Cape Cod And Bornhuetter-Ferguson With Heterogeneous Exposure

Why Exposure and Loss Level Matter

Reserving methods can give misleading results when accident periods are treated as if they had equal exposure or the same expected loss level. One period might represent twice as much premium as another; another might have a different expected loss ratio because of a change in coverage mix or an explicit rate-level adjustment. At the same time, the periods may be at different stages of loss emergence. Cape Cod and Bornhuetter-Ferguson (BF) methods account for emergence differently from a pure projection of observed losses, but both require the exposure and expected loss assumptions to be matched to the period being estimated.

The calculations below use cumulative losses as of the valuation date. Unless stated otherwise, exposure is earned premium measured in thousands of dollars, losses and reserves are also in thousands of dollars, and there is no discounting. The proportion of ultimate losses reported is assumed known for each period. In practice, these proportions are selected from development assumptions appropriate to the data.

Definition: For accident period \(i\), let \(E_i\) be its exposure, \(C_i\) its cumulative reported losses, and \(p_i\) the proportion of ultimate losses expected to be reported by the valuation date. Thus \(0<p_i\leq 1\), and the expected unreported proportion is \(1-p_i\). When the cumulative loss development factor is \(L_i\), the corresponding reported proportion is \(p_i=1/L_i\). Let \(q_i\) denote the selected expected loss ratio for period \(i\). If exposure is earned premium, \(E_iq_i\) is the expected ultimate loss for that period.

Bornhuetter-Ferguson by Accident Period

The BF method combines the observed cumulative losses with an a priori estimate of only the unreported portion. Under the assumptions that \(p_i\) is an appropriate emergence proportion and \(E_iq_i\) is the selected expected ultimate loss, the BF ultimate and reserve are

\widehat U_i^{\mathrm{BF}} = C_i + E_iq_i(1-p_i), \qquad \widehat R_i^{\mathrm{BF}} = E_iq_i(1-p_i).

The expected loss ratio need not be the same for every period. This is important when periods differ in expected loss level, provided each \(q_i\) is selected on a basis consistent with that period's exposure. The BF addition is based on the expected ultimate loss for the period multiplied by the expected unreported proportion; it is not the whole expected ultimate loss.

Worked Example: BF With Changing Exposure and Selected Loss Ratios

A hypothetical insurer has two accident periods. Exposure is earned premium in thousands of dollars. For period 1, \(E_1=80\), \(C_1=36\), \(p_1=0.75\), and the selected expected loss ratio is \(q_1=0.60\). For period 2, \(E_2=120\), \(C_2=54\), \(p_2=0.50\), and \(q_2=0.90\). Calculate each BF ultimate and reserve, then the totals.

Method. Calculate expected ultimate losses \(E_iq_i\), multiply each by its unreported proportion, and add that amount to reported losses. This method avoids applying one loss ratio to periods with different selected expected levels.

For period 1, expected ultimate losses are \(80(0.60)=48\). The unreported proportion is \(1-0.75=0.25\), so the BF reserve is \(48(0.25)=12\). Therefore,

\widehat U_1^{\mathrm{BF}}=36+12=48, \qquad \widehat R_1^{\mathrm{BF}}=12.

For period 2, expected ultimate losses are \(120(0.90)=108\). The unreported proportion is \(1-0.50=0.50\), so the BF reserve is \(108(0.50)=54\). Therefore,

\widehat U_2^{\mathrm{BF}}=54+54=108, \qquad \widehat R_2^{\mathrm{BF}}=54.

The total reported losses are \(36+54=90\), total reserve is \(12+54=66\), and total ultimate is \(90+66=156\). The check also follows by adding period ultimates: \(48+108=156\). All amounts are in thousands of dollars. The common mistake is to add \(E_iq_i\) in full to reported losses; that would count the expected reported portion twice.

Cape Cod With Different Exposure Volumes

A basic Cape Cod estimate uses the observed losses across periods to estimate a common expected loss ratio. However, a period with greater exposure contributes more expected reported losses than a smaller period. It is therefore exposure multiplied by the reported proportion—not exposure alone—that forms the denominator. With a common loss ratio \(q\), the expected reported loss for period \(i\) is \(E_iq p_i\). Equating the sum of these expected reported losses to the sum of observed losses gives the Cape Cod estimate:

\widehat q_{\mathrm{CC}} = \frac{\sum_i C_i}{\sum_i E_i p_i}, \qquad \widehat U_i^{\mathrm{CC}}=E_i\widehat q_{\mathrm{CC}}, \qquad \widehat R_i^{\mathrm{CC}}=\widehat U_i^{\mathrm{CC}}-C_i.

This version assumes a common expected loss ratio on a consistent exposure basis. The resulting common estimate is applied to each period's own exposure. It does not assume the periods have equal premium, and it does not replace the observed losses with a fully prior estimate: the observed losses determine the estimated common loss ratio.

