Why the Exposure Measure Matters
The Cape Cod method uses observed losses to estimate a common expected loss rate, then applies that rate to each period’s exposure. The choice of exposure therefore affects how the method distributes expected ultimate losses among periods. Earned premium, policy counts, and measures such as vehicle-years can all be reasonable in suitable circumstances, but they do not describe the same thing.
The key question is not simply which measure is available. It is whether one unit of the selected measure represents a comparable amount of expected ultimate loss across the periods being analyzed. If the expected loss cost per unit changes materially from period to period, a single fitted Cape Cod rate can produce a misleading allocation.
The reported proportions \(r_i\) come from the selected development pattern. They are between 0 and 1; a less mature period will generally have a smaller reported proportion. The denominator \(\sum_i e_i r_i\) is the exposure-weighted amount expected to be represented in reported losses, not the portfolio’s total exposure. If the exposure measure is earned premium, \(\hat{\theta}\) is a loss ratio. If it is policy-years, \(\hat{\theta}\) is expected loss per policy-year.
What Makes an Exposure Measure Suitable?
A suitable measure should track the volume of insured risk in a consistent way across periods. For a common fitted rate to be meaningful, expected ultimate loss should be approximately proportional to the chosen measure, with a reasonably stable expected loss cost per unit. This is a modeling condition, not something guaranteed by the formula.
| Exposure measure | What it can represent | What to check |
|---|---|---|
| Earned premium | Exposure valued through the premium earned during the period | Whether premium levels are comparable as measures of risk volume, including material changes in coverage, mix, or rating level |
| Policy counts | Number of policies or policy-periods | Whether policies have comparable duration, size, coverage, and expected loss cost |
| Policy-years or vehicle-years | Time at risk for a policy or insured item | Whether a unit represents comparable insured time and risk across periods |
| Payroll or another risk-specific base | A quantity that may track exposure more directly for a particular line | Whether the base is measured consistently and expected losses are proportional to it |
Earned premium is often convenient because it is already measured in money and can reflect the amount and duration of insurance earned. But equal premium amounts need not represent equal risk if coverage, limits, insured mix, or rating levels differ. Conversely, a policy count may be a poor measure when policy terms or insured amounts vary, even if the count is recorded accurately.
A practical review should therefore identify the exposure unit, check its consistency by period, and ask whether material differences in risk characteristics change expected loss per unit. The measure should also align with the losses being modeled: the exposure and losses need to refer to compatible periods, coverage, and populations.
Deriving the Fitted Rate
The Cape Cod rate is chosen so the model’s expected reported losses, summed over periods, match the observed aggregate reported losses. Under the model, expected ultimate loss in period \(i\) is \(\theta e_i\). Since a proportion \(r_i\) of ultimate losses is expected to be reported, expected reported loss is \(\theta e_i r_i\).
This matching argument explains why the reported proportion appears in the denominator. Once the rate is fitted, the expected unpaid amount is the modeled ultimate amount less the modeled reported amount:
Thus the model’s aggregate ultimate loss equals aggregate modeled reported loss plus aggregate expected unpaid loss. In a Cape Cod calculation fitted to the observed aggregate reported losses, the aggregate modeled reported loss equals \(\sum_i R_i\), subject to rounding.
Worked Example: Earned Premium as the Exposure Measure
A hypothetical portfolio has two accident periods. Earned premiums are $200,000 and $300,000. The expected reported proportions are 0.40 and 0.80, and reported losses are $40,000 and $80,000. Use earned premium as the exposure measure. Calculate the fitted loss ratio, each period’s expected unpaid loss, and the aggregate reconciliation. State all amounts in thousands of dollars.
Method. Use earned premium in thousands as \(e_i\), so the fitted rate is expected loss per dollar of earned premium, expressed as a loss ratio. The exposures are \(e_1=200\) and \(e_2=300\), and reported losses are \(R_1=40\) and \(R_2=80\).
Calculate the reported-weighted premium. The two denominator contributions are \(200(0.40)=80\) and \(300(0.80)=240\). Their sum is \(80+240=320\) thousand dollars of reported-weighted premium.
Fit the rate. Total reported losses are \(40+80=120\) thousand dollars. Therefore, $$ \hat{\theta}=\frac{120}{320}=0.375. $$ The fitted loss ratio is 0.375, or 37.5%.
Calculate expected unpaid losses. Period 1 has an unpaid proportion of \(1-0.40=0.60\), so its expected unpaid loss is \(0.375(200)(0.60)=45\) thousand dollars. Period 2 has an unpaid proportion of \(1-0.80=0.20\), so its expected unpaid loss is \(0.375(300)(0.20)=22.5\) thousand dollars. Total expected unpaid losses are \(45+22.5=67.5\) thousand dollars.
