Bornhuetter-Ferguson for Unevenly Earned Exposures

Why Uneven Earned Exposures Matter

Bornhuetter-Ferguson (BF) estimates use an a priori expected ultimate loss and a selected loss emergence pattern to estimate ultimate losses. When accident years have different amounts of earned exposure, the a priori expected loss must reflect each year’s own volume. A loss cost per vehicle-year, for example, cannot be applied to the same assumed number of vehicle-years for every accident year if the earned counts differ.

The calculation is simple once the inputs are aligned: estimate expected ultimate loss separately for each accident year, then apply that year’s unreflected share. The main judgment is ensuring that the exposure measure and the expected loss assumption describe the same risk, period, and level of coverage. The method is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: For accident year $i$, let $E_i$ be its a priori expected ultimate loss, $C_i$ its cumulative loss amount at the valuation date, and $q_i$ the expected share of ultimate losses not yet reflected in the selected cumulative loss measure. The BF ultimate estimate is $C_i+q_iE_i$. If earned exposure is $e_i$ and expected loss cost per exposure unit is $c_i$, then $E_i=e_ic_i$. Alternatively, if earned premium is $P_i$ and a compatible expected loss ratio is $r_i$, then $E_i=P_ir_i$.

The selected cumulative loss amount may be paid or incurred. For paid losses, the unreflected share concerns ultimate losses not yet reflected in cumulative paid amounts. For incurred losses, it concerns ultimate losses not yet reflected in the selected cumulative incurred amount. Incurred amounts include estimates for reported claims, and those estimates may develop further. Therefore, for incurred BF, $q_i$ can include future development on reported claims as well as losses from claims not yet reported; it is not necessarily just the unreported-claim share.

Match the Base Before Applying BF

An exposure base is the quantity to which the selected loss cost or loss ratio applies. The product $e_ic_i$ is meaningful when $e_i$ counts the same exposure unit that $c_i$ prices. If the loss cost is dollars per earned vehicle-year, $e_i$ must be earned vehicle-years. If the loss cost is dollars per earned construction project, $e_i$ must count earned projects on the corresponding basis.

The premium formulation is also a product of a quantity and a rate: earned premium multiplied by an expected loss ratio gives expected loss. The premium and ratio must be compatible. For instance, a ratio selected for premium at a particular rate level should not be applied uncritically to premium on a different basis. The expected loss should also reflect any relevant distinctions in coverage, limits, or risk mix that are built into the selected loss cost or ratio.

Where accident year $i$ contains several risk groups, calculate the expected losses at the matching level and add them. If group $j$ has earned exposure $e_{ij}$ and expected loss cost $c_{ij}$, then $E_i=\sum_j e_{ij}c_{ij}$. Multiplying total exposure by a simple average loss cost can be inappropriate when the groups have different loss costs and different exposure volumes.

Choose the exposure basis.
Identify whether the a priori estimate uses earned exposure and a loss cost, or earned premium and an expected loss ratio.

Calculate expected ultimate loss by year.
Use each accident year’s own earned volume and its corresponding assumption. Sum within the year if the year contains multiple groups.

Apply the unreflected share.
Multiply $E_i$ by $q_i$, the selected share not yet reflected in the cumulative loss measure.

Add the observed cumulative amount.
Calculate ultimate loss as $C_i+q_iE_i$. Under this BF calculation, the estimated unpaid amount above $C_i$ is $q_iE_i$.

Derivation and Interpretation

The exposure scaling follows from the definition of an expected loss cost. If the expected loss per exposure unit is $c_i$, then $e_i$ comparable units have expected aggregate loss $e_ic_i$: expected loss per unit is added across the units. With different groups, add each group’s expected aggregate loss, $\sum_j e_{ij}c_{ij}$. Likewise, if the expected loss ratio is the expected fraction of premium represented by losses, applying it to $P_i$ gives expected loss $P_ir_i$.

To justify the BF adjustment, start with the selected expected ultimate $E_i$. By definition, the expected amount not yet reflected in the cumulative loss measure is the unreflected fraction $q_i$ of that ultimate, or $q_iE_i$. The estimate of ultimate is the amount already reflected, $C_i$, plus the expected amount still to be reflected, $q_iE_i$. Thus:

\widehat{U}_i^{\mathrm{BF}}=C_i+q_iE_i

This construction assumes that $E_i$ is a suitable expected ultimate for the same accident year and loss basis as $C_i$, and that $q_i$ is selected for that year’s maturity and the same cumulative loss measure. It does not imply that all accident years have equal exposures, expected loss costs, or emergence patterns. If $q_i=0$, the formula returns $C_i$; if $q_i=1$, it returns $C_i+E_i$. These boundary cases are useful checks, although a selected $C_i$ and $E_i$ should still be interpreted on their proper bases.