Justification: If the common ultimate loss ratio is \(q\), period \(i\) has expected ultimate losses \(E_iq\), of which the proportion \(p_i\) is expected to be reported. Its expected reported losses are therefore \(E_iq p_i\). Summing over periods gives \(\sum_i E_iq p_i=q\sum_i E_i p_i\). Matching this expected total to the observed total \(\sum_i C_i\) and solving for \(q\) yields \(\widehat q_{\mathrm{CC}}=\sum_i C_i/\sum_i E_i p_i\). The estimate is not appropriate without modification if the periods have materially different expected loss ratios or if their exposures are not on a comparable basis.

Worked Example: Cape Cod Across Unequal Exposures and Maturities

A hypothetical book has three accident periods. Exposure, in thousands of dollars of earned premium, is \(E=(100,200,150)\); reported proportions are \(p=(0.80,0.50,0.20)\); and cumulative losses, in thousands of dollars, are \(C=(72,80,30)\). Calculate the Cape Cod loss ratio, period ultimates, and total reserve.

Method. Use exposure-weighted reported exposure in the denominator, then apply the estimated common loss ratio to each period's own exposure. This avoids giving a small period the same weight as a large one, or treating an immature period as fully reported.

First calculate the denominator:

\sum_i E_ip_i =100(0.80)+200(0.50)+150(0.20) =80+100+30=210.

The cumulative reported losses total \(72+80+30=182\). Thus,

\widehat q_{\mathrm{CC}}=\frac{182}{210}=0.8667

to four decimal places. Applying the unrounded ratio, the period ultimates are \(100(182/210)=86.67\), \(200(182/210)=173.33\), and \(150(182/210)=130.00\), in thousands of dollars. The corresponding reserves are \(86.67-72=14.67\), \(173.33-80=93.33\), and \(130.00-30=100.00\).

The total ultimate is \(86.67+173.33+130.00=390.00\), and the total reserve is \(14.67+93.33+100.00=208.00\). The check is \(390.00-182.00=208.00\). Amounts are rounded to the nearest cent in thousands of dollars. The common mistake is dividing \(182\) by total exposure \(450\), which ignores that only a fraction of each period's ultimate losses is reported.

Adjusting for Heterogeneous Expected Loss Levels

When periods have different expected loss levels, using one common loss ratio on the raw exposure may be inappropriate. One approach is to select period-specific \(q_i\) values and use BF separately by period. A related Cape Cod approach can estimate a common scale parameter while retaining specified relative differences among periods. Suppose \(m_i\) is a selected relative loss-level factor and \(\theta\) is an unknown base loss ratio, so the period's expected loss ratio is \(\theta m_i\). Then expected ultimate losses are \(E_i\theta m_i\), and expected reported losses are \(E_i\theta m_i p_i\). Equating their total to observed losses gives

\widehat\theta_{\mathrm{CC}} =\frac{\sum_i C_i}{\sum_i E_i m_i p_i}, \qquad \widehat U_i=E_i m_i\widehat\theta_{\mathrm{CC}}.

The \(m_i\) values must be selected on a consistent, documented basis; they are not automatically inferred just because exposures differ. This generalized estimate preserves the selected relative levels while using the observed losses to estimate their common scale. If those relative factors are not credible or the common-scale assumption is unsuitable, a single pooled Cape Cod ratio can obscure important differences.

Worked Example: Cape Cod With Selected Relative Loss Levels

A hypothetical portfolio has exposures \(E=(100,150,200)\), reported proportions \(p=(0.75,0.50,0.25)\), relative loss-level factors \(m=(0.80,1.00,1.25)\), and reported losses \(C=(48,60,50)\), all exposure and loss amounts in thousands of dollars. Estimate the common scale \(\theta\), each ultimate, and each reserve.

Method. Estimate the scale using the adjusted reported exposure \(E_im_ip_i\), then calculate ultimate losses using \(E_im_i\). This avoids treating the selected loss-level differences as if they did not exist.

The denominator is

\sum_i E_im_ip_i =100(0.80)(0.75)+150(1.00)(0.50)+200(1.25)(0.25) =60+75+62.5=197.5.

Reported losses total \(48+60+50=158\), so

\widehat\theta_{\mathrm{CC}}=\frac{158}{197.5}=0.8000.

The adjusted exposures \(E_im_i\) are \(80\), \(150\), and \(250\). The corresponding ultimate estimates are \(80(0.80)=64\), \(150(0.80)=120\), and \(250(0.80)=200\). Subtracting reported losses gives reserves of \(64-48=16\), \(120-60=60\), and \(200-50=150\).

