the Premium Basis

A hypothetical insurer estimates paid Cape Cod losses for accident years 2023–2025. At one valuation date, reported paid losses are $24,000, $50,000, and $72,000, with corresponding weights 0.40, 0.65, and 0.80. The selected earned premiums, adjusted to a common rate level and aligned to those accident years, are $80,000, $100,000, and $120,000. Calculate the estimate, then assess the effect of mistakenly substituting calendar-year premiums of $100,000, $120,000, and $80,000 in the same row order.

Method and aligned calculation. The weights represent the expected paid proportions for the stated valuation ages. First calculate the weighted premium denominator: $$ 80{,}000(0.40)+100{,}000(0.65)+120{,}000(0.80) =32{,}000+65{,}000+96{,}000=193{,}000. $$ The observed paid losses sum to $24{,}000+50{,}000+72{,}000=146{,}000$. Therefore, $$ \widehat{ELR}_{CC}=\frac{146{,}000}{193{,}000}=0.7564767. $$ The unobserved premium amounts are $80{,}000(0.60)=48{,}000$, $100{,}000(0.35)=35{,}000$, and $120{,}000(0.20)=24{,}000$, totaling $107,000. Estimated unpaid losses are $107{,}000(0.7564767)=\$80{,}943.01$. Thus estimated ultimate losses are $\$146{,}000+\$80{,}943.01=\$226{,}943.01$.

Mismatch check. Substituting the stated calendar-year premiums gives a weighted denominator of $100{,}000(0.40)+120{,}000(0.65)+80{,}000(0.80)=\$182{,}000$. The resulting ratio is $146{,}000/182{,}000=0.8021978$. The corresponding unobserved premium is $100{,}000(0.60)+120{,}000(0.35)+80{,}000(0.20)=\$118{,}000$, so implied unpaid losses are $\$94{,}659.34$, and implied ultimate losses are $\$240{,}659.34$.

The second calculation is arithmetically valid but not a defensible alternative estimate unless those calendar-year premiums are shown to represent the same exposure as the accident-year losses. The common mistake is to treat a changed denominator as a harmless data substitution; the correction is to establish and use a consistent premium period and basis.

Diagnostic 2: Inspect Development Weights

A development weight is the selected proportion of expected ultimate losses represented by cumulative losses at a particular maturity. For a conventional cumulative emergence pattern on a consistent basis, weights should be between 0 and 1 and should generally be nondecreasing as claims mature. A value of 1 means the selected pattern assumes that all expected losses have emerged; a value below 1 leaves an expected unobserved portion.

These checks are flags, not substitutes for reviewing the data and development method. A seeming reversal may result from a mislabeled age, a different basis, an unusual but explainable pattern, or a data-entry error. Investigate before changing a weight. A correction should be supported by the selected pattern and data, not chosen merely because it increases or decreases the estimate.

Worked Example: A Reversed Maturity Pattern

A hypothetical paid-loss analysis assigns weights 0.45, 0.70, and 0.40 to the youngest, middle, and oldest accident years, respectively. Each year has $100,000 of aligned premium; cumulative paid losses are $30,000, $55,000, and $70,000 in the same order. Diagnose the weight sequence and quantify the effect of one documented alternative in which the oldest-year weight is 0.85.

Method and initial calculation. The sequence is not nondecreasing: the oldest-year weight, 0.40, is below both the middle-year weight 0.70 and the youngest-year weight 0.45. With the entered weights, weighted premium is $45{,}000+70{,}000+40{,}000=\$155{,}000$, equal to observed losses of $30{,}000+55{,}000+70{,}000=\$155{,}000$. Thus $\widehat{ELR}_{CC}=1.0000$. Expected unpaid losses are $100{,}000(0.55)+100{,}000(0.30)+100{,}000(0.60)=\$145{,}000$, giving estimated ultimate losses of $\$300{,}000$.

Investigate and compare. The oldest weight is suspect under the conventional maturity pattern. Check its age label, loss basis, and data entry. If independent support justifies a corrected oldest-year weight of 0.85, while the other weights remain unchanged, weighted premium becomes $45{,}000+70{,}000+85{,}000=\$200{,}000$. The ratio is $155{,}000/200{,}000=0.7750$. Expected unpaid losses become $55{,}000(0.7750)+30{,}000(0.7750)+15{,}000(0.7750)=\$77{,}500$, so estimated ultimate losses are $\$232{,}500$.

The alternative is a sensitivity illustration, not proof that 0.85 is the correct weight. The common mistake is to silently replace 0.40 with a plausible-looking number. The correction is to flag the reversal, investigate its cause, and use an alternative only when supported.

Diagnostic 3: Find Influential Accident Years

Because the pooled expected loss ratio is a ratio of sums, an accident year can affect the estimate through both its observed losses and its weighted premium. A year with a large loss relative to weighted premium can pull the ratio upward; one with a comparatively small loss can pull it downward. A fully developed year may still influence the pooled ratio even though its own expected unpaid amount is zero when $w_i=1$.

A practical diagnostic is a leave-one-year-out calculation: recompute the pooled ratio after excluding one year, then compare it with the full-data ratio. To understand the impact on the full portfolio estimate, apply the alternative ratio to the expected unobserved amounts for all years, including the year omitted only from the ratio calculation. This is a sensitivity test, not a recommendation to discard that year.

