Handling Immature and Sparse Experience with Bornhuetter-Ferguson

Why Immature and Sparse Experience Needs Care

For a very recent accident year, only a small share of the ultimate claims may have emerged. For a small portfolio, the number of observed claims may also be too low for its aggregate losses to provide a stable indication. In either case, the observed amount alone can be a poor guide to ultimate losses. Bornhuetter-Ferguson (BF) addresses this by combining observed losses with an estimate of losses expected to remain unreported or unpaid.

The important interpretive question is not only, “What is the BF estimate?” It is also, “How much of that estimate comes from observed experience, and how much comes from the selected expected losses and emergence pattern?” With very little emergence, the second part can dominate. The result may be less responsive to random early losses than a direct extrapolation of those losses, but it is correspondingly more dependent on the a priori assumptions.

Definition: Let $E$ be the a priori expected ultimate loss for a group, $p$ the selected proportion of ultimate losses expected to have emerged by the valuation date, and $q=1-p$ the unreported share. Let $L$ be the observed cumulative loss on the same basis used to select $p$. The BF estimate of ultimate loss is $U=L+Eq$. The corresponding BF estimate of losses not yet included in $L$ is $R=Eq$.

An a priori expected ultimate loss may be developed from an exposure measure multiplied by a selected expected loss cost, or from premium multiplied by a selected expected loss ratio. The expected loss basis must match the accident period, coverage, and loss definition being estimated. The emergence percentage must also match the observed measure: a paid emergence pattern goes with paid losses, while an incurred emergence pattern goes with incurred losses.

Here, “reserve” means the amount remaining beyond the observed cumulative amount $L$. For paid losses, that is unpaid loss. When $L$ is incurred loss, $U-L$ is the additional amount needed to reach ultimate beyond current incurred loss; it is not, by itself, total unpaid loss, which would require comparing ultimate with paid losses.

What Limited Emergence Does to the Estimate

The BF estimate can be written in a second, useful form:

$$ U=L+E(1-p)=E+(L-Ep) $$

The term $Ep$ is the expected amount emerged by the valuation date under the a priori estimate and selected pattern. Thus BF can be read as the a priori ultimate $E$, adjusted by the difference between actual observed losses $L$ and expected emerged losses $Ep$. This makes the role of early experience explicit: it adjusts the expected ultimate by an observed emergence deviation; it is not simply projected to ultimate.

When $p$ is small, $Ep$ is small and the observed amount is usually only a modest part of the ultimate estimate. The reserve component $Eq$ remains close to the full expected ultimate. This is why sparse or very immature experience does not, by itself, eliminate the need for a credible a priori estimate. “Sparse” describes limited information in the observed data; it does not imply that the unreported losses are zero.

Worked Example: A Very Immature Paid-Loss Estimate

A hypothetical accident year has earned premium of $3,200,000 and a selected a priori loss ratio of 60.0%. At the valuation date, paid losses are $310,000. The selected paid emergence is 20.0%. Calculate the BF ultimate and unpaid reserve. Also compare the result with a direct paid-loss development extrapolation.

Method and assumptions. Use paid losses throughout. Assume the selected loss ratio is appropriate for this accident year and that 20.0% is the appropriate expected paid emergence proportion.

Step 1: Calculate a priori expected ultimate losses.

$$ E=3{,}200{,}000(0.600)=1{,}920{,}000 $$

Step 2: Calculate the unreported share and expected unpaid losses. The unreported share is $q=1-0.200=0.800$, so:

$$ R=Eq=1{,}920{,}000(0.800)=1{,}536{,}000 $$

Step 3: Add observed paid losses to obtain ultimate losses.

$$ U=L+R=310{,}000+1{,}536{,}000=1{,}846{,}000 $$

Check. The estimate reconciles because $U-L=1{,}846{,}000-310{,}000=1{,}536{,}000=R$. A direct development extrapolation would be $310{,}000/0.200=1{,}550{,}000$, substantially below the BF estimate. The BF result is higher because the observed paid amount is below the expected emerged amount, $Ep=1{,}920{,}000(0.200)=384{,}000$, but it retains the expected unpaid component. The common mistake this comparison avoids is treating an observed amount from a very immature period as though it should determine the entire ultimate by simple scaling.