Worked Example: Three Years with Different Earned Counts

A hypothetical insurer estimates losses for a coverage with a constant expected loss cost of $125 per earned unit. Accident years A, B, and C have 800, 1,200, and 500 earned units, respectively. Their cumulative paid losses at the valuation date are $72,000, $110,000, and $28,000. Selected unreflected shares of ultimate paid losses are 40%, 25%, and 60%. Calculate the BF ultimates and unpaid estimates.

Model and assumptions. Use the earned-unit loss-cost basis; the $125 loss cost is appropriate for all three years. The paid amounts and unreflected shares are on a consistent basis. This is an undiscounted calculation, so no interest is applied (0% discounting); mortality is not applicable. No probability distribution is assumed: the given expected loss costs and shares are deterministic selections.

Method: calculate each year’s expected ultimate separately.

For A, $E_A=800\times\$125=\$100{,}000$. Its expected amount not yet reflected is $0.40\times\$100{,}000=\$40{,}000$. Therefore, its BF ultimate is $\$72{,}000+\$40{,}000=\$112{,}000$, and its unpaid estimate above paid losses is $40,000.

For B, $E_B=1{,}200\times\$125=\$150{,}000$. Its unreflected amount is $0.25\times\$150{,}000=\$37{,}500$. The BF ultimate is $\$110{,}000+\$37{,}500=\$147{,}500$, and its unpaid estimate is $37,500.

For C, $E_C=500\times\$125=\$62{,}500$. Its unreflected amount is $0.60\times\$62{,}500=\$37{,}500$. The BF ultimate is $\$28{,}000+\$37{,}500=\$65{,}500$, and its unpaid estimate is $37,500.

Check and trap avoided. In every year, estimated ultimate minus cumulative paid equals the BF unreflected amount. Using 800 units for all three years would ignore the stated earned volumes and produce incorrect a priori estimates. All amounts are exact to the cent; no rounding was needed.

Worked Example: Uneven Earned Premium and Different Loss Ratios

A hypothetical portfolio has earned premiums of $2,400,000, $3,600,000, and $1,500,000 for accident years A, B, and C. The corresponding expected loss ratios are 62%, 58%, and 70%. Cumulative incurred losses are $1,100,000, $1,750,000, and $420,000, and selected unreflected shares of ultimate incurred losses are 35%, 20%, and 50%. Calculate each BF ultimate and the amount added to cumulative incurred losses.

Model and assumptions. Use the stated premium and loss-ratio pairs to estimate expected ultimate losses. The earned premiums, expected loss ratios, incurred amounts, and unreflected shares are assumed compatible for each year. This calculation is undiscounted (0% interest); mortality is not applicable; no probability distribution is assumed because the expected loss ratios and selected shares are deterministic inputs.

Method: scale by each year’s earned premium. For A, $E_A=\$2{,}400{,}000\times0.62=\$1{,}488{,}000$. The unreflected amount is $0.35\times\$1{,}488{,}000=\$520{,}800$. The BF ultimate is $\$1{,}100{,}000+\$520{,}800=\$1{,}620{,}800$.

For B, $E_B=\$3{,}600{,}000\times0.58=\$2{,}088{,}000$. The unreflected amount is $0.20\times\$2{,}088{,}000=\$417{,}600$. The BF ultimate is $\$1{,}750{,}000+\$417{,}600=\$2{,}167{,}600$.

For C, $E_C=\$1{,}500{,}000\times0.70=\$1{,}050{,}000$. The unreflected amount is $0.50\times\$1{,}050{,}000=\$525{,}000$. The BF ultimate is $\$420{,}000+\$525{,}000=\$945{,}000$.

Check and trap avoided. Subtracting each cumulative incurred amount from its BF ultimate gives $520,800, $417,600, and $525,000, exactly the respective unreflected amounts. The incurred basis matters: an unreflected share for incurred losses can include future development of already reported claims, not just claims that have not yet been reported. Applying one year’s premium or loss ratio to another year’s losses would also break the year-specific alignment. All amounts are exact to the cent.

Worked Example: Changing Exposure Volumes and Loss Costs

A hypothetical insurer covers three groups of similar risks over three accident years. The selected earned exposures, expected loss costs, cumulative paid losses, and unreflected paid shares are: A, 4,000 units, $210 per unit, $600,000, and 30%; B, 6,500 units, $225 per unit, $1,200,000, and 15%; C, 3,000 units, $240 per unit, $250,000, and 45%. Calculate the BF ultimate for each year.