Check. Modeled ultimate losses are \(0.375(200)=75\) and \(0.375(300)=112.5\), totaling 187.5 thousand dollars. Modeled reported losses are \(75(0.40)=30\) and \(112.5(0.80)=90\), totaling 120 thousand dollars, equal to observed reported losses. The aggregate reconciliation is \(120+67.5=187.5\) thousand dollars.
Common mistake avoided. Do not use total premium \(200+300=500\) in the fitted-rate denominator. The denominator must weight each period’s premium by its reported proportion. The aggregate modeled ultimate is based on total premium: \(0.375(500)=187.5\) thousand dollars, which reconciles to reported plus expected unpaid losses.
Comparing Premium and Policy Counts
Two measures can produce different allocations because each imposes a different assumption about what drives expected loss volume. A premium-based rate treats premium as proportional to expected ultimate losses across periods. A count-based rate treats the number of counted policies or policy-periods as proportional to expected ultimate losses. Neither choice is automatically preferable: the risk characteristics and the meaning of each unit determine which assumption is more defensible.
Worked Example: Comparing Premium-Based and Count-Based Results
A hypothetical insurer has two periods. Period 1 has $100,000 of earned premium, 100 policy-periods, reported proportion 0.50, and reported losses of $30,000. Period 2 has $400,000 of earned premium, 200 policy-periods, reported proportion 0.25, and reported losses of $40,000. Calculate total expected unpaid losses using (a) earned premium and (b) policy-period counts. Explain what each result assumes. State loss amounts in thousands of dollars.
(a) Earned-premium basis. Use premiums of 100 and 400 thousand dollars. The reported-weighted premium is \(100(0.50)+400(0.25)=50+100=150\) thousand dollars. Total reported losses are \(30+40=70\) thousand dollars, so the fitted loss ratio is \(70/150=0.466666\ldots\).
The period 1 expected unpaid loss is \((70/150)(100)(1-0.50)=23.3333\ldots\) thousand dollars. For period 2 it is \((70/150)(400)(1-0.25)=140\) thousand dollars. The total, calculated before rounding, is \(23.3333\ldots+140=163.3333\ldots\) thousand dollars, or $163,333.33 to the nearest cent.
(b) Count basis. The reported-weighted count is \(100(0.50)+200(0.25)=50+50=100\) policy-periods. The fitted loss cost is \(70/100=0.70\) thousand dollars per policy-period. Expected unpaid losses are \(0.70(100)(0.50)=35\) thousand dollars for period 1 and \(0.70(200)(0.75)=105\) thousand dollars for period 2. The total is \(35+105=140\) thousand dollars.
Interpretation and check. The premium-based result assumes expected ultimate losses are proportional to earned premium; the count-based result assumes they are proportional to policy-period counts. Under the premium basis, modeled ultimate losses total \((70/150)(100+400)=233.3333\ldots\) thousand dollars. Modeled reported losses are 70 thousand dollars, so unpaid losses are \(233.3333\ldots-70=163.3333\ldots\) thousand dollars. Under the count basis, modeled ultimate losses total \(0.70(100+200)=210\) thousand dollars; \(210-70=140\) thousand dollars is unpaid.
Common mistake avoided. Do not conclude that the count basis must be better just because policy counts are available. If the two periods’ policies have substantially different expected loss costs, a constant loss cost per policy-period may not be reasonable. Likewise, premium is not automatically comparable if it reflects changes in price or coverage that do not track risk volume.
Using Other Exposure Measures
A measure tailored to the insured risk can be more meaningful than raw counts. For example, if policies have different terms, policy-years may describe time at risk more consistently than the number of policies. If the amount of insured property varies greatly, a measure related to insured value may be more informative than a policy count. These choices do not remove the need for comparability: the selected unit must still represent a similar expected loss volume across periods.
Worked Example: Vehicle-Years for Different Amounts of Time at Risk
A hypothetical portfolio records exposure in vehicle-years. Period 1 has 600 vehicle-years, a reported proportion of 0.50, and reported losses of $90,000. Period 2 has 900 vehicle-years, a reported proportion of 0.25, and reported losses of $60,000. Calculate the fitted expected loss per vehicle-year and the total expected unpaid loss. State loss amounts in thousands of dollars and the rate both per vehicle-year and in thousands of dollars per vehicle-year.
Method and units. Use \(e_1=600\), \(e_2=900\), \(R_1=90\), and \(R_2=60\), with losses in thousands. The fitted rate will therefore be in thousands of dollars per vehicle-year.
Fit the rate. The reported-weighted exposures are \(600(0.50)=300\) and \(900(0.25)=225\) vehicle-years. Their sum is \(525\) vehicle-years. Total reported losses are \(90+60=150\) thousand dollars, so $$ \hat{\theta}=\frac{150}{525}=\frac{2}{7} =0.285714\ldots\text{ thousand dollars per vehicle-year}. $$ This is $285.714\ldots per vehicle-year.