Quantifying Dependence on the Assumptions

For fixed observed losses $L$, the BF unpaid estimate is $R=E(1-p)$. If expected losses increase, the estimate increases in proportion to the unreported share. If the selected emergence proportion increases, the estimated unreported amount decreases. In particular, a change in $p$ of one percentage point changes $R$ by $0.01E$, if $E$ is held fixed. When $p$ is small, even a modest relative change in $p$ can materially affect the reserve because most expected losses are still classified as unreported.

A useful sensitivity exercise varies one assumption at a time while holding the others fixed. This separates the effect of the expected-loss selection from the effect of the emergence selection. It does not establish which alternative is correct; it shows how much the estimate depends on each selection. Sparse observed experience may also be volatile at the claim level, even when the total BF estimate appears stable. The formula does not make the a priori expected losses or the emergence pattern certain.

Worked Example: Sparse Incurred Experience and Sensitivity

A hypothetical small equipment-repair program has 240 exposure units. Its selected expected loss cost is $850 per unit. By the valuation date, three claims have been reported, with incurred amounts of $14,000, $22,000, and $9,000. The selected incurred emergence is 25.0%. Calculate the BF ultimate and the amount beyond current incurred losses. Then calculate the ultimate if the selected emergence were instead 20.0%, with all other inputs unchanged.

Method and assumptions. Use incurred amounts and an incurred emergence pattern. Treat the three observed claims as the cumulative incurred amount for this group; the claim count is not used as a separate projection model.

Step 1: Calculate a priori expected ultimate losses from exposure.

$$ E=240(850)=204{,}000 $$

Step 2: Sum observed incurred losses.

$$ L=14{,}000+22{,}000+9{,}000=45{,}000 $$

Step 3: Calculate the estimate at 25.0% emergence. The unreported share is $1-0.250=0.750$. Therefore:

$$ R=204{,}000(0.750)=153{,}000 $$

$$ U=45{,}000+153{,}000=198{,}000 $$

The estimate beyond current incurred losses is $153,000. The check is $198{,}000-45{,}000=153{,}000$.

Step 4: Recalculate with 20.0% emergence. The unreported share is now $1-0.200=0.800$:

R=204{,}000(0.800)=163{,}200,\qquad U=45{,}000+163{,}200=208{,}200

Check and interpretation. Reducing emergence from 25.0% to 20.0% raises the amount beyond incurred losses by $10,200, exactly $204{,}000(0.050)$. It raises ultimate by the same amount because observed incurred losses are held fixed. The estimate incorporates the three observed claims, but most of the ultimate comes from the a priori expected losses. The common mistake avoided here is to assume that a small claim count makes the a priori component irrelevant; in this example, it is the main basis for the unobserved portion.

Combining Groups with Different Emergence

A single emergence proportion should not be applied indiscriminately when groups have different patterns. For groups $j=1,\ldots,n$, calculate each expected unreported amount using that group’s own expected ultimate $E_j$ and unreported share $q_j$. The aggregate estimate is the sum of the group estimates. This matters when one group is much more immature than another, even if their observed amounts are similar.

If a single equivalent emergence proportion is needed for the aggregate expected losses, it is expected-loss-weighted: $\bar{p}=\sum_j E_jp_j/\sum_j E_j$. That weighting follows from adding expected emerged amounts, $\sum_j E_jp_j$, and dividing by total expected ultimate losses. It is not generally the simple average of the groups’ percentages.

Worked Example: Aggregating Two Sparse Coverage Groups

A hypothetical insurer estimates two coverage groups separately. Group A has a priori expected ultimate losses of $100,000, paid emergence of 5.0%, and observed paid losses of $4,000. Group B has expected ultimate losses of $300,000, paid emergence of 40.0%, and observed paid losses of $105,000. Calculate the aggregate BF ultimate and reserve. Then show the error from using the unweighted average emergence proportion.

Method and assumptions. Use paid losses and each group’s selected paid emergence. Assume the expected ultimate and emergence selections are appropriate for their respective groups.

Step 1: Calculate each group’s unpaid estimate. Group A has an unreported share of $1-0.050=0.950$:

R_A=100{,}000(0.950)=95{,}000,\qquad U_A=4{,}000+95{,}000=99{,}000

Group B has an unreported share of $1-0.400=0.600$:

R_B=300{,}000(0.600)=180{,}000,\qquad U_B=105{,}000+180{,}000=285{,}000

Step 2: Aggregate the results.

$$ R=95{,}000+180{,}000=275{,}000 $$

$$ U=99{,}000+285{,}000=384{,}000 $$

The check is that total observed paid losses of $4{,}000+105{,}000=109{,}000$, plus the $275,000 reserve, equals the $384,000 ultimate.