Total ultimate is \(64+120+200=384\), total reserve is \(16+60+150=226\), and \(384-158=226\), as required. Amounts are in thousands of dollars. The common mistake is to omit \(m_i\) from the denominator but retain it in the ultimate calculation; that would estimate the scale on one exposure basis and apply it on another.

How the Methods Respond to Selection

Cape Cod and BF both use expected loss assumptions, but the source of the estimate differs. A basic Cape Cod calculation estimates a common loss ratio from the observed losses and emergence-weighted exposure. BF uses a selected expected loss ratio to estimate each period's unreported amount. Consequently, the estimates need not agree, even if the same exposure and emergence assumptions are used. A selected BF loss ratio may reflect external information or an explicit period-level selection; a Cape Cod estimate is anchored to the observed aggregate losses under its pooling assumptions.

The following comparison uses the data from the three-period Cape Cod example. Suppose instead that a common selected BF loss ratio of \(0.75\) is used. The BF reserve for each period is \(E_i(0.75)(1-p_i)\): \(100(0.75)(0.20)=15\), \(200(0.75)(0.50)=75\), and \(150(0.75)(0.80)=90\). The total reserve is \(180\), and the total ultimate is \(182+180=362\). The Cape Cod reserve was \(208\), with ultimate \(390\), because its estimated common loss ratio was \(0.8667\), not \(0.75\). This difference is a direct consequence of the selected loss level, not a calculation inconsistency.

If expected loss ratios genuinely vary by period, period-specific BF selections may be more suitable than a common-ratio Cape Cod estimate. Conversely, a generalized Cape Cod estimate can retain specified relative factors \(m_i\) while estimating an overall scale. Neither method removes the need to assess whether exposures, loss definitions, and emergence assumptions are comparable.

Common Exam Traps

Using total exposure in the Cape Cod denominator. The correct denominator is \(\sum_i E_i p_i\) for a common loss ratio, not \(\sum_i E_i\). It reflects the proportion expected to have emerged.
Applying expected ultimate losses twice. BF adds only \(E_iq_i(1-p_i)\) to reported losses. Adding all of \(E_iq_i\) double counts the expected reported portion.
Confusing a common loss ratio with equal ultimate losses. A common ratio produces period ultimate estimates \(E_iq\), which differ when exposures differ.
Mixing exposure bases. If expected loss levels are adjusted by \(m_i\), use the adjusted basis consistently in both the Cape Cod denominator and ultimate estimate.
Interpreting a selected ratio as data-derived. A BF expected loss ratio is an assumption or selection. A basic Cape Cod ratio is estimated from observed losses divided by emergence-weighted exposure.

Key takeaway: Match each period's exposure, expected loss level, and emergence proportion. BF estimates unreported losses from selected expected ultimate losses; Cape Cod estimates a loss ratio using reported losses divided by exposure weighted for emergence. Any relative loss-level adjustment must be used consistently throughout the calculation.

Check Your Understanding

For each question, treat exposures and losses as thousands of dollars unless otherwise stated. Round dollar amounts to the nearest cent in thousands, and probabilities or ratios to four decimal places where needed.

A period has earned premium \(E=100\), reported losses \(C=45\), reported proportion \(p=0.60\), and selected loss ratio \(q=0.70\). Find its BF reserve and ultimate. (Answer: reserve \(28.00\); ultimate \(73.00\), in thousands of dollars.)
Two periods have \((E_1,p_1,C_1)=(50,0.80,30)\) and \((E_2,p_2,C_2)=(100,0.40,32)\). Calculate the basic Cape Cod ratio and total reserve. (Answer: Cape Cod ratio \(0.7750\); total reserve \(54.25\), in thousands of dollars.)
Use generalized Cape Cod with \(E=(100,200)\), \(m=(0.90,1.10)\), \(p=(0.50,0.25)\), and \(C=(27,44)\). Find \(\widehat\theta\), total ultimate, and total reserve. (Answer: \(\widehat\theta=0.7100\); total ultimate \(220.10\); total reserve \(149.10\), in thousands of dollars.)
Two periods have \(E=(60,80)\), \(p=(0.50,0.25)\), \(C=(20,30)\), and selected loss ratios \(q=(0.50,0.75)\). Calculate the BF reserve for each period and the total ultimate. (Answer: reserves \(15.00\) and \(45.00\); total ultimate \(110.00\), in thousands of dollars.)
For two periods, \(E=(40,160)\), \(p=(0.50,0.25)\), and \(C=(16,24)\). A candidate incorrectly averages the two period ratios \(C_i/(E_ip_i)\) equally. Calculate the correct Cape Cod ratio, total ultimate, and total reserve. (Answer: Cape Cod ratio \(0.6667\); total ultimate \(133.33\); total reserve \(93.33\), in thousands of dollars. The correct calculation pools losses and emergence-weighted exposure.)

What You'll Learn