Calculate expected unpaid losses. Period 1’s unpaid loss is \((2/7)(600)(0.50)=600/7=85.714285\ldots\) thousand dollars. Period 2’s unpaid loss is \((2/7)(900)(0.75)=1{,}350/7=192.857142\ldots\) thousand dollars. Total expected unpaid loss is \(1{,}950/7=278.571428\ldots\) thousand dollars, or $278,571.43 to the nearest cent.
Check. Modeled ultimate losses are \((2/7)(600)=171.428571\ldots\) and \((2/7)(900)=257.142857\ldots\) thousand dollars, totaling \(3{,}000/7=428.571428\ldots\) thousand dollars. Expected reported losses are \(171.428571\ldots(0.50)=85.714285\ldots\) and \(257.142857\ldots(0.25)=64.285714\ldots\), totaling 150 thousand dollars. Thus reported losses plus expected unpaid losses equal modeled ultimate losses.
Common mistake avoided. The rate is not $0.2857 per vehicle-year. Because losses were entered in thousands of dollars, the fitted rate is 0.2857 thousand dollars per vehicle-year, equivalent to about $285.71 per vehicle-year.
Common Exposure-Measure Traps
The most important calculation error is to omit the reported proportion when fitting the rate. That would divide reported losses by total exposure, even though the periods have different amounts of reported development. The resulting rate would not match aggregate expected reported losses to observed reported losses.
For example, in the earned-premium example above, using \(120/(200+300)=0.24\) as the loss ratio would confuse a reported-loss ratio on total premium with the fitted Cape Cod ultimate loss ratio. The correct calculation is \(120/[200(0.40)+300(0.80)]=120/320=0.375\). The check makes the problem visible: the incorrect rate predicts only \(0.24(320)=76.8\) thousand dollars of reported losses, rather than the observed 120 thousand dollars.
- Mixing units: If losses are in thousands and exposure is in policy-years, the rate is thousands of dollars per policy-year, not dollars per policy-year.
- Assuming counts are interchangeable: A count of policies may not measure time at risk or insured size. Specify whether the unit is policies, policy-periods, or policy-years.
- Treating premium as a pure risk measure: Earned premium can change because of pricing, coverage, or portfolio mix. Consider whether those changes also represent changes in expected loss volume.
- Checking the wrong total: Expected unpaid losses alone do not equal modeled ultimate losses. Add modeled reported losses to expected unpaid losses for the reconciliation.
Check Your Understanding
For each problem, identify the exposure basis and keep the stated units throughout the calculation.
- A hypothetical portfolio has earned premiums of $250,000 and $350,000, reported proportions of 0.40 and 0.80, and reported losses of $30,000 and $100,000. Find the fitted loss ratio and total expected unpaid losses. (Answer: fitted loss ratio \(130/380=0.342105\ldots\); expected unpaid losses \(51.3158+23.9474=75.2632\) thousand dollars, or $75,263.16.)
- Two periods have 80 and 160 policy-periods, reported proportions of 0.50 and 0.25, and reported losses of $24,000 and $32,000. Using counts, find the fitted loss cost and total expected unpaid loss. (Answer: \(0.70\) thousand dollars per policy-period; total expected unpaid loss \(28+84=112\) thousand dollars.)
- A hypothetical portfolio has 500 and 1,000 policy-years, reported proportions of 0.60 and 0.30, and reported losses of $90,000 and $45,000. Find the rate per policy-year and total expected unpaid loss. (Answer: \(135/(300+300)=0.225\) thousand dollars per policy-year, or $225 per policy-year; total expected unpaid loss \(45+157.5=202.5\) thousand dollars.)
- Two periods have earned premiums of $100,000 and $300,000, 100 and 200 policies, reported proportions of 0.50 and 0.25, and reported losses of $20,000 and $30,000. Compute total expected unpaid losses using each exposure measure. If the policies are comparable in expected loss cost per policy despite differing premiums, which basis matches that assumption? (Answer: premium basis \(110\) thousand dollars; count basis \(100\) thousand dollars; the count basis matches the stated assumption.)
- A candidate fits a premium-based Cape Cod rate using reported losses of $60,000, premiums of $150,000 and $250,000, and reported proportions of 0.40 and 0.80. They use \(60/(150+250)=0.15\) as the rate. Find the correct rate and the modeled aggregate reported losses implied by the incorrect rate. (Answer: correct rate \(60/[150(0.40)+250(0.80)]=60/260=0.230769\ldots\), or 23.08%; the incorrect rate implies \(0.15(260)=$39,000\) of modeled reported losses, not the observed $60,000.)