Model and assumptions. Each year’s loss cost applies to that year’s earned units and is selected for the same coverage basis as the cumulative paid amount. The calculation is undiscounted (0% interest); mortality is not applicable; no probability distribution is assumed, since the inputs are deterministic estimates.

Method: use the matching exposure and loss cost in each year. For A, $E_A=4{,}000\times\$210=\$840{,}000$. The unreflected amount is $0.30\times\$840{,}000=\$252{,}000$, so the BF ultimate is $\$600{,}000+\$252{,}000=\$852{,}000$.

For B, $E_B=6{,}500\times\$225=\$1{,}462{,}500$. The unreflected amount is $0.15\times\$1{,}462{,}500=\$219{,}375$, so the BF ultimate is $\$1{,}200{,}000+\$219{,}375=\$1{,}419{,}375$.

For C, $E_C=3{,}000\times\$240=\$720{,}000$. The unreflected amount is $0.45\times\$720{,}000=\$324{,}000$, so the BF ultimate is $\$250{,}000+\$324{,}000=\$574{,}000$.

Check and trap avoided. Each BF ultimate minus its cumulative paid amount equals the calculated unreflected amount. A tempting but incorrect shortcut is to apply the average exposure of 4,500 units to every year. That would give expected losses of $945,000, $1,012,500, and $1,080,000, respectively, instead of the correctly aligned $840,000, $1,462,500, and $720,000. The shortcut materially distorts the BF additions because each year has a different volume and loss cost. All amounts are exact to the cent.

Common Mistakes and Practical Checks

Uneven exposure is not itself a problem for BF. The problem is using an expected ultimate that does not represent the accident year being estimated. Before finalizing a calculation, verify the exposure period, the unit definition, the loss-cost or loss-ratio basis, and the cumulative loss measure. In particular, an earned exposure count should not be replaced by written exposure simply because that figure is easier to obtain; the chosen quantity must correspond to the coverage earned for the accident year and to the selected assumption.

Using one volume for all years. Calculate each year’s expected loss from its own earned exposure or premium.
Averaging unlike groups. If group volumes and loss costs both vary, multiply within groups and then sum; an unweighted average cost need not represent the portfolio.
Mixing bases. Do not multiply an exposure count by a loss cost stated per a different unit, or apply an expected loss ratio to premium on an incompatible basis.
Misreading $q_i$ for incurred losses. It is the share not yet reflected in the selected cumulative incurred amount; that can include future development on reported claims.
Applying $q_i$ to the wrong quantity. The BF addition is $q_iE_i$, where $E_i$ is that year’s expected ultimate, not the year’s exposure count or premium alone.

Key takeaway: With uneven earned volumes, preserve the year-by-year link: calculate $E_i$ from accident year $i$’s own matching exposure base and assumption, then add $q_iE_i$ to that year’s cumulative loss amount.

Check Your Understanding

For each problem, use the stated exposure or premium basis and calculate only the requested BF quantity.

A hypothetical accident year has 2,400 earned units, an expected loss cost of $180 per unit, cumulative paid losses of $290,000, and an unreflected paid share of 25%. Find expected ultimate loss and BF ultimate. (Answer: Expected ultimate loss = $432,000; BF ultimate = $398,000.)
A hypothetical year has earned premium of $1,800,000, an expected loss ratio of 65%, cumulative incurred losses of $500,000, and an unreflected incurred share of 40%. Find the BF ultimate. (Answer: $968,000.)
Two hypothetical groups in one accident year have earned exposures of 1,000 and 2,000 units, with expected loss costs of $300 and $150 per unit. Cumulative paid losses are $400,000 and the year’s unreflected paid share is 20%. Find the combined expected ultimate loss and BF ultimate. (Answer: Expected ultimate loss = $600,000; BF ultimate = $520,000.)
Two hypothetical accident years have expected loss costs of $200 per earned unit. Year A has 700 earned units, cumulative paid losses of $80,000, and $q_A=0.50$. Year B has 1,300 earned units, cumulative paid losses of $190,000, and $q_B=0.20$. Find each BF ultimate and the combined estimate. (Answer: Year A = $150,000; Year B = $242,000; combined = $392,000.)
A hypothetical year has earned premium of $2,000,000 and cumulative incurred losses of $900,000. An analyst uses an expected loss ratio of 60% and an unreflected incurred share of 30%, but mistakenly applies the ratio to $1,500,000 of written premium instead. Calculate the correct BF ultimate and the erroneous one, and state the overstatement. (Answer: Correct = $1,260,000; erroneous = $1,170,000; the erroneous estimate is an understatement of $90,000.)

What You'll Learn