Step 3: Evaluate the unweighted-average error. The unweighted average emergence is $(5.0\%+40.0\%)/2=22.5\%$. Applying its 77.5% unreported share to total expected losses of $400,000 gives a reserve of $400{,}000(0.775)=310{,}000$, and an ultimate of $109{,}000+310{,}000=419{,}000$. This overstates the group-by-group BF reserve by $35,000. The correct expected-loss-weighted emergence is:

\bar{p}=\frac{100{,}000(0.050)+300{,}000(0.400)}{400{,}000} =\frac{125{,}000}{400{,}000}=31.25\%

Its unreported share is 68.75%, which gives $400{,}000(0.6875)=275{,}000$, matching the group-by-group reserve. The common mistake avoided is averaging emergence percentages without regard to how much expected loss each group contributes.

Exam Interpretation and Practical Checks

For very immature or sparse experience, the BF result is often best understood as an expected-loss-based estimate with an adjustment for observed emergence. The adjustment may be small when only a small share of losses is expected to have emerged. Consequently, a result can be calculated correctly yet still be highly dependent on the selected a priori losses and emergence pattern. Sensitivity analysis is especially informative in that situation.

Before accepting an estimate, check the following:

Match the bases. Confirm that observed losses, emergence proportions, and expected losses refer to compatible loss definitions, coverage, and periods.
Check the unreported share. Verify that $q=1-p$, and that $p$ is a proportion rather than a development factor.
Reconcile the amounts. Confirm that $U=L+R$, so the ultimate less observed losses equals the estimated amount remaining.
Assess sensitivity. For a small $p$, examine reasonable alternatives for $E$ and $p$; the observed data may not strongly constrain the result.
Preserve group differences. Aggregate group reserves calculated on appropriate group-specific assumptions, rather than imposing a simple average emergence percentage.

A frequent exam trap is to calculate the reserve as $L(1-p)/p$. That is the reserve implied by extrapolating observed losses using a development proportion; it is not the BF reserve. In the first worked example, it gives $310{,}000(0.800)/0.200=1{,}240{,}000$, whereas BF gives $E(1-p)=1{,}536{,}000$. The two methods answer the question differently: BF bases the unobserved portion on expected losses, while direct extrapolation bases it on observed losses.

Key takeaway: When emergence is very limited, BF ultimate losses are close to the a priori expected ultimate, adjusted for observed losses relative to expected emergence. The method can moderate the effect of sparse early experience, but it cannot remove dependence on the selected expected losses or emergence pattern.

Check Your Understanding

For each problem, use the stated loss basis and emergence assumption.

A hypothetical liability group has expected ultimate losses of $720,000, paid emergence of 35.0%, and observed paid losses of $190,000. Calculate the BF reserve and ultimate. (Answer: Reserve $468,000; ultimate $658,000.)
A small hypothetical repair portfolio has 500 exposure units, an expected loss cost of $420 per unit, incurred emergence of 10.0%, and $18,500 of observed incurred losses. Calculate expected ultimate losses, the BF amount beyond current incurred losses, and BF ultimate. (Answer: Expected ultimate $210,000; amount beyond incurred $189,000; ultimate $207,500.)
A hypothetical accident year has expected ultimate losses of $1,250,000, paid emergence of 40.0%, and observed paid losses of $350,000. Find the reserve and ultimate. Then find the revised reserve and ultimate if emergence is 32.0% and all other inputs are unchanged. (Answer: At 40.0%, reserve $750,000 and ultimate $1,100,000; at 32.0%, reserve $850,000 and ultimate $1,200,000.)
Two hypothetical groups have expected ultimate losses of $80,000 and $220,000, emergence proportions of 10.0% and 50.0%, and observed losses of $9,000 and $105,000, respectively. Calculate the group-by-group aggregate reserve and ultimate, and the expected-loss-weighted aggregate emergence. (Answer: Aggregate reserve $182,000; ultimate $296,000; weighted emergence 39.33%.)
A hypothetical new coverage has expected ultimate losses of $640,000, paid emergence of 5.0%, and no observed paid losses. Calculate the BF reserve and ultimate, and explain why both are nonzero. (Answer: Reserve $608,000 and ultimate $608,000; the expected unreported portion remains even though no paid losses have emerged.)

What You'll Learn