Worked Example: Testing an Influential Mature Year

For three hypothetical accident years, each with $100,000 of aligned premium, selected paid weights are 0.50, 0.75, and 1.00. Reported paid losses are $40,000, $60,000, and $140,000. Calculate the full Cape Cod estimate and assess the effect of leaving out the oldest year when estimating the pooled ratio.

Full-data method. Weighted premium is $50{,}000+75{,}000+100{,}000=\$225{,}000$. Observed losses total $\$240{,}000$, so $\widehat{ELR}_{CC}=240{,}000/225{,}000=1.0666667$. The expected unobserved premium amounts are $50,000, $25,000, and $0. Estimated unpaid losses are therefore $75{,}000(1.0666667)=\$80{,}000$, and estimated ultimate losses are $\$240{,}000+\$80{,}000=\$320{,}000$.

Leave out the oldest year from the pooled ratio. The remaining observed losses are $\$40{,}000+\$60{,}000=\$100{,}000$, and the remaining weighted premium is $\$50{,}000+\$75{,}000=\$125{,}000$. The alternative ratio is $100{,}000/125{,}000=0.8000$. Applying it to the full portfolio's $75,000 of unobserved premium gives unpaid losses of $\$60{,}000$, and an ultimate estimate of $\$240{,}000+\$60{,}000=\$300{,}000$.

The oldest year shifts the estimate by $20,000 in this sensitivity, despite having no expected unpaid amount at weight 1. Its observed loss of $140,000 is part of the full pooled ratio. Investigate whether that experience is representative or explained by a specific feature; do not remove it solely because the estimate changes.

Diagnostic 4: Test the Pooled Experience Selection

The selected years used to estimate the pooled ratio should have a defensible relationship to the years being estimated. Useful checks include recalculating with a reasonable alternative experience period, excluding a demonstrably noncomparable year, or separately examining subsets with materially different exposure or claim conditions. Any alternative must still use consistent premium and loss periods and appropriate weights.

Report both the selected estimate and the sensitivity, including which years were included and why. A material change signals dependence on the pooled experience selection; it does not identify which result is correct. If the selected population is heterogeneous, one common expected loss ratio may not adequately represent every year. Conversely, a small change across reasonable alternatives supports robustness but does not validate every underlying assumption.

Establish the basis.
Confirm the valuation date, accident-year definitions, loss measure, premium measure, and any common rate-level basis.

Validate weights.
Check age labels, bounds, maturity ordering, and consistency with the paid or incurred basis used.

Measure influence.
Recalculate the pooled ratio after omitting one year at a time and identify material changes.

Document sensitivity.
Compare defensible experience selections, explain the rationale, and retain the selected estimate with its limitations.

Common Exam Traps and Practical Interpretation

Mixing periods: accident-year losses paired with an unrelated calendar-year premium series may produce a number but not a coherent Cape Cod ratio.
Reversing the maturity check: with weights listed from youngest to oldest, a conventional pattern generally increases, not decreases. For example, 0.45, 0.70, 0.40 has an oldest weight of 0.40, which is below 0.70 and requires investigation.
Ignoring mature years: a year with $w_i=1$ has no expected unobserved amount in the ultimate formula, but its losses and weighted premium still affect the pooled ratio.
Treating sensitivity as selection: a leave-one-out change diagnoses influence; it does not justify excluding the year. Exclusion requires a substantive reason such as noncomparability or a data problem.
Changing weights without recalculating both effects: a weight changes the pooled denominator and the expected unobserved amount for that year. Both consequences must be reflected.

These checks make a Cape Cod estimate more interpretable by showing which data and assumptions drive it. A well-documented result states the selected basis, flags unresolved concerns, and shows whether reasonable alternatives materially change the indication. This diagnostic discipline is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Check Your Understanding

For each problem, identify the diagnostic issue and calculate the requested quantity. Unless stated otherwise, amounts are in dollars.

A hypothetical accident-year analysis has losses of $18,000 and $42,000, weights 0.30 and 0.70, and matching earned premiums of $50,000 and $80,000. Calculate the pooled expected loss ratio. (Answer: $60{,}000/(15{,}000+56{,}000)=0.8451$, or 84.51%.)
Weights are listed from youngest to oldest as 0.55, 0.78, and 0.72. Identify the issue and give the conventional direction of investigation. (Answer: The oldest weight 0.72 is below the middle weight 0.78, violating the usual nondecreasing maturity pattern; verify age labels, basis, and data before selecting a supported older weight at least 0.78 and no greater than 1.)
Three years each have premium $100,000, weights 0.40, 0.70, 1.00, and losses $30,000, $60,000, $120,000. Calculate the full pooled ratio and unpaid estimate. (Answer: Weighted premium $40{,}000+70{,}000+100{,}000=\$210{,}000$; loss total $210,000; ratio 1.0000; unpaid $60{,}000+30{,}000+0=\$90{,}000$.)
For the data in Question 3, omit the oldest year from the pooled ratio, then apply the alternative ratio to the full portfolio's unobserved premium. What is the resulting ultimate estimate? (Answer: Alternative ratio $90{,}000/110{,}000=0.8182$; full unobserved premium is $90,000, giving unpaid $73,636.36 and ultimate $210{,}000+73{,}636.36=\$283{,}636.36$.)
A team pairs accident-year losses with calendar-year premium, then obtains a ratio different from its aligned-premium result. Is the difference itself evidence that one result is correct? State the required diagnostic response. (Answer: No. Reconcile the exposure and period definitions, use premium consistent with the accident-year losses and the intended level basis, and then recalculate; the numerical difference alone does not validate either result.)

What You'll Learn